Systems and methods for just-in-time characterization of raw materials - Patents.com
The system addresses the challenge of raw material quality variations by implementing real-time material characterization and predictive models to optimize industrial plant performance and output quality.
Patent Information
- Application Number
- JP2022511355
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2019-08-20
- Filing Date
- 2020-08-20
- Publication Date
- 2024-11-13
- Estimated Expiration
- 2040-08-20
AI Technical Summary
Conventional industrial control systems fail to account for variations in raw material quality, leading to unpredictable and non-linear deviations in plant performance, as they operate based on design materials without real-time quality measurements, affecting the efficiency and output quality of industrial processes.
A processor-implemented method and system for real-time material characterization using hardware processors to detect changes in raw materials, classify them, predict properties, and associate predictive models to optimize plant performance by generating recommendations for optimal operation.
Enables real-time optimization of industrial plant performance by accurately tracking raw material transitions and adjustments, improving efficiency and output quality through predictive models and actionable insights.
Smart Images

Figure 0007583028000002 
Figure 0007583028000003 
Figure 0007583028000004
Abstract
Description
[Technical field]
[0001] This application was filed on August 20, 2019. Special Priority is claimed from patent application No. 201921033532. This is a national phase application in Japan of International Application No. PCT / IN2020 / 050726, filed on August 20, 2020.
[0002] The disclosure herein relates generally to industrial process and equipment monitoring, and more particularly to data analysis for just-in-time (JIT) characterization of raw materials in any process industry. [Background technology]
[0003] In process industries where different types of raw materials are processed to produce one or more outputs, the quality of the raw materials being processed directly affects the quality of the output and the efficiency of the plant. When two or more raw materials are mixed together, the performance of the plant may deviate in an unpredictable and non-linear manner. For example, the performance of a coal-fired power plant shifts non-linearly when two different coals are mixed as fuel.
[0004] To obtain optimal output, plant settings need to be adjusted depending on the quality of the raw materials used. However, traditionally, control systems that monitor and control industrial plants work based on the design materials to determine the control parameters / logic, and variations in the materials' quality and properties are not fully taken into account. As a result, the quality of the plant's output and the overall performance of the plant deviate with variations in the quality of the raw materials used for processing.
[0005] In many cases, the quality of the raw materials being consumed is not measured in real time or is not available in a usable form. Even if quality measurements are possible, they can be very expensive to perform in real time. Without real-time information on the quality of the raw materials, it is very difficult to operate a plant optimally.
[0006] Furthermore, plants often consist of chains of equipment (serial or parallel architectures) where raw materials are consumed at different rates and at different times. The raw material supply lines for each piece of equipment may be different, resulting in uneven raw material usage across the equipment and altering the operation of downstream processing equipment. For example, in a thermal power plant, different pulverizers are fed through their respective coal feeders, which may be loaded with different levels and types of coal and therefore process different coals at different times. This may alter the performance of a boiler that is fed with different coals across its ports. Therefore, it is very important to track these raw material quality changes even at the individual equipment level.
[0007] Often the raw materials are in the form of a heterogeneous mixture of solid fine materials. The transition from raw material A to raw material B may take several hours to complete. The performance of the plant during this transition period of the raw materials has to be addressed in a different way, as the quality of the raw materials is constantly changing during this period. For example, different coals A and B are loaded one after the other into a coal bunker (e.g. in a thermal power plant). This change in coal is gradually reflected in the crusher, as the initial high concentration of coal A decreases over time and the initial high concentration of coal B gradually increases. However, this itself may happen in a very non-linear manner over a period of as much as 20 hours, as the material flow is not always streamlined. This directly affects the boiler performance. Therefore, this transition period of the raw materials needs to be tracked in real time and recommendations for the operation of the plant are provided accordingly.
[0008] Despite the advanced predictive model-based control systems currently available, identifying / building the right kind of predictive model is a challenge given the inherent variation in material quality and its impact on the plant output under various operating conditions. Existing systems do not adequately meet the need to manipulate plant operations by detecting, identifying, and utilizing raw material information for optimal plant operation. Summary of the Invention [Means for solving the problem]
[0009] The embodiments of the present disclosure provide technical improvements as solutions to one or more of the aforementioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a processor-implemented method for material characterization based on plant data is provided. In the method, first, via one or more hardware processors, plant data is received as input from an industrial plant. Further, by processing the plant data via one or more hardware processors, a change in one or more raw materials used in the industrial plant is determined, where the change in the one or more raw materials is detected at least at a plant level or an equipment level. Further, via one or more hardware processors, at least one class that matches each of the one or more raw materials is determined using at least one material class identification model, where the determined at least one class is a newly defined class or is from a set of predefined classes. Further, via one or more hardware processors, material properties are predicted for each of the raw materials. Then, via the one or more hardware processors, at least one of the plurality of predictive models associated with at least one of the predicted material properties of the one or more raw materials and the determined at least one class is selected, and via the one or more hardware processors, performance of the industrial plant is predicted using the selected at least one predictive model. Actual performance of the plant is also measured, and if the actual performance is below a performance threshold, at least one recommendation is generated via the one or more hardware processors to optimize performance of the industrial plant based on the predicted performance.
[0010] In another aspect, a system for material data characterization based on plant data is provided. The system includes one or more hardware processors, a communication interface, and a memory storing a plurality of instructions. When the plurality of instructions are executed, the one or more hardware processors receive the plant data as an input from an industrial plant. Further, by processing the plant data via the one or more hardware processors, a change in one or more raw materials used in the industrial plant is determined, where the change in the one or more raw materials is detected at least at a plant level or an equipment level. Further, at least one class that matches each of the one or more raw materials is determined via the one or more hardware processors using the at least one material class identification model, where the determined at least one class is a newly defined class or is from a set of predefined classes. Further, material properties are predicted for each of the raw materials via the one or more hardware processors. Thereafter, at least one of a plurality of predictive models associated with at least one of the predicted material properties and the determined at least one class of the one or more raw materials is selected via the one or more hardware processors, and the performance of the industrial plant is predicted using the selected at least one predictive model via the one or more hardware processors. An actual performance of the plant is also measured, and if the actual performance is below the performance threshold, then at least one recommendation is generated, via the one or more hardware processors, for optimizing performance of the industrial plant based on the predicted performance.
