Multi-stage process modeling method

Abstract

Methods for modeling a process, including: Dividing the process into several process stages, wherein the several process stages include at least a first process stage and a second process stage; and Developing several models, each corresponding to one of several process stages, wherein the several models include at least one model of the first process stage and one model of the second process stage; wherein the model corresponding to each process stage is developed using data from one or more passes of that process stage and output quality data relating to one or more passes of that process stage, and wherein the model corresponding to each process stage is adapted to generate an output quality prediction related to that process stage; and wherein the output quality prediction generated by the first process stage model is used to develop the second process stage model, characterized in that the model corresponding to the individual process stages is adapted to generate an output quality prediction that includes a prediction of either a stage end-product quality or a batch end-product quality.

Description

RELATED REGISTRATIONS

[0001] This application claims, pursuant to 35 USC §119(e), the benefit of the preliminary US patent application with serial number 61 / 347,244 entitled “Multi-Stage Process Modeling Method”, filed on May 21, 2010, the entire disclosure of which is hereby incorporated into the present subject matter. AREA OF REVELATION

[0002] This patent relates generally to process control system modeling and in particular to methods for modeling a batch or continuous process that is divided into several process stages. GENERAL STATE OF THE ART

[0003] Process control systems, such as those used in chemical, petroleum, or other processes, typically feature one or more process control devices and input / output (I / O) devices that communicate via analog, digital, or combined analog / digital buses with at least one host or operator workstation and one or more field devices. The field devices, which may include valves, valve positioners, switches, and sensors (e.g., temperature, pressure, and flow rate sensors), perform control functions within the process, such as opening or closing valves and measuring process control parameters.The process control units receive signals indicating process measurements from the field devices and use this information to implement a control routine. They then generate control signals that are sent to the field devices via buses or communication lines to control process operation. In this way, the process control units can execute and coordinate control strategies with the field devices via the buses and / or other communication lines.

[0004] Process information from the field devices and control units can be provided to one or more applications (i.e., software routines, programs, etc.) executed by the operator workstation (e.g., a processor-based system) to enable an operator to perform desired functions related to the process, such as recalling the current process state (e.g., via a graphical user interface), evaluating the process (e.g., via a visual object diagram), and so on. Many process control systems also include one or more application computers (e.g., workstations), typically implemented using a personal computer, laptop, or similar device, and connected via a local area network (LAN) to the control units, operator workstations, and other systems within the process control system.Each application computer can have a graphical user interface that displays process control information, including values ​​of process variables, values ​​of quality parameters related to the process, process fault detection information and / or process state information.

[0005] Typically, displaying process information on the graphical user interface is limited to showing a single value for each process variable related to the process. Additionally, some process control systems can characterize simple relationships between certain process variables to determine process-related quality metrics. However, in cases where a resulting product of the process fails to meet predefined quality control metrics, the process and / or process variables can only be analyzed after the completion of a batch, process, and / or assembly of the resulting product.Although considering the process and / or quality variables after the process is complete allows for the implementation of improvements in the manufacture or processing of subsequent products, these improvements cannot rectify currently completed products that do not meet specifications.

[0006] This problem is particularly noticeable in batch processes, i.e., in batch process control systems that implement batch processes. As is known, batch processes typically proceed in such a way that a common set of base materials is processed together as a "batch" in a varying number of stages or steps to produce a product. Many stages or steps of a batch process can be carried out in the same piece of equipment, such as a tank, while other stages or steps can be carried out in other pieces of equipment.Because the same base materials are processed differently at various stages or steps over time, often in a shared piece of equipment, it is difficult to accurately determine at any stage or step of the batch process whether the material in that batch is being processed in a way that is likely to produce the final product with the desired or sufficient quality metrics. That is, because the temperature, pressure, consistency, pH, or other parameters of the processed materials change over time during the execution of the batch, often while the material remains in the same location, it is difficult to determine whether the batch processes at any given time during the batch run are proceeding in a way that is likely to produce the final product with the desired or sufficient quality metrics.

[0007] A well-known method for determining whether a currently executed batch is progressing normally or within the desired specifications (and therefore likely to result in a final product with the desired or sufficient quality metrics) compares various process variable measurements taken during the execution of the current batch with similar measurements taken during the execution of a "golden batch." In this case, a golden batch is a pre-defined, previously executed batch that represents the normal or expected execution of the batch and results in a final product with the desired quality metrics. However, batch runs of a process typically vary in length, i.e.,The time required to complete a batch makes it difficult to determine which time within the "golden batch" most closely corresponds to the currently measured parameters of the ongoing batch. Furthermore, batch process variables can vary significantly during batch execution compared to a selected batch without resulting in a substantial deterioration in the quality of the final product. Therefore, it is often difficult or even practically impossible to identify a specific batch run that can be used as a "golden batch" in all cases and against which all other batch runs should be compared.

[0008] A method for analyzing the results of ongoing batch processes, which overcomes one of the problems of using a golden batch, involves generating a statistical model of the batch. This method involves collecting data for individual process variables from a set of process variables (batch parameters) across several different batch runs of a batch process and identifying or measuring quality metrics for each of these batch runs. The collected batch parameters and quality data are then used to generate a statistical model of the batch, where the statistical model represents the "normal" execution of the batch that yields the desired quality metrics.This statistical model of the batch can then be used to analyze the relationship between different process variable measurements taken during a particular batch run and the same measurements within the batch runs used to develop the model. For example, this statistical model can be used to provide an average or mean value for each measured process variable and a standard deviation assigned to each measured process variable at any given time during the batch run, against which the currently measured process variables can be compared. Furthermore, this statistical model can be used to predict how the current state of the batch affects, or relates to, the final quality of the batch product manufactured at the end of the run.

[0009] In general terms, this type of batch modeling requires the acquisition of vast amounts of data from various sources, including sensors, control loops, analyzers, virtual sensors, computational blocks, and manual inputs. Most of this data is stored in continuous data repositories. However, process management systems typically also contain significant amounts of data, particularly manual inputs. Data extraction from these two types of systems must be merged to meet the modeling requirements. Furthermore, as mentioned above, a batch process typically goes through several stages, steps, or phases that are very different from both a technical and modeling perspective. Therefore, a batch process is typically divided into phases, and a model can be created for each phase.In this case, data from many batch runs for the same phase or stage are grouped to develop the statistical model for that phase or stage. The purpose of such data arrangement is to eliminate or mitigate nonlinearities in the process. Another reason for developing separate batch models on a stage, phase, or other basis is that different process parameters are active at different stages and are used for modeling. In this way, a stage model can be created with a specific set of parameters relevant to that stage, ensuring that only those process parameters relevant to that particular batch stage are used or considered.For example, additives can be added to the main batch load at a certain stage, and process parameters related to these additives do not need to be considered at previous batch stages, but are relevant for the batch stage at which the additives are added.

[0010] However, when creating this statistical batch model, it must still be considered that different batch runs typically cover different time periods. This phenomenon is due to a number of factors, such as varying waiting times associated with manual operator activities within the batch runs, different environmental conditions requiring longer or shorter heating or other processing times, variations in the composition of base material leading to longer or shorter processing times during a batch run, and so on. In fact, it is normal for the data trend for a given process variable to cover a different duration in different batch runs, which is why common batch markers exhibit temporally shifted positions in the different batch processes.To create a valid statistical model, the data for each stage, operation, or phase of a batch must be aligned with comparable data for the same stage, operation, or phase from the other batches used to create the model. Therefore, before data measured during the runs of a batch process can be used to create a statistical model for modeling and analyzing the batch process, it is necessary to align the batch data from the different batch runs to a common timeframe. Methods for performing such alignment of batch data are disclosed in U.S. Patent Application No. 12 / 784,689 entitled "On-Line Alignment Of A Process Analytical Model With Actual Batch Operation," filed on May 21, 2010, the disclosure of which is hereby incorporated into the present subject matter.After alignment, the batch data can be used in conjunction with analysis tools such as principal component analysis (PCA) and latent structure projection (PLS) to develop models of the batch process that can be used to model and analyze further runs of the batch process.

[0011] The uninterrupted ("online") use of analytical tools such as PCA and PLS methods for defect detection and prediction of quality parameters has often been limited to continuous processes where a single product is manufactured. In such cases, the process is frequently treated as a single unit with a fixed set of measurements and laboratory analysis. For these process types, a single PCA or PLS model can be developed and applied to an online environment. However, to meet the requirements of continuous or batch processes where a large number of products are manufactured using one or more pieces of equipment, each with its own set of instruments and quality parameters, a more general approach is needed to develop a model with process interruption ("offline") and then apply it to online analysis.

[0012] Applying online analysis tools to continuous and batch processes presents several challenges. First, in a batch environment, a product may be manufactured using numerous pieces of equipment operating in series, in parallel, or in a hybrid configuration where some equipment operates serially and some in parallel. The equipment used in manufacturing and related processes depends on the product being manufactured. At various points in the manufacturing process, different laboratory and field measurements may be required for one product than for another, or different laboratory and field measurements may be used to manufacture different products, thus complicating model development.Similarly, a continuous operating environment can also include a variety of equipment components arranged in different configurations. The processing associated with each piece of equipment, along with related process measurements and controls, may vary in some cases, while the processing conditions change with throughput or the product being processed.

[0013] Tools designed to support online analysis for process modeling must therefore consider the manufactured product, the equipment setups that can be used to manufacture the product, and the different operating conditions and associated field and laboratory measurements necessary for producing the product. Previous modeling approaches used a single, aggregated model for a process, which did not allow for changes in operating conditions and associated field and laboratory measurements, as are necessary in modeling processes involving a variety of equipment or the production of a wide range of different products.

[0014] A method according to the preamble of claim 1 is known from DE 102 03 320 A1. SUMMARY

[0015] The present invention comprises, in various aspects, a method with the features of claim 1, a method with the features of claim 11, a computer-based model of a process with the features of claim 16, a computer-based method with the features of claim 26, a computer-based medium with the features of claim 30 or claim 40, a method with the features of claim 45, a method with the features of claim 49, a computer-based model with the features of claim 52, a method with the features of claim 56, and finally a computer-based medium with the features of claim 61. Advantageous developments of the inventive concept are the subject of the respective dependent claims.

