MÉTODO IMPLEMENTADO POR COMPUTADOR PARA DETERMINAR AÇÕES PARA MELHORAR A OPERAÇÃO DE UM SISTEMA INDUSTRIAL, SISTEMA E MEIO LEGÍVEL POR COMPUTADOR

A machine learning-based method predicts future states and suggests corrective actions in industrial systems, addressing data complexity issues to enhance operator decision-making and optimize processes.

BR112025024223A2Pending Publication Date: 2026-11-10GEA GROUP AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
BR112025024223
Authority / Receiving Office
BR · BR
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-11
Filing Date
2024-05-09
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Operators in industrial systems, such as food and beverage processing plants, struggle to react effectively to process variations and upstream changes due to complex data overload from SCADA systems, leading to missed correlations and suboptimal corrective actions.

Method used

A computer-implemented method using a trained machine learning model to predict future system states, detect deviations, and suggest mitigating control actions through a user interface, enhancing operator decision-making with relevant data refinement and real-time optimization.

Benefits of technology

Improves industrial system efficiency by allowing operators to quickly understand and address inefficiencies before they escalate, optimizing processes based on up-to-date information and future forecasts.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to determining actions to improve operation of an industrial system. Reference data indicative of one or more states of the industrial system is obtained and it is determined whether a deviation is present within the reference data. If a deviation is present, the effect of the deviation on the industrial system is assessed and one or more mitigating control actions are selection from a plurality of predetermined control actions based on the assessed effect of the deviation. A representation of the impact of performing one or more mitigating control actions is output to a user interface, enabling a user of the user interface to select an operation to initiate one or more mitigating control actions to change one or more processes of the industrial system.
Need to check novelty before this filing date? Find Prior Art

Description

1 / 124 INDUSTRIAL SYSTEM FIELD OF TECHNIQUE

[0001] This disclosure refers to the assisted operation of an industrial system. Specifically, but not exclusively, this disclosure relates to providing operational assistance to an industrial system operator. Specifically, but not exclusively, this disclosure relates to determining actions to improve the operation of an industrial system based on the assessed effect of a deviation detected in the data associated with the industrial system. BACKGROUND

[0002] Process operators in the food and beverage industry are traditionally equipped with classic supervisory, control, and data acquisition (SCADA) systems to operate an industrial system, such as a processing plant. These systems have very detailed overviews of the plants, with hundreds of sensor, actuator, and alarm values. Local controllers and database-level sequence handling systems, such as programmable logic controller (PLC) systems, manage the basic process operations. The operator's role, therefore, is limited to acting reactively and in response to local events / alarms, or operators rely on PLC systems to perform assigned tasks.

[0003] The challenge with the above is that PLC systems are often programmed for a fixed set of parameters, regardless of process variation and upstream variations, for example, quality and / or quantity of raw materials, seasonal changes in Petition 870250101780, dated 06 / 11 / 2025, page 15 / 167 2 / 124 Environmental conditions and changes in the supply of public services affect the process. Although the SCADA system indicates these variations, the complexity of the information provided to operators is beyond human capacity to assimilate. Correlations and impact analyses are frequently missed, and operators are unable to take up-to-date corrective actions to optimize the operation for productivity, quality, and production sustainability, all in real time. A decision support system is applied to an industrial system, such as a plant geared towards the food and beverage industry, for example, brewery or liquid dairy processes, to assist operators in performing these real-time optimization operations.

[0004] This disclosure aims to collect various types of data from different sections of the process, monitor the development of key performance indicators, predict future development, assess the impact of deviations, form a conclusive set of information for the operator, and suggest corrective actions. SUMMARY OF REVELATION

[0005] The disclosure is directed to methods for predicting a future state of an industrial system and providing operating recommendations for altering one or more processes of the industrial system.

[0006] One aspect of the present disclosure relates to a computer-implemented method for determining actions to improve the operation of an industrial system. The method comprises obtaining reference data indicative of one or more states of the system. Petition 870250101780, dated 06 / 11 / 2025, page 16 / 167 3 / 124 industrial, where the reference data are associated with data collected from one or more sensors of the industrial system. It is determined whether a deviation greater than a predetermined limit is present in the reference data, where the predetermined limit is associated with a process of the industrial system. If there is a deviation greater than the predetermined limit, the effect of the deviation on one or more outputs of the industrial system is evaluated based on the difference between the performance of the industrial system with the deviation present at an initial moment and the projected performance of the industrial system at an initial moment without the presence of the deviation.The method further comprises selecting one or more mitigating control actions from a plurality of predetermined control actions, wherein each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system, and wherein the selection comprises identifying a problem and the associated control action. The problem is identified from a plurality of predetermined problems as causing the deviation, based on the assessed effect of the deviation. Based on the problem identified as causing the deviation, one or more control actions from the plurality of predetermined control actions are identified, wherein one or more of the identified control actions are selected to reduce the effect of the deviation on one or more outputs of the industrial system.Subsequently, the impact of implementing each mitigating control action on one or more outputs of the industrial system is calculated, and a representation of the impact of implementing one or more mitigating control actions is generated. Petition 870250101780, dated 06 / 11 / 2025, page 17 / 167 4 / 124 sent to a user interface. The representation allows a user of the user interface to select an operation to initiate one or more mitigating control actions to alter one or more processes of the industrial system.

[0007] Beneficially, the complex information associated with the industrial system is refined into the information most relevant to production operation. The efficiency of the industrial system is increased as operator decision-making is improved due to targeted communication of relevant consolidated data. Subsequently, a user of the user interface, such as an operator of the industrial system, can quickly understand the main current or future impacts on one or more processes of the industrial system. The operator can decide with greater accuracy and reliability whether to perform a recommended corrective action to optimize the industrial system process based on up-to-date information about the industrial system processes.This process allows small inefficiencies to be detected and corrected, or otherwise overcome, at an earlier stage than would otherwise be possible, potentially before larger inefficiencies can occur.

[0008] According to another aspect of the present disclosure, the disclosure relates to a computer-implemented method for predicting the performance of an industrial system. The method comprises obtaining a first state value of the system from a trained machine learning model, wherein: the trained machine learning model predicts, based on performance Petition 870250101780, dated 06 / 11 / 2025, page 18 / 167 5 / 124 of the industrial system over a first time period, the performance of the industrial system over a second time period, wherein the performance of the industrial system over the first time period is determined using data collected in the first time period from a set of sensors of the industrial system; and the first system state value is associated with the predicted performance of the industrial system over the second time period and is indicative of a future state of the industrial system. Real-time system information is also obtained, wherein the real-time system information is indicative of the current state of the industrial system and comprises data collected by the set of sensors of the industrial system.Using a prediction model configured to forecast the industrial system's performance in the second time period based on real-time system information and the first system state value, a second state value is determined. The method further comprises the calculation of a third system state value using data collected from the industrial system's sensor array throughout the second time period, where the third system state value is associated with the industrial system's performance in the second time period and is indicative of the industrial system's state in the second time period. The trained machine learning model is subsequently tuned to reduce the difference between the second state value determined by the prediction model and the third system state value.

[0009] Beneficially, the process uses a Petition 870250101780, dated 06 / 11 / 2025, page 19 / 167 6 / 124 A combination of a trained machine learning model and the results of a predictive model produces an initial forecast. The initial forecast is subsequently modified to take into account real-time data associated with the forecast. Furthermore, the trained machine learning is kept up-to-date by including a continuous learning loop, so that subsequent forecasts are therefore increasingly accurate and reliable. Additionally, the inclusion of the predictive model allows real-time variabilities associated with relevant processes to be incorporated into the forecast, so that the forecast more closely matches the state of the industrial system at the predicted time point. These forecasts can then be used by an operator, or optimization systems, of the industrial system to determine operations to increase or maintain the efficiency of the industrial system, potentially before inefficiencies affect the industrial system's processes.

[0010] According to another aspect of this disclosure, the disclosure relates to a computer-implemented method for automated monitoring of an industrial system and subsequent generation of operating recommendations for one or more processes of the industrial system. The method comprises obtaining a first state value from a trained machine learning model, wherein: the trained machine learning model predicts, based on the performance of the industrial system over a first time period, the performance of the industrial system over a second time period, wherein the performance of the industrial system over the first time period is determined using data collected in the first time period. Petition 870250101780, dated 06 / 11 / 2025, page 20 / 167 7 / 124 time from a set of sensors in the industrial system; and the first system state value is associated with the predicted performance of the industrial system over the second time period and is indicative of a future state of the industrial system. Real-time system information is also obtained, in which real-time system information indicates the current state of the industrial system and comprises data collected by the industrial system's sensor array. A second system state value is then determined using a forecasting model configured to predict the performance of the industrial system in the second time period based on the first system state value and the real-time system information.

[0011] Using the second state value, it is determined whether a deviation greater than a predetermined limit is present in the predicted performance of the industrial system in the second time period, where the predetermined limit is associated with a process of the industrial system. If there is a deviation greater than the predetermined limit, the method comprises evaluating the effect of the deviation on one or more outputs of the industrial system based on the difference between the predicted performance of the industrial system in the second time period and the projected performance of the industrial system without the deviation. One or more mitigating control actions are selected from a plurality of predetermined control actions, where each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system. The selection comprises: evaluating a difference between the real-time performance of the industrial system and the Petition 870250101780, dated 06 / 11 / 2025, page 21 / 167 8 / 124 performance of the industrial system over the first time period; determine a problem from a plurality of predetermined problems that cause the deviation based on the assessed difference; and identify one or more control actions from the plurality of predetermined control actions based on the problem that causes the deviation, where one or more identified control actions mitigate the difference between the predicted performance of the industrial system in the second time period and the actual performance of the industrial system.

[0012] One or more mitigation control actions are prioritized based on the calculation of the impact of executing each mitigation control action on the industrial system. Furthermore, a representation of the impact of executing one or more mitigation control actions is displayed in a user interface, allowing the user, through the interface, to select an operation to initiate one or more mitigation control actions to alter one or more processes of the industrial system.

[0013] In this embodiment, predicted states of the industrial system are used to determine operations that increase or maintain the efficiency of the industrial system. Deviations in the prediction data, potentially undetectable by human operators, can be used to identify current or future problems with the industrial system. Such problems may affect or impact the performance of the industrial system. Impact calculations provide the operator with relevant information so that he can determine whether or not to perform the recommended actions to overcome a future problem. This Petition 870250101780, dated 06 / 11 / 2025, page 22 / 167 9 / 124 Automated monitoring and operational support can result in a highly efficient industrial system, optimized based on real-time variations and the potential consequences of trend deviations.

[0014] Additional aspects and achievements of the present systems are disclosed, and the above aspects and achievements should not be interpreted as limiting the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To facilitate a more complete understanding of this disclosure, reference is made to the accompanying drawings, in which similar elements are referenced with similar numerals. These drawings should not be interpreted as limiting this disclosure, but are intended to be illustrative only.

[0016] Figure 1 shows a flowchart that illustrates a method for determining actions to improve the functioning of an industrial system;

[0017] Figure 2 shows a high-level system architecture diagram, according to the present description, to determine actions aimed at improving the operation of an industrial system;

[0018] Figure 3 shows an example of representing the impact of a problem on the performance of an industrial system in a user interface of that system;

[0019] Figure 4 shows a flowchart illustrating a method for selecting one or more mitigating control actions based on their impact on the performance of an industrial system;

[0020] Figure 5 shows a flowchart that Petition 870250101780, dated 06 / 11 / 2025, page 23 / 167 Figure 10 / 124 illustrates a method for improving the selection of one or more mitigating control actions based on feedback data;

[0021] Figure 6 shows a flowchart that illustrates a method for predicting the system performance of an industrial system;

[0022] Figure 7 shows a high-level system architecture diagram, according to the present description, for predicting the performance of an industrial system;

[0023] Figure 8 shows a flowchart illustrating a method for updating a trained machine learning model, configured to predict the performance of an industrial system based on data collected from a set of sensors in the industrial system;

[0024] Figure 9 shows a flowchart illustrating a method for using quality labeling in updating a trained machine learning model configured to predict industrial system performance;

[0025] Figures 10A-10B show a flowchart illustrating a method for automated monitoring of an industrial system and generation of operating recommendations for the industrial system.

[0026] Figure 11 shows a high-level system architecture diagram, according to the present description, for automated monitoring of an industrial system and generation of operating recommendations for the industrial system; and

[0027] Figure 12 shows an example of a computer system to operate, alone or in combination, any of the methods disclosed here. Petition 870250101780, dated 06 / 11 / 2025, page 24 / 167 11 / 124 DETAILED DESCRIPTION

[0028] In the present description, references to singular items should be understood as including plural items, and vice versa, unless explicitly stated otherwise or the context makes it clear. Grammatical conjunctions are intended to express any and all disjunctive and conjunctive combinations of clauses, phrases, words, and the like, unless stated otherwise or the context makes it clear. Thus, the term "or" should generally be understood as "and / or," and so forth. The use of all examples or illustrative language (e.g., such as, including, or similar) provided in this document is merely for the purpose of better illustrating the modalities and does not impose a limitation on the scope of the modalities or claims.

[0029] The expert versed in the area will understand that the systems and methods of the present description are not limited to a single programming language or paradigm. In fact, the systems and methods of the present description are applicable to any suitable programming language or environment, including, but not limited to, Java, C, C++, any suitable ensemble language, Python, C#, scripting language code such as Javascript, Ruby, PHP and the like.

[0030] Some embodiments described in this document may refer to a computer storage product with a computer-readable, transient or non-transient medium (also called processor-readable, transient or non-transient medium), containing instructions or computer code to execute various Petition 870250101780, dated 06 / 11 / 2025, page 25 / 167 12 / 124 operations implemented by computer. Computer-readable media (or processor-readable media) is non-transient in the sense that it does not include transient propagating signals themselves (e.g., an electromagnetic wave propagating and carrying information in a transmission medium, such as space or a cable). Computer media and code (also called code) are those designed and constructed for a specific purpose or purposes.Examples of non-transient computer-readable media include, but are not limited to, magnetic storage media such as hard disks, floppy disks, and magnetic tapes; optical storage media such as CDs / DVDs, CD-ROMs, and holographic devices; magneto-optical storage media such as optical discs; carrier wave signal processing modules; and hardware devices specially configured to store and execute program code, such as application-specific integrated circuits (ASICs), programmable logic devices (PLDs), read-only memories (ROMs), and random access memories (RAMs). Other embodiments described in this document refer to a transient computer program product, which may include, for example, the instructions and / or computer code discussed in this document.

[0031] Some embodiments and / or methods described in this document may be performed by software (running on hardware), hardware, or a combination of both. Hardware modules include, for example, a general-purpose processor, an FPGA (Field Programmable Gate Array), and / or Petition 870250101780, dated 06 / 11 / 2025, page 26 / 167 13 / 124 an ASIC (Application Specific Integrated Circuit). Software modules (running on hardware) can be expressed in various programming languages ​​(e.g., computer code), including C, C++, Java, Ruby, Visual Basic, Python, and / or other object-oriented, procedural, or other programming languages, as well as development tools. Examples of computer code include, but are not limited to, microcode or microinstructions, machine instructions such as those produced by a compiler, code used to generate a web service, and files containing high-level instructions that are executed by a computer using an interpreter. For example, implementations can be made using imperative programming languages ​​(such as C, Fortran, etc.), functional programming languages ​​(Haskell, Erlang, etc.), logic programming languages ​​(such as Prolog), object-oriented programming languages ​​(such as Java, C++, etc.).) or other suitable programming languages ​​and / or development tools. Other examples of computer code include, but are not limited to, control signals, encrypted code, and compressed code.

[0032] Figures 1 to 5 are directed towards operational decision support, in which operational recommendations are automatically provided to a user through an interface, allowing the selection of operations to alter one or more processes of the industrial system. Figures 6 to 9 are directed towards forecasting the future performance of the industrial system. Figures 10A to 11 are directed towards applying the forecast of future performance of the industrial system from Figures 6 to 9 to enable the Petition 870250101780, dated 06 / 11 / 2025, page 27 / 167 14 / 124 operational decision support from Figures 1 to 5. The 1200 computer system in Figure 12 is an example of a system for implementing any of the methods described here, although an expert in the field understands that an alternative system could be used.

[0033] The industrial system illustrated in Figures 1 to 11 is a processing plant configured to perform at least one process, such as a chemical or physical process. The processing plant comprises at least one component, such as machines necessary for the execution of the process. The industrial system also comprises one or more sensors, which collect data associated with the process and / or component of the industrial system. The industrial system process results in one or more inputs that are used to generate one or more outputs of the industrial system. The industrial system is preferably monitored and / or operated, at least partially, by a qualified person, such as an operator. The operator is responsible for performing operations that result in a change in the process to increase efficiency, production quality, production quantity, plant sustainability, or other performance metric.

[0034] Monitoring an industrial system, whether by automated systems or by the operator, involves determining or obtaining a plant state value. The state value is associated with efficiency, production quality, production quantity, plant sustainability, or other industrial system performance metrics. The state value indicates the state of the industrial system at a specific point in time or over a period of time. Petition 870250101780, dated 06 / 11 / 2025, page 28 / 167 15 / 124 is determined and associated with the performance of the industrial system. For example, a state value is obtained directly from data collected from one or more sensors of the industrial system, such as a state value associated with the quality of the industrial system's production over a predefined period of time, for example, the protein density in production. Another state value is calculated from data collected from one or more sensors of the industrial system, such as the progress of completion of the industrial system's process at a point in time, for example, 30% complete. One or more state values ​​are used as indicators of whether the process and / or component of the industrial system is functioning optimally, for example, [insert example here], whether the process is efficient, whether the component is in good condition, whether the production yield is high, or whether another performance metric is optimized.

[0035] One or more sensors in an industrial system typically collect large amounts of data during their operation. Traditional analysis and monitoring of this complex data collection can be time-consuming, limiting the operator's role to reacting to local events detected or analyzed from the collected data, based on static boundary conditions associated with the process and / or component, such as safe operating constraints defined by the component manufacturer.