[0011] In yet another aspect, a non-transitory computer readable medium for material characterization based on plant data is provided. The non-transitory computer readable medium comprises a plurality of executable instructions stored in a memory. When the plurality of instructions are executed, one or more hardware processors execute the following method for material characterization based on plant data. In the method, first, plant data is received as an input from an industrial plant via the one or more hardware processors. Further, a change in one or more raw materials used in the industrial plant is determined by processing the plant data via the one or more hardware processors, where the change in the one or more raw materials is detected at least at a plant level or an equipment level. Further, at least one class that matches each of the one or more raw materials is determined via the one or more hardware processors using at least one material class identification model, where the determined at least one class is a newly defined class or is from a set of predefined classes. Further, material properties are predicted for each of the raw materials via the one or more hardware processors. Then, via the one or more hardware processors, at least one of the plurality of predictive models associated with at least one of the predicted material properties of the one or more raw materials and the determined at least one class is selected, and via the one or more hardware processors, performance of the industrial plant is predicted using the selected at least one predictive model. Actual performance of the plant is also measured, and if the actual performance is below a performance threshold, at least one recommendation is generated via the one or more hardware processors to optimize performance of the industrial plant based on the predicted performance.
[0012] The system stores information about a plurality of classes in a database, where each of the plurality of classes includes material properties, plant data, a predictive model associated with the class, a plurality of optimal settings associated with the class, clustering information, extracted classification rules, and one or more secondary information, where the term "secondary information" refers to any information, other than the examples given above, used at any stage of the raw material characterization process performed by the system.
[0013] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
[0014] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles. [Brief description of the drawings]
[0015] [Figure 1] 1 illustrates an exemplary system for material characterization, according to some embodiments of the present disclosure. [Figure 2A] FIG. 2 is a block diagram illustrating an implementation of the system of FIG. 1 in accordance with some embodiments of the present disclosure. [Figure 2B] FIG. 2B is a block diagram illustrating an example deployment of the system of FIG. 2A in an industrial plant configured as a digital twin, according to some embodiments of the present disclosure. [Diagram 3] 3A and 3B (collectively referred to as FIG. 3) are a flow diagram illustrating steps involved in a process for generating recommendations based on material characterization using the system of FIG. 1 according to some embodiments of the present disclosure. [Figure 4] 4A, 4B, and 4C (collectively referred to as FIG. 4) are a flow diagram illustrating steps involved in a process of material characterization using the system of FIG. 1 according to some embodiments of the present disclosure. [Diagram 5]FIG. 2 is a flow diagram illustrating steps involved in a process of automatic stepwise clustering performed by a material class extraction model using the system of FIG. 1, according to some embodiments of the present disclosure. [Figure 6] FIG. 2 illustrates an example implementation of a database shared among multiple digital twin systems for multiple plants in the system of FIG. 1 in accordance with some embodiments of the present disclosure. [Figure 7A] FIG. 7A is an example diagram illustrating the transfer of raw materials according to some embodiments of the present disclosure. [Figure 7B] FIG. 7B is an example diagram illustrating the transfer of ingredients, according to some embodiments of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0016] Exemplary embodiments will now be described with reference to the accompanying drawings. In the figures, the leftmost digit(s) of a reference number identifies the figure in which the reference number first appears. Whenever convenient, the same reference numbers will be used throughout the drawings to refer to the same or similar parts. Examples and features of the disclosed principles are described herein, however, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments. It is intended that the following detailed description be considered merely as exemplary, with the true scope being indicated by the following claims.
[0017] Referring now to the drawings, and more particularly to FIGS. 1-7B, where like reference characters indicate corresponding features consistently throughout the figures, preferred embodiments are shown and will be described in conjunction with the following exemplary systems and / or methods.
[0018] FIG. 1 illustrates an exemplary system for characterizing raw materials, according to some embodiments of the present disclosure. The memory module 101 may include any computer readable medium known in the art, including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and / or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memory, hard disk, optical disk, magnetic tape, and the like. In one embodiment, one or more modules (not shown) of the system 100 may be stored in the memory 101. The memory module 101 may be further configured to store a plurality of instructions, which, when executed, cause the one or more hardware processors 102 to perform different operations related to the free space identification being processed by the system 100. The memory module 101 may be further configured to store any data, such as input sensor data, any intermediate output (such as generated occupancy grid map, seed information, etc.), and output related to the material quality detection being processed by the system 100 (i.e., data related to the identified free space).
[0019] The one or more hardware processors 102 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, graphic controllers, logic circuits, and / or any device that manipulates signals based on operational instructions. Among other functions, the processor is configured to fetch and execute computer readable instructions stored in memory. In one embodiment, the system 100 may be implemented in a variety of computing systems, such as laptop computers, notebooks, handheld devices, workstations, mainframe computers, servers, network clouds, and the like. The one or more hardware processors 102 may be configured to execute data and control signal processing at different stages of free space detection, as needed.
[0020] The communication interface 103 may include various software and hardware interfaces, e.g., web interfaces, graphical user interfaces, etc., and may facilitate communications within a wide variety of network N / W and protocol types, including wired networks, e.g., LAN, cable, and wireless networks, e.g., WLAN, cellular, or satellite. In one embodiment, the communication interface 103 may include one or more ports for connecting several devices to each other or to another server. The communication interface may be configured to provide one or more channels with appropriate protocols for the system 100 to communicate with one or more external systems. For example, the communication interface 103 may interact to gather inputs required for free space detection from one or more sensors, which may or may not be part of the system 100. The communication interface 103 may further provide a suitable user interface for authorized users to interact with the system 100. The communication interface 103 may further be configured to provide one or more channels for different components of the system 100 to interact with each other to execute and / or control the transfer of data.