[0016] Modeling a batch or continuous manufacturing process can be supported by dividing the process into different manufacturing stages necessary to produce a specific product. In this context, a manufacturing "stage" can be characterized by the type of equipment required for processing, the field and laboratory measurements necessary to monitor or control the process, the process operating conditions that must be maintained, and the impact on the manufactured end product. The stage concept can be applied to both continuous and batch processes in the development and application of online analytics. Once the different stages associated with a product have been defined, an analytical model can be created for each stage.The effort required to develop analysis models can be reduced because the offline analysis tools used for model development can be designed to leverage the stage definition to extract data from an online data repository. The online analysis application can be designed to automatically select the appropriate PCA and PLS model for online use based on the modeled processing stage. As a result, stage models can be flexibly configured to accommodate a changing manufacturing configuration.

[0017] A method for modeling a process implemented by a process control system involves dividing the process into several process stages, wherein the several process stages include at least a first process stage and a second process stage, and developing several models, each corresponding to one of the several process stages. The several models include at least one model of the first process stage and one model of the second process stage, and the model corresponding to each process stage is developed using data from one or more iterations of that process stage and output quality data associated with the iteration(s) of the process stage.Furthermore, the model corresponding to each process stage is adapted to generate an output quality prediction related to the process stage, and the output quality prediction generated by the model of the first process stage is used to develop the model of the second process stage.

[0018] The output quality data associated with the pass(s) of a process stage can include stage-end product quality or batch-end product quality. The model corresponding to each process stage can be adapted to generate an output quality prediction that includes a prediction of either stage-end product quality or batch-end product quality.

[0019] Developing multiple models corresponding to multiple process stages of a batch process involves collecting data during each of multiple runs of the process to generate multiple batches, including measurements for each of the multiple process variables during each process stage of each run of the process.

[0020] The modeling data for a process stage of a batch process forms a three-dimensional array containing multiple values ​​measured during the process stage for each of several batches at multiple times for each of several variables. This three-dimensional data array can be unfolded at each of the multiple times during the multiple process stages into a two-dimensional data array containing values ​​of the process variables for multiple batches. In this way, the three-dimensional array can be dimensioned by variables, time, and multiple batches.

[0021] Information derived from the first-stage model of a batch process and used by the second stage of the batch process can include a quality prediction for a batch produced by the batch process and can be used as initial conditions for the second-stage model. A forgetting factor (i.e., filtering) can be applied to at least one portion of the information derived from the first-stage model of a batch process and used by the second stage of the batch process.

[0022] When modeling a continuous process, developing multiple models corresponding to the various stages of the continuous process can involve collecting data during each of several time periods during the process's implementation. This data collection can include measuring values ​​for each of several process variables during each time period. The values ​​measured for a stage of a continuous process can comprise a three-dimensional data array containing multiple values ​​measured during that stage for each of several variables at each of several time periods during the process's implementation. This three-dimensional data array can then be unfolded into a two-dimensional data array containing values ​​of the process variables at each of several time periods during the continuous process's implementation.Developing multiple models can include creating a projection onto latent structures or a PLS model of a process stage.

[0023] A batch process implemented with (or executed by) a process control system can be analyzed by dividing the process into several process stages, including at least a first process stage and a second process stage, and by developing several models, each corresponding to a specific process stage. The multiple models include at least one model of the first process stage and one model of the second process stage. The second process stage model uses information derived from the first process stage model, and the multiple models are then used to predict the value of a process parameter.

[0024] In one embodiment, a process with a first process stage and a second process stage can be modeled by developing a model of the first process stage using a training data set corresponding to multiple iterations of the process, applying at least one section of the training data set as input to the model of the first process stage to generate an output quality prediction for the first process stage, and developing a model of the second process stage using at least a second section of the training data set and the output quality prediction for the first process stage. The model of the first process stage can further generate an indicator of the reliability of the output quality prediction for the first process stage, and the indicator of the reliability of the output quality prediction for the first process stage can be used to develop the model of the second process stage.Preferably, a model is developed for each process stage, at least one section of the training data set is applied as input to the model of each process stage to generate an output quality prediction for the process stage, and at least one section of the training data set and the output quality prediction for the preceding process stage are applied as input to the model of each process stage following the first process stage.

[0025] Such a model can be used by applying a first data set, obtained from a first run of the multi-stage process, to the model of the first process stage to generate an output quality prediction for the first process stage, and a second data set, obtained from a second run of the multi-stage process, and the output quality prediction for the first process stage to the model of the second process stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] They show: Fig. 1 A representation of a process control network with a control unit and field devices that can be used to implement batch processes. Fig. 2 a block diagram representing an exemplary process control system with an exemplary operations management system that can implement an online batch analysis system for analyzing batch processes. Fig. 3. A flowchart of an exemplary procedure for determining a statistical batch model for a batch process. Fig. 4 A representation of a data structure of an exemplary batch process, which includes process variable measurements and quality variable measurements related to the batch process. Fig. 5 a representation of a data structure that illustrates batch data for a number of different batch runs of a batch process and includes process variables and respective quality variables for each batch run. Fig. 6. A representation of a data structure illustrating batch data for a number of different batch runs of a batch process, showing process variables and respective quality variables after aligning the data from the batch runs in an offline alignment process. Fig. 7 a block diagram illustrating a multi-stage batch process. Fig. 8 A diagram illustrating the batch-wise unfolding of process data for a single process stage. Fig. 9 a diagram illustrating the batch-wise unfolding of process data for multiple process stages. Fig. 10 a diagram that represents data structures for single-block and multi-block PLC modeling. Fig. 11 a representation of data structures for multi-stage multi-block modeling, whereby batch final quality or stage final calculation term values ​​are transferred from one process stage to the subsequent process stage. Fig. 12 a representation illustrating two alternative modeling processes for multi-level, multi-block models. Fig. 13 a representation illustrating a PLS model that uses a forgetting factor to model a three-stage process. Fig. 14 a representation illustrating the development of a PLS model from a multi-stage batch process. Fig. 15 a representation illustrating the use of the PLS model, as in Fig. 14 was shown and developed. DETAILED DESCRIPTION

[0027] Fig. Figure 1 shows an exemplary process control system 10 with a process control unit 11 connected to a data archive 12 and one or more host workstations or computers 13 (which can be any type of personal computer, workstation, etc.), each having a display screen 14. The control unit 11 is also connected to field devices 15-22 via input / output (I / O) cards 26 and 28 and can be functional to implement one or more batch passes of a batch process using the field devices 15-22. The data archive 12 can be any desired type of data acquisition unit with any desired type of memory and any desired or known software, hardware, or firmware for storing data. The data archive 12 can be separate from the workstations 13 (as shown in Figure 1). Fig. 1 shown), or part of one of the same. The control unit 11, which is, for example, a DeltaV sold by Emerson Process Management. ® The control unit 11 can communicate with the host computers 13 and the data archive 12, for example, via an Ethernet connection or any other desired communication network 23. The control unit 11 can also be operated using any desired hardware and associated software, such as standard 4-20 mA devices and / or any intelligent communication protocol, such as the FOUNDATION ® -Fieldbus protocol, the HART ® -protocol, the WirelessHART™ protocol, etc., communicating with field devices 15-22.

[0028] The field devices 15-22 can be any type of device, such as sensors, valves, transducers, positioners, etc., while the I / O cards 26 and 28 can be any type of I / O device conforming to any desired communication or control protocol. In the embodiment shown Fig. 1. Field devices 15-18 are standard 4-20 mA devices or HART devices that communicate with the I / O card 26 via analog lines or combined analog and digital lines, while field devices 19-22 are intelligent devices such as FOUNDATION ®These are fieldbus devices that communicate with the I / O card 28 via a digital bus using a fieldbus communication protocol. Of course, the field devices 15-22 can also conform to any other standards or protocols, including those developed in the future.

[0029] The control unit 11 comprises a processor 30 that implements or monitors one or more process control routines (stored in a memory 32), which may have control loops, and communicates with the devices 15-22, the host computers 13, and the data archive 12 to control a process in any desired manner. It should be noted that each control routine or control module described herein may have parts that are implemented or executed by other control units or devices, if desired. Likewise, the control routines or modules described herein, to be implemented by the process control system 10, may take any form, including software, firmware, hardware, etc.Control routines can be implemented in any desired software format, such as using object-oriented programming, ladder logic, sequential function diagrams, function block diagrams, or any other software programming language or design paradigm. Likewise, the control routines can be hard-coded, for example, into one or more EPROMs, EEPROMs, application-specific integrated circuits (ASICs), or any other hardware or firmware elements. In this way, the control device 11 can be configured to implement a control strategy or control routine in any desired manner.

[0030] In some embodiments, the control device 11 implements a control strategy using so-called function blocks, wherein each function block is an object or other part (e.g., a subroutine) of an overall control routine and operates (via communication paths called links) in conjunction with other function blocks to implement process control loops in the process control system 10. Function blocks typically execute a function from an input function, such as one associated with a transducer, sensor, or other process parameter measuring device; a control function, such as one associated with a control routine that performs PID control, fuzzy logic control, etc.; or an output function that controls the operation of a device such as a valve, in order to perform a physical function within the process control system 10.Of course, hybrid or other types of function blocks also exist. Function blocks can be stored in and executed by the control device 11, which is typically the case when these function blocks are used for standard 40-20 mA devices and some types of intelligent field devices such as HART devices, or they can be implemented in and executed by the field devices themselves, which can be the case with fieldbus devices.

[0031] Like the pulled-apart block 40 from Fig. As illustrated in Figure 1, the control device 11 can have a number of single-loop control routines, represented as routines 42 and 44, and can, if desired, implement one or more extended control loops, such as multi-input / multi-output control routines, represented as control loops 46. Each such loop is typically referred to as a control module. In the illustration, the single-loop control routines 42 and 44 perform single-loop control using a single-input / single-output fuzzy logic control block and a single-input / single-output PID control block, respectively, each connected to a suitable analog input (AE) and analog output (AA) function block, which are associated with process control devices such as valves, measuring devices such as temperature and pressure sensors, or any other device within the process control system 10.The extended control loop 46, as shown in the diagram, has inputs that are communicatively connected to one or more AE function blocks and outputs that are communicatively connected to one or more AA function blocks, although the inputs and outputs of an extended control block 48 can also be connected to any other function blocks or control elements to receive other types of inputs and provide other types of control outputs. The extended control block 48 can be any type of model predictive control (MPC) block, neural network modeling block or control block, a multivariable fuzzy logic control block, a real-time optimizer block, etc., or an adaptively tuned control block, etc. It is understood that the function blocks consist of... Fig. 1, which have the extended control block 48, can be executed by the control unit 11 or alternatively be arranged in any other processing device and can be executed by it, for example in one of the workstation computers 13 or even in one of the field devices 19-22.