[0036] A decision support method and / or system, detailed in Figures 1 to 5 below, can assist operators in optimizing the performance of the industrial system by providing relevant information for this purpose. Petition 870250101780, dated 06 / 11 / 2025, page 29 / 167 16 / 124 optimization, along with recommendations for action. Decision support is based on the evaluation of information comprising one or more state values ​​and / or real-time data from one or more sensors in the industrial system, which would otherwise be presented directly to the operator or obtained by PLC systems. Determining whether the information is relevant to the operator is an important step in decision support and can be achieved by detecting deviations in the information. This provides an objective measure to determine the relevance of the information.

[0037] A deviation is a variation in the information associated with the state of the industrial system compared to what is expected. The expectation may be based on historical trends or predicted from multiple data points, or on single data points associated with specific conditions and / or values ​​(i.e., a deviation corresponds to a variation of one or more values ​​in the reference data compared to one or more expected values). In other words, a deviation is detected by comparing the reference data with the expected value of that data (e.g., expected values ​​based on historical data or future state data). Since there is natural variation in the information associated with the industrial system, for example, due to fluctuations in environmental conditions, the quality of inputs, etc., detecting the occurrence of a deviation consists of determining whether it is greater than a predetermined limit.The predetermined limit can be dynamic, such as being based on a standard deviation calculated from historical data or future data. Petition 870250101780, dated 06 / 11 / 2025, page 30 / 167 17 / 124 projected, or static, as based on the specification of an industrial system component.

[0038] In an example with multiple data points, the information associated with the industrial system comprises a state value related to the progress of a process in the industrial system. The progress is calculated using data collected from one or more sensors in real time, providing an indicator of the state of the industrial system at a given instant.

[0039] For example, if the calculated state value is 82% at a given time, this value is compared to historical state values ​​at previous times. In the present example, if the calculated state value follows the trend present in the historical state values, the calculated value would be 90%, resulting in a deviation of 8%. The predetermined limit for this state value is based on the historical variation of progress over multiple process cycles, resulting in a predetermined limit of + / - 5%. Since the 8% deviation exceeds the predetermined limit, it is determined that there is a deviation in the data. The information associated with the state of the industrial system is not limited to current or historical data, and deviations can be detected in information associated with a future state of the industrial system.

[0040] Once a discrepancy exceeding a predetermined limit is detected, the impact of that discrepancy is calculated. The impact of the discrepancy is the likely effect it will have on the system if it continues to operate without intervention. For example, the impact of the discrepancy could be a reduction in production, an increase in energy consumption, Petition 870250101780, dated 06 / 11 / 2025, p. 31 / 167 18 / 124 etc. Thus, the impact of the discrepancy can also be referred to as the effect, probable effect, result, or consequence of the discrepancy. If there is no impact associated with the discrepancy, it may not be relevant to the operator.

[0041] Furthermore, the impact of the discrepancy helps in assessing its cause, such as, for example, an obstructed component. With the cause of the deviation identified, an action can be determined to mitigate or resolve the problem, where this mitigating action is linked to a corresponding problem that also needs to be mitigated. The relationships between problems and actions are based on prior knowledge of the processes and components of the industrial system, such as process know-how and system configuration(s). Prior knowledge, therefore, depends on the specific industrial system and may include information obtained from system experts, such as operators or component manufacturers.

[0042] The actions with the highest calculated impact on the problem causing the deviation are selected and presented to the operator, along with the calculated impact. The selected mitigation actions represent the most relevant course of action for the operator, without requiring them to analyze all the information associated with the industrial system used throughout the decision support process. Furthermore, the calculated impact is the most relevant analysis needed for the operator to decide whether to execute any of the selected mitigation control actions. This is especially important in scenarios where multiple deviations are detected and / or multiple mitigation control actions. Petition 870250101780, dated 06 / 11 / 2025, page 32 / 167 19 / 124 are selected, and the operator must decide which one, if any, to execute when optimizing the industrial system's performance, based on their experience and technical knowledge.

[0043] For example, a component in an industrial system has an ideal performance range. Exceeding this range decreases the component's performance and results in reduced process efficiency. In this example, a deviation in the information associated with a predicted future state of the system is detected based on a comparison with historical data associated with the component. The impact of the deviation is assessed as being that the ideal performance range will be exceeded in two process cycles. The problem is identified as an increase in the amount of byproduct associated with the next process cycles due to changes in the quality of an input resource. Control actions associated with the problem include mitigation actions such as adjusting the composition of input resources, reducing process speed, or rescheduling the next process cycles.The impact of the selected mitigation actions is calculated, such as a decrease in process efficiency resulting from changes in the input composition or an increase in the time required to complete the next process cycles. Both the selected mitigation actions and their impacts are presented to the operator, who decides whether to implement a mitigation action, such as rescheduling the next process cycles.

[0044] While correlations and impact analyses are neglected in traditional industrial systems, operators are unable to implement up-to-date corrective actions to optimize the Petition 870250101780, dated 06 / 11 / 2025, page 33 / 167 In a 20 / 124 system, the inclusion of decision support methods and / or systems allows the operator to receive the most relevant impact analysis, along with predetermined corrective actions. In this way, the operator can quickly, accurately, and reliably decide whether to execute the recommended corrective actions and optimize the industrial system based on real-time information and / or future forecasts.

[0045] Figure 1 shows a flowchart illustrating a method for determining actions to improve the operation of an industrial system. Specifically, Figure 1 illustrates a computer-implemented method 100, compatible with the processes of methods 400, 500, 600, 800, 900 and / or 1000.

[0046] Figure 1 represents method 100, which is a support process for operators of an industrial system, in which corrective actions are suggested to the industrial system operator to minimize any deviant trends detected in the system's performance. The execution of these suggested corrective actions aims to mitigate any negative impact on productivity, performance quality, production quality, sustainability, efficiency, and / or other operations related to the industrial system's performance.

[0047] For each individual state value indicative of industrial system performance, forecast data, historical data, and / or real-time data from various sources can be evaluated to identify possible deviations in the state value compared to an expected state value. Method 100 can be adapted based on the individual state value, where different Petition 870250101780, dated 06 / 11 / 2025, page 34 / 167 21 / 124 combinations of forecast data, historical data, and / or real-time data are used to reliably identify deviations. Once deviations are identified, the impact is calculated so that associated problems and corrective actions can be correctly identified. These corrective actions are presented to a user of the industrial system interface, such as an operator, who can review the findings and suggested actions to decide whether to implement the corrective action to mitigate the impact of the deviation. An example of a system for implementing method 100 is shown in Figure 2.

[0048] Method 100 comprises steps 102, 104, 106, 108, 110, and 112. Step 102 comprises obtaining baseline data in an initial time period; step 104 comprises determining the presence of a deviation in the baseline data; step 106 comprises evaluating the effect of the deviation on the industrial system; step 108 comprises selecting one or more mitigating control actions from a plurality of predetermined control actions; step 110 comprises calculating the impact of implementing each control action; and step 112 comprises generating a representation of the impact.

[0049] Step 102 consists of obtaining reference data, such as data indicative of one or more states of the industrial system. Reference data are relevant to the active process(es) of the industrial system that are being actively monitored to maintain optimal system performance. This data is associated with information collected from one or more sensors of the industrial system and is described in more detail below. One Petition 870250101780, dated 06 / 11 / 2025, page 35 / 167 22 / 124 once obtained, the reference data can be combined, refined, or processed in other ways using standard data processing methods, allowing for a more reliable evaluation of the reference data in step 104. This can be done, for example, using data fusion techniques, including a computational many-to-many full join approach to form a multidimensional dataset or a computational inner join approach, in which only the intersection of the combined data is retained, or even data cleaning techniques such as filtering.

[0050] For example, reference data comprises real-time data collected from one or more sensors in the industrial system. In another example, reference data comprises a historical state value associated with historical data accessed from a database or other storage medium. In yet another example, reference data comprises a future state value projected based on current or historical state values, or a future state value predicted based on the output of a machine learning model trained and configured to predict the future performance of the industrial system, such as the Method 600 trained machine learning model described in detail below.

[0051] Reference data comprise a combination of at least one of the above examples, wherein the composition of the reference data is tailored to specific state values ​​that are being monitored to determine whether the industrial system is operating optimally. State values ​​are indicative of the state Petition 870250101780, dated 06 / 11 / 2025, page 36 / 167 23 / 124 and / or industrial system performance, including: values ​​associated with the progress of a process occurring in the industrial system, such as the percentage of progress completion; values ​​associated with the availability of the industrial system, such as batch durations and phase / cycle durations; values ​​associated with the quality performance of the industrial system, such as the quality of the outputs of the industrial system, the average temperatures of specific processes, or the quantities of mass and / or energy; values ​​associated with the productivity of the industrial system, such as process rates, safety metrics, or performance indicators of equipment and / or components; and the efficiency of the industrial system, such as energy recovery rates and percentages or resource consumption.

[0052] Reference data may optionally include any state values ​​referenced above and / or data collected from one or more industrial system sensors used to calculate the respective state value. For example, reference data may include state values ​​associated with the specific monitored state value or the data used to calculate or obtain the included state values. This results in reference data that encompasses all potential variables that may impact the monitored state value, which improves the accuracy and reliability of deviation detection, as described in step 104 below.

[0053] One or more sensors, whether physical, motion, proximity, temporal, virtual, detectors and / or transducers, include physical sensors, motion sensors, proximity sensors, sensors Petition 870250101780, dated 06 / 11 / 2025, page 37 / 167 24 / 124 time-dependent detectors and / or transducers. In addition, the reference data comprise constant values ​​(e.g., time-independent), variable values ​​(e.g., time-dependent), or a combination of both.

[0054] Step 104 consists of determining whether there is a deviation in the reference data, such as determining whether there is a deviation greater than a predetermined limit when comparing the reference data with the expected values, where the predetermined limit is associated with a process of the industrial system. A deviation is a variation in the information associated with the state of the industrial system compared to what is expected, where the expectation may be based on historical or predicted trends based on multiple data points or on single data points associated with specific conditions and / or values. For example, the deviation may be an outlier, such as a global, contextual, or collective outlier, or a value that exceeds a range of expected values.

[0055] Deviation detection is based on sequences of data or real-time data measurements, such as within a predetermined window of a defined number of most recent measurements or a predetermined time window. Depending on the type of deviation, determining whether a deviation exists requires different predetermined thresholds, such as a threshold based on processed measurements, for example, the estimated variation in reference data relative to a measured value. For example, the threshold may be based on process knowledge, such as recipe parameters, equipment specifications, and / or historical performance.

[0056] In an example where the reference data Petition 870250101780, dated 06 / 11 / 2025, p. 38 / 167 25 / 124 comprises a state value associated with a component within the industrial system; the predetermined limit used for deviation detection is a constant value based on a safe and efficient operating range for the component within the industrial system. In another example where the reference data comprises a state value associated with a component within the industrial system, the predetermined limit used is variable, and deviation detection is based on historical and / or predicted data trends. Deviation detection allows the identification of specific deviations for each component and process, even if the reference data is still within standard safety ranges.Deviations can be detected early, such as a batch process that takes longer to complete on a specific component than expected, or that is historically normal for that component, even when the total time spent is within standard operating times for the component in general operation. Therefore, determining if a deviation is present results in greater efficiency of the industrial system due to early detection specific to each process and component.

[0057] Step 106 consists of evaluating the effect of the deviation on the industrial system, that is, if there is a deviation greater than the predetermined limit, as determined in step 104, evaluate the effect of the deviation on one or more outputs of the industrial system based on the difference between the performance of the industrial system at a first point in time with the present deviation and the projected performance of the industrial system at the first point in time without the deviation. Petition 870250101780, dated 06 / 11 / 2025, page 39 / 167 26 / 124 The first point in time can be a point in the present time or a point in the future time.

[0058] If the first point in time is a future point, the predicted performance of the industrial system at that future point, considering the present deviation, is based on a predicted future state of the industrial system, derived from the state of the industrial system contained in the reference data. The predicted performance is obtained through machine learning processes, such as the 600 method described below, or mathematical processes, such as averaging and trend forecasting. If the first point in time is a current point, the performance of the industrial system is represented by a current state of the industrial system associated with data collected in real time or over a current time period, from one or more sensors of the industrial system.

[0059] The projected performance of the industrial system at the first instant is a mathematical projection of the state of the industrial system, based on the current state of the system, if the deviation is constant, and / or on historical states of the system, if the deviation varies over time, such as increasing with time. The mathematical projection is determined using standard mathematical processes, such as the arithmetic mean over a predetermined time interval or trend forecasting methods, such as linear regression techniques. Alternatively, the current and / or historical performance of the industrial system is used, or a virtual performance of the industrial system generated based on a theoretical optimal performance and data collected from one or more sensors of the industrial system. Petition 870250101780, dated 06 / 11 / 2025, page 40 / 167 27 / 124

[0060] Step 108 consists of selecting one or more mitigating control actions from a plurality of predetermined control actions, where each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system, and the selection consists of identifying a problem and subsequently identifying one or more control actions based on the identified problem. Mitigation actions are a series of predetermined actions that are executed by the system to solve the identified problems.

[0061] A problem is identified from a plurality of potential problems predetermined as causing the deviation, based on the assessed effect of the deviation. Additionally, one or more control actions, from multiple predetermined control actions, are identified based on the problem identified as causing the deviation. These control actions are selected to reduce the effect of the deviation on one or more outputs of the industrial system. The plurality of predetermined problems, the effects that would result from the plurality of predetermined problems, and the plurality of predetermined control actions are based on process knowledge and industrial system configurations, and are stored in a corrective action database, which is a database associated with the industrial system and may be local or not, such as an edge computing-based approach or a cloud database. Each problem is linked to an impact.Each problem is also linked to a control action in a one-to-one relationship. Petition 870250101780, dated 06 / 11 / 2025, page 41 / 167 28 / 124 to-many or many-to-one. The corrective action database is described in more detail below, with reference to the corrective action database 220 of the operational support system 200.

[0062] For example, a heat recovery unit in an industrial system has a limited heat storage capacity. When the amount of heat recovered is greater than that used, the unit will eventually reach its maximum storage capacity. In this case, the deviation detection in step 104 determines the presence of a deviation in the reference data, and the evaluation in step 106 concludes that the deviation is due to a higher-than-typical heat recovery, based on historical trends. The mitigating control action selected in step 108 consists of reprogramming production to utilize the stored heat and obtain the maximum heat recovery potential, based on the calculated difference in energy efficiency between operating the heat recovery unit at full capacity and operating with idle capacity.

[0063] Step 110 consists of calculating the impact of implementing each mitigating control action, such as calculating the impact of implementing each mitigating control action on one or more outputs of the industrial system. The impact calculation considers one or more types of impacts based on the potential consequences associated with the industrial system. The types and / or values ​​of impact include: production measures, operational measures, environmental measures, safety measures, or efficiency measures. Continuing with the previously mentioned example of a heat recovery unit, the impact of Petition 870250101780, dated 06 / 11 / 2025, page 42 / 167 29 / 124 rescheduling production to utilize stored heat, including indirect impacts such as the rescheduling of subsequent processes and the energy used by the system during idle time.

[0064] For example, the calculated impact may include production impacts related to changes in process times, changes in the quality and / or quantity of production, changes in the resources needed to maintain the quality and / or quantity of production, changes in sustainability, changes in energy consumption, and / or changes in the environmental footprint. When the calculated impact comprises more than one type of impact, a total impact can be calculated based on the weighting of each type of impact. The weighting of the impact types results in a customized final impact, such as a customized total impact for each type of deviation, each process of the industrial system, each component of the industrial system, each operation of the focus of the industrial system, for example, water conservation during a drought, and / or each industrial system.Optionally, one or more selected control actions are prioritized or ranked based on total impact, as in the 400 method process.

[0065] Step 112 consists of generating a representation of the impact, such as generating a representation of the impact of executing one or more mitigating control actions in a user interface, thus allowing a user of the interface to select an operation to initiate one or more mitigating control actions to alter one or more processes of the industrial system, for example, using a visual representation based on the example. Petition 870250101780, dated 06 / 11 / 2025, page 43 / 167 30 / 124 of Figure 300.

[0066] For example, the user interface displays the total impact and / or at least one type of calculated impact, such as the most relevant impact type for the deviation, the problem, and / or one or more mitigating control actions. The interface user utilizes technical knowledge associated with the operation of the industrial system to determine whether to initiate one or more mitigating control actions, no action, or another action to alter one or more processes of the industrial system with the objective of mitigating the impact of the deviation on the industrial system's performance. Preferably, the interface user's determination is stored and used to improve the reliability and / or accuracy of step 104, step 106, step 108, or step 110, for example, by using any of the steps in method 500.

[0067] If control actions are prioritized, the operation to change one or more processes of the industrial system is the highest priority mitigating control action that, when selected, causes a change in one or more processes of the industrial system in order to mitigate the impact of the deviation on the industrial system. Alternatively, the operation to change one or more processes of the industrial system may be different from the highest priority mitigating control action.

[0068] Optionally, step 106 is performed in conjunction with step 110, where the assessment of the effect of the deviation includes the calculation of the impact of the deviation. Alternatively or additionally, step 110 is performed in conjunction with step 106, where the calculation of the impact of Petition 870250101780, dated 06 / 11 / 2025, page 44 / 167 31 / 124 The execution of each mitigating control action includes calculating the impact of the deviation based on an assessment of the deviation's effect.

[0069] Preferably, method 100 is employed for a single state value. For example, for a plurality of state values, method 100 is repeated serially or in parallel, where, for example, the predetermined limit of step 104 depends on the specific state value. An example concerning a single state value using the steps of method 100 is detailed below. Alternatively, method 100 can be executed for a plurality of state values, such as where a set of state values ​​is associated and a single branch classifier can be used for each state value in the set.