[0021] The system 100 can be implemented in a variety of ways, even though the basic functionality, i.e., characterization of raw materials and therefore the generation of recommendations for plant / process optimization, remains the same. One example of an implementation of the system of Figure 1 is shown in Figure 2. The various steps involved in the process of raw material characterization are shown in the flow diagrams in Figures 3 to 5.
[0022] FIG. 2A is a block diagram illustrating an implementation of the system of FIG. 1 according to some embodiments of the present disclosure. The system implementation as seen in FIG. 2A includes a data pre-processing module 201, a change detection module 202, a class identification module 203, a class extraction module 204, a simulation module 205, a predictive model update / creation module 206, a predictive model selection module 207, a plant advisory module 208, and a database that stores various information required for material characterization and data generated during and as a result of material characterization and recommendation generation by the system 100. To explain the operation of the system 100, a scenario is considered in which the system 100 is deployed in an industrial plant (alternatively referred to as a "plant") and the steps / different phases are explained below. The deployment of the system 100 in an industrial plant environment is shown in FIG. 2B. In this implementation example, the system 100 can function as a digital twin of the plant. The operation of the system 100 is explained below. The remaining blocks / components illustrated in Figure 2B (i.e., except for system 100) may be standard components of any industrial plant or may vary depending on the type of industrial plant in which system 100 is deployed. Also illustrated in Figures 3 and 4 are the process steps involved in characterizing the feedstock and generating recommendations. The operation of system 100 and each component of system 100 will now be described with reference to the steps illustrated in method 300 and vice versa.
[0023] The system 100 interacts with the plant using one or more suitable channels provided by the I / O interface 103 to collect / receive (302) plant data as input. The plant data collected as input can include one or more of a) data collected from industrial plant sensors, b) experimental data based on sample measurements of material properties or material batch information, and c) soft sensor data and synthetic data generated through multiple simulators, d) plant design data and maintenance data, and e) environmental data. The synthetic data is generated based on the identified operating state of the industrial plant and multiple material properties extracted from the determined at least one class. The generated synthetic data, experimental data, and real-time data collected from the industrial plant, plant maintenance data, and environmental data are used to train multiple predictive models of the industrial plant. The system 100 can also fetch information regarding the operation of the plant at a past time, which is stored in a database in the memory 101 as historical information / data. This plant data includes data obtained from plant sensors, raw material properties from laboratory information management systems, environmental conditions, plant / equipment maintenance information, plant / equipment design information. In one embodiment, in addition to the plant data, the system 100 may collect one or more user inputs and / or instructions required for material characterization, recommendation generation, and / or any other processes being processed by the system 100. The data pre-processing module 201 merges, cleans, filters, and processes the plant data in real time and stores it in a database for further use.
[0024] The change detection module 202 receives the pre-processed plant data and processes the received plant data to identify (304) whether a raw material has changed, where the raw material change may be identified by the system 100 by comparison to a list of previously used raw materials and / or based on information regarding the initial state of the raw materials used in the plant. The change may occur due to the addition and / or removal of raw materials as a result of chemical reactions within the plant, or due to the change of raw materials from one form to another.
[0025] Upon detecting that a change has occurred, the class identification module 203 determines (306) one or more classes of the raw material from existing classes available in the material database in the memory 101. Since each raw material has multiple characteristics, each material can be classified into one or more classes. The system 100 can determine the change in the raw material based on the change in the pattern observed in the collected plant data. The set of raw materials being processed by the industrial plant, the process being performed, and other factors at any given time form a pattern. Any change / variation from the pattern is considered by the system to determine the change in the raw material. The system 100 can also predict / quantify (308) the material properties. If the raw material under consideration does not belong to any of the multiple predefined classes, the system 100 triggers the class extraction module 204. The class extraction module 204 is configured to extract characteristics of the raw material under consideration and further use the extracted characteristics to define a new class or update one or more existing classes (i.e., predefined classes). Some example scenarios where one or more raw materials may be identified as not belonging to any of the predefined classes are: a) a plant using previously unseen / used materials, b) due to changes in plant behavior due to equipment modification or aging, c) changes in plant operating conditions. The class extraction module 204 uses different types of data, such as but not limited to real-time and historical operating data, material properties, maintenance data, design data, ambient condition data, and simulated / soft-sensed data, to automatically identify the different classes of materials used. The system 100 can use the simulation module 205 to generate synthetic operating data for the plant based on the operating regime and known properties of the raw materials. The system 100 can use the generated synthetic data for class identification and class extraction purposes at any later point in time. Once a sufficient amount of actual operating data for the new classes is accumulated, the weight of the simulated data can be reduced during use.The material change detection model, the material class identification model, and the material class extraction model can automatically retune themselves.
[0026] Further, the predictive model update / creation module 206, together with the predictive model selection module 207, is used to select (310) at least one predictive model that matches the class of the raw material. Some of the inputs to the predictive models are, but are not limited to, raw material characteristics, plant operating set points, environmental conditions, plant design and maintenance information. The predictive models can predict plant parameters including parameters related to process, efficiency, cost, energy, product quality, throughput, emissions, and safety. Furthermore, the predictive models can consist of models for detecting and diagnosing anomalies, predicting trends, and estimating remaining useful life of the process / equipment. During the selection of the predictive models, the system determines the accuracy of each of the predictive models with respect to the given inputs, and for any predictive models, if the accuracy is identified to be below the accuracy threshold, the predictive model is updated with new plant data collected in the database. The selected predictive model is then used to predict (312) the performance of the industrial plant. The system 100 also obtains as input (314) information regarding the actual performance of the industrial plant. The actual performance of the industrial plant is compared to the performance threshold and if the actual performance is below the performance threshold, the plant advisory module 208 is triggered. If the actual performance is identified as exceeding the performance threshold, the process is repeated from step 302. It should be noted that this trigger is not required by the system 100 to begin monitoring the industrial plant / process / equipment.