[0032] Furthermore, how Fig. Figure 1 illustrates that one or more process analysis routines 50 are stored in and executed by various devices of the process control system 10. In the illustration, the process analysis routines 50 are stored in one or more computer-readable memories 52 for execution on processors 54 of the workstations 13; however, the routines 50 can also be stored in and executed by other devices. Each process analysis routine 50 is communicatively linked to one or more control routines, such as control routines 42, 44, 46, and / or the data archive 12, to receive one or more measured process variable measurements. Each process analysis routine 50 can be used to develop a statistical process model and to analyze a running or online batch process based on this model.The analysis routines 50 can also display information about the online or ongoing batch, as implemented by the process control system 10, to users, such as batch operations personnel.

[0033] Fig. Figure 2 is a block diagram illustrating another example of a process control environment 100, which includes an operations management system (OSS) 102, also referred to as a process monitoring and quality prediction system (PPS), and which can be used to implement an online batch process modeling and analysis system. The OSS 102 is located within a factory 104, which has a process control system 106 that forms the process control network 10. Fig. 1 or parts thereof. The exemplary Factory 104 can be any type of manufacturing plant, process plant, automation plant, and / or any other type of process control structure or system. In some examples, Factory 104 may have a variety of plants located at different sites, and although Factory 104 consists of Fig. While the illustration shows a single process control system 106, the factory 104 may also have other process control systems.

[0034] The process control system 106, which is coupled to a control unit 108 via a data bus 110 and is part of this system, can include any number of field devices (e.g., input and / or output devices) for implementing process functions such as executing physical functions within the process or taking measurements of process variables. The field devices can be any type of process control component capable of receiving inputs, generating outputs, and / or controlling a process. For example, the field devices can be devices such as valves, pumps, fans, heaters, coolers, and / or mixers for controlling a process.Furthermore, the field devices can be output devices such as thermometers, pressure gauges, concentration meters, fluid level gauges, flow meters, and / or steam sensors that measure process variables within a process or parts of a process. The input devices can receive instructions from the control unit 108 to execute a specific command and effect a change in the process. Additionally, the output devices measure process data, environmental data, and / or input device data and transmit the measured data to the control unit 108 as process control information. This process control information can include the values ​​of variables (e.g., measured process variables and / or measured quality variables) that correspond to a measured output from the individual field devices.

[0035] In the example shown from Fig. 2. The control unit 108 can communicate with the field devices within the process control system 106 via the data bus 110, which can be coupled to intermediate communication components within the process control system 106. These communication components can include field terminal boxes that connect field devices in a command area to the data bus 110 for communication. These communication components can also include distribution cabinets that organize the communication paths to the field devices and / or the field terminal boxes. Furthermore, the communication components can include I / O cards for receiving data from the field devices and converting the data into a communication medium that can be received by the exemplary control unit 108. These I / O cards can convert data from the control unit 108 into a data format that can be processed by the corresponding field devices.In one example, the data bus 110 can be implemented using the Fieldbus protocol or other types of wired and / or wireless communication protocols (e.g. Profibus protocol, HART protocol, etc.).

[0036] The control unit 108 from Fig. 2 (which may be a PC or any other type of control device) manages one or more control routines for managing the field devices within the process control system 106. The control routines may include process monitoring applications, alarm management applications, process trend generation and / or history applications, batch processing and / or action management applications, statistical applications, video streaming applications, advanced control applications, and so on. Furthermore, the control device 108 may forward process control information to the PVS 102. The control routines may be implemented to ensure that the process control system 106 produces specified quantities of a desired product within a certain quality threshold. For example, the process control system 106 may be configured as a batch system that produces a product upon completion of a batch.In other examples, the process control system 106 can feature a manufacturing system with a continuous process.

[0037] The process control information from the control unit 108 can contain values ​​corresponding to measured process and / or quality variables originating from the field devices within the process control system 106. In other examples, the PVS 102 can parse values ​​within the process control information into the corresponding variables. The measured process variables can be associated with process control information originating from the field devices that measure parts of the process and / or characteristic curves of the field devices. The measured quality variables can be associated with process control information related to the measurement of process characteristic curves that are assigned to at least a part of a finished product.

[0038] For example, the process might involve a chemical reaction in a tank that produces a concentration of a chemical in a liquid. In this example, the concentration of the chemical in the liquid could be a quality variable. The liquid's temperature and the liquid flow rate into the tank could be process variables. The PVS 102 can use process control modeling and / or monitoring to determine that the liquid concentration in the tank depends on the liquid's temperature and the liquid flow rate. In other words, the measured process variables contribute to or influence the quality of the measured quality variables. The PVS 102 can use statistical processing to determine the extent to which each process variable influences and / or contributes to the quality variables.

[0039] Furthermore, the PVS 102 can model and / or determine relationships between the measured process variables and / or quality variables assigned to the process control system 106. These relationships between the measured process variables and / or quality variables enable the creation of one or more calculated quality variables. A calculated quality variable can be a multivariate and / or linear algebraic combination of one or more measured process variables, measured quality variables, and / or other calculated quality variables. The PVS 102 can also determine an overall quality variable from a combination of the measured process variables, measured quality variables, and / or other calculated quality variables. The overall quality variable can correspond to a quality determination of the entire process and / or to a predicted quality of a resulting product of the process.

[0040] As in Fig. As shown in Figure 2, the PVS 102 features an analysis processor 114 that uses descriptive modeling, predictive modeling, and / or optimization to generate feedback on the state and / or quality of the process control system 106. The analysis processor 114 can detect, identify, and / or diagnose process operating errors and predict the impact of any errors on quality variables and / or an overall quality variable related to the quality of a resulting product of the process control system 106. Furthermore, the analysis processor 114 can monitor the quality of process operation by statistically and / or logically combining quality and / or process variables into an overall quality variable related to the overall quality of the process.The analysis processor 114 can then compare the values ​​calculated from the overall quality variable and / or values ​​associated with the other quality variables to their respective thresholds. These thresholds can be based on predefined quality limits of the overall quality variable at different times in the process. For example, if an overall quality variable associated with a process exceeds a threshold for a certain period of time, the predicted final quality of the resulting product may not meet the quality metrics associated with the finished product.

[0041] If the overall quality variable and / or any other quality variable deviates from its respective threshold, the Analysis Processor 114 can display a fault indicator in a process summary graph and / or a process variation graph, showing an explained and / or unexplained variation (or deviation) related to the overall quality variable, and / or showing a variable that generated the process fault. The Example Analysis Processor 114 manages the analysis to determine the cause of one or more process faults by providing functions that allow an operator to generate process quality graphs (e.g., combination graphs, micrographs, process variation graphs, variable trend graphs, charts, etc.) that can display current and / or past values ​​of measured process variables, measured quality variables, and / or calculated quality variables.Furthermore, in some cases the analysis processor 114 generates these graphs while the process is running and continuously updates and / or recalculates multivariate statistics related to each graph as further process control information is received from the PVS 102.

[0042] To execute these functions for batch processes, the PVS 102 acquires process data for a number of different process variables for each of a number of different batch runs. This data can be acquired from the control unit 108 or the field devices within the control network 110, from a data archive (e.g., archive 12). Fig. 1) which may have already recorded and stored process data for different batch runs, or which may be recorded from any other data source. The PVS 102 then processes this data to generate one or more statistical batch models and stores the statistical batch models, for example, in a memory such as a computer-readable memory of the PVS 102 or in one of the memories of the workstations 13. Fig. 1. The statistical batch models can then be retrieved as needed to analyze future ongoing or online batch runs. In particular, the PVS 102 can use the stored batch models to analyze data collected during the online or ongoing operation of a specific batch run, or to allow a user to analyze this data.

[0043] To analyze data from a batch run while the batch is being executed online, the PVS 102 must first determine the precise stage at which the online batch is being executed with respect to the batch model. That is, the PVS 102 must determine which point of the batch model to compare with the online batch data so that it can determine other factors related to the online batch, such as whether parameters of the online batch are irregular or deviate from the specifications for those parameters in the batch model, whether the output of the online batch will meet desired quality metrics, and so on. In fact, an analysis of the online data using the statistical batch model must first determine the point within the statistical batch model that is most applicable to the online data.Only once the online data has been aligned with the statistical batch model can further analyses be performed, such as providing an operator with screen views to illustrate the results of a comparison between the online batch and the batch model, conducting statistical analysis to determine whether the batch is being produced normally or within limits, or whether the batch is being produced irregularly, and / or whether the output of the batch is predicted to meet the desired quality metrics, such as desired consistency, concentrations, etc.

[0044] Once the data for the current online batch has been aligned to a specific point within the batch model, the PVS 102's Analysis Processor 114 can, for example, provide the user with a series of different graphs or other displays to enable them to determine the current operational stage or the suitability of the online batch run. Some of these graphs or displays are discussed below; it is understood that other displays, analyses, or information may also be provided to a user, such as an operator, maintenance personnel, etc., either as well or alternatively.

[0045] As an example, the analysis processor 114 can generate a contribution graph by calculating the contributions of process variables and / or quality variables to the overall quality variable or to the multivariate statistical error indicators of modeled and unmodeled process variations. The contributions of the process and / or quality variables can be displayed as a modeled and / or unmodeled variation of each variable, as a contribution to the variation associated with the overall quality variable and / or the quality variable associated with the error.

[0046] Furthermore, the Analysis Processor 114 can generate variable trend graphs for any of the selected process and / or quality variables, along with a defined threshold. The variable trend graph can show values ​​associated with the variable over a period of time in the process relative to values ​​of the variable during similar periods in previous processes, such as the model variable values. By generating the contribution graph and / or variable trend graphs, the Analysis Processor 114 can also identify potential process corrections to compensate for the detected batch process error. The variable trend graph can help an operator determine the cause of a process error by providing an overlay of historical data plots from the batches used to create the batch model across associated variations (e.g., standard deviations), with the current value aligned to the same timescale.

[0047] The analysis processor 114 can also generate a quality prediction graph to determine the impact of the correction(s), if implemented, on the overall process quality. If the correction(s) maintain or improve the overall quality within the specified thresholds, the analysis processor 114 can instruct the PVS 102 to implement the correction(s). Alternatively, the analysis processor 114 can send instructions to the control unit 108 to implement the correction(s).