[0070] For the single-state value, step 102 involves obtaining a predicted state value, historical state values, and additional data from multiple data sources for use in evaluating potential deviations. Depending on the single-state value, various types of evaluations are selected, such as multiple trained machine learning models, where the input is a combination of real-time data sources, historical data, and estimated data. The result of such an analysis could be, for example, “Inefficient heating rate in the wort boiler”. In step 104, the deviation is detected based on real-time data sequences or measurements (possibly within a window of the most recent measurements or measurements from the last few minutes). Based on the type of deviation, individual deviation classifiers require different measurements or characteristics. A characteristic is a Petition 870250101780, dated 06 / 11 / 2025, page 45 / 167 32 / 124 pre-processed measurement, such as the variance estimated from a measured value. Optionally, at the same time, a relevant graphical illustration of the symptom is constructed, as shown in Figure 3. Such an illustration is, for example, a time series, batch performance, or any other relevant mapping of performance data that illustrates the deviation.

[0071] Once a deviation is triggered, in step 106, the consequences for the value of the impacted single state are calculated and the problem is identified in step 108. This analysis helps the plant operator determine whether it is better to continue operations according to the current plan or if a more economical path exists. Calculating the consequences can be based on efficiency calculations, which indicate how much the time and energy cost of the operation increases or the production rate decreases compared to a viable alternative, such as the clean-in-place (CIP) process, in which one or more relevant components, processes, and / or equipment of the industrial system are shut down for component cleaning, for example, cleaning a filter without removing it. This knowledge is based, for example, on equipment specifications, product costs, resource costs, and the state of the system, such as the degree of fouling. For example, with a certain degree of fouling, the equipment consumes more electrical energy. The average allocation of electrical energy for this specific process in the industrial system is 20% of the total allocation. The additional hourly operating allowance is 5% per hour. The average additional allocation required for a CIP cleaning is 15%. Consequently, the assessed effect of the diversion is a Petition 870250101780, dated 06 / 11 / 2025, page 46 / 167 33 / 124 higher electricity consumption and performing a CIP cleaning would be the most efficient action in three hours of operation or more.

[0072] Once the problem is determined, a corrective action is identified in step 108 based on process knowledge and plant configurations, for example, using the corrective action database 220, described in detail below with reference to Figure 2. The impact of executing the action is also determined, for example, using step 110. Several control actions are stored in a corrective action database, used to select the action with the greatest impact on correcting the problem. A result of such an action identifier is, for example, “Heat exchanger to be cleaned, when possible”. The appropriate set of actions for a specific problem is usually known a priori. In this case, the possible set of actions is known based on the detected deviation, i.e., based on rules.In the above case of fouling in the heat exchanger, it is known by the technician, as well as the operator of the industrial system, that when the deviation of the heat transfer coefficient is too low, a CIP (clean-in-place) cleaning of the unit is recommended. Another case is a compensation tank between three lines (P, C1 and C2). One line (P) is a producer that supplies the intermediate storage tank, and the other two lines (C1 and C2) perform subsequent processing of the product from the intermediate storage tank, removing it, for example, through UHT treatment. When there is no balance between the inflow and outflow of the tank, it will either fill up or overflow. Petition 870250101780, dated 06 / 11 / 2025, page 47 / 167 34 / 124 will empty. Three appropriate actions if the tank is at risk of becoming full would be: a. reschedule the process to increase the production rate of the subsequent process, for example, C1, if possible; b. reschedule to increase the production rate of the subsequent process, for example, C2, if possible; or c. reschedule to decrease the production rate of the upstream process (P). Another case is a heat recovery unit with limited heat storage capacity. When more heat is recovered than used, the unit reaches its maximum storage capacity. In this case, the solution is to reschedule production to utilize the stored heat and take full advantage of the heat recovery potential, similar to the previous example. Preferably, step 108 and / or step 110 consist of an analysis based on process knowledge and rule-based operational models.

[0073] In step 112, the above conclusions are applied to each divergent state value. Optionally, the results from the previous steps are fed into prioritization algorithms that order the problems for operator focus. Prioritization is a plant goal-based ranking that takes into account a custom weighted combination of production, environmental, safety, or other measures. For example, if an observed problem has 3 sets of consequences, C, namely: C1: an initial value based on the production loss resulting from batches with longer production times instead of batches with optimized times; C2: a second value based on product quality rework due to yield losses, which means Petition 870250101780, dated 06 / 11 / 2025, page 48 / 167 35 / 124 excessive use of expensive raw materials; C3: a third value based on sustainability, energy consumption, and environmental footprint.

[0074] For each consequence, a weighting index, “I”, can be applied to calculate the total cost, Ct. The Ct of the consequences is therefore Yk=1(Ck4). The problems and suggested actions are then ranked according to the total cost, Ct, of the problem in ascending order. The final list of ranked problems is presented to the operator.

[0075] Optionally, the reference data for step 102 comprises data collected over a first time period from one or more sensors of the industrial system, wherein the data collected over the first time period are indicative of a historical state of the industrial system. Additionally or alternatively, the reference data comprises real-time data collected by one or more sensors of the industrial system, wherein the real-time data are indicative of a current state of the industrial system. Additionally or alternatively, the reference data comprises a predicted state value of the industrial system, wherein the predicted state value is associated with a predicted performance of the industrial system in the second time period and is based on the data collected over the first time period, and wherein the predicted state value is indicative of a future state of the industrial system.The advantage is that the different compositions of reference data allow the 100 method to be applied based on monitoring a real-time value, a historical trend, and / or a value predicted for the future. Petition 870250101780, dated 06 / 11 / 2025, page 49 / 167 36 / 124 This adaptability results in an operational support system that can be applied to a wide range of industrial systems, including those where complex processes make predictions unreliable or those where real-time data collection is not feasible.

[0076] If the reference data comprise data collected over an initial period, determining whether deviation is present may optionally include evaluating the data collected during that initial period, where such data are time-dependent, and identifying one or more trends and / or temporal variations in the evaluated data collected over the initial period.

[0077] Additionally and optionally, the data collected from one or more sensors of the industrial system associated with the reference data may be collected at a predefined time interval, such as ten minutes, the duration of a batch and / or the duration of one or more process cycles of the industrial system, or may be collected in real time. Alternatively, or additionally, the data collected from one or more sensors of the industrial system may comprise a data sequence of predetermined size, such as ten data points or the last one hundred data points collected from a specific sensor.

[0078] Preferably, step 104 is performed by a deviation classifier. The deviation classifier is based on a first-principles model and / or consists of one or more trained machine learning models, where each trained machine learning model is configured to detect one or more deviation types above. Petition 870250101780, dated 06 / 11 / 2025, p. 50 / 167 37 / 124 of a predetermined limit. For example, the one or more trained machine learning models can be supervised, unsupervised, or semi-supervised machine learning models, including, for example: a support vector machine, a neural network, a decision tree-based algorithm, a set-based algorithm, a generalized likelihood ratio detector, an auto-encoder, a Bayesian network, a K-nearest neighbor algorithm, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm, or a cumulative sum detector. A trained machine learning model, from one or more trained machine learning models, is selected based on one or more states and / or state values ​​associated with the reference data.Additionally or alternatively, the deviation classifier is configured to indicate whether or not there is a deviation based on an input that includes reference data indicative of one or more states of the industrial system. Optionally, the predetermined limit is determined by the deviation classifier.

[0079] In an exemplary implementation, the deviation classifier is based on an unsupervised automatic encoder neural network.

[0080] The auto-encoder is a non-recurrent feedforward neural network composed of an input layer, three dense hidden layers, and an output layer. The input layer and the output layer have the same number of nodes, so the auto-encoder generates a reconstruction of the input data. The auto-encoder Petition 870250101780, dated 06 / 11 / 2025, page 51 / 167 38 / 124 is trained to learn a first transformation from the high-dimensional input space to a lower-dimensional space (represented by the intermediate layer of the three hidden dense layers) and a second transformation from the lower-dimensional space back to the high-dimensional input space. By training the auto-encoder with data associated with the normal operation of the industrial system, it learns to reconstruct this normal operation data with a high degree of confidence, while reconstructing anomalous operation data (i.e., deviations) with a lower degree of confidence. The reconstruction obtained by the auto-encoder is therefore used to classify whether a state value from the reference data represents a deviation.

[0081] As mentioned above, the automatic encoder is trained with data representing the normal operational state, such as data collected from one or more sensors of the industrial system during a state in which the system was operating optimally. During training, the automatic encoder learns the interactions between the various variables and how to reconstruct them to the original variables (as output). When a process or component of the industrial system is affected by a problem, the interaction between the variables is impacted, for example, by changes in pressure, temperature, or other values ​​associated with the state values. This results in an increase in the error in the output of the automatic encoder, as it can no longer reconstruct the input (the impacted data) based on the relationships learned during training (normal operational data). By monitoring the reconstruction error, it is possible to obtain an indicator that the data Petition 870250101780, dated 06 / 11 / 2025, p. 52 / 167 39 / 124 of the input data are deviating from normal operational data; for example, the error will increase as the equipment deteriorates. Next, a probability distribution of the reconstruction error is used to identify whether a data point represents a deviation and, consequently, whether there is a deviation from the reference data.

[0082] In the implementation described herein, the training data comprises one or more datasets, where each dataset consists of individual files containing snapshots of 1-second infrared signals, recorded at specific intervals associated with the output quality of a process. Each file contains 20,000 points, with a sampling rate of 10 kHz. The file name indicates when the data was collected. Each record (line) in the data file represents a data point. Larger intervals of time records (indicated in the file names) indicate the resumption of the experiment on the next working day. It is assumed that process degradation occurs gradually over time as the process input composition degrades; therefore, one data point every 10 minutes is used for subsequent analyses. Each data point every 10 minutes is aggregated using the average absolute value of the infrared photon sensor recordings across the 20,000 data points in each file, which are then combined to generate a single data frame.

[0083] In this example, the objective is to identify output quality degradation based on snapshots of infrared signals, resulting in predictive mitigation actions to be selected to avoid output quality control failures (e.g., inserting features). Petition 870250101780, dated 06 / 11 / 2025, page 53 / 167 40 / 124 different processes or repeat it until the output shows superior quality). The datasets, each composed of data collected from four infrared photon sensors, are used to train the automatic encoder. As the equipment is operated until the process is repeated several times and the output quality degrades completely, the data from the first operating cycles are used as training data to represent the normal operation associated with high output quality. The remaining part of the datasets, referring to the period before output degradation, is then used as test data to evaluate whether the different methods can detect the deviation before complete output degradation. After loading the infrared data, the index is transformed into date and time format and ordered chronologically to be accessed throughout the training.

[0084] A standard split is performed on the training data to isolate a portion for testing. If the auto-encoder is being updated with new training data, such as when a new dataset is added to the previous training data, a split is not necessary, as the original test data can be used. Standard data preprocessing is performed on both training and test data to scale the data for optimal training, such as a Scikit-learn MinMaxScaler so that the data is in the range [0,1].

[0085] As mentioned above, the automatic encoder compresses the infrared sensor readings into a low-dimensional representation, which captures the correlations and interactions. Petition 870250101780, dated 06 / 11 / 2025, page 54 / 167 41 / 124 among the various variables that impact sensor readings over time. Using the aforementioned neural network with three hidden layers, the central hidden layer has fewer nodes than the adjacent layers. More specifically, the first hidden layer consists of 15 nodes, the intermediate hidden layer of 5 nodes, and the third hidden layer of 15 nodes. The mean squared error is used as the loss function, and the auto-encoder is trained using the “Adam” optimizer, which is a stochastic gradient descent method based on adaptive estimation of first and second-order moments. Five percent of the training data is isolated for validation after each epoch (validation_split = 0.05). This helps determine if the auto-encoder is being trained successfully, for example, if it is converging to a global minimum or if it is suffering from underfitting.

[0086] The calculated loss distribution, as the average of the absolute differences between the actual value and the predicted value (mean absolute error), in the training set is collected and used to identify a suitable predetermined threshold value to detect a deviation. This helps ensure that the predetermined threshold is above the noise level and that any signaled deviations are statistically significant above the background noise and therefore relevant to an industrial system operator. From the calculated loss distribution, an example of a predetermined threshold of 0.4 is identified. The loss in the test set is then calculated to check when the output crosses the predetermined threshold. The same metrics are also calculated for the training set and all Petition 870250101780, dated 06 / 11 / 2025, page 55 / 167 42 / 124 the data are combined to form a single data frame. A standard graph of reconstruction loss as a function of date and time is optionally plotted to visualize the output of the auto-encoder.

[0087] Alternatively or additionally, data associated with one or more processes of the industrial system, and / or data associated with one or more components of the industrial system, are used to determine if a deviation exists. For example, process- and / or component-specific data are used to select the trained machine learning algorithm, to preprocess the reference data before inserting it into the deviation classifier, and / or as input to the deviation classifier.

[0088] Optionally, steps 106 and 108 are interchangeable. For example, problem determination in step 106 further involves the use of a rule-based model associated with one or more processes of the industrial system, and involves the analysis of the plurality of predetermined control actions of step 108 based on the rule-based model.

[0089] Optionally, step 112 also includes generating a visual representation of the deviation impact, determining a relevant graphical interpretation of the deviation impact from a list of graphical interpretations based on mapping the data associated with the deviation impact to one or more state values ​​obtained or calculated from the reference data.

[0090] Figure 2 shows a high-level system architecture diagram, according to the present description, to determine actions aimed at improving Petition 870250101780, dated 06 / 11 / 2025, page 56 / 167 43 / 124 the operation of an industrial system. Specifically, Figure 2 illustrates an operation support system 200 configured to execute method 100 and directed to determine actions aimed at improving the operation of an industrial system.

[0091] The operational support system 200 comprises a decision module 202a and a feedback module 202b. The decision module 202a comprises a deviation detection unit 204, an impact analysis unit 206, and an action identification unit 208. The deviation detection unit 204 obtains reference data 210 based on real-time data 212 or trend data 214. The impact analysis unit 206 optionally obtains resource data 216 and / or equipment data 218. The action identification unit 208 obtains data from the corrective action database 220, which stores a plurality of predetermined problems 222 and a plurality of predetermined control actions 224. The feedback module 202b comprises feedback data 226, which can be obtained from the operations interpreter unit 228. The decision module 202a and the feedback module 202b communicate with a user interface 230.Optionally, the operation support system 200 also includes a priority unit 232, which obtains graphical problem data 234, problem symptom data 236, impact state value data 238 and / or corrective action data 240.

[0092] The deviation detection unit 204 obtains reference data 210, for example, by performing step 102 of method 100. Reference data 210 comprises data Petition 870250101780, dated 06 / 11 / 2025, page 57 / 167 44 / 124 real-time 212 and / or trend data 214. Real-time data 212 is associated with data collected by one or more sensors of the industrial system, indicating the current state of the industrial system. Trend data 214 is associated with data collected by one or more sensors at a first instant, indicating the historical state of the industrial system, or with data collected by one or more sensors at a second instant, indicating the future state of the industrial system. For example, trend data 214 comprises historical data and / or forecast data, such as predicted future data obtained using method 600 detailed below. Optionally, the deviation detection unit 204 obtains inputs comprising forecast data, values ​​collected from one or more sensors of the industrial system, and calculated state values ​​indicating the state of the industrial system.

[0093] The deviation detection unit 204 determines whether a deviation is present in the reference data 210, for example, by executing step 104 of method 100. The deviation detection unit 204 identifies deviations based on data sequences, such as trend data 214, or on real-time measurements, such as real-time data 212. The deviation detection unit 204 consists of one or more deviation classifiers, where each classifier is configured to determine whether a deviation is present. The deviation classifier(s) are based on first-principles models and / or machine learning models, such as support vector machines, neural networks, decision tree-based methods, ensemble-based methods, Likelihood Ratio detectors. Petition 870250101780, dated 06 / 11 / 2025, page 58 / 167 45 / 124 Generalized Likelihood Ratio (GLR) or cumulative sum detectors. For example, the deviation detector unit 204 is a model trained on both normal data, such as data where a deviation is not present or is present and below a threshold, and on failure scenario data, such as data where a deviation is present and above a threshold. The threshold is a predetermined limit, and each deviation classifier has a corresponding threshold.

[0094] Impact analyzer unit 206 obtains information from deviation detector unit 204 about the presence of a deviation and evaluates the effect of this deviation on one or more outputs of the industrial system, based on the difference between the performance of the industrial system with the deviation present at an initial instant and the projected performance of the industrial system at the initial instant without the deviation, for example, using step 106 of method 100. Impact analyzer unit 206 calculates the consequences for any impacted state values ​​of the industrial system.

[0095] The action identification unit 208 obtains the assessment from the impact analysis unit 206 and selects one or more mitigating control actions from the corrective action database 220, as performed in step 108 of method 100. The selection comprises identifying a problem from several predetermined problems 222 based on the assessment from the impact analysis unit 206 and identifying one or more control actions from the various predetermined control actions 224 based on the corresponding problem. The control action(s) are Petition 870250101780, dated 06 / 11 / 2025, page 59 / 167 46 / 124 selected to reduce the effect of the detected deviation assessed in the impact analysis unit 206.

[0096] The corrective action database 220 is an organized collection of predetermined problems, impacts that would result from those problems, and corresponding predetermined control actions, accessible by the operation support system 200 and by a database user associated with the industrial system, such as an operator. Each predetermined problem is linked to an impact and associated with at least one predetermined control action, based on rules related to process knowledge and the industrial system configuration(s). The corrective action database 220 is therefore specific to the industrial system. The links between predetermined problems and predetermined control actions can be static, based on historical knowledge obtained from the database user, or dynamic, changing according to operational decisions and / or feedback from an industrial system user.

[0097] For example, an industrial system operator lists all problems that may occur in the system, such as each potential problem associated with each process or component of the system, and the resulting impacts of each problem. Then, the operator lists each control action that would mitigate or resolve each listed problem. The listed problems and control actions are then organized into a plurality of predefined problems 222, with the impact associated with each problem, and into a plurality of predefined control actions 224 in the corrective action database 220. Alternatively, it is generated Petition 870250101780, dated 06 / 11 / 2025, page 60 / 167 47 / 124 a list of all control actions that could be performed on the industrial system. One or more problems that would be impacted by each control action are also listed. The potential impact of each problem is then listed or stored with the associated problem. The corrective action database 220 is then formed by organizing the listed control actions into a plurality of predetermined control actions 224 and an associated plurality of predetermined problems 222, where each problem is associated with one or more impacts on the performance, state, or process of the industrial system. Additionally or alternatively, each control action is associated with one or more impacts.