[0027] The plant advisory module 208 then generates recommendations for real-time optimization of plant performance or abnormal events based on the predicted performance of the industrial plant using at least one predictive model. The actionable insights / recommendations are returned to the communication interface 103, which then provides the recommendations to the user as an output of the system 100.
[0028] Information such as, but not limited to, all data required by the system 100 to perform the material classification, the results of the material classification, and the generated recommendations are stored in a database. Furthermore, the database allows the recording and reuse of various types of data and information such as, but not limited to, material properties and usage (historical and planned), operational data, processed data, simulated data, models (material change detection models, material identification models, material class extraction models, plant performance prediction models that may comprise first principles models, data-driven models, and knowledge-based models), algorithms, optimization and other decisions, expert knowledge, equipment and maintenance records, environmental conditions, and plant information, among others. The database can be configured to collect, store, and utilize data from multiple plants at a time. This is illustrated in FIG. 6. The various steps described in the description of FIG. 2A are illustrated in FIG. 3. The steps described in the method 300 are performed in the same order as shown, or in any alternative order that is technically feasible. Also, one or more steps in the method 300 may be omitted.
[0029] 4A, 4B, and 4C (collectively referred to as FIG. 4) are a flow diagram illustrating steps involved in a process of material characterization using the system of FIG. 1 according to some embodiments of the present disclosure.
[0030] The change detection performed by the system 100 involves checking the current state of the plant through various means such as identifying if the operating conditions have changed (e.g., load change or activity crusher change in a thermal power plant, change in the product grade produced in an alumina plant). This may also involve algorithms checking if the operating parameters have changed due to some anomaly or due to aging / failure of a particular piece of equipment. If the condition or equipment has changed, the user is notified and other digital twin services are triggered, which may help mitigate challenges arising from such changes.
[0031] However, if no such condition / equipment change is detected, the material change detection model continuously monitors a set of key performance indicators from plant sensors to identify if a material change has occurred. The material change detection model runs for each piece of plant equipment individually or jointly to determine if a material change has occurred at a particular piece of equipment and to identify that piece of equipment. The material change detection model may consist of data-driven models, physics-based models combined with knowledge-based models. An illustrative example covering the details of the operation of the material change detection model is given below.
[0032] The properties of the consumed materials often leave their traces in the operation of the equipment / process. For example, in a coal-fired boiler, coals that are more difficult to grind always leave a trace in the grinder in terms of power consumption or pressure parameters. Similarly, coals with more moisture require more air and higher temperature air so that the moisture can evaporate before entering the boiler. The material change detection algorithm monitors these important material indicator parameters to check if there is a change in the raw material. As shown in the example of Figure 7A, the change may be very slow or very fast. The transition from one material to another may only appear for certain parameters and not for others. For example, three materials are used in equipment A over a 24-hour period. Material 1 and Material 2 have similar hardness but different chemical composition. Material 3 has a very different hardness but a very similar composition to Material 1. The material change detection model is pre-built based on historical data of past plant performance parameters and related measured material properties. Alternatively, the material change detection model can be extracted automatically from the class extraction process. The material change detection model, as shown in the figure, tracks key material signature parameters and, after identifying a significant change in one or more of them, assigns the change to a corresponding change in material property that indicates a change in material class.
[0033] Another complicated part of the process is that each piece of equipment may undergo a material change at a different time. As shown in Figure 7B, equipment A and equipment B see a material change from I to II starting at 6:00, while equipment C continues to use the previous material I until 18:00. This could be due to multiple reasons, including shutting down a particular piece of equipment for maintenance or redundancy in the system. However, the material change detection model monitors each piece of equipment individually and identifies the equipment where a material change is occurring.
[0034] Furthermore, some changes may appear faster and some slower. For example, the transition from material I to II is quicker and the transition from material III to III is slower. The material change detection model identifies the material transition periods. This can be achieved by a combination of physics, knowledge rules, and advanced data-driven time series techniques. For example, advanced time series clustering techniques such as dynamic time warping can be applied to a selected set of measured parameters in the plant (previously identified based on knowledge) and a set of physics-based calculated parameters (not directly measured). An example of physics-based calculated parameters can be heat balance calculations in a pulverizer or boiler that serve as a moisture signature for the materials used. The material change detection model observes trends / patterns in multiple plant parameters over previously identified time periods, called windows, and then continues to compare these successive windows to each other in real time. When the difference between successive windows of such multiple parameters exceeds a threshold, a raw material change is detected. The threshold can be selected based on domain expertise or statistical parameters based on historical data of the plant.
[0035] Detection of material transition periods is important as it has a significant impact on plant performance and optimization strategies. If the material change detection model identifies that the material transition is not complete, it passes this information to a different set of transition models. However, if the material change detection model identifies that the material change is complete and the plant is stabilized, it passes that information to the selected predictive model. The transition models may be a separate set of material-identifying plant predictive models built specifically to handle transition periods between two materials in process.
[0036] While performing material change detection, in step 402, the system 100 processes the pre-processed data to detect if there is a change in operating conditions or an equipment change / deterioration. For example, if the system 100 is used for material characterization of a plant producing product grade A and is currently being used while processing product grade B, this may be identified by the system 100 as a change in operating conditions. Similarly, if one or more of the equipment is modified or if any of the equipment deteriorates beyond a set deterioration threshold, this is also considered a "change". If there is no change in conditions / equipment, in step 404, the system 100 identifies if the raw material has changed and if so, uses the material change detection model to identify the equipment whose raw material has changed (called a material change). If a material change is detected, in step 406, the system 100 checks the state of the material change. The state of the material change is either "material change completed" and "material change in transition". Some material changes may take time. If the material transformation is complete, the system 100 uses one or more material class identification models to identify the class of the raw material (i.e., the raw material present after the transformation) in step 410. The material class identification models may be pre-built based on historical data from the plant and may be stored in a database.