[0048] Furthermore, the analysis processor 114 can generate a micrograph when it identifies a defect related to an overall quality variable and / or another quality variable. The micrograph can display values ​​of the process and / or quality variables at a specified time (e.g., a time associated with the process defect) relative to a mean and / or standard deviation for each variable predicted by the batch model. Additionally, the micrograph can include word graphs indicating previous values ​​related to each process and / or quality variable within the model.From the microdiagram, the exemplary analysis processor 114 can enable the operator to determine and / or select one or more corrective actions regarding the process and / or to determine whether one of the corrections improves the process to such an extent that it is predicted that the overall quality variable will be within the specified limits.

[0049] The PVS 102 manages access to and control of process control data, including process variation graphs, contribution graphs, variable trend graphs, quality prediction graphs, and / or micrographs, via an online data processor 116. Furthermore, the online data processor 116 provides operators with access to retrieve, change, and / or modify process control data and / or generate instructions for field devices within the process control system 106.

[0050] To provide access to online analytics, the factory 104 identifies Fig. Figure 2 shows a router 120 and a local workstation 122, which are connected to the online data processor 116 via a local area network (LAN). Furthermore, the router 120 can connect any other workstations (not shown) within the factory 104 to the LAN 124 and / or the online data processor 116. The router 120, which can be connected to the other workstations wirelessly and / or via a wired connection, can have any type of wireless and / or wired router as an access hub for the LAN 124 and / or the online data processor 116.

[0051] The LAN 124 can be implemented using any desired communication medium and / or protocol. For example, the LAN 124 can be based on a wired or wireless Ethernet communication scheme. However, any other suitable communication medium and / or protocol can also be used. Furthermore, although a single LAN is shown, more than one LAN and suitable communication hardware can be used in the workstation 122 to provide redundant communication paths between the workstation 122 and any similar workstation (not shown).

[0052] In the diagram, LAN 124 is also connected to a firewall 128, which uses one or more rules to determine whether communication from remote workstations 130 and / or 132 is permitted to reach factory 104. The remote workstations 130 and 132 can provide access to resources in factory 104 for operators who are not physically present there. The remote workstations 130 and 132 are connected to firewall 128 via a wide area network (WAN) 134.

[0053] The workstations 122, 130, and / or 132 can be configured to access, modify, and / or correct one or more processes within the process control system 106 based on the online analysis performed by the PVS 102, or these workstations can directly implement the online process analysis applications and procedures described herein. For example, the workstations 122, 130, and / or 132 can have a user interface 136 that formats and / or displays process control information generated by the PVS 102. As another example, the user interface 136 can receive graphs and / or diagrams generated by the PVS 102, or alternatively, data for generating a process control graph and / or diagram.When the graph and / or chart data are received in the respective workstations 122, 130 and / or 132, the user interface 136 can generate a display of a graph and / or chart 138 that is relatively easy for an operator to understand. The example configuration from . Fig. Figure 2 shows the workstation 132 with the analytical user interface 136. However, workstations 122 and / or 130 can have two analytical user interfaces 136.

[0054] Furthermore, the user interface 136 can inform a process control operator about the occurrence of any process control errors within the process control system 106 and / or any other process control systems within the factory 104, as determined by the online analysis described herein. The user interface 136 can also guide a process control operator through an analysis process to determine the source of a process error and predict its impact on the quality of the resulting product. When a process error occurs, the user interface 136 can provide the operator with statistical information about the process control, thus enabling the operator to make any necessary adjustments to the process to correct any errors. By correcting errors during the process, the operator can maintain the quality of the resulting product.

[0055] Furthermore, the user interface 136 can display detection, analysis, corrective action, and quality prediction information via the exemplary PVS 102. For example, the user interface 136 can display a process overview diagram, a process variation graph, a micrograph, a contribution graph, a variable trend graph, and / or a quality prediction graph (e.g., graph 138). When the operator views these graphs 138, they can select additional graphs 138 to access multivariate and / or statistical process information and determine a cause of the process failure. The user interface 136 can also display possible corrective actions for a process failure. The user interface 136 can then allow an operator to select one or more corrective actions.After selecting a correction, the user interface 136 can transmit the correction to the PVS 102, which sends an instruction to the control unit 108 so that it makes the appropriate correction in the process control system 106.

[0056] The workstations 122, 130 and / or 132 from Fig. 2. Workstations can include any computing device, such as a personal computer, laptop, server, control unit, personal digital assistant (PDA), microcomputer, etc. Workstations 122, 130, and / or 132 can be implemented using any suitable computer or processing system. For example, workstations 122, 130, and / or 132 can be implemented using a single personal computer, a workstation with one or more processors, etc.

[0057] The process control environment 10 from Fig. 1 and Fig. 100 out Fig. Section 2 is intended to illustrate system types within which the exemplary methods and devices described in more detail below can be advantageously used. However, the exemplary methods and devices described here can also be advantageously used, if necessary, in other systems of greater or lesser complexity than in the exemplary process control environments 10 and 100 and / or the process control system 106. Fig. 1 and Fig. 2 and / or are used in systems related to process control activities, business management activities, communication activities, etc.

[0058] Currently, many process control systems provide analytical and / or statistical analysis of process information. However, these systems generally implement offline tools to determine the cause and possible corrective actions for process defects that may affect the quality of the resulting products. These offline tools may include process studies, laboratory studies, business studies, troubleshooting, process improvement analysis, and / or Six Sigma analysis. While these tools can correct the process for subsequent products, they cannot restore and / or correct process quality once the defect occurs. Thus, these offline tools do not prevent the production of substandard products.

[0059] In contrast, the exemplary online batch process control system analyses described here can be used within a process control system to provide detection, analysis, and / or correction information for errors within the process and to enable an operator to correct a process error while the product is being manufactured. In other words, process corrections can be implemented in response to predicted errors, at the time an error occurs, or essentially immediately after an error occurs. Although the exemplary methods and devices described here can be used to predict and / or correct process errors to improve the process quality of a batch and / or continuous process, they are described in detail with regard to batch processes.Additionally or alternatively, the exemplary methods and devices can be used to correct product quality by predicting product quality and correcting corresponding process errors and / or correcting detected process errors.

[0060] Fig. Figure 3 shows an example flowchart 150 of a procedure that can be implemented by the PVS 102 (which is the routine 50 from Fig. 1 can execute), to develop a statistical batch model for a batch process and then use this statistical batch model to analyze data from an online batch process run. At a block 152, the PVS 102 acquires batch data for a specific batch. This batch data can include measured, calculated, or estimated process variable values ​​for a number of different process variables for a specific batch run of the process, including, for example, input variables such as base material compositions and other initial conditions for the batch, variables of the ongoing process such as temperatures, flow rates, levels, or other process variable measurements, estimated process variables, environmental data such as humidity, ambient temperature, etc., laboratory data including data measured or obtained offline in one or more laboratory analyses, etc. This data is retrieved from a data archive (such as Archive 12). Fig. 1) for a previously executed batch. If required, a user or operator can select a specific batch run, the data of which is stored in a data archive, for use in the modeling process. After completion of the batch run for which the data was captured at block 152, and / or at the end of various stages, operations, or phases of the batch, the PVS 102 captures quality measurements or quality data for the batch run at block 154.Quality data can include any type of measurement or indication of the quality of the batch output or the output from any of the batch's stages, operations, or phases. Examples include material consistency, concentrations of a particular chemical or element, pH, material compositions and ratios, and / or any other quality data that indicates the success of the batch process in producing an acceptable or desirable output. Naturally, the specific quality data to be collected will depend on the type of product being manufactured, and this quality data can be measured online, determined through laboratory analysis, ascertained by visual inspection (and user input), calculated based on parameters, or determined in any other known manner.Furthermore, this quality data can be obtained from a batch archive if this data is stored therein, or via online processes or offline laboratory analyses.

[0061] Fig. Figure 4 shows a data structure 200 for an exemplary batch run (e.g., Batch No. 1) with measured process variables 202 and calculated or otherwise measured or determined quality variables 204, which may include one or more overall quality variables obtained at the end of the batch run through measurements or observations. Batch processes typically have one or more process stages, with each stage having one or more operations and each operation having one or more phases. Thus, the exemplary measured process variables 202 may include process variables (also called process parameters) of a single phase, a single operation, or a single stage, or process variables that encompass multiple phases, operations, or stages of the batch process. For example, the variable P1 in Fig. Variable 4 corresponds to a liquid flow rate (e.g., a process variable), while variables P2-P8 can correspond to temperature, pressure, another flow rate, etc. Variable 204 can correspond to quality variables such as concentration, etc. Although the batch process consists of Fig. While Figure 4 shows eight measured process variables 202 and two quality variables 204, the batch process in other examples may have fewer or more process variables and / or more or fewer quality variables. Additionally, the batch process data is recorded during a time period represented by the t-axis (in Figure 4). Fig. 4 marked “Time”).

[0062] The data graph 200 from Fig. Figure 4 shows that some of the process variables 202 are only relevant for specific times during the batch process. For example, the process variable P1 is relevant from the beginning of the batch until a point in the middle of the batch (or the stage, operation, or phase for which data was collected). Thus, if the variable P1 is associated with a liquid flow rate, liquid can only flow within the batch process from the beginning of the batch to a point in the middle of the batch. After this point, the batch must not use any liquid flow, which is why the variable P1 is not relevant to the batch process at this time. In contrast, the variable P4 from Fig. 4 is relevant for the entire batch process.

[0063] The exemplary quality variables 204 can be assigned to the entire batch process or to a specific phase or stage of the batch process. The quality variables 204 can be the result of a multivariate, statistical, and / or algebraic relationship between the measured process variables 202 and / or other quality variables 204, can be measured or determined in any known manner, or can be entered by a user. For example, the quality variable Q1 can correspond to a compositional quality of a product resulting from the batch process. Q1 is a quality variable, although it may not be directly measurable within the process control system 106. Instead, the compositional quality variable Q1 can be modeled and / or determined from a multivariate combination of the measured variables 202, or determined with some delay by laboratory analysis.

[0064] Referring again to Fig. Step 3 of PVS 102 then determines at block 156 whether batch data has been collected for a sufficient number of process runs to create an appropriate statistical model for the batch. If not, block 156 returns control to block 152 to collect further process variable data for another run of the batch process. If so, block 158 aligns the batch data from the stored batch models.