[0098] Details on problems, impacts (effects) and possible control actions can be found in various technical sources, such as Brewery utilities (manual of good practice)” by EBC Technology and Engineering Forum, Getränke-Fachverlag Hans Carl, Nuremberg, (1997), Clean-in Place for Biopharmaceutical Processes” ed. by DA Seiberling, Drugs and Pharmaceutical Sciences 173, pp. 302-313 (2007) and Handbook of Milk Powder Manufacture (2nd Edition)” by J Pisecky, GEA Process Engineering A / S p. 222 (2012).

[0099] By knowing or somehow obtaining an impact, such as the impact of a given deviation using the impact analyzer unit 206, the corrective action database 220 can be accessed to obtain an associated problem from the multiple predetermined problems 222 (i.e., identifying the problem that has a similar or corresponding associated impact to the impact). Petition 870250101780, dated 06 / 11 / 2025, p. 61 / 167 48 / 124 determined). As the corrective action database 220 is configured to link each predetermined problem to one or more control actions from the multiple predetermined control actions 224, the associated control action can be subsequently obtained from the corrective action database 220. If a new process, component, or action-problem link is made available to the industrial system, the corrective action database 220 is updated with new predetermined problems, control actions, and optionally, the associated impacts.

[00100] User interface 230 obtains one or more mitigating control actions from the action identifier unit 208, along with a calculation of the impact of each mitigating control action on the outputs of the industrial system, as performed in step 110 of method 100. User interface 230 provides a representation of the impact of executing one or more mitigating control actions to the interface user, as performed in step 112 of method 100. The impact representation comprises calculations obtained from the impact analyzer, such as the impact state value data 238. This analysis assists the interface user, such as an industrial plant operator, in knowing whether it is better to continue operation according to the current plan or whether there is a more economical path, thus allowing the user to select an operation to initiate one or more mitigating control actions in order to alter one or more processes of the industrial system.

[00101] Optionally, user interface 230 obtains one or more mitigating control actions from the action identification unit 208 through the unit of Petition 870250101780, dated 06 / 11 / 2025, page 62 / 167 49 / 124 priority 232. Priority unit 232 is configured to prioritize mitigating control actions based on the calculated impact of executing each action. Priority unit 232 obtains problem 234 graphical data, problem 236 symptom data, impact state 238 value data, and / or corrective action 240 data, and uses the obtained data to prioritize control actions. Problem 234 graphical data comprises data relevant to generating a graphical representation of the impact of a detected deviation, such as relevant time series data, batch performance data, or any other relevant mapping of data associated with industrial system performance and the deviation. An example of a graphical representation is illustrated in Figure 3.

[00102] The operation interpretation unit 228 depends on the type and / or feedback data 226 required. For example, the operation interpretation unit 228 is configured to interpret whether one of the selected control actions was implemented or whether an alternative control action was implemented, initiated, and / or completed. The alternative control action includes the absence of any action over a predetermined period. Additionally or alternatively, the operation interpretation unit 228 is configured to label the selected control actions, indicating whether a control action is relevant or not, whether a problem was the cause of a deviation or not, and / or whether the deviation was relevant or not.

[00103] In an example application of the operational support system 200, the occupancy time of a batch in a first industrial process is expected to be fifteen Petition 870250101780, dated 06 / 11 / 2025, page 63 / 167 50 / 124 minutes longer than normal, based on real-time data 212 and trend data 214. Minimum and maximum occupancy times are predetermined, based on technical knowledge, at 105 minutes and 135 minutes, respectively. The historical average values ​​contained in the trend data 214 confirm the above. The impact analysis unit 206 assesses that occupancy time affects the productivity and product quality of the first industrial process. Based on equipment data 218, the capacity associated with the first industrial process is 2,000 hectoliters. Based on resource data 216, the reduction in product quality associated with the first industrial process has an impact of 100 weighted units per hectoliter.

[00104] Impact analysis unit 206 calculates two impacts: loss of productivity due to the first industrial process being occupied for additional time, with resource data 216 resulting in 30,000 weighted units per hour, therefore the effect is 7,500 weighted units; and lower quality yield leading to process repetitions, resulting in the mixing of the batch associated with the first industrial process with other batches of lower quality yield, therefore resulting in a loss of margin in the higher quality batch, with a total effect of 20,000 weighted units. Together, impact analysis unit 206 assesses the two impacts, totaling 27,500 weighted units.

[00105] The action identification unit 208 analyzes the causes, for example, using real-time data 212, such as the input and output flow of a plurality of predetermined problems 222, which are analyzed together Petition 870250101780, dated 06 / 11 / 2025, p. 64 / 167 51 / 124 with the corresponding historical values ​​of trend data 214. It is determined that the inflow and outflow is not the cause of the occupancy time being longer than normal. The heating rate, another problem from the various predetermined problems 222, calculated using real-time data 212, is considered lower than the historical rates of the first industrial process, based on trend data 214. This comparison between the actual heating rate and the expected heating rate, considering volume, composition, equipment information, heating surface and steam supply properties obtained from resource data 216 and equipment data 218, together with real-time data 212, is presented. The main reason for this discrepancy is related to the heating rate.Using the corrective action database 220, a corresponding control action, from the various predetermined control actions 224, is selected as reviewing the steam valve settings in the mechanical configuration.

[00106] Priority unit 232 obtains not only data associated with the heating rate with a calculated impact of 27,500 weighted units, but also other problems with associated calculated impacts. Priority unit 232 classifies the problems according to the calculated impacts. A user of the user interface 230 then determines that the suggested problem and / or action are relevant, which is recorded by the operation interpretation unit 228. The consequence is the readjustment of one or more predetermined limits of the deviation detection unit 204, or alternative proposed actions for the same problem in Petition 870250101780, dated 06 / 11 / 2025, pp. 65 / 167 52 / 124 corrective action database 220, for example, “Increase the steam supply setpoint”.

[00107] Preferably, the operation support system executes any of the steps of method 100, 400 and / or 500 and supports the display of visual representation 300 of Figure 3.

[00108] Figure 3 shows an example of representing the impact of a problem on the performance of an industrial system in a user interface of that system. Specifically, Figure 3 illustrates an example of a visual representation 300 to be displayed to the user of the interface to graphically convey the impact of an identified problem in the industrial system.

[00109] The visual representation 300 comprises a display 302 of the user interface, a graphic illustration 304, impact information 306 associated with the evaluated effect of the deviation, a first icon 308 associated with a first control action, a second icon 310 associated with a second control action, and a third icon 312 associated with return data or feedback.

[00110] Display 302 is a full or partial display of the user interface, which may be directly or indirectly connected to the industrial system and accessible to an interface user, such as an industrial system operator. Display 302 is configured to provide a graphical illustration 304 and impact information 306 to the interface user. The graphical illustration 304 and the impact information 306 are associated with a detected deviation and an estimated effect of that deviation on the industrial system. The graphical illustration 304 is considered the Petition 870250101780, dated 06 / 11 / 2025, pp. 66 / 167 53 / 124 is the most relevant graphical representation of the impact of the deviation, and therefore both graphical illustration 304 and impact information 306 can be displayed in the same way.

[00111] The user interface, such as interface 230 in Figure 2, is configured to receive data associated with a user selection. The user selection corresponds to the first icon 308, the second icon 310, or the third icon 312. The first icon 308 is associated with a first control action, such as a mitigating control action with the highest priority, which, when executed, reduces the estimated effect of the deviation. The second icon 310 is optional and is associated with a second control action, such as the second highest priority mitigating control action, or a control action associated with not executing the highest priority control action, such as a control action that, when executed, does not result in any change in one or more relevant processes of the industrial system.The third icon 312 represents an option for the user to contribute feedback associated with the control action of the first icon 308, such as whether the mitigating control action of the first icon 308 is appropriate in relation to the impact information 306, with the feedback based on the user's technical knowledge and the information provided by the visual representation 300.

[00112] Figure 4 shows a flowchart illustrating a method for selecting one or more mitigating control actions based on the impact on the performance of an industrial system. Specifically, Figure 4 illustrates a computer-implemented method 400, compatible with the processes of methods 100, 500, 600, 800, 900 and / or 1000. Petition 870250101780, dated 06 / 11 / 2025, pp. 67 / 167 54 / 124

[00113] Figure 4 represents the 400 method, which is a process for selecting and prioritizing one or more mitigating control actions. The process can be performed in addition to or in place of step 108 of the 100 method. First, one or more mitigating control actions are selected from a plurality of predetermined control actions. In this example, the selection comprises a two-step process: identifying a problem that is likely causing a detected deviation in a state of the industrial system; and identifying at least one control action known to resolve or mitigate the identified problem. Then, the impact of each mitigating control action is calculated so that the result of executing each of the mitigating control actions on the industrial system can be determined. This calculation can then be used to prioritize the control actions. An example of a system to execute the 400 method is shown here in Figure 2.

[00114] Method 400 comprises steps 402, 404, 406, 408, and 410. Step 402 comprises selecting one or more mitigating control actions from a plurality of predetermined control actions; step 404 comprises identifying a problem from a plurality of predetermined problems that cause a deviation; step 406 comprises identifying one or more control actions based on the identified problem; step 408 comprises calculating the impact of each mitigating control action on the industrial system; and step 410 comprises prioritizing the mitigating control action(s). Petition 870250101780, dated 06 / 11 / 2025, pp. 68 / 167 55 / 124

[00115] Step 402 consists of selecting one or more mitigating control actions from a plurality of predetermined control actions, where each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system. The selection of one or more mitigating control actions is a process based on technical knowledge, in which it is known that problems associated with the industrial system result in one or more alterations in one or more processes of the industrial system. Similarly, alterations in one or more processes of the industrial system can therefore be associated with at least one predetermined problem. Thus, the selection of one or more mitigating control actions is based on historical and a priori knowledge of how to alter one or more processes of the industrial system in order to mitigate an undesirable effect of a detected deviation.

[00116] Step 404 consists of identifying a problem from several predetermined problems that cause a deviation, where the identification of a problem, from several predetermined problems, as the cause of the deviation is based on the evaluated effect of the deviation. The various predetermined problems are stored in a database of corrective actions and include problems related to the processes, equipment, environment, resources, and / or operation of the industrial system. Each problem is linked to at least one control action in a one-to-one, many-to-one, or one-to-many relationship.

[00117] Step 406 consists of identifying one or more control actions from the multiple actions of Petition 870250101780, dated 06 / 11 / 2025, pp. 69 / 167 56 / 124 predetermined control actions are selected based on the identified problem causing the deviation. The identified control actions are selected to reduce the effect of the deviation on one or more outputs of the industrial system. These actions represent operations that, once initiated and / or executed, cause a change in one or more processes of the industrial system, in order to mitigate the deviation.

[00118] Step 408 consists of calculating the impact of each mitigating control action on the industrial system, based on the calculation of one or more impact values ​​from a plurality of impact values ​​that include any production measures, operational measures, environmental measures, safety measures, or efficiency measures. The impact values ​​are combined to form a total impact calculated for each mitigating control action, for example, by weighting each impact value based on the specific process and / or component of the industrial system. The weighted impact values ​​can then be summed, multiplied, averaged, or mathematically combined using other standard methods.Weighting the impact values ​​results in a customized final impact, such as a customized total impact for each type of deviation, each process in the industrial system, each component of the industrial system, each focus operation of the industrial system, for example, water conservation during a drought, and / or each industrial system. Optionally, step 408 can be adapted to a specific state value, such as the state value associated with a detected deviation, for example, the deviation detected in step 404. Petition 870250101780, dated 06 / 11 / 2025, page 70 / 167 57 / 124

[00119] Step 410 involves prioritizing one or more mitigating control actions, such as prioritizing based on the mitigating control action that, according to calculations, will have the greatest impact on mitigating the deviation or improving the performance of the industrial system, as performed in step 408. Optionally, step 410 may not be performed, such as when only one mitigating control action is selected and / or appropriate.

[00120] Beneficially, one of the advantages of the 400 method is that the user of the interface, such as an operator of the industrial system, is provided with the most relevant information and associated data for a quick and efficient determination of the need to perform an operation to change one or more processes of the industrial system, helping to optimize system performance with up-to-date and relevant operational recommendations.

[00121] Figure 5 shows a flowchart illustrating a method for improving the selection of one or more mitigating control actions based on feedback data. Specifically, Figure 5 illustrates a computer-implemented method 500, compatible with the processes of methods 100, 400, 600, 800, 900 and / or 1000.

[00122] Figure 5 represents method 500, which is a process to improve method 100 and / or method 400 and / or operational support system 200. The process involves determining and storing whether a selected mitigating control action has been executed, such as whether an industrial system operator initiated the mitigating control action. This information can then be used to improve deviation detection and analysis. An example of a system for Petition 870250101780, dated 06 / 11 / 2025, page 71 / 167 58 / 124 executing method 500 is shown in Figure 2.

[00123] Method 500 comprises step 502, step 504, and one or more of steps 506, 508, or 510. Step 502 comprises determining whether a mitigating control action has been taken; step 504 comprises storing the feedback data associated with the determination; step 506 comprises using the feedback data to determine whether there is a deviation; step 508 comprises using the feedback data to identify the problem; step 510 comprises using the feedback data to identify a control action; and step 512 comprises using the feedback data to adjust a deviation classifier in order to improve the accuracy of the deviation determination.

[00124] Step 502 consists of determining whether a mitigating control action has been executed, such as the mitigating control action selected in method 400. Determining whether a mitigating control action has been executed comprises determining whether the action has been initiated, is in progress, or has been completed. This determination can be made based on data collected from one or more sensors of the industrial system or based on input from an operator of the industrial system.

[00125] For example, an interface user confirms whether they have initiated the selected control action. In another example, the user confirms whether the problem and / or the suggested control action is relevant to the deviation and / or operation of the industrial system. Alternatively, user input is not required, such as when the selected control action is not relevant or does not represent the best current course of action for an operator, and the operator initiates a different action, or no control action at all, such as performing a Petition 870250101780, dated 06 / 11 / 2025, page 72 / 167 59 / 124 manual readjustment. In this example, the determination is therefore that the mitigating control action was not performed.

[00126] Step 504 consists of storing the feedback data associated with the determination. Feedback data consists of any relevant data, such as the data used to perform Step 502, for example, user input or data collected from one or more sensors. Optionally, storing feedback data includes labeling the selected control actions, such as whether a control action is relevant or not relevant, whether an associated problem was relevant, such as the cause of a deviation, or not relevant, such as not being the cause of a deviation, and / or whether the deviation was relevant or not relevant.

[00127] Step 506 involves using feedback data to determine if there is a deviation. For example, if the feedback data shows that the deviation was not relevant, the predetermined threshold associated with that deviation is adjusted to avoid detecting irrelevant deviations. Further determining whether there is a deviation involves checking for a new deviation in new reference data, which is now a more reliable determination, such as by performing step 104 of method 100.

[00128] Step 508 involves using feedback data to better identify the problem. Further problem identification includes adjusting how problems are associated with the calculated impacts of the deviation, for example, by incorporating other proposed problems and their associated impact on the industrial system, or by altering how impacts are calculated, for example, by adjusting the weightings applied to specific impact values. A Petition 870250101780, dated 06 / 11 / 2025, page 73 / 167 60 / 124 additional identification also includes identifying a new problem based on a new deviation, for example, by performing step 106 of method 100 and / or step 406 of method 400.

[00129] Step 510 involves using feedback data to better identify a control action. Further identification of a control action includes adjusting how problems and control actions are linked or otherwise associated in the corrective action database. For example, alternative control actions may be linked to a previously detected problem based on feedback data. Further identification also includes identifying a new control action based on a new problem, such as by performing step 106 of method 100 and / or step 408 of method 400.

[00130] Step 512 consists of updating at least one deviation classifier based on feedback data and / or using feedback data as an additional input to the deviation classifier. Optionally, step 512 is included within step 506, where adjusting a predetermined threshold consists of updating the associated deviation classifier.

[00131] Incorporating the processes of method 500, for example, into method 100 or operational support system 200, results in a feedback loop for continuous improvement of operational recommendations, enabling a precise, reliable and optimized process for specific processes, components or operations of the industrial system.

[00132] For operational decision support Petition 870250101780, dated 06 / 11 / 2025, page 74 / 167 61 / 124 detailed above, in reference to Figures 1 to 5, to be applied accurately and reliably in predicting future deviations based on reference data associated with the predicted future performance of the industrial system, it is necessary that the future performance of the industrial system be predicted accurately and reliably. Figures 6 to 11 detail performance prediction methods and / or systems that use a combination of a trained machine learning model, a prediction model, and a continuous learning loop. The inclusion of a trained machine learning model results in an accurate prediction based on historical data associated with the industrial system. The inclusion of the prediction model results in an updated prediction based on real-time data associated with the industrial system.The continuous learning cycle allows forecasts to be continually improved or adjusted to more closely match the measured future performance of the industrial system.

[00133] Figure 6 shows a flowchart illustrating a method for predicting the performance of an industrial system. Specifically, Figure 6 illustrates a computer-implemented method, method 600, compatible with the processes of methods 100, 400, 500, 800, 900 and / or 1000.

[00134] Figure 6 represents method 600, which is a process for predicting how one or more processes in the industrial system will develop or change in the future. Each state value, like the state values ​​described above in reference to method 100, has a dedicated prediction process. The process uses a combination of trained machine learning data and real-time data. Petition 870250101780, dated 06 / 11 / 2025, page 75 / 167 62 / 124 to increase explanatory ability, in addition to an integrated learning cycle to increase reliability. The 600 method, therefore, represents an explainable and reliable process for predicting the performance of the industrial system, being used, for example, to monitor the industrial system and / or by an operator of the industrial system for decision support purposes. An example of a system for the execution of the 600 method is shown here in Figure 7.