[0037] The system 100 can identify the class of raw materials using one or more suitable material class identification models.
[0038] Each material class identification model receives real-time data of the plant and predicts the class to which the material belongs. The class of the material can be pre-determined based on the material class extraction model. The material class identification model can be a combination of pre-built data-driven classification models, physics-based models, and knowledge-based models. For example, the material class identification model may include a machine learning based advanced classification model built using historical data of the plant. However, the material class identification model may utilize physics-based models / calculations to improve accuracy. As an illustrative example, a heat balance model is used to identify high or low moisture materials in a thermal power plant. Furthermore, the heat balance model may use plant specific knowledge of the control loop to build rules that can be incorporated into the heat balance model. The development of the material identification model is tightly coupled with the material class extraction model and the material quality prediction model.
[0039] If the material class identification model classifies an existing material into a set of predefined classes, it assigns the identified classes to the material and then triggers the material quality prediction model to quantify the material properties. The material quality prediction model receives real-time data (process data and measured properties) from the plant and associates this information with the identified classes to estimate the material properties.
[0040] Some examples of raw material properties that can be predicted / measured are: Chemical composition (concentrations of certain elements such as carbon, nitrogen, and volatile matter in coal in thermal power plants) Physical properties (concentration of magnetite and hematite in iron ore at steel plants) Physical properties (hardness of coal in thermal power plants, moisture, tumbler index for coke in coke plants) Material shape / size (size distribution of coke pellets in coke plants, fineness of pulverized coal in thermal power plants) - Form / condition (condition of coke, iron ore when loading into blast furnace in steel plant) - Specific chemical properties (kinetic parameters of coal combustion, coal ash melting temperature)
[0041] Some of the properties may not be directly measurable by the system 100. In such cases, the system 100 uses certain soft sensors that can combine and provide a measurement or indication of the properties that cannot be directly measured. For example, moisture in a raw material manifests itself in the form of heat absorption in the plant where the material is burned. Parameters that represent heat transfer in the plant or other soft sensing parameters that represent the heat balance in the plant may be used by classification and class extraction algorithms to separate materials with different moisture contents. Similarly, in a thermal power plant, the ash melting temperature of the coal mixture is not measured during the operation of the boiler. However, it is important because ash melting can cause permanent damage to the boiler if the furnace temperature exceeds the melting temperature. A material quality prediction model can use operational data and offline coal property data such as oxide composition to predict the melting temperature in real time to aid in the operation of the plant.
[0042] Additionally, advanced sensors may be placed in appropriate locations within the plant so that these measurements can be used for change detection and material class identification. For example, a coal-fired power plant may have real-time gas analyzers to measure the composition of the flue gases and furnace temperatures. Measuring In the furnace, an infrared temperature analyzer can be installed to measure the temperature of the coal. Measurement of various elements in the flue gases such as unburned carbon, ash composition etc. can be used to detect the coal class and predict its properties in real time. Measurement of the furnace temperature can provide insight into the chemistry of the ash, thereby allowing a better classification of coals for power plants.
[0043] Such material quality prediction models can use a combination of data-driven models using machine learning / deep learning regression, physics-based models, and expert knowledge models.
[0044] An example of a material quality prediction model using a basic model based on first principles is described below. Some of the properties of raw materials, such as kinetic parameters (which describe how these materials react in a chemical reaction), cannot be measured in real time. Other examples are the particle size distribution and physical composition of the material. For example, the fineness of iron ore cannot be measured in real time, or the mix ratio of two iron ores from two different mines with completely different properties cannot be measured online. However, these properties have a significant impact on the plant performance. These properties are quantified using the following methods:
[0045] As an illustrative example, a first principles-based predictive model is comprised of a physics-based simulation model that predicts the plant performance depending on the raw material properties and the plant operating conditions. The properties identified by the other material quality predictive models are fed into this first principles-based predictive model along with data related to the unknown properties. The first principles-based predictive model reads the operating conditions in real time and predicts the plant performance in real time. This plant performance is compared to the measured plant performance obtained in real time from the sensors. The unknown properties are adjusted so that the predicted plant performance matches the actually measured one with the desired accuracy. This is done via an inner optimization loop. These identified properties are stored in a database for future use. Thus, the system learns and adjusts the raw material properties, which are further required for the overall plant optimization.
[0046] If the material identification model is unable to classify the current material into the set of existing classes, the material class extraction model is triggered.
[0047] If in step 412 the raw material is identified as belonging to one of the pre-identified classes stored in the database, in step 414 the system 100 quantifies the raw material properties using one or more material quality prediction models. The material quality prediction models can be a combination of data-driven models, physics-based fundamental models, and knowledge models. The material quality prediction models may use real-time operational data and material property information from the plant. In another embodiment, the material quality prediction models may use stored simulated data from the database as well as historical plant data from various databases. Materials can be classified into a specific class or shared among multiple classes with appropriate similarity index assigned per class. Multiple materials can be classified into the same class. The material quality prediction models consist of either data-driven models, first principles-based models, knowledge-based models, or a combination thereof.
[0048] If in step 412 the raw material is identified as not belonging to any of the predefined classes, then in step 422 the system 100 uses the material class extraction model to create / define a new class of material with all its associated information or reclassify an existing class. Further data required for the newly defined class is generated in step 424 using the physics-based model and the plant performance data. Using the generated further data, one or more of the existing classes are updated or a new class is defined in step 426. Reclassification may be necessary due to different operating conditions or aging of the plant equipment.