[0065] To illustrate this point, Fig. 5 represents a data structure 300 in connection with a set of exemplary batch runs, which are found in blocks 152-156. Fig. 3. These variables are obtained and can be stored in memory at the beginning of block 158. As shown, the data structure 300 has process variables 302 and corresponding quality variables 304 for each of a number of batch runs. The batch runs (e.g., BATCHES 1-N) indicate that this particular batch process has four stages (e.g., STAGES 1-4) that are executed sequentially. For example, STAGE 1 might correspond to combining and mixing chemicals in a batch, while STAGE 2 might correspond to baking the mixed chemicals in the batch. These stages might be further subdivided into operations, phases, and / or levels. Furthermore, the quality variables 304 might correspond to the measured process variables 302 of each stage, phase, operation, or level of the batch and / or might correspond to the end of the batch.

[0066] The exemplary data structure from Fig. Figure 5 shows that each individual batch can differ in duration, with the start and end times of each stage of the batch also varying from batch to batch. For example, BATCH 2 is completed in a shorter time than BATCH 1, while BATCHES 3 and 4 are completed after a longer time than BATCH 1. Furthermore, BATCH 1 requires more time to complete STAGE 1 than BATCH 2.

[0067] Referring again to Fig. 3, block 158 aligns the batch data Fig. 5 are aligned to enable the creation of the batch model. In some cases, this data can be aligned by expressing the relevant duration of each variable (not shown) as proportional to the time span for the corresponding stage(s). In this way, the varying time required to complete batches and / or stages can be overcome using the measured process variable 302. In another example, the well-known DTW method discussed above can be used to align the batch data to a constant or normalized timeframe, which could be, for example, a mean timeframe of all batches, an average timeframe of all batches, or another timeframe, such as that assigned to a control batch or a selected batch. The aligned batch data would be expressed as in data structure 350. Fig. Figure 6 shows the timeframe of each batch normalized to be exactly the same, and all stages are aligned. The actual data points within each batch are time-distorted by stretching or shortening them to fit the normalized timeframe used in the batch model. Of course, the time within each stage (phase or operation) can be time-distorted differently depending on the times of those stages (phases or operations) relative to the normalized time for that stage (phase or operation), so that all stages are aligned separately. In any case, any known method, such as the DTW method, can be used to time-align the batch data from the different batch runs before processing the batch data or developing a statistical model.

[0068] It will be understood that due to the expansion and contraction of the timeframe within the different batch runs for creating the data structure from Fig. Six more or fewer data points can be provided for each batch run, or assigned to the individual different levels of normalized data. If necessary, this data can be converted into a fixed number or set of normalized data points (e.g., by linear or nonlinear interpolation) so that each batch used to create the batch model has the same number of data points, or so that data points are available for each of the same times within the normalized timeframe of the batch model. Of course, as mentioned above, the values ​​of the points within the data can be obtained by interpolation between a large number of points using linear interpolation or any other desired interpolation, such as nonlinear interpolation.Furthermore, it is understood that each data point acquired for the various batch runs can be a statistical data point, such as an average, mean, etc., of a set of successively acquired raw data points. For example, a single data point for a batch run can be created as a statistical combination (typically, but not necessarily, an average) of any desired number of raw data points (e.g., 10, 100, or any other number) to reduce the model size and processing times associated with model processing. Naturally, the number of raw data points used to create a particular statistical data point across the batch runs can be based on the raw data measurement frequency relative to the total duration within the batch, and so on.

[0069] Once the batch data from the different batch runs have been aligned, as in Fig. As shown in section 6, a block of 160 is created ( Fig. 3) A statistical batch model, consisting of a stage model derived from the aligned data, to define, from a statistical perspective, the normal or expected operation of the batch process as defined by the data collected from the different batch runs in steps 152-156. A method for creating a statistical batch model establishes one or more model process variable trajectories for each of the process variables within the batch runs, each such model process variable trajectory identifying or expressing the expected or normal operation of a process variable for the period during which the process variable is relevant to batch execution. This period may be, for example, the entire length of the batch, one or more batch stages, phases, operations, levels, etc.As an example, each model process variable trajectory can define the expected value of the process variable as, for instance, the average or mean value of the process variable (calculated from the collected batch data) at the respective time point during the normalized time frame of the model. If required, each model process variable trajectory can also include one or more standard deviations assigned to the collected batch data at a given time point to indicate the typical variation of the batch data for that variable at that time point.

[0070] Referring again to Fig. 3. Alternatively, Block 160 can also develop other statistical models, such as a PCA (principal component analysis) model or a PLS (latent structure projection) model. As is known, the PCA modeling procedure develops a set of principal components for the batch modeling data and a PCA model matrix, which can then be used to analyze other batch data, such as data from an online batch. Furthermore, the PLS modeling procedure performs a PLS calculation to map the process variable data, which were collected or determined for the batch runs, to the quality variables that were measured, calculated, or otherwise determined for the batch runs used in the model. A PLS model can then be used to predict the quality variables of future batches based on the statistical values ​​of variables within the measured batches.PCA and PLS modeling methods are generally known and are therefore not described in detail here. Of course, other types of statistical batch models can also be derived from the aligned batch data if required. Fig. 6. After one or more statistical batch models have been created, a block 162 stores these models in a computer-readable memory, for example in a memory 52. Fig. 1, for later use.

[0071] Fig. Figure 7 shows an exemplary processing plant in which an exemplary multi-stage batch process (or continuous process) can be implemented, and in connection with which multi-stage modeling can be carried out. The illustrated exemplary processing plant 400 has several loading or storage tanks 401, 402, 403, 404, a reactor A 406, a reactor B 408, a filtration unit 410 and a storage tank 412. As indicated by the arrows 414 in Fig. As shown in Figure 7, in the illustrated exemplary batch process, reactor A 406 is loaded with base material process inputs from loading tanks 401 and 402, and reactor B 408 is loaded with base material process inputs from loading tanks 402 and 404. These materials are processed in reactor A 406 and reactor B 408, and the resulting products are directed to the filtration unit 410, where they are filtered to produce a filtered product, which is then loaded into storage tank 412 and stored there for subsequent further processing or use.

[0072] To facilitate modeling and enable more accurate modeling, this exemplary batch process, which may involve any number of operations or stages (e.g., loading, converting, filtering, storing, etc.), can be divided into several process phases or stages, each comprising one or more operations or stages. For example, a first process stage (stage 1) may involve feeding base materials into reactor A 406 and reactor B 408 and subsequently processing these base materials to generate reaction products. A second process stage (stage 2) may involve feeding these reaction products into filtration unit 410.A third process stage (stage 3) may include a series of filtration operations carried out within the filtration unit 410 to produce a desired filtered product, and a fourth process stage (stage 4) may include a further series of operations to direct the filtered product into the storage tank 412.

[0073] In general, a model of the entire batch process can be developed by creating separate models that correspond to the individual process stages. In the example process from Fig. 7. A model can be developed for each of the stages 1-4. The model corresponding to each stage can be developed by collecting data from each of several batches produced by the batch process. To differentiate the data collected in connection with separate batches, each produced batch can be assigned a unique batch identifier (416) and linked to all data collected in connection with the batch process that led to the production of that batch, such as data stored in data archives.

[0074] Fig. Figure 8 is a diagram that graphically represents the batch-wise unfolding of process data for a single process stage. The process data can be obtained from a number of batch runs, which can be used as a training dataset to develop an analytical model of the batch process or, in the case of a multi-stage process, an analytical model of the individual stages of the batch process.

[0075] As shown, a data file 420, used to store batch data, comprises a three-dimensional array of data for each of I batches or batch passes of an industrial batch process, with J variables and K sampling periods. The data file 420 stores values ​​for each of the J variables used in a batch pass. Values ​​can be obtained during all or some of the K sampling periods and of each I batch for all or some of the J variables. For a simple single-stage batch process, this type of unfolding is sufficient. For multi-stage batches, however, the data should be unfolded separately for each stage, as shown in Fig. 9 is shown and described in more detail below.

[0076] Before developing an analytical model of the process, a data file 420 is unfolded into a two-dimensional array 422 of dimensions I × KJ, as shown in Fig. Figure 8 shows that each of the K sampling periods is represented as a two-dimensional array 424 of dimensions I × J, which may contain variables for some or all of the variables J from some or all of the I batches. When the data file 420 is unfolded, the K two-dimensional arrays 424 of data (i.e., one for each of the K sampling periods) are arranged side by side to form the two-dimensional array 422, as shown.

[0077] A similar data unfolding scheme can be applied to a multi-stage batch process, as in Fig. Figure 9 shows that data for each process stage can initially be stored in a three-dimensional array 430, similar to the three-dimensional array 420. Fig. The three-dimensional array 430 has dimensions Ii × Ji × Ki, where 1 ≤ i ≤ S, and S is the total number of stages comprising the entire batch process. In this way, the data is represented in S three-dimensional arrays 430, with each array 430 storing data for J variables for I batches and for K sampling periods for a given process stage i between 1 and S. The data in each three-dimensional array i 430 is unfolded into a two-dimensional array 432 with dimensions Ii × KiJi, with Ki two-dimensional arrays 434 of dimensions Ii × Ji again arranged side by side.

[0078] Fig. Figure 10 shows several single-block and multi-block data structures that can be used for PLC modeling of a multi-stage industrial process. The simplest data structure (shown in the upper left section of Figure 10) is shown in Figure 10. Fig. 10) uses a single block X for all data used to develop a model of a process and a block Y for a batch-end-product quality indicator that can be calculated or measured for the batches used to generate the data for block X. In this single-block, single-data structure, a batch is treated as a stage, and all data for all stages are unfolded and synchronized as described above, after which a PLC model is developed for the batch as if the process were a single-stage process. Since most processes or process stages are strongly influenced by the conditions prevailing before the process or process stage entered the batch or stage with respect to final product quality, all data from the beginning of the batch or stage up to the present time are often used to generate quality predictions.These conditions, referred to as "starting conditions" in process modeling, include indications of the quality or condition of base or pre-processed materials used in the process, or of other permanent conditions known at the start of the stage that affect the stage-end or batch-end product quality.