[00135] Method 600 comprises steps 602, 604, 606, and 608. Step 602 comprises obtaining a first state value from a trained machine learning model; step 604 comprises obtaining real-time system information, comprising data collected from a set of industrial system sensors; step 606 comprises determining a second state value using a predictive model based on the first state value and real-time system information; and step 608 comprises updating the trained machine learning model.

[00136] Step 602 consists of obtaining a first state value from a trained machine learning model, where the trained machine learning model predicts, based on the industrial system's performance over a first time period, the industrial system's performance over a second time period. The industrial system's performance over the first time period is determined using data collected during the first time period from a set of sensors in the industrial system. The first system state value is associated with the predicted performance of the industrial system. Petition 870250101780, dated 06 / 11 / 2025, page 76 / 167 63 / 124 over the second time period and is indicative of a future state of the industrial system.

[00137] For example, the trained machine learning model uses historical data associated with the industrial system to predict its future performance. The trained model is dedicated to a specific state value and is therefore trained to generate one or more specific state values. The input data corresponds to the first state value, where different state values, from different trained machine learning models, use the same input data or different data depending on the output state value. Alternatively, a single trained machine learning model is configured to predict multiple initial state values.The input data comprises data collected in the first time period from a set of sensors in the industrial system, which are indicative of the system's historical states; data associated with one or more relevant processes of the industrial system; data associated with one or more relevant components, such as equipment, of the industrial system; and / or data associated with the trained machine learning model, such as model configuration specifications. Additionally, feedback data is fed into the trained machine learning model to increase the reliability of the outputs obtained.

[00138] Step 604 involves obtaining real-time system information, including data collected from a set of industrial system sensors, where real-time system information indicates the Petition 870250101780, dated 06 / 11 / 2025, page 77 / 167 64 / 124 current state of the industrial system and comprise data collected by the industrial system's sensor array. Real-time system information indicates the current state of the industrial system, and a current state value is calculated based on the data collected by the sensors over a given time period. In addition to the data collected by the industrial system's sensors, real-time system information comprises data from various sources, depending on the target state value for which method 600 is intended.

[00139] Examples of real-time data sources include programmable logic controllers (PLCs) of process units; laboratory devices associated with one or more processes of the industrial system; upstream data associated with resource availability and operations planning; batch progress estimates associated with process cycle time; output quality and quantity indicators; and other calculated or measured values ​​associated with the industrial system.

[00140] Step 606 consists of determining a second state value using a forecasting model, in which the forecasting model uses the first state value and real-time system information to determine a second state value. The first state value obtained in step 602 constitutes a starting point for tracking specific properties and / or characteristics of the industrial system and predicting the future performance of the industrial system. Real-time system information adjusts or adapts the forecast so that it more accurately matches the future state of the industrial system. Petition 870250101780, dated 06 / 11 / 2025, page 78 / 167 65 / 124

[00141] For example, for a specific industrial system process, such as a process unit in focus and its operating properties, a comprehensive set of sensor and / or actuator values ​​from the process unit's PLC is collected. In addition, available quality data is collected from various laboratory devices capable of mirroring the data at the PLC level. Furthermore, upstream data impacting the unit's behavior for predetermined process requirements is also collected. Additionally, other calculated data (e.g., virtual sensors for quality measurements) are also fed into the prediction model. For batch operations, post-batch values ​​indicative of the industrial system's performance are also used in the current batch prediction process. To assess the quality of the predictions against the actual values, the batch progress status is also fed into the prediction model.Based on this real-time system information, the prediction model alters the value of the first state to account for factors impacting the process, which may be too variable, unpredictable, or rare to be incorporated into the training data of the machine learning model trained in step 602. The advantage is that using a prediction model results in a prediction that is increasingly specific to the process and more reliable than a more generalized approach based solely on a trained machine learning model.

[00142] Step 608 consists of updating the trained machine learning model, for example: calculating a third system state value using Petition 870250101780, dated 06 / 11 / 2025, page 79 / 167 66 / 124 data collected from the industrial system's sensor array over the second period, where the third system state value is associated with the industrial system's performance in the second period and is indicative of the industrial system's state in that second period; and adjusting the trained machine learning model to reduce the difference between the second state value determined by the prediction model and the third system state value.

[00143] The third state value is obtained from real-time system information or calculated using additional data associated with system performance in the second period. The third state value represents a measure, an unpredicted state value, with which a predicted state value can be compared, such as the second or first state value. By adjusting the trained machine learning model based on the difference between the second and third state values, the entire prediction system is updated so that the predicted performance more closely matches the measured future performance of the industrial system. Method 900 is an example of a process for updating the trained machine learning model.

[00144] Preferably, the first time period precedes the second time period, such that the first time period is a current and / or historical time period and the second time period is a future time period. For example, one or more process cycles of the industrial system, originating from a sequence of process cycles, occur within the second time period.

[00145] Optionally, the third value of Petition 870250101780, dated 06 / 11 / 2025, page 80 / 167 The 67 / 124 state is used to obtain an additional first state value from the trained machine learning model, so that method 600 is repeated, as performed in method 800. The third state value is used directly or obtained from storage, where the second and third state values ​​are optionally stored for later use as inputs to the trained machine learning model or prediction model and / or use as feedback data.

[00146] The calculation of a third state value is repeated at predetermined time intervals throughout the second period if systematic quality labeling is performed. For example, the second state value is assigned a first quality label, and the third state value is assigned a second quality label. Subsequent tuning of the trained machine learning model consists of determining whether the first quality label is consistent with the second quality label. Updating the trained machine learning model is performed remotely or in person, where, optionally, the update of the trained machine learning model can be based on a comparison between a plurality of second state values ​​and a plurality of third state values.

[00147] In the examples where a plurality of third-state values ​​is used to update the machine learning model trained in step 608, step 608 further comprises filtering the plurality of third-state values ​​and determining whether the filtered plurality of third-state values ​​meets a predefined criterion, wherein the Petition 870250101780, dated 06 / 11 / 2025, page 81 / 167 The 68 / 124 predefined criterion is associated with the number of filtered third-state values. For example, the plurality of third-state values ​​is used to update the trained machine learning model only when there are enough high-quality values ​​to update the trained machine learning model without introducing inaccuracies or, for example, bias. If the predefined criterion is not met, the update of the trained machine learning model is canceled, so that the trained machine learning model is restored to a pre-updated state, such as the state of the trained machine learning model when the first state value of step 602 was obtained. Additionally or alternatively, if the predefined criterion is not met, step 608 further comprises generating one or more first state values ​​using an alternative model, such as a model based on the average of a plurality of filtered third-state values.

[00148] The trained machine learning model is either a narrow artificial intelligence (AVI) model, for example, configured for the specific value of the first state, or a general artificial intelligence (AGI) model, for example, configured to predict the future performance of the industrial system on a broader scale. Preferably, the machine learning model trained in step 602 is one of a plurality of trained machine learning models, where each state value associated with a specific component or process of the industrial system corresponds to a model of Petition 870250101780, dated 06 / 11 / 2025, page 82 / 167 69 / 124 specific trained machine learning. For example, the plurality of trained machine learning models comprises any of the following: a linear multivariate regression model, a polynomial regression model, a random forest regression model, a neural network model, or an expert model, and step 602 comprises: receiving a focus indicator from a plurality of focus indicators, where each focus indicator is associated with a component, a process, an input, or an output of the industrial system; and selecting the trained machine learning model from the plurality of trained machine learning models based on the focus indicator.

[00149] Optionally, the inputs to the trained machine learning model and / or the prediction model comprise information associated with a component, a process, an input or an output of the industrial system and / or data collected by the industrial system's sensor array, including any of the following: temperature data, humidity data, pressure data, consumption data, process rates or operating conditions of the industrial system.

[00150] For an example of implementing method 100, sensor and actuator data, historical data, product recipe data, equipment design specification data, and model design specifications, such as a process knowledge-based specification to determine a model type, for example, a gray box model, are used to train a representation of a given process behavior. Petition 870250101780, dated 06 / 11 / 2025, page 83 / 167 70 / 124, for example, filtration time or thermal efficiency of the process, represented by a state value. Depending on the nature of the state value in question, one of the following modeling approaches is used: linear multivariable regression model, polynomial regression models, random forest regression models, neural network models, and / or expert models, which are manual models defined by experts, linear or non-linear.

[00151] To fit one or more trained machine learning models, which are optionally gray-box models, a mathematical expression, f, is specified for a state value, y, with specified input variables / features, x, and parameters, p. y = ί(χ,ν)

[00152] The input variables / characteristics can be upstream data, such as associated state values, setpoints and / or process variables, and historical data, such as historical state values, setpoints and process variables, for the target process(es) associated with the state value y. All the data mentioned above is stored in a central database of an edge device.

[00153] Using a target value for the state at the time of the forecast, which is typically the final batch value of the state, and the values ​​of the input variables, the model parameters are adjusted by minimizing the mean squared error or another standard mathematical approach to adjust the error between the expected target value and the model output value. Preferably, the learning model of Petition 870250101780, dated 06 / 11 / 2025, page 84 / 167 71 / 124 The trained machine is non-linear, meaning that a numerical iterative method must be used. A non-limiting example of this is Gradient Descent Optimization. However, any optimization can be used to fit the model using the mean squared error as the objective function. In the case of linear models, an analytical solution can be found.

[00154] Throughout the model fitting process, data is filtered based on quality, such as missing data (gaps in the data) for a batch, or data such as the state value that exhibits flaws based on the state of the processes and / or problems in data processing. In the case of flawed and / or low-quality data, where the fitting algorithm does not have sufficient data, the model fitting is canceled for the current state value y, and the model fitting is terminated, and / or the prediction of the next state value is fitted / optimized instead. A predefined criterion is used to ignore the fitted model and use an alternative strategy, such as a model based on an average state value, which is then fitted and used in the real-time data processing algorithm.

[00155] The mathematical expression, as well as the input data and parameter values, are stored in a database and associated with the specific state value y for which the model should be used. This state value may refer to a specific recipe and / or process.

[00156] Another example of a gray box model is a hybrid model, which uses a combination of a known mathematical expression with variables / characteristics. Petition 870250101780, dated 06 / 11 / 2025, page 85 / 167 72 / 124 input variables and parameters, and also a black-box layer to estimate the parameter values ​​from the same input variables / characteristics and / or additional variables / characteristics. Examples of black-box models include neural networks, random forests, and / or polynomial regression models. To fit the black-box model, no specific process knowledge / experience is required, and, for example, a standard deep learning algorithm is used to automatically find the input characteristics / variables from upstream data and historical data of the relevant process.

[00157] In one example, the black box model comprises data filtering processes, such as removing incorrect ranges from the entire dataset, for example, if batch data is incorrect due to missing data / gaps in the batches or if a state value is incorrect / faulty based on the state of the processes and / or problems in data processing. The best fit of the selected model is determined using the currently selected inputs / features. The selected black box model may have configuration parameters. For example, a neural network model needs specifications such as the number of layers, the number of neurons / nodes in the layers, and the activation model, and a random forest model (random decision forests) requires information such as the number of depths, the number of estimators, or the specification of the fitting algorithm.The configuration parameters are stored in a database and associated with a specific state value. Petition 870250101780, dated 06 / 11 / 2025, page 86 / 167 73 / 124 Preferably, cross-validation is performed between the model and the data to analyze the fit. Based on this analysis, a predefined criterion is used to ignore the fitted model and employ a reserve or fallback strategy, as described previously. For some types of black-box models, the most significant features / inputs are determined during fitting, such as when the model is a linear regression and / or random forest. Using this determination, features are selected based on a criterion, and the model is readjusted with the new feature selection until all meet the criterion. The mathematical expression, as well as the inputs and parameter values ​​to be used, are stored in a database corresponding to the assigned state value for which the model should be used. This state value preferably represents a specific recipe and / or process.

[00158] The trained machine learning model, along with its associated parameters, serves as a starting point for tracking the designed properties and predicting their future behavior in real time. To do this, considering a specific process unit and its operational properties, a comprehensive set of sensor and PLC actuator values ​​from the unit is collected. In addition, available quality data from various laboratory devices or equivalent PLC-level data are collected. Upstream data impacting the unit's behavior for the recipe in question are also collected. Additionally, other calculated data (e.g., virtual sensors for quality measurements) are fed into the prediction model. For batch operations, Petition 870250101780, dated 06 / 11 / 2025, page 87 / 167 74 / 124 post-batch status values ​​are used in the current batch forecasting process. To assess the quality of the forecasts against actual values, the batch progress status is entered into the forecasting model.

[00159] All predicted values ​​are stored, and the user interface is updated with the latest values. To improve the quality and reliability of the predictions, a periodic quality check is performed for each state value. The results are stored with a snapshot of the related input values, and necessary updates to the trained machine learning model and / or the training associated with the trained machine learning model are performed manually (in the cloud) by subject matter experts. Feedback is based on systematic sampling of the quality of the state values. Sampling consists of the manual labeling of the state value according to its quality by the responsible operator. Sampling is based on a fixed / variable frequency for specific processes or batches of a production run, in order to encompass a representative sample of the prediction quality.

[00160] The format or type of data used by the trained machine learning model and / or the prediction model depends on the model's real-time evaluation. In an example that includes real-time evaluation, some values ​​are constant and others are variable / time-dependent, within the real-time system information. Considering a machine learning model trained for the occupancy example described above, related to a maturation process: Petition 870250101780, dated 06 / 11 / 2025, page 88 / 167 75 / 124 [Total Occupancy Time of the Mixing Tank] = aO * [Progress] + al * [Sequence Step] + a2 * [Malt Charge] + a3 * [Average Heating Rate]

[00161] The above prediction uses a multivariate linear model. The PLC data (obtained from PLC storage, collected by different PLC and OPC protocols) and the real-time internal calculations (obtained from RAM storage) are time-dependent values / data, representing, respectively, the sequence step and the average heating rate. The upstream data (obtained from database storage), the post / historical batch state value data, and the laboratory device data (obtained from database storage) are constant values / data. In this example, the malt load is an upstream value. The batch progress estimate is a time-dependent value / data, being a prediction itself (stored in RAM or the database [8]). αθ, al, a2, and a3 are parameters adjusted for the prediction model.

[00162] In an example of fitting a trained machine learning model, the input data is structured as vectors / matrices. For model fitting, a target value is provided, not calculated. The following table presents an example of a dataset setup (Total samples M, Total batches B, and total number of entries / features N, including the progress estimate): TABLE 1: Target Value (Final Value) Lot Number / ID Progress Entry 2 (variable) Entry 3 Entry N Petition 870250101780, dated 06 / 11 / 2025, page 89 / 167 76 / 124 (Constant or variable progress per batch) Constant) Variable) Target Value 1 1 0% 5 10 -10000 Target Value 2 1 1% 1 10 2323 Target Value 3 1 2% 100 10 123123 ... 1 Target Value 100 1 99% 10 10 -2342 Target Value 101 1 100% 0 10 53453 Target Value 102 2 0% 2 15 123 Target Value 103 2 1% 10 15 -324234 Target Value 104 2 2% 4 15 4234 ... 2 Target Value 250 2 99% 130 15 0 Target Value 251 2 100% 3 15 32 ... Value Target M-1000 B 0% 8 12 234 Target Value M-999 B 1% 10 12 423424 Target Value M-998 B 2% 34 12 0 ... B Target Value Ml B 99% 23 12 123 Target Value MB 100% 1 12 123 Petition 870250101780, dated 06 / 11 / 2025, p. 90 / 167 77 / 124 Example 1. Dataset aimed at predicting the Total Occupancy Time of the Mixing Tank: TABLE 2: Total occupancy time [minutes] (Constant final value for batch number / Batch ID) Progress Sequence Step (variable) Malt Load (constant) Average Heating Rate (variable) 123 1 0% 1 10000 0.5 123 1 1% 1 10000 0.2 123 1 2% 2 10000 0.8 ... 1 123 1 99% 50 10000 0.4 123 1 100% 51 10000 0.4 130 2 0% 1 13000 0.5 130 2 1% 1 13000 0.1 130 2 2% 1 13000 0.9 ... 2 130 2 99% 49 13000 0.5 130 2 100% 51 13000 0.5 110 B 0% 1 11000 0.4 110 B 1% 2 11000 0.05 110 B 2% 2 11000 0.4 110 B 99% 50 11000 0.6 110 B 100% 51 11000 0.6 Petition 870250101780, dated 06 / 11 / 2025, page 91 / 167 78 / 124 Example 2. Dataset aimed at predicting the progress of the mixing tank: TABLE 3: Target Value (Progress) Number / ID. Batch Sequence Stage (variable) Malt Charge (constant) Average Heating Rate (variable) 0% 1 1 10000 0.5 1% 1 1 10000 0.2 2% 1 2 10000 0.8 ... 1 99% 1 50 10000 0.4 100% 1 51 10000 0.4 0% 2 1 13000 0.5 1% 2 1 13000 0.1 2% 2 1 13000 0.9 ... 2 99% 2 49 13000 0.5 00% 1 3000 0.5 ... 0% B 1 11000 0.4 1% B 2 11000 0.05 2% B 2 11000 0.4 ... B 99% B 50 11000 0.6 100% B 51 11000 0.6

[00163] Typical input and output values ​​for the trained machine learning model and component algorithms are floating-point numbers and, Petition 870250101780, dated 06 / 11 / 2025, page 92 / 167 79 / 124 generally, there are no limits to possible values. Only the status values ​​associated with progress (Target Value or predictor) are limited between 0 and 100%.

[00164] For expert-based models, all real-time system information is essential for both real-time tuning and evaluation. The algorithm interrupts tuning or performs real-time evaluation if specific data is not available. For black-box models, all data available to the process unit are optional, since the tuning algorithm in the examples above performs feature selection independently of expert analysis. Data / inputs are available during real-time evaluation; otherwise, the prediction is interrupted and / or canceled.

