Industrial Process Model Generation System

Abstract

The present invention relates to an industrial process model generation system including an input unit and a processing unit. The input unit is configured to receive a plurality of input value trajectories, including operational input value trajectories and simulated input value trajectories, associated with the industrial process. The processing unit is configured to implement a simulator for the industrial process. The processing unit is configured to generate a plurality of industrial process behavioral data, the industrial process behavioral data being generated for at least some of the plurality of input value trajectories, and generating the industrial process behavioral data for at least some of the plurality of input value trajectories includes utilizing the simulator. The processing unit is configured to implement a machine learning algorithm for modeling the industrial process. The processing unit is configured to train the machine learning algorithm. The processing unit is configured to process first behavioral data of the plurality of behavioral data with the machine learning algorithm to determine a first modeled outcome. The processing unit is configured to decide whether to train or not train the machine learning algorithm using the first behavioral data, the decision including comparing the first modeled outcome to a performance condition. The processing unit is configured to process second behavioral data of the plurality of behavioral data with the machine learning algorithm to determine a second modeled outcome. The processing unit is configured to decide whether to train or not train the machine learning algorithm using the second behavioral data, or whether to further train or not further train the machine learning algorithm using the second behavioral data, where the decision includes comparing the second modeled results to the performance condition.

Description

[Technical field]

[0001] The present invention relates to an industrial process model generation system, an industrial process model selection and generation system, an industrial process model generation method, and an industrial process model selection and generation method. [Background technology]

[0002] Industrial applications of machine learning typically suffer from the availability of sufficient (labeled) training data for training machine learning models. High-fidelity simulators can be used to generate additional data to improve the quality of such machine learning models. However, performing high-fidelity simulations is time-consuming and computationally expensive. Blind generation of data is very expensive and has no guaranteed impact on the performance of machine learning algorithms.

[0003] There is a need to improve model development for industrial processes. Summary of the Invention

[0004] Therefore, it would be advantageous to have improved techniques for generating models for industrial processes.

[0005] The object of the present invention is solved by the subject matter of the independent claims, with further embodiments being incorporated in the dependent claims.

[0006] In a first aspect, an industrial process model generation system is provided, comprising: an input unit; - a processing unit. The input unit is configured to receive a plurality of input value trajectories including an operation input value trajectory and a simulation input value trajectory related to an industrial process. The processing unit is configured to implement a simulator of the industrial process. The processing unit is configured to generate a plurality of industrial process behavior data. The industrial process behavior data is generated for at least some of the plurality of input value trajectories, and the generation of the industrial process behavior data for at least some of the plurality of input value trajectories includes the use of the simulator. The processing unit is configured to implement a machine learning algorithm for modeling the industrial process. The processing unit is configured to train the machine learning algorithm. The processing unit is configured to process the first behavior data among the plurality of behavior data using the machine learning algorithm to determine a first modeled result. The processing unit is configured to determine whether to train or not train the machine learning algorithm using the first behavior data, and the determination includes a comparison between the first modeled result and a performance condition. The processing unit is configured to process the second behavior data among the plurality of behavior data using the machine learning algorithm to determine a second modeled result. The processing unit is configured to determine whether to train or not train the machine learning algorithm using the second behavior data, or whether to further train or not further train the machine learning algorithm using the second behavior data, and the determination includes a comparison between the second modeled result and a performance condition.

[0007] In one example, the input value trajectory is actual input data or simulated input data.

[0008] Thus, a set of industrial process behavior data is generated by the simulator based on input value trajectories that may be actual values or simulated, the machine learning algorithm is trained on a subset of this data, and the system determines which subset of the data to actually use in that training.

[0009] The machine learning algorithm can be trained on "actual" industrial process behavior data in addition to the industrial process behavior data generated by the simulator, and the multiple industrial process behavior data can include those generated by the simulator and "actual" data. However, all of the multiple industrial process behavior data may be those generated by the simulator.

[0010] In this way, when a new input trajectory that is actual input data or simulated input data is provided, the machine learning model of the industrial process can be trained, or the relevant behavior data generated from the simulator can be used without training the machine learning model. This continues step by step as the behavior data is generated, and the machine learning algorithm is continuously trained with appropriate data but not with inappropriate data.