[0079] Another data structure, shown in the upper right section of Fig. 10 uses blocks X1, X2, ... Xi for data, each used to develop models of individual process stages 1, 2, ... i. This data can be combined as a single block and used together with an associated batch-end-product quality indicator Y, as shown in the upper right section of Fig. Figure 10 shows that these can be used to generate a PLS model of the process, or separate stage end-quality indicators Y1, Y2 ... Yi can be used, as shown in the lower left section of Figure 10. Fig. Figure 10 is shown to create the PLS model of the process. If no stage end-of-stage quality indicators are available, a batch end-of-stage quality prediction can be calculated at the end of the stage and used as a substitute, or it can be used in addition to stage end-of-stage quality indicators if these are available.

[0080] Instead of using stage data blocks X1, X2, ... Xi and a PLS model of the process, stage-end-product quality indicators Y1, Y2, ... Yi, the PLS model can also be created using rating matrices or latent structural ratings T1, T2, ... Ti for each of the i stages together with the stage data blocks X1, X2, ... Xi, as shown in the lower right section of Fig. Figure 10 shows that the evaluation matrices or latent structure evaluations T1, T2, ... Ti can be combined to form a latent structure TS, which in turn can be used to generate a batch-end product quality prediction Y for creating a PLS model of the multi-stage process, as shown. Alternatively or additionally to these latent structure evaluations, stage-end quality or batch-end quality predictions can be generated at the end of the stage.

[0081] Fig. Figure 11 is a representation of data structures for multi-stage, multi-block modeling, where batch-end quality or stage-end calculation term values ​​are developed in one process stage and transferred to the subsequent process stage. As shown, a data structure A for a multi-block batch model (shown in the left section of Figure 11) has the following characteristics: Fig. 11) Data blocks X1, X2, etc., for stage data, each obtained during passes through process stages 1, 2, etc. Furthermore, data structure A includes an additional block I1, I2 for initial conditions applicable to each stage. For stages following the first stage, these additional blocks may also contain batch-end quality predictions Y, stage-end calculation terms, or other parameter prediction information from the immediately preceding stage. Thus, the additional block for data block X2, corresponding to the second process stage, may contain initial conditions I2 and also a batch-end product quality prediction Y, developed by the model from the first process stage.

[0082] An alternative data structure B, shown in the right-hand section of Fig. 11, is similar to data structure A, except that the additional blocks, in addition to or as an alternative to the stage-end-product quality predictions used in data structure A described above, contain quality calculation terms (rating numerator - denominator) and possibly confidence interval calculation terms T from the preceding stage for stages after the first stage.

[0083] The batch-end product quality predictions Y, or more precisely the quality-calculated deviation from the means, can be rescaled using a scaling factor λ, where 0 < λ < 1, to mitigate the impact of information from the previous stage on the batch-end quality prediction relative to the impact of information from the current stage. Principal components can also be used, and confidence interval calculation terms derived from stage data can be passed to subsequent stages as additional parameters via the add-on blocks, as described above.

[0084] Fig. Figure 12 is a diagram illustrating two alternative modeling operations for multi-block modeling of multi-stage processes. In the example shown, a multi-stage batch process is run multiple times to obtain process data and batch-end quality data for use in building the model. Data blocks X1, X2, etc., store data obtained from each stage of the multi-stage batch process, as described above. A quality prediction block Y stores data related to the batch-end product quality for each run of the multi-stage batch process. Once all this data has been collected and aligned as described above, one of the two operations described in Figure 12 can be used, for example, to create a model of the multi-stage batch process. Fig. Figure 12 shows how a PLS model of the process can be created.

[0085] Process (1), which in the left section of Fig. As shown in Figure 12, this involves creating a series of respective models M1, M2, ... Mn for the n stages of the multi-stage batch process. A PLS model M1 for the first stage of the multi-stage batch process is derived from the data block X1 containing process data from stage 1, an initial condition block I1 containing data indicating valid initial conditions for stage 1 of the multi-stage batch process, and a batch end-product quality data block Y containing data indicating the batch end-product quality achieved during the iterations of the multi-stage batch process from which the data for creating the PLS model were obtained.Once the PLS model M1 has been created for Stage 1, the PLS model M1 is executed using the Stage 1 data block X1 and the Stage 1 output condition block I1 to generate a predicted batch-end product quality indicator Yp for Stage 1 and, optionally, a corresponding confidence interval CI, which is a well-known statistical measure of the accuracy of the predicted batch-end product quality indicator Yp. The confidence interval CI provides a way to measure the accuracy of the model's predictive capability, both in terms of how well the model predicts the outcome of the process and in terms of how well the actual batch data match the training data used in developing the model (i.e., whether disturbances or other inaccuracies cause the batch data to deviate from the data expected based on the model).The predicted batch-end product quality indicator Yp and, if desired, the confidence interval CI are included in the initial condition block I2, which is used when creating a PLS model M2 of the second stage of the multi-stage batch process.

[0086] The PLS model M2 is generated from stage 2 data block X2, stage 2 output condition block I2 (which contains the predicted batch-end product quality indicator Yp from stage 1 PLS model M1 and its associated confidence interval), and the predicted batch-end product quality data block Y. The generated PLS model M2 is then executed using stage 2 data block X2 and stage 2 output condition block I2 (which contains the Yp and CI generated by model M1) to produce a predicted batch-end product quality indicator Yp for stage 2 and calculate a corresponding confidence interval CI. The predicted batch-end product quality indicator Yp and, if desired, the confidence interval CI from stage 2 can then be included in the output condition block I3 for use in generating a third-stage PLS model M3 of the multi-stage batch process.This process is repeated for each stage of the multi-stage batch process, using the initial condition block Ii for the model of stage i and incorporating the predicted batch-end product quality indicator Yp and the confidence interval CI from the preceding stage. Likewise, the initial conditions Ii of each stage may also incorporate the initial conditions of each preceding stage.

[0087] Procedure (2) for the multi-block modeling of a multi-stage batch process, which is described in the right-hand section of Fig. As shown in Figure 12, process (1) is similar, except that model M1 is created using data blocks X1, X2, ... Xn for all n stages of the multi-stage batch process, instead of using only data block X1 containing the data from stage 1. However, process (2) proceeds similarly to process (1) by creating a PLC model from data blocks X1, X2, ... Xn, a batch-end product quality data block Y, and an initial condition block I1. The created PLC model M1 is then executed using data block X1 and initial condition block I1 as inputs to generate a predicted batch-end product quality indicator Yp and an associated confidence interval, which are then included in the initial condition block I2, based on data blocks X2, X3, ...Xn, but not data block X1, together with the batch-end product quality data block Y and the initial condition block I2, is used to create a PLC model M2 of the second process stage. As in step (1), this process is repeated in step (2) for each of the n stages of the multi-stage batch process to create the PLC model of the entire multi-stage batch process.

[0088] Fig. Figure 13 illustrates a PLS model that uses a forgetting factor to model a three-stage process. A so-called "forgetting factor" is a normalization factor used to reduce the impact of process data obtained in the early stages of a process on, for example, batch-end-product quality predictions made in later stages of the process. The rationale for using a forgetting factor is that while each stage of a process can affect the quality of the product ultimately produced by the entire process, the impact of each processing stage is greatest during and immediately after that stage and decreases as the process progresses to later stages.

[0089] In the case of the exemplary three-stage batch process from Fig. 13. A process control system (PCS) model can be developed using a data block X1 containing data generated at stage 1 of several process passes, data blocks X2 and X3 containing data generated at stages 2 and 3 of these passes, respectively, and a batch-end-product quality data block Y containing data related to the batch-end quality of the products manufactured during these passes. Instead of weighting data blocks X1, X2, and X3 equally for stages 1, 2, and 3, a forgetting factor is continuously applied to each data block to give more weight to process data from later stages of the process than to process data from earlier stages.

[0090] In the example shown, when creating a model of stage 1, a forgetting factor of 1 (i.e., no forgetting) is applied to data block X1. When creating a model of stage 2, a forgetting factor of 1 is applied to data block X2, while a forgetting factor of 1 / 2 is applied to data block X1. When creating a model of stage 3, a forgetting factor of 1 is applied to data block X3, while a forgetting factor of 2 / 3 is applied to data block X2, and a forgetting factor of 1 / 3 is applied to data block X1. In this way, the process data of each stage contributes more to the batch-end quality predictions associated with that process stage than the process data from earlier stages of a multi-stage process. A forgetting factor can also be continuously applied to all stages of a process model.

[0091] Fig. Figure 14 shows an example of an exemplary approach for developing a process control system (PCS) model of a multi-stage batch process from training data collected from multiple runs of the multi-stage batch process (or extracted from a data archive containing data from previous batch runs, as explained above). Once developed, such a PCS model of a multi-stage batch process can, for example, be used to perform one of the two exemplary process modeling operations described in Fig. 12 are shown.

[0092] As in Fig. As shown in Figure 14, a training dataset for stage 1 is acquired (or extracted from a data archive if necessary), comprising initial condition data I1, process variable data X1 (acquired at the first stage of the process), and final batch output quality Y1. PLS modeling calculations are then performed on this training dataset to generate a first-stage PLS batch model M1 of the multi-stage batch process. The stage 1 PLS model M1 is the pass, where initial condition data I1 and the process variable data X1 are applied as inputs to the PLS model M1 to generate a final batch quality prediction Yp1. The PLS model M1 may additionally provide an estimate of the reliability or accuracy of the prediction Yp1, such as a confidence interval CI1, or such an estimate of reliability or accuracy may be calculated separately, as is known to the average professional in the field.The batch-end quality prediction Yp1 and / or the confidence interval CI1 are then used as inputs when developing a PLS model M2 of stage 2 of the batch process, as in . Fig. Figure 14 shows that once the PLS model M2 (using the quality prediction Yp1 and / or the confidence interval CI1 developed by the Stage 1 PLS model as inputs, along with the Stage 2 data X2 and the batch output quality measurement Y) has been created, the PLS model M2 is then run with inputs that include the initial condition data I2 and the process variable data X2 for Stage 2 of the batch process, as well as the final batch quality prediction Yp1 and the corresponding confidence interval CI1 output by Stage 1 model M1 of the batch process (developed using the training data as inputs) to develop the final batch quality prediction Yp2 and the corresponding confidence interval CI2 for the second stage of the model.This process is repeated for stage 3 using the batch-end quality prediction Yp2 and the confidence interval CI2 from model M2, and so on for each subsequent stage of the multi-stage batch process, as shown. In this way, each stage model is created using the outputs of preceding stage models, including prediction and / or confidence outputs from the previous stage models, only as an example. Although... Fig. Figure 14 illustrates a model development using the procedure (1), which is shown on the right-hand side of Figure 14. Fig. As shown in Figure 12, the stage models M1, M2, M3, etc. can of course also be created using the method (2) on the left side of Figure 12. Fig. 12 are developed so that the first stage model M1 is developed using the training data X1, X2, etc. by all subsequent stages of the process, the second stage model M2 is developed using the training data X2, X3, X4, etc. by all subsequent stages of the process, and so on.