[0049] Once the material classes and properties are identified, the system 100 further checks in step 416 whether one or more suitable predictive models exist for the identified class of raw material. If a matching predictive model exists, the system 100 selects the matching / associated predictive model in step 420. A predictive model captures the relationship between various plant parameters with respect to a particular plant output parameter of interest. For example, a predictive model may receive raw material properties, plant operating settings as inputs to predict product quality. A predictive model may consist of either a data-driven model, a first principles based model, a knowledge-based model, or a combination thereof. Different predictive models may be built for different raw material classes or a single (generic) model may be built to accommodate all raw material classes. In the case of different predictive models, once the raw material classes are identified, the appropriate predictive model is selected for further use. In the case of a single model, the appropriate plant parameters identified from the class extraction algorithm are fed as inputs to the predictive model so that the predictive model works for that particular class of raw material. In another embodiment, the predictive model may also include predictive maintenance type models (fault detection, RUL estimation, forecasting).
[0050] If multiple suitable / appropriate predictive models are identified, the system 100 determines the accuracy of each of the predictive models and selects the most accurate predictive model from among the predictive models and can be used for optimizing the plant. If the predictive models are sufficiently accurate, the predictive models can be further used for optimizing the plant. As part of the plant optimization, the system 100 can generate alerts and provide operational recommendations to authorized users. Once a suitable predictive model is obtained, an optimization model of the system 100 is used to identify optimal operational settings for the plant. The optimization model uses the selected predictive model along with various existing optimization algorithms from the database to identify optimal settings for the plant. The optimal settings or recommendations from the system are stored in the database for future use. Alternatively, the identified classes and characteristics can also be used as inputs to operational advisory and predictive maintenance models. They can help predict the performance of the equipment and also detect and diagnose abnormal operation of the process / equipment. If no predictive model exists for the particular type of raw material class, or the existing predictive models are not suitable for the identified class, then in step 418, the system 100 creates a new class-specific predictive model or one or more of the existing predictive models are updated with the latest data and identified characteristics of the raw material. The updated / created predictive model is stored in the database along with the appropriate tagging of the raw material class and associated tagged input parameters.
[0051] If no suitable predictive model exists and if a new class-specific predictive model cannot be generated (due to any reason, e.g., due to unavailability of required data), the system 100 may generate an alert to alert the user and details and results of the process are displayed to one or more users via a user interface. The user may verify one or more outcomes / results from the system via the user interface. User inputs are stored in a database for future use. The steps shown in method 400 may be performed in the same order as shown or in any alternate order that is technically feasible. Also, one or more steps in method 400 may be omitted.
[0052] FIG. 5 is a flow diagram illustrating the steps involved in the process of automatic stepwise clustering performed by the material class extraction model using the system of FIG. 1 according to some embodiments of the present disclosure.
[0053] First, a master list of material signature parameters (pre-selected based on plant and domain knowledge) is retrieved from the database in step 502. Material signature parameters are plant process variables that directly or indirectly indicate the type of material used or its properties. For example, in a coal-fired power plant, the amount of sulfur oxides in the exhaust gas indicates the level of sulfur in the raw coal, or the power consumed by the coal pulverizer indicates the hardness of the coal. Or some physics-based soft sensors, such as the heat balance across the pulverizer, may indicate the amount of moisture in the coal.
[0054] Further, in step 504, the system creates / prepares a number of possible permutations and combinations of material signature parameters. This can be either exhaustive or selective. Some parameters may be grouped together when creating the combinations. For example, in a coal-fired power plant, the heat balance of the entire mill and the parameters of the mill such as air temperature and air flow rate may be grouped together since they both significantly indicate the moisture of the material. All the sets of parameter combinations are created and stored in the memory. The stored operational and property data available in the database are separated for various operational states. For example, the operational data for when the plant is running at 100% load is separated from the operational data for 50% load. The data is then properly normalized. Further, in step 506, the system 100 uses machine learning based clustering on the set of processed plant data to obtain clustering results for all possible permutations and combinations of the material signature parameters prepared in step 504. This data consists of the data received from the sensors and the data stored in the database. If sufficient amount of such data is not available, material property data from the LIMS can also be used for clustering. Clustering is performed for all sets of parameter combinations varying hyperparameters such as number of clusters and type of clustering (density / distance based). The best set of clustering results is obtained by comparing the separation index for each of the combinatorial clustering results. The clustering result with the maximum separation index is identified as the best clustering at that stage in step 508.
[0055] Separation indices ~ Functions (statistical clustering indices such as completeness score, percentage of separation between plant data points associated with different material batches, degree of separation between plant data points associated with each of the material batches)
[0056] The separation index is a number that indicates how well the plant data is separated into various clusters, both from a machine learning / statistical perspective and from a ground truth perspective. Ground truth here refers to the information of the raw material batch or characteristics available to each operational plant data point user for clustering. First, for each combination of plant data parameters, the best clustering results are obtained by varying the number of clusters. This is obtained based on statistical tests such as the silhouette method and the elbow bend method, which are well known in the field of machine learning. Then, each of these clustering results is compared with each other based on how well the plant data belonging to each of the material batches were separated. As an illustrative example, if more than 90% of all the plant data associated with a material batch accumulates in one of the clusters, those plant data receive a high separation score. For example, the plant data associated with material batch 5 is divided into multiple clusters, so that material batches 1, 2, 3, and 4 receive a high separation score, while material batch 5 receives a low separation score (see Table 1 for an illustrative example). This can happen for multiple reasons, including incorrect association of the plant data with a particular material batch, or various characteristics of the materials in the batch itself. However, the algorithm accommodates this by separating the data. A separation index is then calculated for all plant data points across the clustering results. It can either be the average, sum, or some other function of the individual separation indexes. In step 508, the clustering result with the maximum separation index is selected as the best of all combinations.