[0011] The operating input value trajectory is any input that the plant operator performs on the production process. These are, for example, set values (target values for automation / control loops), parameters for actuators (e.g., the opening or closing ratio of valves), and digital inputs (pump on / off).

[0012] In addition to the inputs made by the operator (operating input value trajectory), there are also inputs for configuring and controlling the simulation. Examples are the initial plant state at the start of the simulation, raw material composition / quality, simulation of specific types of faults (e.g., valve faults (leakage, sticking, etc.), rotating equipment faults (pumps, compressors), etc.). These are called simulation input value trajectories.

[0013] Therefore, the input value trajectory is an input that controls (feed-forwards) the simulation process, and the simulator can generate industrial process behavior data including, for example, one or more of process data (e.g., simulated temperature, pressure, level, flow rate values), actuator data (simulated valve position, heat exchanger inflow, motor current, etc.), setpoint values (e.g., target values of a PID (Proportional Integral Derivative) controller).

[0014] Note that "a processor" and "the processor" do not mean that the system must use only one processor. For example, a processor can implement a simulator to generate data, and a second processor can implement a machine learning algorithm.

[0015] In one example, the plurality of behavior data includes one or more of process data, temperature data, pressure data, flow rate data, level data, voltage data, current data, power data, actuator data, valve data, sensor data, controller data.

[0016] Therefore, the input value trajectory is an input that controls (feed-forwards) the simulation process, and the simulator can generate industrial process behavior data including, for example, one or more of an initial simulation state (e.g., simulated temperature, pressure, level, flow rate values), actuator data (simulated valve position, heat exchanger inflow, motor current, etc.), setpoint values (e.g., target values of a PID (Proportional Integral Derivative) controller).

[0017] In one example, the modeled results include control or monitoring outputs of an industrial process.

[0018] Therefore, the machine learning algorithm is trained, for example, to generate a machine learning model that realizes task control or monitors the actual industrial process being simulated.

[0019] In one example, the processing unit is configured to select at least a portion of a plurality of input value trajectories.

[0020] Accordingly, the system can select or determine which input value trajectories should be provided to the simulator in order to generate new behavior data that can be used for training, which may optimize computational efficiency since the simulator may not need to be invoked.

[0021] In one example, the selection of at least some of the plurality of input value trajectories includes utilization of the determined sensitivity of a trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0022] In one example, the determination of whether to train or not train a machine learning algorithm that uses first behavior data includes determination of the sensitivity of a trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0023] In one example, the determination of whether to train or not train a machine learning algorithm that uses second behavior data includes determination of the sensitivity of a trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0024] In one example, the determination of whether to further train or not further train a machine learning algorithm that uses second behavior data includes determination of the sensitivity of a trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0025] In one example, the determination of the sensitivity of a trained machine learning algorithm to at least a portion of the plurality of behavior data includes analysis of the loss function of a trained machine learning algorithm with respect to at least a portion of the plurality of behavior data.

[0026] In one example, the processing unit is configured to determine to stop training a machine learning algorithm, the determination including determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0027] Accordingly, the machine learning algorithm (model) is analyzed to identify which motion input trajectories and simulation input value trajectories should be used to generate new behavior data for training the machine learning algorithm such that newly generated simulated data "significantly" changes the machine learning model when the machine learning algorithm is (additionally) trained on new data. And when the machine learning algorithm has not changed significantly, training can be stopped and the final trained machine learning algorithm can be provided to model the process.

[0028] In a second aspect, an industrial process model selection and generation system is provided, - an input unit, and - a processing unit. The input unit is configured to receive a plurality of input value trajectories including an operation input value trajectory and a simulation input value trajectory. The processing unit is configured to implement a simulator for an industrial process. The processing unit is configured to generate a plurality of industrial process behavior data. The industrial process behavior data is generated for at least some of the plurality of input value trajectories, and the generation of the industrial process behavior data for at least some of the plurality of input value trajectories includes the use of the simulator. The processing unit is configured to implement a plurality of machine learning algorithms for modeling the industrial process. The processing unit is configured to process first behavior data among the plurality of behavior data using a first machine learning algorithm among the plurality of machine learning algorithms to determine a first modeled result of the first machine learning algorithm. The processing unit is configured to determine to train the first machine learning algorithm using the first behavior data or to implement a second machine learning algorithm among the plurality of machine learning algorithms, and the determination includes a comparison between the first modeled result of the first machine learning algorithm and a performance condition.