[0093] Once the models M1, M2, ... Mn have been developed for the n stages of the multi-stage batch process, as described above with reference to Fig. As described in section 14, the multi-stage model can be used to model the batch process in order to, for example, predict the output quality for batches in manufacturing and thus make decisions as to whether a run of the batch process should be continued or aborted, or to adjust inputs for subsequent stages in order to try to correct defects in the batch end product quality that were predicted by modeling earlier stages of the batch process. Fig. Figure 15 shows an exemplary use of the multi-stage batch process.

[0094] As in Fig.As shown in Figure 15, initial condition data I1 and process variable data X1, measured or obtained in an actual run of stage 1 of the batch process, are applied to the stage 1 PLC model M1. This data can be acquired online or retrieved from a data archive if needed. Running the stage 1 PLC model M1 on this data generates an output quality prediction Yp1 and a corresponding confidence interval CI1. The output quality prediction Yp1 and the confidence interval CI1 can then be used to estimate the final quality of the batch run based on the performance of this stage of the batch. Based on the actual performance of the first stage of the batch, a decision can be made as to whether the batch should be continued, stopped, or whether changes should be made to the batch process at subsequent stages to improve the batch quality.

[0095] Similarly, the outputs of the PLS model M1 can then be applied as inputs to the PLS model M2 of stage 2, along with the output condition data I2 and the process variable data X2 measured in stage 2 of the batch, once stage 2 of the batch is complete. Model M2 generates a final batch quality prediction Yp2 and a confidence interval CI2, which are then applied to the PLS model of the next subsequent stage of the batch process, and so on, until all stages of the multi-stage batch process have been modeled, with each model utilizing the final batch quality prediction and confidence interval generated by the PLS model of the preceding stage.

[0096] Naturally, the multi-stage model described here can be executed at any time during a batch run (e.g., while the batch is online), so that all preceding stages of the online batch run can be used to predict the final output quality of the batch run at the current stage or execution point of the batch. Likewise, the multi-stage models developed and executed as described here can be run after a batch run to determine potential changes for a future batch run and to achieve better batch output quality. Of course, the procedures described here for developing and executing the multi-stage modeling can be used for any desired purpose in numerous situations and apply to both batch and continuous processes.

[0097] As noted above, at least some of the exemplary methods and / or devices described above can be implemented by one or more software and / or firmware programs running on a computer processor. However, dedicated hardware implementations can also be constructed, including, without limitation, application-specific integrated circuits, programmable logic arrays, and other hardware devices to implement some or all of the exemplary methods and / or devices described above, in whole or in part. Furthermore, alternative software implementations can also be constructed, including, without limitation, distributed processing or component / object-distributed processing, parallel processing, or processing by a virtual machine to implement the exemplary methods and / or systems described.

[0098] It should also be noted that the exemplary software and / or firmware implementations described herein are stored on a physical storage medium, such as a magnetic medium (e.g., a magnetic disk or magnetic tape), a magneto-optical or optical medium such as an optical disc, or a solid-state medium such as a memory card or other enclosure containing one or more (non-volatile) read / write memories, read / write memories, or other rewritable (volatile) memories. Accordingly, the exemplary software and / or firmware described herein may be stored on a physical storage medium such as the one described above or a subsequent storage medium. Insofar as the foregoing description describes exemplary components and functions with reference to specific standards and protocols, it is understood that the scope of this patent is not limited to these standards and protocols.The respective standards for internet and other packet-switched network transmission (e.g., Transmission Control Protocol (TCP) / Internet Protocol (IP), User Datagram Protocol (UDP) / IP, HyperText Markup Language (HTML), HyperText Transfer Protocol (HTTP)) represent examples of the current state of the art. Standards of this kind are regularly replaced by faster or more efficient equivalents with the same general functionality. Accordingly, replacement standards and protocols with the same functions are equivalents provided for by this patent and fall within the scope of the accompanying claims.

[0099] Although this patent describes exemplary methods and devices, including software or firmware executed on hardware, it should be noted that these systems serve only for illustration and are not to be understood as limiting. For example, it is intended that the hardware and software components, in whole or in part, may be embodied exclusively in hardware, exclusively in software, exclusively in firmware, or in any combination thereof. Thus, although the foregoing description describes exemplary methods, systems, and / or machine-accessible media, the examples are not the only way to implement these methods, systems, and machine-accessible media. Therefore, although certain exemplary methods, systems, and machine-accessible media have been described, the scope of this patent is not limited to them.