[0057] The next step 510 involves identifying the cluster with the best homogeneity score / index within the identified clustering results. The homogeneity index / score indicates the similarity of plant data points and associated material properties within the cluster.
[0058] Homogeneity index~function (degree of uniformity within the plant data points associated with each cluster, degree of uniformity within the material properties associated with the plant data points within each cluster). [Table 1]
[0059] For example, cluster K3 in Table 1 has a high homogeneity score since the plant data points associated with it are homogeneous and the properties of material batch 1 and material batch 4 are similar. Such homogeneous clusters, both in terms of plant data and material properties, are separated as new material classes. On the other hand, clusters K2 and K3 have low homogeneity scores and materials 2, 3, and 5 have diverse properties and therefore need to be separated further. In step 514, steps 504 to 510 are repeated for all clusters with homogeneity indices below the homogeneity index threshold until all clusters are assigned to material classes. At this stage, the aforementioned clustering procedure is repeated with different parameters of the plant data associated with clusters K1, K2 (individually) and new material classes are obtained based on the separation scores and homogeneity scores as shown in the illustrative Table 1.
[0060] Furthermore, class names are automatically assigned to the pure clusters in step 510. In step 512, the system 100 generates a material discrimination model based on the parameters, the number of clusters, the clustering technique, and the specific clustering parameters used. The material discrimination model consists of a combination of data-driven and knowledge-driven classification. The knowledge here refers to the rules extracted from the clustering and separation exercises performed in the previous step. The rules encode the class extraction steps performed and the corresponding criteria used to separate the classes based on the separation index. The system 100 then stores data related to the identified pure clusters to which the material classes are assigned in a database. This data may include: i. Raw material classes formed ii. The specific material signatures and other operational data used at each stage of class extraction iii. Operating data associated with a range of properties for each class of raw materials, and simulated data associated with each of the classes; iv. The relationships between the characteristics of each of the classes (for each particular operating state and plant configuration) and the plant operating data. v. Numerical and statistical information regarding the operating data for each subclass and class of raw material. This consists of basic statistics such as measures of central tendency such as the median or the mean / standard deviation for each group of materials. vi. Generated material identification model
[0061] The system 100 checks whether all clusters have been assigned as material classes. If so, the class extraction is complete. If not, it extracts unclassified clusters and repeats the clustering process again with the unclassified plant data points until no further classification is possible, as in step 514.
[0062] In the absence of sufficient plant operating data, either simulated data based on raw material properties or predictive models (run in offline mode) can be used. For example, synthetic data can be generated using predictive models by providing previously sampled and measured raw material properties. Alternatively, models from a different but similar design plant can be used with appropriate modifications to the design and specific tuning parameters.
[0063] In some cases, materials may share multiple classes with a specific similarity index for each class. For closely related materials that may not belong to a particular class, a distance / density based proximity index can be provided to identify the materials. If there is a sufficiently large difference between an existing class and the new property, a new class can be created. The steps shown in method 500 may be performed in the same order as shown, or in any alternate order that is technically feasible. Also, one or more steps of method 500 may be omitted.
[0064] The written description explains the subject matter herein to enable one of ordinary skill in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims, and may include other modifications that occur to those of ordinary skill in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims, or if they include equivalent elements that do not differ substantially from the literal language of the claims.
[0065] The embodiments of the present disclosure herein address the unsolved problem of characterizing feedstock materials using plant data. Thus, the embodiments provide a mechanism for characterizing feedstock materials using plant data. Additionally, the embodiments of the present disclosure further provide a mechanism for generating optimization recommendations based on information about the material characterizations that have been performed.
[0066] It should be understood that the scope of protection extends to such programs and further to computer readable means having messages therein. Such computer readable storage means includes program code means for performing one or more steps of the method when the program is executed on a server or a mobile device or any suitable programmable device. The hardware device may be any type of device that can be programmed, including any type of computer such as a server or a personal computer, or any combination thereof. The device may also include means that can be hardware means, such as, for example, an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or a combination of hardware means and software means, such as, for example, an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components disposed therein. Thus, the means may include both hardware means and software means. The embodiments of the methods described herein may be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented in different hardware devices, for example using multiple CPUs.
[0067] Examples of the present disclosure may include hardware and software elements. Examples implemented in software include, but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented with other components or combinations of other components. For purposes of this description, a computer usable or computer readable medium may be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
[0068] The illustrated steps are set out to illustrate the illustrated exemplary embodiments, and ongoing technological developments should be expected to change the manner in which a particular function is performed. These examples are presented herein for purposes of explanation, not limitation. Moreover, the boundaries of functional components are arbitrarily defined herein for convenience of description. Alternative boundaries can be defined so long as the specified functions and relationships therebetween are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc. of those described herein) will be apparent to those skilled in the relevant art based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Additionally, the terms "comprise," "have," "include," "includes," and other similar forms are intended to be non-limiting in that they are equivalent in meaning and do not imply that an item or items following any one of these terms is an inclusive list of such item or items, or is limited to only the listed item or items. It should also be noted that as used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise.
[0069] Additionally, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory capable of storing information or data readable by a processor. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor to perform steps or stages consistent with the embodiments described herein. The term "computer-readable medium" should be understood to include tangible items and exclude carrier waves and transient signals, i.e., non-transient. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, non-volatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and other known physical storage media.
[0070] It is intended that the disclosures and examples be considered as exemplary only, with the true scope of the disclosed embodiments being indicated by the following claims.