[00165] Preferably, multiple batches / runs should be included in the training dataset of the trained machine learning model to result in a more general understanding of process behavior. A minimum of 3 batches is recommended for statistical calculations and algorithm tuning, since the dataset for each batch consists of multiple temporal samples throughout the batch, such as a sampling frequency every 10 seconds. For example, for tuning the state value prediction using data from 3 batches, where the batch duration is 2 hours and uses 10 inputs / features, and the data is ideal (without filtered data), the estimated amount of data would be: [batch number] * [batch duration] * [sample frequency] * ([number of entries] + 1) Petition 870250101780, dated 06 / 11 / 2025, page 93 / 167 80 / 124 = 3 * 2 * 60 * (60 / 10) * (10 + 1) = 23760 data points or a dataset with 2160 rows and 11 columns (10 entries and 1 Target Value).

[00166] Examples of data preprocessing steps performed throughout method 100 include: selecting batch data (sensor values, logic / automation values, setpoints, state values) for the relevant process, for example, the mixing vat in the example above (all data unrelated to the batch is removed); locating batch indices / numbers within the current dataset and then appending all upstream process data (setpoints and state values) with the same batch ID / number, where the upstream data is defined as constant values ​​for the individual batches and, in the example above, this would be the mill and the mill malt silo; creating a column for progress, which is a linear interpolation from 0 to 1 in time between the start and end of the batch and, optionally, is the Value Target or an additional input / predictor; Create a Target Value column that, for each batch, contains the Target Value. For a state value forecast, the Target Value is a constant value corresponding to the final state value for the batch, and for a progress forecast, the Target Value is a time-dependent value, between 0 and 1, for the batch, based on relative time as mentioned earlier. The selection of features / inputs (columns) is done as follows: in the black box example, multiple fitting iterations are performed based on a feature selection algorithm (input / feature significance threshold), where some columns will be Petition 870250101780, dated 06 / 11 / 2025, page 94 / 167 81 / 124 excluded from the dataset. In the expert model example, only the specified features / inputs are included, in addition to the Target Value (specified in the database for the state-revenue value combination). The removal of any rows containing NaN in any of the columns and verification of the dataset quality (number of rows and batch number) are performed, interrupting the algorithm if the quality is too low. K-Fold techniques and other types of data splitting (training / test sets) are also used. Preferably, this should be included in the tuning and evaluation algorithm.

[00167] Optionally, real-time system information is stored in a database and / or RAM, and method 100 also includes database connection, such as queries and processes. The first, second, and third state values, as well as any associated calculated values, are stored in RAM and / or local storage for retrieval purposes. PLC data, such as sensory, logical, or automation values, are stored in the PLC, with the values ​​read by various standard means based on the industrial system's communication protocols.

[00168] Figure 7 shows a high-level system architecture diagram, according to the present description, for predicting the performance of an industrial system. Specifically, Figure 7 illustrates a performance prediction system 700 configured to execute method 600 and aimed at predicting state values ​​indicative of a future state of the industrial system. Petition 870250101780, dated 06 / 11 / 2025, p. 95 / 167 82 / 124

[00169] The performance prediction system 700 comprises a static prediction unit 702, a dynamic prediction unit 704, and a learning unit 706. The static prediction unit 702 comprises a trained machine learning model 708 and inputs to the trained machine learning model 708, such as historical data 710, process data 712, component data 714, and model specification data 716. The dynamic prediction unit 704 comprises a prediction model 718 and real-time data 720. The real-time data 720 comprises data from multiple sources, such as a PLC process unit 722 and laboratory devices 724, and includes data such as upstream data 726, batch progress estimates 728, post-batch status values ​​730, and / or calculated status values ​​732. The dynamic prediction unit 704 optionally comprises stored data. obtained from the static forecasting unit, static forecasting data 734.Additionally, the dynamic forecasting unit 704 stores the output of the forecasting model 718 as dynamic forecasting data 736. The learning unit 706 comprises feedback data 738 and, optionally, a learning data labeling unit 740.

[00170] The static prediction unit 702 and the dynamic prediction unit 704 operate together for accurate and reliable predictions specific to the location, process, component, and / or state value. For example, the static prediction unit 702 is configured for offline design and tuning, with maintenance and / or updates performed remotely by machine learning experts or a tuning algorithm. The prediction unit Petition 870250101780, dated 06 / 11 / 2025, page 96 / 167 83 / 124 dynamic 704 is configured to be site-specific. The combination of remote and local prediction units allows the use of a larger training dataset, such as from multiple processes, for the trained machine learning model 708, which can be subjected to supervised data cleaning and processing methods. Furthermore, the impacts of updating the trained machine learning model 708 can be analyzed before its implementation in industrial system monitoring. The local prediction model 718 then transforms the first state value 742 from a more generic future prediction into a customized real-time future prediction, combining the first state value 742 with real-time data 720 to generate a second state value 744. The second state value 744 is optionally stored as dynamic prediction data 736.

[00171] Dynamic prediction data 736 is optionally sent to a user interface 746, allowing an interface user, such as an industrial system operator, to select an operation to change one or more industrial system processes based on the second state value 744. Alternatively or additionally, dynamic prediction data 736 is used by the learning unit 706 to update the trained machine learning module 708. For example, a third state value 748 is calculated in the learning data labeling unit 740 based on data collected from the industrial system's sensor array. The third state value 748 is associated with the industrial system's performance over the time period associated with the second state value. Petition 870250101780, dated 06 / 11 / 2025, page 97 / 167 84 / 124 state 744 and is stored with or without the second state value 744 as feedback data 738.

[00172] Optionally, the data associated with the third state value 748 and the second state value 744 are stored as feedback data 738 and / or the difference between the third state value 748 and the second state value 744. The trained machine learning module 708 is tuned to reduce the difference between the second state value 744 and the third state value 748. Optionally, the third state value 748 is an input to the static prediction unit 702, as in combination with historical data 710, and / or an input to the dynamic prediction unit 704, for example. where the third state value 748 is included in the real-time data 720. The use of the third state value 748 as real-time data 720 efficiently updates the dynamic forecasting unit 704, so that the second state value 744 becomes more accurate and reliable due to a constantly updated performance forecasting system 700.Furthermore, using the third state value 748 to update the static forecasting unit 702 results in a continuous learning cycle, in which the performance forecasting system 700 is improved over time.

[00173] Figure 8 shows a flowchart illustrating a method for updating a trained machine learning model, configured to predict the performance of an industrial system based on data collected from a set of sensors in the industrial system. Specifically, Figure 8 illustrates a computer-implemented method 600, compatible with the processes of methods 100, 400, 500, 600, Petition 870250101780, dated 06 / 11 / 2025, page 98 / 167 85 / 124 900 and / or 1000.

[00174] Figure 8 represents the 800 method, which is a process for carrying out a continuous learning cycle to predict the performance of an industrial system with greater accuracy. This additional prediction is more accurate and / or reliable, indicating a future state of the industrial system. The 600 method is repeated to form a continuous monitoring process of the system. An example of a system for applying the 800 method is shown in Figure 7.

[00175] Method 800 comprises steps 802, 804, 806, and 808. Step 802 comprises calculating a third state value using data collected from a set of sensors in the industrial system; step 804 comprises fitting a trained machine learning model based on the third state value; step 806 comprises obtaining a fourth state value from the trained machine learning model; step 808 comprises determining a fifth state value using the third and fourth state values; and step 810 comprises updating the trained machine learning model.

[00176] Step 802 consists of calculating a third system state value using data collected from the industrial system's sensor array over the second time period, where the third system state value is associated with the industrial system's performance in the second time period and is indicative of a measured state of the industrial system in the second time period. For example, step 802 consists of determining, at a future point in time, the state of the industrial system such that Petition 870250101780, dated 06 / 11 / 2025, page 99 / 167 86 / 124 a previously obtained predicted state of the industrial system for the same future point in time can be compared.

[00177] Step 804 involves tuning the trained machine learning model to reduce the difference between the second state value determined by the prediction model and the third state value of the system. For example, step 804 involves using the third state value as input to the trained machine learning model, resulting in more accurate and reliable output predictions. In another example, step 804 involves using the third state value as additional training data for the third state value, resulting in more accurate and reliable output predictions. In a further example, step 804 involves altering the mathematical processes, parameters, or hyperparameters of the trained machine learning model, either through a remote machine learning expert or a tuning algorithm, based on the difference between the third and second state values.

[00178] Step 806 involves obtaining a fourth system state value from a trained machine learning model. This model predicts, based on the industrial system's performance in a new current period, the industrial system's performance in a subsequent future period. The industrial system's performance in the new current period is determined using data collected in the current period from a set of sensors in the industrial system. For example, the fourth state value is obtained based on the third state value from step 802. The fourth Petition 870250101780, dated 06 / 11 / 2025, pp. 100 / 167 The 87 / 124 system state value is associated with the predicted performance of the industrial system in a new future period that occurs after the period in which the third state value was calculated and therefore indicates a future state of the industrial system.

[00179] Step 808 consists of determining a fifth system state value using a forecasting model configured to predict the industrial system's performance in the new future period, based on the fourth system state value and real-time system information, which optionally includes the third state value from step 802. For example, the fourth state value is adjusted using recently measured data and associated values, such as the calculated third state value, to form a fifth state value. The fifth state value is therefore an updated and more reliable forecast of the new future state of the industrial system.

[00180] Step 810 consists of updating the trained machine learning model; this step is optional. Updating the trained machine learning model involves calculating a sixth state value, equivalent to a new third state value, using data collected from the industrial system's sensor array, and adjusting the trained machine learning model to reduce the difference between the fifth state value determined by the prediction model and the sixth state value calculated by the system. For example, step 810 is a repetition of steps 802 and 804 in a new future period. Preferably, as the industrial system process progresses, method 800 is continuously repeated. Petition 870250101780, dated 06 / 11 / 2025, pp. 101 / 167 88 / 124 at systematic time intervals, in a cycle that continuously updates and improves the predictions of the trained machine learning model and / or the prediction model.

[00181] Figure 9 shows a flowchart illustrating a method for using quality labeling in updating a machine learning model trained and configured to predict industrial system performance. Specifically, Figure 9 illustrates a computer-implemented method 900, compatible with the processes of methods 100, 400, 500, 600, 800, and / or 1000. For example, method 900 is executed between step 802 and step 804 of method 800.

[00182] Figure 9 represents method 900, which is a process for applying quality labels to predicted state values ​​for use in updating the trained machine learning model and for determining whether the available data is of sufficient quality to be used in this update. An example system for executing method 900 is shown in Figure 7.

[00183] Method 900 comprises steps 902, 904, 906, 908 or 910, and 912. Step 902 comprises assigning a first quality label and a second quality label to a second state value and a third state value; step 904 comprises filtering a plurality of third state values; step 906 comprises determining whether the filtered plurality of third state values ​​meets a predefined criterion; step 908 comprises canceling the update of the trained machine learning model; step 910 comprises continuing the model update. Petition 870250101780, dated 06 / 11 / 2025, pp. 102 / 167 89 / 124 trained machine learning; and step 912 comprises the output of the second state value to a user interface. The second state value and the third state value refer to a predicted state value and a calculated and / or measured state value, such as the second state value and the third state value of method 600 and method 800.

[00184] Step 902 consists of assigning a first quality label and a second quality label to a second and a third state value. The first quality label indicates the quality of the second state value. For example, the first quality label is assigned to the second state value based on the difference between the second and third state values. The first quality label represents a binary label, such as relevant or irrelevant, good or bad, and / or correct or incorrect, or the first quality label represents a non-binary label, such as a percentage difference between the second and third state values ​​and / or the opinion of an industrial system operator on the quality of the second state value. The second quality label indicates the quality of the third state value. For example, the third state value is calculated based on data collected from a set of sensors in the industrial system.If the calculation and / or data collected from the sensor array are below a quality threshold or exhibit errors and / or anomalies, the third state value receives a second quality label indicating low quality. Optionally, only the second state value receives a first-quality label or only the third value. Petition 870250101780, dated 06 / 11 / 2025, pp. 103 / 167 90 / 124 state receives a second-quality label.

[00185] Step 904 involves filtering a plurality of third-state values, such as filtering each third-state value from a plurality of third-state values ​​based on the second quality label assigned to each third-state value. Alternatively, step 904 is performed before step 902, where the plurality of third-state values ​​is filtered and then a second quality label is assigned to the filtered third-state values. The third-state values ​​and / or the data associated with those values ​​are filtered based on quality. The data associated with the third-state values ​​comprises data collected from one or more sensors in the industrial system. Quality is assessed based on multiple factors, such as missing data (data gaps) for a batch, or whether the third-state value is defective / bad based on the state of the processes and / or problems in data processing.

[00186] Step 906 consists of determining whether the filtered plurality of third-state values ​​meets a predefined criterion. The predefined criterion depends on the quality label and quality assessment. For example, the predefined criterion is a predefined number of third-state values ​​within the filtered plurality of third-state values. If the data quality is low, a tuning algorithm, or, for example, a machine learning expert, does not have enough data to accurately and / or efficiently update the machine learning model trained by method 600. For example, updating the machine learning model trained with Petition 870250101780, dated 06 / 11 / 2025, pp. 104 / 167 91 / 124 small amounts of data can introduce bias or inaccuracies.

[00187] Step 908 involves canceling the update of the trained machine learning model. For example, tuning the trained machine learning model based on the filtered plurality of third-state values ​​is canceled for the current target third-state value. Optionally, the update of one or more trained machine learning models is canceled, in which case the model(s) are not tuned or are reset to a previous condition if the update has already begun. Alternatively, updating, tuning, or any other type of optimization of the trained machine learning model(s) for a different target third-state value can be continued.

[00188] Step 910 consists of continuing to update the trained machine learning model. Continuing to update the trained machine learning model includes initiating the tuning and / or optimization of at least one machine learning model. Furthermore, continuing to update the trained machine learning model includes optimizing one or more trained machine learning models for a different target third state value.

[00189] Step 912 consists of sending the second state value to a user interface. Step 912 is optional. Alternatively, steps 906, 908, and / or 910 are repeated. Optionally, step 912 also consists of sending one or more third filtered state values ​​to the user interface, corresponding to the second value of Petition 870250101780, dated 06 / 11 / 2025, pp. 105 / 167 92 / 124 state, allowing an interface user to compare the second state value with the corresponding third state value. Furthermore, the second state value and, optionally, the third state value are subsequently used to select one or more control actions, allowing an interface user to select an operation to alter a process in the industrial system, as detailed in method 1000.

[00190] The advantage of determining whether to update the trained machine learning model or cancel it results in using only high-quality data as inputs and / or training data for the model. Therefore, Method 900 improves the accuracy and reliability of the trained machine learning model, since the continuous learning cycle cannot result in model degradation, and inadequate training data, such as data associated with point variations in measurements or inaccurate sampling, do not impact the trained machine learning model.

[00191] Figures 10A-10B show a flowchart illustrating a method for automated monitoring of an industrial system and generation of operating recommendations for the industrial system. Specifically, Figures 10A-10B illustrate a computer-implemented method 1000, compatible with the processes of methods 100, 400, 500, 600, 800 and / or 900.

[00192] Figures 10A-10B represent method 1000, which is a process for predicting a state value indicative of a future state of the industrial system and, based on the predicted state value, determining whether a Petition 870250101780, dated 06 / 11 / 2025, pp. 106 / 167 93 / 124 deviation associated with the predicted state value and select one or more mitigating control actions to reduce the impact of the deviation. An example system for implementing method 1000 is shown in Figures 2, 7, and 11.

[00193] Method 1000 comprises steps 1002, 1004, 1006, 1008, 1010, 1012, 1014, 1016, 1018, 1020, and 1022. Step 1002 comprises obtaining a first state value from a trained machine learning model; step 1004 comprises obtaining real-time system information, comprising data collected by a set of sensors in the industrial system; step 1006 comprises determining a second state value using a predictive model; step 1008 comprises determining the presence of a deviation using the second state value; step 1010 comprises evaluating the effect of the deviation; step 1012 comprises selecting one or more mitigating control actions; step 1014 comprises evaluating the difference in the industrial system's performance; Step 1016 involves determining the problem that caused the deviation based on the assessment; Step 1018 consists of identifying one or more control actions based on the problem;Step 1020 consists of prioritizing the mitigating control action(s); and step 1022 consists of generating a visual representation.

[00194] Step 1002 consists of obtaining a first state value from a trained machine learning model, for example, by executing step 602 of method 600. The trained machine learning model predicts, based on the industrial system's performance over an initial period, the system's performance. Petition 870250101780, dated 06 / 11 / 2025, pp. 107 / 167 94 / 124 industrial system performance over a second period. The industrial system's performance over the first period is determined using data collected during that period from a set of sensors in the industrial system. Furthermore, the first system state value is associated with the predicted performance of the industrial system over the second period and indicates a future state of the industrial system.

[00195] Step 1004 involves obtaining real-time system information, comprising data collected by a set of sensors in the industrial system, where the real-time system information is indicative of the current state of the industrial system and comprises data collected by the set of sensors in the industrial system, for example, by performing step 604 of method 600.

[00196] Step 1006 consists of determining a second state value using a forecasting model configured to predict the industrial system's performance in the second time period based on the system's first state value and real-time system information, for example, by executing step 606 of method 600.

[00197] Step 1008 consists of determining, using the second state value, whether there is a deviation greater than a predetermined limit in the predicted performance of the industrial system in the second time period, where the predetermined limit is associated with a process of the industrial system, such as, for example, executing step 104 of method 100.

[00198] Step 1010 consists of evaluating the effect if there is a deviation greater than the predetermined limit. Petition 870250101780, dated 06 / 11 / 2025, pp. 108 / 167 95 / 124 of the deviation in one or more outputs of the industrial system based on the difference between the predicted performance of the industrial system in the second period and a projected performance of the industrial system without the deviation, such as, for example, performing step 106 of method 100.