Claims

[1] Methods for modeling a process, comprising: Dividing the process into several process stages, wherein the several process stages include at least a first process stage and a second process stage; and Developing several models, each corresponding to one of several process stages, wherein the several models include at least one model of the first process stage and one model of the second process stage; wherein the model corresponding to each process stage is developed using data from one or more passes of that process stage and output quality data relating to one or more passes of that process stage, and wherein the model corresponding to each process stage is adapted to generate an output quality prediction related to that process stage; and where the output quality prediction generated by the first-stage process model is used to develop the second-stage process model, characterized by that the model corresponding to the individual process stages is adapted to generate an output quality prediction that includes a prediction of either a stage-end product quality or a batch-end product quality. [2] The method of claim 1, wherein the output quality data relating to the pass(s) of this process stage comprise stage end product quality and batch end product quality. [3] Method according to claim 1, wherein the process is a batch process, and wherein the step of developing multiple models corresponding to the multiple process stages comprises acquiring data during each of the multiple runs of the process to produce multiple batches, wherein the step of acquiring data includes measuring values ​​for each of multiple process variables during each process stage of the individual runs of the process. [4] Method according to claim 1, wherein the process is a batch process, and wherein the values ​​measured for a process stage comprise a three-dimensional data array with multiple values ​​measured during the process stage for each of several variables for each of several time periods during the implementation of the process, and further comprising unfolding the three-dimensional data array into a two-dimensional data array comprising values ​​of the process variables in each of several time periods during the several process stages. [5] Method according to claim 1, wherein the process is a batch process, and wherein the information derived from the model of the first process stage and used by the second process stage comprises a quality prediction for a batch produced by the batch process. [6] Method according to claim 1, wherein the process is a continuous process, and wherein the step of developing multiple models corresponding to the multiple process stages comprises acquiring data during each of the multiple time periods during the implementation of the process, wherein the step of acquiring data includes measuring values ​​for each of multiple process variables during each time period. [7] Method according to claim 1, wherein the process is a continuous process, and wherein the values ​​measured for a process stage comprise a three-dimensional data array with multiple values ​​measured during the process stage for each of several variables for each of several time periods during the implementation of the process, and further comprising unfolding the three-dimensional data array into a two-dimensional data array comprising values ​​of the process variables in each of several time periods during the implementation of the continuous process. [8] Method according to claim 1, wherein the information derived from the model of the first process stage is used as input conditions for the model of the second process stage. [9] Method according to claim 1, wherein the step of developing multiple models includes creating a model with projection onto latent structures (PLS model) of a process stage. [10] Method according to claim 1, wherein a forgetting factor is applied to at least one section of the information derived from the model of the first process stage of a batch process and used by the second process stage of the batch process. [11] Methods for analyzing a batch produced by a process, comprising: Dividing the process into several process stages, wherein the several process stages include at least a first process stage and a second process stage; and Developing several models, each corresponding to one of several process stages, wherein the several models include at least one model of the first process stage and one model of the second process stage; where the second-stage process model uses information derived from the first-stage process model; and Using multiple models to predict the value of a process parameter. [12] Method according to claim 11, wherein the information derived from the model of the first process stage includes an output quality prediction for the first process stage. [13] Method according to claim 12, wherein the model of the second process stage further uses a measure of the reliability of the output quality prediction for the first process stage. [14] Method according to claim 13, wherein the reliability measure is a confidence interval related to the output quality prediction for the first process stage. [15] Method according to claim 11, wherein at least one of the model for the first process stage and the model for the second process stage is a PLS model. [16] Computer-based model of a process, wherein the process can be divided into several process stages, which have at least a first process stage and a second process stage, wherein the computer-based model comprises the following: multiple models, each corresponding to one of several process stages, wherein the multiple models include at least one model of the first process stage and one model of the second process stage; wherein the model corresponding to each process stage is developed using data from one or more passes of that process stage and output quality data relating to one or more passes of that process stage, and wherein the model corresponding to each process stage is adapted to generate an output quality prediction related to that process stage; and wherein the output quality prediction generated by the first process stage model is used to develop the second process stage model, wherein the model corresponding to each process stage is adapted to generate an output quality prediction that includes a prediction of either a stage end-product quality or a batch end-product quality. [17] Computer-based model according to claim 16, wherein the output quality data relating to the pass(s) of a process stage comprise stage end-product quality and batch end-product quality. [18] Computer-based model according to claim 16, wherein the process is a batch process, and wherein the step of developing multiple models corresponding to the multiple process stages comprises acquiring data during each of the multiple passes of the process to produce multiple batches, wherein the step of acquiring data includes measuring values ​​for each of several process variables during each process stage of the individual passes of the process. [19] Computer-based model according to claim 16, wherein the process is a batch process, and wherein the values ​​measured for a process stage comprise a three-dimensional data array with multiple values ​​measured during the process stage for each of several variables for each of several time periods during the implementation of the process, and further comprising unfolding the three-dimensional data array into a two-dimensional data array comprising values ​​of the process variables in each of several time periods during the several process stages. [20] Computer-based model according to claim 16, wherein the process is a batch process, and wherein the information derived from the model of the first process stage and used by the second process stage comprises a quality prediction for a batch produced by the batch process. [21] Computer-based model according to claim 16, wherein the process is a continuous process, and wherein the step of developing multiple models corresponding to the multiple process stages comprises acquiring data during each of the multiple time periods during the implementation of the process, wherein the step of acquiring data includes measuring values ​​for each of multiple process variables during each time period. [22] Computer-based model according to claim 16, wherein the process is a continuous process, and wherein the values ​​measured for a process stage comprise a three-dimensional data array with multiple values ​​measured during the process stage for each of several variables for each of several time periods during the implementation of the process, and further comprising unfolding the three-dimensional data array into a two-dimensional data array comprising values ​​of the process variables in each of several time periods during the implementation of the continuous process. [23] Computer-based model according to claim 16, wherein the information derived from the model of the first process stage is used as input conditions for the model of the second process stage. [24] Computer-based model according to claim 16, wherein the step of developing multiple models includes creating a model with projection onto latent structures (PLS model) of a process stage. [25] Computer-based model according to claim 16, wherein a forgetting factor is applied to at least one section of the information derived from the model of the first process stage of a batch process and used by the second process stage of the batch process. [26] Computer-based method for analyzing a batch produced by a process, comprising: Dividing the process into several process stages, wherein the several process stages include at least a first process stage and a second process stage; and Developing several models, each corresponding to one of several process stages, wherein the several models include at least one model of the first process stage and one model of the second process stage; where the second-stage process model uses information derived from the first-stage process model; and Using multiple models to predict the value of a process parameter. [27] Computer-based method according to claim 26, wherein the information derived from the model of the first process stage includes an output quality prediction for the first process stage. [28] Computer-based method according to claim 27, wherein the model of the second process stage further uses a measure of the reliability of the output quality prediction for the first process stage. [29] Computer-based method according to claim 26, wherein at least one of the model for the first process stage and the model for the second process stage is a PLS model. [30] Computer-readable medium on which machine-executable instructions for modeling a process are stored, wherein the process is divisible into several process stages, which have at least a first process stage and a second process stage, wherein the machine-executable instructions comprise: a first set of instructions for developing multiple models, each corresponding to one of several process stages, wherein the multiple models include at least one model of the first process stage and one model of the second process stage; wherein the first set of instructions develops the model corresponding to a given process stage using data from one or more passes of that process stage and output quality data relating to one or more passes of that process stage, and wherein the model corresponding to a given process stage is adapted to generate an output quality prediction related to that process stage; and wherein the first set of instructions uses the output quality prediction generated by the first process stage model to develop the second process stage model, wherein the first set of instructions includes further instructions that cause the model corresponding to each process stage to generate an output quality prediction that includes a prediction of stage-end product quality and batch-end product quality. [31] Computer-readable medium according to claim 30, wherein the output quality data relating to the pass(s) of a process stage comprise stage end-product quality and batch end-product quality. [32] Computer-readable medium according to claim 30, wherein the process is a batch process, and wherein the first set of instructions includes instructions for collecting data during each of several passes of the process in order to produce multiple batches, including measurements for each of the multiple process variables during each process stage of each pass of the process. [33] Computer-readable medium according to claim 30, wherein the process is a batch process, and wherein the values ​​measured for a process stage comprise a three-dimensional data array with multiple values ​​measured during the process stage for each of several variables for each of several time periods during the implementation of the process, and further comprising unfolding the three-dimensional data array into a two-dimensional data array comprising values ​​of the process variables in each of several time periods during the several process stages. [34] Computer-readable medium according to claim 30, wherein the process is a batch process, and wherein the information derived from the model of the first process stage and used by the second process stage comprises a quality prediction for a batch produced by the batch process. [35] Computer-readable medium according to claim 30, wherein the process is a continuous process, and wherein the first set of instructions comprises instructions for developing multiple models corresponding to the multiple process stages in order to acquire data during each of the multiple time periods during the implementation of the process, wherein the step of acquiring data includes measuring values ​​for each of multiple process variables during each time period. [36] Computer-readable medium according to claim 30, wherein the process is a continuous process, and wherein the values ​​measured for a process stage comprise a three-dimensional data array with multiple values ​​measured during the process stage for each of several variables for each of several time periods during the implementation of the process, and wherein the first set of instructions further comprises instructions for unfolding the three-dimensional data array into a two-dimensional data array comprising values ​​of the process variables in each of several time periods during the implementation of the continuous process. [37] Computer-readable medium according to claim 30, wherein the first set of instructions uses information derived from the model of the first process stage as input conditions for the model of the second process stage. [38] Computer-readable medium according to claim 30, wherein the first set of instructions for developing multiple models comprises instructions for creating a model with projection onto latent structures (PLS model) of a process stage. [39] Computer-readable medium according to claim 30, wherein the first set of instructions further comprises instructions for applying a forgetting factor to at least one section of the information derived from the model of the first process stage of a batch process and used by the second process stage of the batch process. [40] A computer-readable medium on which machine-executable instructions are stored for analyzing a batch produced by a process, wherein the process is divisible into several process stages, having at least a first process stage and a second process stage, wherein the machine-executable instructions comprise: a first set of instructions for developing multiple models, each corresponding to one of several process stages, wherein the multiple models include at least one model of the first process stage and one model of the second process stage; where the second-stage process model uses information derived from the first-stage process model; and a second set of instructions for using the multiple models to predict the value of a parameter of the process. [41] Computer-readable medium according to claim 40, wherein the information derived from the model of the first process stage includes an output quality prediction for the first process stage. [42] Computer-readable medium according to claim 41, wherein the model of the second process stage further uses a measure of the reliability of the output quality prediction for the first process stage. [43] Computer-readable medium according to claim 42, wherein the reliability measure is a confidence interval related to the output quality prediction for the first process stage. [44] Computer-readable medium according to claim 40, wherein at least one of the model for the first process stage and the model for the second process stage is a PLS model. [45] Method for modeling a process with a first process stage and a second process stage, comprising: Developing a first-stage process model using a training dataset corresponding to multiple iterations of the process; Applying at least one first section of the training dataset as input for the first-stage process model to generate an output quality prediction for the first-stage process; Developing a second-stage process model using at least one second section of the training dataset and the output quality prediction for the first-stage process, wherein the first-stage process model further generates an indication of the reliability of the output quality prediction for the first-stage process, and wherein the indication of the reliability of the output quality prediction for the first-stage process can be used to develop the second-stage process model. [46] The method of claim 45, further comprising developing a model for each process stage, applying at least one section of the training data set as input to the model of each process stage to generate an output quality prediction for the process stage, and applying at least one section of the training data set and the output quality prediction for the preceding process stage to the model of each process stage following the first process stage. [47] Method according to claim 45, wherein at least one of the models for the first and second process stages is a model with projection onto latent structures (PLS model). [48] ​​Method according to claim 45, wherein the process is one of a batch process and one of a continuous process. [49] Method for using a model of a multi-stage process with a first and a second process stage, wherein the model has a model for each of the first and second process stages, comprising: Applying an initial data set obtained from a first pass of the multi-stage process to the model of the first process stage to generate an output quality prediction for the first process stage; Applying a second data set, obtained from a second pass of the multi-stage process, and the output quality prediction for the first process stage to the model of the second process stage, further comprising applying a measure of the reliability of the output quality prediction for the first process stage to the model of the second process stage. [50] Method according to claim 49, wherein the reliability measure is a confidence interval related to the output quality prediction for the first process stage. [51] Method according to claim 49, wherein at least one of the model for the first process stage and the model for the second process stage is a PLS model. [52] Computer-based model of a process with a first process stage and a second process stage, wherein the computer-based model comprises: a model for a first process stage, developed using a training dataset corresponding to multiple iterations of the process, and adapted to generate an output quality prediction for the first process stage when at least a first section of the training dataset is applied as input to the first process stage model; and a second-stage process model developed using at least one second section of the training data set and the output quality prediction for the first-stage process, wherein the first-stage process model further generates an indication of the reliability of the output quality prediction for the first-stage process, and wherein the second-stage process model is developed using the indication of the reliability of the output quality prediction for the first-stage process. [53] Computer-based model according to claim 52, further comprising a model of each process stage, wherein the model of the respective process stages after the first process stage receives as input at least a section of the training data set and an output quality prediction of the preceding process stage. [54] Computer-based model according to claim 52, wherein at least one of the models for the first and second process stages is a model with projection onto latent structures (PLS model). [55] Computer-based model according to claim 52, wherein the process is a batch process and a continuous process. [56] Method for using a computer-based model of a multi-stage process with a first and a second process stage, wherein the model has a model for each of the first and second process stages, comprising: Applying an initial data set obtained from a first pass of the multi-stage process to the model of the first process stage to generate an output quality prediction for the first process stage; Applying a second data set, obtained from a second pass of the multi-stage process, and the output quality prediction for the first process stage to the model of the second process stage, including an indication of the reliability of the output quality prediction for the first process stage to the model of the second process stage. [57] Method according to claim 56, wherein at least one of the models for the first and second process stages is a PLS model. [58] Method according to claim 56, wherein at least one of the models for the first and second process stages is a principal component analysis (PCA) model. [59] Method according to claim 56, wherein the first data set corresponds to process variables obtained during the first process stage of a run of the multi-stage process, and wherein the second data set corresponds to process variables obtained during the second process stage of the run of the multi-stage process. [60] Method according to claim 56, wherein the display of the reliability of the output quality prediction for the first process stage is a confidence interval for the output quality prediction for the first process stage. [61] Computer-readable medium on which machine-executable instructions for modeling a process with a first process stage and a second process stage are stored, wherein the machine-executable instructions comprise: a first set of instructions for developing a first-stage process model using a training dataset corresponding to multiple iterations of the process; a second set of instructions for applying at least one first section of the training dataset as input to the first-stage process model in order to generate an output quality prediction for the first-stage process; and a third set of instructions for developing a second-stage process model using at least a second section of the training data set and the first-stage output quality prediction, wherein the first-stage process model further generates an indication of the reliability of the first-stage output quality prediction, and wherein the indication of the reliability of the first-stage output quality prediction can be used to develop the second-stage process model. [62] Computer-readable medium according to claim 61, further comprising developing a model for each process stage, applying at least one section of the training data set as input to the model of each process stage to generate an output quality prediction for the process stage, and applying at least one section of the training data set and the output quality prediction for the preceding process stage to the model of each process stage following the first process stage. [63] Computer-readable medium according to claim 62, wherein at least one of the models for the first and second process stages is a model with projection onto latent structures (PLS model). [64] Computer-readable medium according to claim 62, wherein the process is a batch process and a continuous process. [65] Computer-readable medium according to claim 62, further comprising: a fourth set of instructions for applying an initial data set, obtained from a first pass of the multi-stage process, to the model of the first process stage in order to generate an output quality prediction for the first process stage; a fifth set of instructions for applying a second data set, obtained from a second pass of the multi-stage process, and the output quality prediction for the first process stage to the model of the second process stage.

Images