Claims
1. 1. A method (300) for plant data based material characterization, comprising: One or more hardware processors receiving (302) the plant data as input from an industrial plant; determining (304) a change in one or more raw materials used in the industrial plant by processing the plant data with the one or more hardware processors, pre-processing the plant data, the plant data including at least one of: a) data collected from sensors in an industrial plant; b) experimental data based on sampled measurements of material properties or material batch information; c) soft sensor data and synthetic data generated via multiple simulators; d) plant design data and maintenance data; and e) environmental data; filtering the pre-processed plant data; determining the change in one or more of the raw materials based on a change in an observed pattern of the filtered plant data for at least one piece of equipment; checking the pre-processed plant data to detect whether there are changes in the operating conditions of the industrial plant, equipment changes, or equipment degradation; identifying the changes in one or more of the raw materials in the absence of changes in operating conditions of the industrial plant, equipment changes, or equipment degradation using a material change detection model, the material change detection model continuously monitoring a set of key performance indicators from the filtered plant data; identifying a material transformation state as completed or in transition by monitoring the one or more raw materials over a period of time; determining the change in one or more of the raw materials by determining (304) a change in one or more raw materials used in the industrial plant, where the change in the one or more raw materials is detected at least at a plant level or an equipment level by determining (306), using at least one material class identification model, by the one or more hardware processors at least one class that corresponds to each of the one or more raw materials when the material change state is complete, the at least one determined class being a newly defined class or being from a set of predefined classes; predicting (308) material properties for each of the one or more raw materials by the one or more hardware processors; selecting (310) at least one of a plurality of predictive models associated with at least one of the predicted material properties and the determined at least one class of the one or more raw materials; predicting (312) performance of the industrial plant using the at least one selected predictive model by the one or more hardware processors; If the obtained actual performance of the industrial plant is below a performance threshold, generating (318) at least one recommendation by the one or more hardware processors to optimize performance of the industrial plant based on the predicted performance; The method includes:
2. The step of determining the at least one class that matches the raw material comprises: identifying a class as a matching class if the plant data for the raw material matches plant data associated with a class of a plurality of existing classes; defining a new class if no match is found for the plurality of plant data of the raw material in any of the existing classes, and determining the new class as a class that matches the plurality of plant data associated with the raw material; The method of claim 1 , comprising:
3. The step of determining the newly defined class as the at least one class that matches the at least one raw material comprises: determining that the plant data associated with at least one of the raw materials does not match plant data associated with any of the predetermined classes; extracting a characteristic of the at least one raw material from the plant data; defining the new classes using the extracted characteristics of the at least one raw material and associated plant data; The method of claim 2 , comprising:
4. 2. The method of claim 1, wherein information about a plurality of said classes is stored in a database, each of said plurality of classes including material properties, plant data, a predictive model associated with said class, a plurality of optimal settings associated with said class, clustering information, extracted classification rules, and one or more secondary information.
5. A system (100) for plant data based material data characterization, comprising: one or more hardware processors (102); A communication interface (103); A memory (101) for storing a plurality of instructions; wherein, when the instructions are executed, the one or more hardware processors: receiving said plant data as input from an industrial plant; determining a change in one or more raw materials used in the industrial plant by processing the plant data; pre-processing the plant data, the plant data including at least one of: a) data collected from sensors in the industrial plant; b) experimental data based on sampled measurements of material properties or material batch information; c) soft sensor data and synthetic data generated via multiple simulators; d) plant design data and maintenance data; and e) environmental data; filtering the preprocessed plant data; determining the change in one or more of the raw materials based on a change in an observed pattern of the filtered plant data for at least one piece of equipment; checking the pre-processed plant data to detect whether there are changes in the operating conditions of the industrial plant, equipment changes, or equipment degradation; identifying the changes in one or more of the raw materials in the absence of changes in operating conditions of the industrial plant, equipment changes, or equipment degradation using a material change detection model, the material change detection model continuously monitoring a set of key performance indicators from the filtered plant data; Identifying a material transformation state as completed or in transition by monitoring the one or more raw materials over a period of time; determining the change in one or more of the raw materials by detecting the change in the one or more raw materials at least at a plant level or equipment level; using at least one material class identification model to determine at least one class that corresponds to each of the one or more raw materials when the material change state is complete, the determined at least one class being a newly defined class or from a set of predefined classes; predicting material properties for each of said one or more raw materials; selecting at least one of a plurality of predictive models associated with at least one of the predicted material properties and the determined at least one class of the one or more raw materials; predicting performance of the industrial plant using the selected at least one predictive model; generating at least one recommendation for optimizing performance of the industrial plant based on the predicted performance if the measured actual performance of the industrial plant is below a performance threshold; system.
6. The system comprises: identifying a class as a matching class if the plant data for the raw material matches plant data associated with a class of a plurality of existing classes; defining a new class if no match is found for the plurality of plant data of the raw material in any of the existing classes, and determining the new class as a class that matches the plurality of plant data associated with the raw material; The system of claim 5 , further comprising: determining the at least one class that matches the raw material by:
7. The system comprises: determining that the plant data associated with at least one of the raw materials does not match plant data associated with any of the predetermined classes; extracting characteristics of the at least one raw material from the plant data; defining the new classes using the extracted characteristics of the at least one raw material and associated plant data; The system of claim 6, wherein the newly defined class is determined as the at least one class that matches the at least one raw material by:
8. 6. The system of claim 5, wherein the system stores information about a plurality of the classes in a database, each of the plurality of classes including material properties, plant data, a predictive model associated with the class, a plurality of optimal settings associated with the class, clustering information, extracted classification rules, and one or more secondary information.
Citation Information
Patent Citations
Production control method for iron and steel industry
JP1993012303A
Method for predicting material of a1 alloy plate
JP2002224721A
Analyzer for analyzing relation between operation and quality in production process, analytical method, program, and computer-readable recording medium
JP2009151383A
Operation quality prediction device, method for predicting operation quality, computer program, and computer-readable recording medium
JP2013114362A
Quality management and intelligent manufacturing with labels and smart tags in event-based product manufacturing
US20060149407A1