[00199] Step 1012 comprises the selection of one or more mitigating control actions from a plurality of predetermined control actions, where each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system, such as, for example, performing step 108 of method 100.

[00200] Step 1014 consists of evaluating the difference between the real-time performance of the industrial system and the performance of the industrial system over the first period, for example, by performing steps 402 and 404 of method 400.

[00201] Step 1016 involves determining a problem from a plurality of predetermined problems that cause the deviation based on the evaluated difference, such as, for example, performing step 404 of method 400.

[00202] Step 1018 involves identifying one or more control actions from the plurality of predetermined control actions based on the problem causing the deviation, where the identified control action(s) mitigate the difference between the predicted performance of the industrial system in the second period and the actual performance of the industrial system, for example, by executing step 406 of method 400. Petition 870250101780, dated 06 / 11 / 2025, pp. 109 / 167 96 / 124

[00203] Step 1020 consists of prioritizing one or more mitigating control actions based on the calculation of the impact of executing each mitigating control action on the industrial system, such as, for example, executing step 410 of method 400.

[00204] Step 1022 consists of generating a visual representation, such as generating a representation of the impact of executing one or more mitigating control actions in a user interface, for example, by executing step 112 of method 100. Advantageously, the visual representation allows a user of the user interface to select an operation to initiate one or more mitigating control actions to alter one or more processes of the industrial system.

[00205] Optionally, step 1022 also includes collecting user feedback on the visual representation and associated mitigating control actions. The feedback is used to manually adjust the trained machine learning model and / or the prediction model, for example, by adjusting hyperparameters, feature selection, and / or the model type. Collection is preferably automated, so that if the user rating associated with the visual representation is too low, the trained machine learning model and / or the prediction model will disable prediction in the user interface and any associated evaluation / adjustment.

[00206] Beneficially, one of the advantages of the 1000 method is the possibility of detecting and correcting deviations in the estimated values ​​of the current state or in the predicted values ​​of the future state, or even mitigating them, while the Petition 870250101780, dated 06 / 11 / 2025, pp. 110 / 167 97 / 124 impacts are still minimal or even occur before problems begin. Therefore, the 1000 method results in an optimized industrial system, with forecast-based monitoring that can be configured as an early warning system for inefficiencies in equipment or processes. Furthermore, the 1000 method allows operations to be updated based on changes in conditions, process variations, and / or equipment deficiencies that occur in real time and / or as a result of future processes.

[00207] Figure 11 shows a high-level system architecture diagram, according to the present description, for automated monitoring of an industrial system and generation of operating recommendations for the industrial system. Specifically, Figure 11 illustrates a semi-automated industrial system 1100 configured to execute method 1000 and directed to monitoring the industrial system 1100 by predicting state values ​​indicative of a future state of the industrial system and recommending operations for the industrial system 1100 based on the predicted state values.

[00208] The industrial system 1100 comprises input resources 1102, a first sensor 1104, an operating unit 1106, a process unit 1108, a second sensor 1110, outputs 1112, a third sensor 1114, data collected by one or more sensors 1116, a monitoring module 1118, a performance prediction system 1120, an operation support system 1122, a corrective action database 1124, an edge-based cloud server 1126, a user interface 1128, a Petition 870250101780, dated 06 / 11 / 2025, pp. 111 / 167 98 / 124 operating recommendation 1130 and a control action 1132.

[00209] Monitoring module 1118 obtains a first state value from the performance prediction system 1120, which is optionally the performance prediction system 700 of Figure 7. The performance prediction system 1120 comprises the machine learning model trained and configured to predict, based on the performance of the industrial system 1100 over a first period, the performance of the industrial system 1100 over a second period. The performance of the industrial system 1100 over the first period is determined using data collected by one or more sensors 1116 in that first period, where the sensor set comprises the first sensor 1104, the second sensor 1110, and / or the third sensor 1114. The first system state value is associated with the predicted performance of the industrial system 1100 over the second period and indicates a future state of the industrial system 1100.The monitoring module 1118 also obtains real-time system information, which indicates the current state of the industrial system 1100 and includes data collected by one or more sensors 1116.

[00210] Monitoring module 1118 determines a second state value of industrial system 1100 using performance prediction system 1120. Performance prediction system 1120 is configured to predict the performance of industrial system 1100 in the second time period, based on the first system state value and real-time system information. Using the second state value, monitoring module 1118 determines if there is a deviation greater than Petition 870250101780, dated 06 / 11 / 2025, pp. 112 / 167 99 / 124 a predetermined limit on the predicted performance of the industrial system in the second time period, where the predetermined limit is associated with a process of process unit 1108.

[00211] If a deviation exceeding the predetermined limit occurs, monitoring module 118 evaluates the effect of the deviation on one or more outputs 1112 of the industrial system, based on the difference between the predicted performance of the industrial system in the second period and the projected performance of the industrial system without the deviation. Then, monitoring module 1118 selects one or more mitigating control actions from several predetermined control actions, where each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system.

[00212] The selection of one or more control actions is performed by the operational support system 1122 and consists of evaluating the difference between the real-time performance of the industrial system and the performance of the industrial system over the first period. The evaluation of the difference is based on any of the following: data collected from the first sensor 1104, which monitors the input resources required for the process unit 1108; data collected from the second sensor 1110, which monitors one or more processes of the process unit 1108; or data collected from the third sensor 1114, which monitors the outputs associated with the process unit 1108.

[00213] The 1122 operational support system determines a problem by identifying it from several predefined problems that cause the deviation, based on Petition 870250101780, dated 06 / 11 / 2025, pp. 113 / 167 100 / 124 difference assessed. The various predefined problems are stored in a corrective action database 1124. One or more control actions, from the various predefined control actions, are then identified based on the problem causing the deviation, where the identified control actions mitigate the difference between the predicted performance of the industrial system in the second period and the actual performance of the industrial system.

[00214] The operational support system 1122 is additionally configured to prioritize one or more mitigating control actions based on the calculation of the impact of each action on the industrial system. The operational recommendation 1130, which represents the impact of each mitigating control action, is then issued by the monitoring module 118 to the user interface 1128. The operational recommendation 1130 allows the user of the interface 1128 to select an operation in the operational unit 1106 to initiate one or more mitigating control actions and modify one or more processes of the process unit 1108.

[00215] User interface 1128 is configured to analyze operation recommendation 1130 and subsequent operation selection in the operation unit 1106, such as executing method 500. The analysis is used to update the corrective action database 1124 and is used by the cloud server 1126 to update the monitoring module 1118. Optionally, the cloud server 1126 can be a local server and / or a computing device.

[00216] Figure 12 shows an example of a computer system for orchestrating the organization, Petition 870250101780, dated 06 / 11 / 2025, pp. 114 / 167 101 / 124 implementation and / or updating of policy management rules. Specifically, Figure 12 shows a block diagram of a computer system implementation according to the implementations exemplified in this description.

[00217] The 1200 computer system can be configured to perform any of the operations described herein, such as, for example, any of the operations discussed with reference to methods 100, 400, 500, 600, 800, 900, 1000, to the operational support system 200 of Figure 2 and / or the performance prediction system 700 of Figure 7. The computing system includes one or more computing devices 1202. The computing devices 1202 of the computing system 1200 comprise one or more processors 1204 and memory 1206. One or more processors 1204 may be any general-purpose processors configured to execute a set of instructions. For example, one or more processors 1204 may be one or more general-purpose processors, one or more FPGAs (Field Programmable Gate Arrays), and / or one or more ASICs (Application Specific Integrated Circuits). In one implementation, one or more processors 1204 include a single processor. Alternatively, one or more processors 1204 include a plurality of processors that are operationally connected.One or more 1204 processors are communicatively coupled to the 1206 memory via the 1208 address bus, the 1210 control bus, and the 1212 data bus. The 1206 memory can be a random access memory (RAM), a read-only memory (ROM), or a persistent storage device. Petition 870250101780, dated 06 / 11 / 2025, pp. 115 / 167 102 / 124 such as a hard disk, an EPROM (Errable-Programmable Read-Only Memory) and / or similar. The computing device(s) 1202 further comprise an I / O interface 1214 communicatively coupled to the address bus 1208, the control bus 1210 and the data bus 1212.

[00218] Memory 1206 may store information that can be accessed by one or more processors 1204. For example, memory 1206 (e.g., one or more non-transient computer-readable storage media, memory devices) may include computer-readable instructions (not shown) that can be executed by one or more processors 1204. The computer-readable instructions may be software written in any suitable programming language or may be implemented in hardware. Additionally, or alternatively, the computer-readable instructions may be executed in logically and / or virtually separate threads on one or more processors 1204.For example, memory 1206 may store instructions (not shown) that, when executed by one or more processors 1204, cause one or more processors 1204 to perform operations such as any operations and functions for which the computing system 1200 is configured, as described in this document. Additionally, or alternatively, memory 1206 may store data (not shown) that may be obtained, received, accessed, written, manipulated, created, and / or stored. The data may include, for example, the data and / or information described in this document in relation to this disclosure. In some implementations, the computing device(s) 1202 may obtain and / or... Petition 870250101780, dated 06 / 11 / 2025, pp. 116 / 167 103 / 124 store data in one or more remote memory devices of the computing system 1200.

[00219] The computing system 1200 further comprises a storage unit 1216, a network interface 1218, an input controller 1220 and an output controller 1222. The storage unit 1216, the network interface 1218, the input controller 1220 and the output controller 1222 are communicatively coupled to the central control unit via the I / O interface 1214.

[00220] Storage unit 1216 is a computer-readable medium, which may be transient or non-transient, comprising one or more programs. These programs contain instructions that, when executed by processors 1204, cause the computer system 1200 to execute the steps of the method described in this description. Alternatively, storage unit 1216 may be a transient computer-readable medium. Storage unit 1216 may be a persistent storage device, such as a hard disk, a cloud storage device, or any other suitable storage device.

[00221] Optionally, the industrial system 1100 and the processing unit 1108 are directed to any of the following equipment, or a combination thereof, without limitation: mixers and other feed preparation devices, atomizing devices, gas dispersers, drying chambers, fluidized beds and other dryers, spray freezing systems, pneumatic evaporators and crystallizers, centrifuges and separators, coolers and heat pumps, compressors, Petition 870250101780, dated 06 / 11 / 2025, pp. 117 / 167 104 / 124 homogenizers, granulators, cyclones and / or filters, heaters, pumps, blowers, condensers, valves, on-site cleaning and sanitizing systems, dust handling systems, brewing systems, coolers, freezers, distillation and fermentation systems, emission control systems, agricultural equipment, filling, bottling and packaging systems, food processing systems, ice machines, membrane filtration systems, milking systems, liquid processing systems, pumps and valves, tablet presses, vacuum systems, bioreactor systems, perfusion and feeding systems, sterilization systems, culture medium recovery systems, harvesting systems, scaffolding equipment design, interconnecting ducts.and / or piping and other related auxiliary equipment, as well as associated processes, process parameters, control systems and bioprocess sensors.

[00222] Network interface 1218 can be a Wi-Fi module, a network interface card, a Bluetooth module, and / or any other suitable wired or wireless communication device. In one implementation, network interface 1218 is configured to connect to a network, such as a local area network (LAN), a wide area network (WAN), the Internet, or an intranet.

[00223] At this point, it should be noted that the orchestration of the methods of this disclosure, as described above, may involve, to some extent, the processing of input data and the generation of output data. This processing of input data and the generation of Petition 870250101780, dated 06 / 11 / 2025, pp. 118 / 167 105 / 124 Output data are implemented in hardware or software. For example, specific electronic components are employed in a control module or similar or related circuit to implement the functions associated with the methods, according to this disclosure, as described above. Alternatively, one or more processors, operating according to instructions, may implement the functions associated with the methods, according to this disclosure, as described above. In this case, it is within the scope of this disclosure that such instructions are stored in one or more processor-readable transient or non-transient storage media (for example, a magnetic disk or other storage media), or transmitted to one or more processors by means of one or more signals embodied in one or more carrier waves.

[00224] The present description should not be limited in scope by the specific embodiments described herein. In fact, other embodiments and modifications of the present description, beyond those described herein, will be evident to those skilled and knowledgeable in the art from the preceding description and the accompanying drawings. Therefore, such other embodiments and modifications are included within the scope of the present description. Furthermore, although the present description has been described herein in the context of at least one specific implementation in at least one specific environment for at least one specific purpose, those skilled and knowledgeable in the art will recognize that its usefulness is not limited to this and Petition 870250101780, dated 06 / 11 / 2025, pp. 119 / 167 106 / 124 that the present description can be successfully implemented in any number of environments for any number of purposes. Consequently, the statements presented herein should be interpreted considering the scope and spirit of the present description, as described herein. Numbered Statements of Invention

[00225] 1 - A computer-implemented method for determining actions to improve the operation of an industrial system, the method comprising: to obtain reference data indicative of one or more states of the industrial system, where the reference data are associated with data collected from one or more sensors of the industrial system; to determine if there is a deviation greater than a predetermined limit in the reference data, where the predetermined limit is associated with a process of the industrial system; If a deviation exceeds the predetermined limit, evaluate the effect of the deviation on one or more outputs of the industrial system based on the difference between the performance of the industrial system with the deviation present at an initial instant and the projected performance of the industrial system at an initial instant without the deviation present; Select one or more mitigating control actions from a plurality of predetermined control actions, where each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system. Petition 870250101780, dated 06 / 11 / 2025, pp. 120 / 167 107 / 124 and in which the selection includes: Identify one problem, from among several predetermined problems, as the cause of the deviation based on the assessed effect of the deviation; to identify one or more control actions from a plurality of predetermined control actions based on the identified problem causing the deviation, where the identified control action(s) are selected to reduce the effect of the deviation on one or more outputs of the industrial system; Calculate the impact of implementing each mitigating control action on one or more outputs of the industrial system; and display a representation of the impact of executing one or more mitigating control actions in a user interface, allowing the interface user to select an operation to initiate one or more mitigating control actions to alter one or more processes of the industrial system.

[00226] 2 - The method of statement 1, in which the reference data comprise: data collected over an initial period of time from one or more sensors of the industrial system, where the data collected over an initial period of time are indicative of a historical state of the industrial system; or real-time data collected by one or more sensors of the industrial system, where the real-time data are indicative of the current state of the industrial system; or a predicted state value of the industrial system, in Petition 870250101780, dated 06 / 11 / 2025, pp. 121 / 167 108 / 124 that the predicted state value is associated with a predicted performance of the industrial system in the second time period and is based on data collected over the first time period, and where the predicted state value is indicative of a future state of the industrial system.

[00227] 3 - The method according to the statement 2, where determining the presence of the deviation involves: evaluating the data collected over the first period, where this data is time-dependent; To identify one or more trends and / or temporal patterns in the evaluated data collected over the first period.

[00228] 4 - The method in accordance with any of statements 1 to 3, additionally comprising the prioritization of one or more mitigating control actions based on the calculated impact of implementing each mitigating control action.

[00229] 5 - The method according to the statement 4, where the operation to alter one or more processes of the industrial system is the highest priority mitigating control action that, when selected, causes a change in one or more processes of the industrial system in such a way that the impact of the deviation on the industrial system is mitigated.

[00230] 6 - The method according to the statement 4, where the operation to alter one or more processes of the industrial system is different from the highest priority mitigating control action.

[00231] 7 - The method according to any of statements 1 to 6, additionally comprising Petition 870250101780, dated 06 / 11 / 2025, pp. 122 / 167 109 / 124 to determine if any of the one or more mitigating control actions have been implemented.

[00232] 8 - The method according to the statement 7, in which feedback data is additionally stored, which feedback data is associated with determining whether one or more of the mitigating control actions have been performed.

[00233] 9 - The method according to the statement 8, in which feedback data is additionally used to determine if there is any deviation.

[00234] Statement 10 - The method in accordance with statement 8 or 9, in which feedback data is used to better identify the problem causing the deviation.

[00235] 11 - The method according to any of statements 8 to 10, in which feedback data is used to further identify one or more control actions from the plurality of predetermined control actions.

[00236] 12 - The method according to any of the declarations 1 to 11, additionally comprising: updating the reference data, wherein the updated reference data comprises an additional predicted state value from the industrial system user interface; To determine if there is a new deviation between the additional predicted state value and an updated measured performance of the industrial system based on updated data collected from one or more sensors of the industrial system; If a deviation exists, calculate the impact of the new deviation; Petition 870250101780, dated 06 / 11 / 2025, pp. 123 / 167 110 / 124 and select one or more additional control actions based on the impact of the new deviation.

[00237] 13 - The method in accordance with any of statements 1 to 12, in which the determination of the existence of a deviation is carried out by a deviation classifier.

[00238] 14 - The method according to the statement 13, in which the deviation classifier is based on first-principles models.

[00239] 15 - The method according to statement or 14, where the deviation classifier is one or more trained machine learning models.

[00240] 16 - The method according to the statement 15, where one or more trained machine learning models are any of the following: a support vector machine, a neural network, a decision tree-based algorithm, a set-based algorithm, a generalized likelihood ratio detector, or a cumulative sum detector.

[00241] 17 - The method according to statement or 16, additionally comprising selecting a machine learning model trained from one or more machine learning models trained based on one or more industrial system states associated with the obtained reference data.

[00242] 18 - The method according to any of statements 15 to 17, where the deviation classifier is configured to indicate whether or not there is a deviation based on an input comprising indicative reference data Petition 870250101780, dated 06 / 11 / 2025, pp. 124 / 167 111 / 124 from one or more states of the industrial system.

[00243] 19 - The method according to any of statements 1 to 18, in which data collected from one or more sensors of the industrial system are collected over a predefined period of time.

[00244] 20 - The method according to statement 19, where the predefined time window is 10 minutes.

[00245] 21 - The method according to any of statements 1 to 20, in which data collected from one or more sensors of the industrial system are collected in real time.

[00246] 22 - The method according to any of statements 1 to 21, wherein the data collected from one or more sensors of the industrial system comprise a sequence of data of predetermined size.