[0029] In one example, the processing unit is configured to process second behavior data among the plurality of behavior data using the first machine learning algorithm to determine a second modeled result of the first machine learning algorithm. The processing unit is configured to determine to train the first machine learning algorithm using the second behavior data or to implement the second machine learning algorithm, and the determination includes a comparison between the second modeled result of the first machine learning algorithm and the performance condition.

[0030] In one example, the processing unit is configured to process the first behavior data using a second machine learning algorithm to determine a first modeled result of the second machine learning algorithm. The processing unit is configured to determine to train the second machine learning algorithm using the first behavior data or to implement a third machine learning algorithm among a plurality of machine learning algorithms, and the determination includes a comparison between the first modeled result of the second machine learning algorithm and performance conditions.

[0031] In this way, when a new input trajectory, which is actual input data or simulated input data, is provided, the machine learning model can be trained using the related behavior data generated from the simulator, or different machine learning models of the industrial process can be selected and trained. This continues step by step as the behavior data is generated, and the machine learning algorithm is continuously trained using appropriate data, or different machine learning algorithms are selected and trained using appropriate data.

[0032] In one example, the processing unit is configured to select at least some of a plurality of input value trajectories.

[0033] Therefore, the system can select or determine which input value trajectories should be provided to the simulator to generate new behavior data that can be used for training, which may optimize computational efficiency since the simulator may not need to be called.

[0034] In one example, the selection of at least some of the plurality of input value trajectories includes utilizing the determined sensitivity of a machine learning algorithm trained on at least a portion of the plurality of behavior data.

[0035] In one example, the decision to train an existing machine learning algorithm or implement a new machine learning algorithm using behavioral data includes determining the sensitivity of the existing trained machine learning algorithm to at least a portion of the plurality of behavioral data.

[0036] In one example, the processing unit is configured to determine to stop training an existing machine learning algorithm, and the determination includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavioral data. In one example, determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavioral data includes analyzing the loss function of the trained machine learning algorithm with respect to at least a portion of the plurality of behavioral data.

[0037] Thus, the machine learning algorithm (model) is analyzed to identify which behavioral input trajectories and simulation input value trajectories should be used to generate new behavioral data for training the machine learning algorithm such that newly generated simulated data "significantly" changes the machine learning model when the machine learning algorithm is (additionally) trained on new data. Also, a more advanced machine learning algorithm can be implemented when the machine learning algorithm is not being executed as needed, and training can be stopped when the machine learning algorithm has not changed significantly, providing the final trained machine learning algorithm to model the process.

[0038] In one example, the first machine learning algorithm is the simplest machine learning algorithm among the plurality of machine learning algorithms.

[0039] In one example, the second machine learning algorithm is the second simplest machine learning algorithm among the plurality of machine learning algorithms.

[0040] In one example, the third machine learning algorithm is the third simplest machine learning algorithm of the plurality of machine learning algorithms.

[0041] In a third aspect, there is provided a method for generating an industrial process model, comprising: receiving a plurality of input value trajectories, including operational input value trajectories and simulation input value trajectories associated with an industrial process; and implementing a simulator of the industrial process by the processing unit.