[00247] 23 - The method in accordance with any of statements 1 to 22, additionally comprising the generation of a visual representation of the impact of the deviation, determining a relevant graphical interpretation of the impact of the deviation from a list of graphical interpretations based on mapping the data associated with the impact of the deviation to the predicted state value.

[00248] 24 - The method according to any of statements 1 to 23, in which the determination of the problem is carried out using a rule-based model associated with one or more processes of the industrial system, and includes the analysis of the plurality of predetermined control actions based on the rule-based model.

[00249] 25 - The method in accordance with any of statements 1 to 24, where the impact of the deviation comprises Petition 870250101780, dated 06 / 11 / 2025, pp. 125 / 167 112 / 124 a plurality of impact values, including any of the following measures: production measures, operational measures, environmental measures, safety measures or efficiency measures.

[00250] 26 - The method in accordance with statement 25, whereby prioritizing one or more control actions involves ordering one or more control actions based on a weighted combination of the plurality of impact values.

[00251] 27 - The method in accordance with statement 26, in which the weighted combination is adapted to the predicted state value.

[00252] 28 - The method in accordance with any of statements 1 to 27, additionally comprising the implementation of at least one or more mitigating control actions.

[00253] 29 - The method in accordance with any of statements 1 to 28, additionally comprising displaying the visual representation of the impact of the deviation on a user interface screen.

[00254] 30 - The method of any of statements 1 to 29, wherein the plurality of predetermined control actions and the plurality of predetermined problems are stored in a database of corrective actions, and wherein each problem is assigned to one or more predetermined control actions.

[00255] 31 - The method described in any of statements 1 to 30, in which data collected from one or more sensors of an industrial system over a prior period are used to determine if there is a deviation. Petition 870250101780, dated 06 / 11 / 2025, pp. 126 / 167 113 / 124

[00256] 32 - The method according to any of statements 1 to 31, in which data associated with one or more processes of the industrial system are used to determine if a deviation exists.

[00257] 33 - The method according to any of statements 1 to 32, in which data associated with one or more components of the industrial system are used to determine if a deviation exists.

[00258] 34 - A system comprising one or more processors and a memory that stores instructions which, when executed by the processor(s), cause the system to execute the steps of any of statements 1 to 33.

[00259] 35 - A computer-readable medium containing computer-executable instructions which, when executed by one or more processors, perform the steps of any of statements 1 to 33.

[00260] 36 - The computer-readable medium of declaration 35, where the medium is not transitory.

[00261] 37 - A computer-implemented method for predicting the performance of an industrial system, wherein the method comprises: Obtaining an initial system state value from a trained machine learning model, where: The trained machine learning model predicts, based on the industrial system's performance over a first period, the industrial system's performance over a second period, where the industrial system's performance over the first period is... Petition 870250101780, dated 06 / 11 / 2025, pp. 127 / 167 114 / 124 determined using data collected in the first period from a set of sensors in the industrial system; and the first system state value is associated with the predicted performance of the industrial system over the second period and is indicative of a future state of the industrial system; Obtaining real-time system information, where real-time system information is indicative of the current state of the industrial system and includes data collected by the industrial system's sensor array; The determination of a second system state value using a forecasting model configured to predict the industrial system's performance in the second time period based on the first system state value and real-time system information; and the updating of the trained machine learning model, where the update comprises: To calculate a third system state value using data collected from the industrial system's sensor array over the second time period, where the third system state value is associated with the industrial system's performance in the second time period and is indicative of the industrial system's state in the second time period; and to adjust the trained machine learning model to reduce the difference between the second state value determined by the prediction model and the third system state value. Petition 870250101780, dated 06 / 11 / 2025, pp. 128 / 167 115 / 124

[00262] 38 - The method according to statement 37, where the first time period is prior to the second time period.

[00263] 39 - The method according to statement 37 or 38, where the third state value is used to obtain an additional first state value from the trained machine learning model.

[00264] 40 - The method according to any of statements 37 to 39, in which one or more process cycles of the industrial system of a sequence of process cycles occurs within the second time period.

[00265] 41 - The method in accordance with any of statements 37 to 40, additionally comprising the storage of the second state value and the third state value.

[00266] 42 - The method in accordance with any of statements 37 to 41, in which the calculation of a third state value is repeated at predetermined time intervals throughout the second time period.

[00267] 43 - The method according to any of statements 37 to 42, where the second status value receives a first-quality label and the third status value receives a second-quality label.

[00268] 44 - The method according to statement 43, in which the adjustment of the trained machine learning model understands the determination of whether the first quality label is consistent with the second quality label.

[00269] 45 - The method in accordance with any of statements 37 to 44, where the update of the trained machine learning model is performed remotely. Petition 870250101780, dated 06 / 11 / 2025, pp. 129 / 167 116 / 124

[00270] 46 - The method according to any of statements 37 to 45, wherein the update of the trained machine learning model is additionally based on a comparison between a plurality of second-state values ​​and a plurality of third-state values.

[00271] 47 - The method in accordance with statement 46, additionally comprising filtering out plurality of third-party state values.

[00272] 48 - The method according to statement 47, additionally comprising determining whether the filtered plurality of third-state values ​​meets a predefined criterion, wherein the predefined criterion is associated with the number of filtered third-state values.

[00273] 49 - The method according to statement 48, in which if the predefined criteria are not met, the update of the trained machine learning model is canceled, so that the trained machine learning model is restored to a pre-updated state.

[00274] 50 - The method in accordance with statement 47 or 48, additionally comprising, if the predefined criteria are not met, the generation of one or more first-state values ​​using an alternative model.

[00275] 51 - The method according to statement 50, where the alternative model is based on the average of one or more stored third-state values.

[00276] 52 - The method according to any of statements 37 to 51, where one or more first-state values ​​obtained from the trained machine learning model are associated with a component or process of the industrial system. Petition 870250101780, dated 06 / 11 / 2025, pp. 130 / 167 117 / 124

[00277] 53 - The method according to any of statements 37 to 52, where one or more first-state values ​​obtained from the trained machine learning model are associated with an input and / or an output of the industrial system.

[00278] 54 - The method in accordance with any of statements 37 to 53, wherein the trained machine learning model is one of a plurality of trained machine learning models.

[00279] 55 - The method according to statement 54, wherein the plurality of trained machine learning models comprises any of the following: a linear multivariate regression model, a polynomial regression model, a random forest regression model, a neural network model, or a specialized model.

[00280] 56 - The method according to statement 54, additionally comprising: Receive a focus indicator from several focus indicators, where each focus indicator is associated with a component, process, input, or output of the industrial system; and select the machine learning model trained from the various machine learning models trained based on the focus indicator.

[00281] 57 - The method in accordance with any of statements 37 to 56, wherein the industrial system sensor array comprises one or more virtual sensors.

[00282] 58 - The method in accordance with any of statements 37 to 57, wherein the sensor array of the industrial system comprises one or more actuators. Petition 870250101780, dated 06 / 11 / 2025, pp. 131 / 167 118 / 124

[00283] 59 - The method in accordance with any of statements 37 to 58, wherein the real-time system information additionally comprises information associated with a component, process, input or output of the industrial system.

[00284] 60 - The method according to statement 59, in which the information associated with a component, a process, an input or an output of the industrial system is a prediction based on data collected from one or more virtual sensors.

[00285] 61 - The method according to statement 59 or 60, where the information associated with a component, a process, an input or an output of the industrial system is a prediction based on one or more outputs of the trained machine learning model.

[00286] 62 - The method in accordance with any of statements 37 to 61, wherein the data collected by the industrial system's sensor array comprise any of the following: temperature data, humidity data, pressure data, consumption data, process rates or operating conditions of the industrial system.

[00287] 63 - The method in accordance with any of statements 37 to 62, wherein the second state value comprises information associated with any of the following: a quality indicator for an output of the industrial system; a quality indicator for an input of the industrial system; a duration indicator for a process cycle of the industrial system; a productivity indicator for a process of the industrial system; a performance indicator for a component of the system. Petition 870250101780, dated 06 / 11 / 2025, pp. 132 / 167 119 / 124 industrial; or a state of a process in the industrial system.

[00288] 64 - The method in accordance with any of statements 37 to 63, wherein real-time system information is obtained from a storage database.

[00289] 65 - The method in accordance with statement 64, in which real-time system information is stored using RAM and / or PLC data storage.

[00290] 66 - The method according to statement 64 or 65, where the storage database is on an edge device or on a cloud server.

[00291] 67 - The method according to any of statements 37 to 66, wherein data collected from a set of industrial system sensors in a third time period are an input to the trained machine learning model.

[00292] 68 - The method in accordance with any of statements 37 to 67, wherein the trained machine learning model is configured to predict the performance of the industrial system in a third time period, based on data associated with one or more processes of the industrial system.

[00293] 69 - The method according to any of statements 37 to 68, wherein the trained machine learning model is configured to predict the performance of the industrial system in a third time period, based on data associated with one or more components of the industrial system.

[00294] 70 - The method in accordance with any of statements 37 to 69, where the forecasting model is a Petition 870250101780, dated 06 / 11 / 2025, pp. 133 / 167 120 / 124 multivariable linear model.

[00295] 71 - The method according to any of statements 37 to 70, wherein the trained machine learning model predicts the performance of the industrial system over a third time period using the second state value, wherein the third time period is later than the second time period.

[00296] 72 - The method according to any of statements 37 to 71, wherein the trained machine learning model predicts the performance of the industrial system over a third time period using the third state value, wherein the third period is later than the second time period.

[00297] 73 - The method according to statement 71 or 72, additionally comprising obtaining a first additional state value from the trained machine learning model and generating a second additional state value using the prediction model, wherein the second additional state value is associated with the industrial system performance in the third time period.

[00298] 74 - The method in accordance with any of statements 37 to 73, additionally comprising the generation of the trained machine learning model.

[00299] 75 - The method in accordance with any of statements 37 to 74, where the first state value is time-dependent and / or variable.

[00300] 76 - The method according to any of statements 37 to 75, where the first state value is time-independent and / or constant. Petition 870250101780, dated 06 / 11 / 2025, pp. 134 / 167 121 / 124

[00301] 77 - The method according to any of statements 37 to 76, wherein the first state value is a first state value of a plurality of first state values.

[00302] 78 - The method of any of the statements 37 to 77, in which the second state value is issued to a user interface of the industrial system.

[00303] 79 - A system comprising one or more processors and a memory that stores instructions which, when executed by the processor(s), cause the system to execute the steps of any of statements 37 to 78.

[00304] 80 - A computer-readable medium containing stored computer-executable instructions that, when executed by one or more processors, performs the steps of any of statements 37 to 79.

[00305] 81 - The computer-readable medium of declaration 80, where the medium is not transitory.

[00306] 82 - A computer-implemented method for automated monitoring of an industrial system and subsequent generation of operating recommendations for one or more processes of the industrial system, the method comprising: Obtaining an initial state value from a trained machine learning model, where: The trained machine learning model predicts, based on the industrial system's performance over a first period, the industrial system's performance over a second period, where the industrial system's performance over the first period is... Petition 870250101780, dated 06 / 11 / 2025, pp. 135 / 167 122 / 124 determined using data collected in the first period from a set of sensors in the industrial system; and the first system state value is associated with the predicted performance of the industrial system over the second period and is indicative of a future state of the industrial system; Obtaining real-time system information, where real-time system information is indicative of the current state of the industrial system and includes data collected by the industrial system's sensor array; Determining a second system state value using a forecasting model configured to predict the industrial system's performance in the second time period based on the first system state value and real-time system information; The determination, using the second state value, of whether there is a deviation greater than a predetermined limit in the predicted performance of the industrial system in the second time period, where the predetermined limit is associated with a process of the industrial system; If a deviation exceeding the predetermined limit is present, evaluate the effect of the deviation on one or more outputs of the industrial system based on the difference between the predicted performance of the industrial system in the second period and the projected performance of the industrial system without the deviation. select one or more mitigating control actions from a plurality of control actions Petition 870250101780, dated 06 / 11 / 2025, pp. 136 / 167 123 / 124 predetermined, where each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system, and where the selection comprises: to evaluate the difference between the real-time performance of the industrial system and the performance of the industrial system over the first period; To determine a problem from several predetermined problems that caused the deviation, based on the evaluated difference; Identify one or more control actions from the various predetermined control actions, based on the problem that caused the deviation, where the identified control action(s) mitigate(s) the difference between the predicted performance of the industrial system in the second period and the actual performance of the industrial system; To prioritize one or more mitigating control actions based on the calculation of the impact of executing each mitigating control action on the industrial system; and to display a representation of the impact of executing one or more mitigating control actions in a user interface, allowing the interface user to select an operation to initiate one or more mitigating control actions to alter one or more processes of the industrial system.

[00307] 83 - A system comprising one or more processors and a memory that stores instructions which, when executed by the processor(s), cause the system to execute the steps of statement 82. Petition 870250101780, dated 06 / 11 / 2025, pp. 137 / 167 124 / 124

[00308] 84 - A computer-readable medium containing stored computer-executable instructions that, when executed by one or more processors, performs the steps of any of the statements 82.

[00309] 85 - The computer-readable medium of declaration 84, where the medium is not transitory. Petition 870250101780, dated 06 / 11 / 2025, pp. 138 / 167

Claims

1 / 6 CLAIMS 1. A computer-implemented method for determining actions to improve the operation of an industrial system, characterized in that it comprises: obtaining reference data indicative of one or more states of the industrial system, wherein the reference data are associated with data collected from one or more sensors of the industrial system; determining whether a deviation greater than a predetermined limit is present in a comparison of the reference data with the expected values ​​of the reference data, wherein the predetermined limit is associated with a process of the industrial system; in the event that a deviation greater than the predetermined limit is present, evaluating the effect of the deviation on one or more outputs of the industrial system based on the difference between the performance of the industrial system with the deviation present at a first point in time and the projected performance of the industrial system at the first point in time without the deviation present;Select one or more mitigating control actions from a plurality of predetermined control actions, wherein each predetermined control action represents a predetermined operation to alter one or more processes of the industrial system, and wherein the selection comprises: identifying a problem, from a plurality of predetermined problems, as the cause of the deviation based on the assessed effect of the deviation; and Petition 870250101780, dated 06 / 11 / 2025, page 139 / 167 2 / 6 identifying one or more control actions from the plurality of predetermined control actions based on the problem identified as the cause of the deviation, wherein the one or more identified control action(s) are selected to reduce the effect of the deviation on one or more outputs of the industrial system; calculating the impact of executing each mitigating control action on one or more outputs of the industrial system;[e] to display a representation of the impact of executing one or more mitigating control actions in a user interface, allowing the interface user to select an operation to initiate at least one of the mitigating control actions to alter one or more processes of the industrial system.

2. Method according to claim 1, characterized in that the reference data comprises: data collected over a first time period from one or more sensors of the industrial system, wherein the data collected over the first time period are indicative of a historical state of the industrial system; or real-time data collected by one or more sensors of the industrial system, wherein the real-time data are indicative of a current state of the industrial system; or a predicted state value of the industrial system, wherein the predicted state value is associated with a predicted performance of the industrial system in a second time period and is based on the data collected over the first time period, and wherein the predicted state value is indicative of a future state of the industrial system.

3. Method according to claim 2, characterized in that the determination of the deviation being present comprises: evaluating the data collected over the first time period, wherein the data collected over the first time period are time-dependent; and identifying one or more time-dependent trends and / or patterns over the first time period in the evaluated data collected over the first time period.

4. A method according to any one of claims 1 to 3, characterized in that it additionally comprises prioritizing one or more mitigating control actions based on the calculated impact of implementing each mitigating control action.

5. A method according to any one of claims 1 to 4, characterized in that it additionally comprises determining whether any one or more mitigating control actions have been performed.

6. A method according to claim 5, characterized in that it additionally comprises the storage of feedback data, wherein the feedback data is associated with determining whether any one or more mitigating control actions have been performed.

7. Method according to claim 6, characterized in that the feedback data are used to: determine more precisely whether a deviation exists; identify more precisely the problem causing the deviation from several predetermined problems; or identify more precisely one or more control actions from several predetermined control actions.

8. A method according to any one of claims 1 to 7, characterized in that it further comprises: updating the reference data, wherein the updated reference data comprises an additional predicted state value of the industrial system's user interface; determining whether there is a new deviation between the additional predicted state value and an updated measured performance of the industrial system based on updated data collected from one or more sensors of the industrial system; in the event that a deviation is present, calculating the impact of the new deviation; and selecting one or more additional control actions based on the impact of the new deviation.

9. A method according to any one of claims 1 to 8, characterized in that the determination of the existence of a deviation is performed by a deviation classifier, wherein the deviation classifier is one or more trained machine learning models.

10. Method according to claim 9, characterized in that it further comprises selecting a machine learning model trained from one or more machine learning models Petition 870250101780, dated 06 / 11 / 2025, page 142 / 167 5 / 6 trained based on one or more industrial system states associated with the obtained reference data.

11. A method according to any one of claims 1 to 10, characterized in that it additionally generates a visual representation of the deviation impact by determining a relevant graphical interpretation of the deviation impact from a list of graphical interpretations based on mapping the data associated with the deviation impact to the predicted state value.

12. A method according to any one of claims 1 to 11, characterized in that the problem determination is carried out using a rule-based model associated with one or more processes of the industrial system, and comprises the analysis of a plurality of predetermined control actions based on the rule-based model.

13. A method according to any one of claims 1 to 12, characterized in that the impact of the deviation comprises a plurality of impact values, including any of the following measures: production measures, operational measures, environmental measures, safety measures, or efficiency measures.

14. A system comprising one or more processors and a memory that stores instructions, characterized in that, when said instructions are executed by one or more processors, the system performs the steps of any one of claims 1 to 13.

15. Computer-readable medium containing stored computer-executable instructions, Petition 870250101780, dated 11 / 06 / 2025, pp. 143 / 167 6 / 6 characterized in that, when said instructions are executed by one or more processors, they perform the steps of any of claims 1 to 13. Petition 870250101780, dated 11 / 06 / 2025, pp. 144 / 167