[0042] generating, by a processing unit, a plurality of industrial process behavior data, the industrial process behavior data being generated for at least some of the plurality of input value trajectories, the generating of the industrial process behavior data for at least some of the plurality of input value trajectories including utilizing a simulator; Implementing, by a processing unit, a machine learning algorithm that models the industrial process, the processing unit configured to train the machine learning algorithm; processing, by a processing unit, first behavioral data of the plurality of behavioral data with a machine learning algorithm to determine a first modeled outcome; determining, by a processing unit, whether to train or not train a machine learning algorithm using the first behavioral data, the determination including comparing the first modeled results to a performance condition; processing, by a processing unit, second behavioral data of the plurality of behavioral data with a machine learning algorithm to determine a second modeled outcome; and determining, by the processing unit, whether to train or not train the machine learning algorithm using the second behavioral data, or whether to further train or not train the machine learning algorithm using the second behavioral data, wherein the determination includes comparing the second modeled results to the performance condition.

[0043] In one example, the method includes selecting, by a processing unit, at least some of a plurality of input value trajectories.

[0044] In one example, selecting at least some of a plurality of input value trajectories includes using a determined sensitivity of a trained machine learning algorithm for at least a portion of a plurality of behavior data.

[0045] In one example, determining whether to train or not to train a machine learning algorithm using first behavior data includes determining a sensitivity of a trained machine learning algorithm to at least a portion of a plurality of behavior data.

[0046] In one example, determining whether to train or not to train a machine learning algorithm using second behavior data includes determining a sensitivity of a trained machine learning algorithm to at least a portion of a plurality of behavior data.

[0047] In one example, determining whether to further train or not to further train a machine learning algorithm using second behavior data includes determining a sensitivity of a trained machine learning algorithm to at least a portion of a plurality of behavior data.

[0048] In one example, determining a sensitivity of a trained machine learning algorithm to at least a portion of a plurality of behavior data includes analyzing a loss function of a machine learning algorithm trained with respect to at least a portion of a plurality of behavior data.

[0049] In one example, the method includes determining, by a processing unit, to stop training of a machine learning algorithm, and determining includes determining a sensitivity of a trained machine learning algorithm to at least a portion of a plurality of behavior data.

[0050] In a fourth aspect, an industrial process model selection and generation method is provided, receiving a plurality of input value trajectories including an operation input value trajectory and a simulation input value trajectory related to an industrial process; implementing a simulator for the industrial process by a processing unit; generating a plurality of industrial process behavior data by a processing unit, the industrial process behavior data being generated for at least some of the plurality of input value trajectories, and the generation of the industrial process behavior data for at least some of the plurality of input value trajectories including using the simulator; implementing a first machine learning algorithm among a plurality of machine learning algorithms for modeling the industrial process by a processing unit; processing first behavior data among the plurality of behavior data using the first machine learning algorithm by a processing unit so as to determine a first modeled result of the first machine learning algorithm; including determining by the processing unit to train the first machine learning algorithm using the first behavior data or to implement a second machine learning algorithm among the plurality of machine learning algorithms, the determination including a comparison between the first modeled result of the first machine learning algorithm and a performance condition.

[0051] In one example, the method includes processing second behavior data among the plurality of behavior data using the first machine learning algorithm by the processing unit so as to determine a second modeled result of the first machine learning algorithm, and determining by the processing unit to train the first machine learning algorithm using the second behavior data or to implement the second machine learning algorithm, the determination including a comparison between the second modeled result of the first machine learning algorithm and the performance condition.

[0052] In one example, the method includes using a processing unit to process first behavior data with a second machine learning algorithm to determine first modeled results of the second machine learning algorithm, and determining, by the processing unit, whether to train the second machine learning algorithm using the first behavior data or implement a third machine learning algorithm among a plurality of machine learning algorithms, the determination including a comparison of the first modeled results of the second machine learning algorithm with performance conditions.

[0053] In one example, the method includes selecting, by a processing unit, at least some of a plurality of input value trajectories.

[0054] In one example, selecting at least some of a plurality of input value trajectories includes using a determined sensitivity of a trained machine learning algorithm for at least a portion of a plurality of behavior data.

[0055] In one example, determining whether to train an existing machine learning algorithm using behavior data or implement a new machine learning algorithm includes determining a sensitivity of the existing trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0056] In one example, the processing unit is configured to determine to stop training an existing machine learning algorithm, the determination including determining a sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0057] In one example, determining a sensitivity of a trained machine learning algorithm to at least a portion of the plurality of behavior data includes analyzing a loss function of the trained machine learning algorithm with respect to at least a portion of the plurality of behavior data.

[0058] Therefore, the machine learning algorithm (model) is analyzed to identify which motion input trajectories and simulation input value trajectories should be used to generate new behavior data for training the machine learning algorithm so that when the machine learning algorithm is (additionally) trained on new data, the newly generated simulated data "significantly" changes the machine learning model. Also, when the machine learning algorithm is not being executed as needed, a more advanced machine learning algorithm can be implemented, and when that machine learning algorithm has not changed significantly, training can be stopped and the final trained machine learning algorithm can be provided to model the process.

[0059] The above aspects and examples will become apparent from and be clarified with reference to the embodiments described below.

Brief Description of the Drawings

[0060] Exemplary embodiments are described below with reference to the accompanying drawings. [Figure 1] FIG. 1 shows an overview workflow of a new process for model training and training data generation. [Diagram 2] FIG. 2 shows an overview workflow of a new process for searching for an appropriate model architecture, i.e., an appropriate machine learning algorithm. [Figure 3] FIG. 3 shows a detailed workflow of a standard process for model training and training data generation. [Figure 4] FIG. 4 shows a detailed workflow of a new process for model training and training data generation. [Figure 5] FIG. 5 shows a detailed workflow of a new process for searching for an appropriate model architecture.

Best Mode for Carrying Out the Invention

[0061] Figures 1 to 5 relate to an industrial process model generation system, an industrial process model selection and generation system, an industrial process model generation method, and an industrial process model selection and generation method.

[0062] Industrial process model generation system An example of an industrial process model generation system includes an input unit and a processing unit. The input unit is configured to receive a plurality of input value trajectories including operation input value trajectories and simulation input value trajectories related to an industrial process. The processing unit is configured to implement a simulator for the industrial process. The processing unit is configured to generate a plurality of industrial process behavior data. The industrial process behavior data is generated for a plurality of input value trajectories, and the generation of the industrial process behavior data for the plurality of input value trajectories includes the use of the simulator. The processing unit is configured to implement a machine learning algorithm for modeling the industrial process. The processing unit is configured to train the machine learning algorithm. The processing unit is configured to process first behavior data among the plurality of behavior data using the machine learning algorithm to determine a first modeled result. The processing unit is configured to determine whether to train or further train the machine learning algorithm using the first behavior data, and the determination includes a comparison between the first modeled result and performance conditions. The processing unit is configured to process second behavior data among the plurality of behavior data using the machine learning algorithm to determine a second modeled result. The processing unit is configured to determine whether to train or not train the machine learning algorithm using the second behavior data, or whether to further train or not train the machine learning algorithm using the second behavior data, and the determination includes a comparison between the second modeled result and performance conditions.

[0063] In other words, it can be determined whether the machine learning algorithm can be improved by further training it using the generated data, or whether the machine learning model does not need to be improved using the generated data.

[0064] In one example, the performance conditions include one or more of a target accuracy, a target false detection rate, and a target missed detection rate.

[0065] According to one example, the plurality of input value trajectories include one or more of process data, temperature data, pressure data, flow rate data, level data, voltage data, current data, power data, actuator data, valve data, sensor data, and controller data.

[0066] According to one example, the modeled result includes a control or monitoring output of an industrial process.

[0067] According to one example, the processing unit is configured to select at least some of the plurality of input value trajectories.

[0068] In one example, the selection of at least some of the plurality of input value trajectories includes using the determined sensitivity of a trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0069] According to one example, the determination of whether to train or not train the machine learning algorithm using the first behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0070] According to one example, the determination of whether to train or not train the machine learning algorithm using the second behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0071] According to one example, the determination of whether to further train or not further train a machine learning algorithm using the second behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0072] According to one example, the determination of the sensitivity of the trained machine learning algorithm to the plurality of behavior data includes analyzing the loss function of the trained machine learning algorithm for at least a portion of the plurality of behavior data.

[0073] According to one example, the processing unit is configured to determine to stop training the machine learning algorithm, and the determination includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0074] Industrial Process Model Selection and Generation System An example of an industrial process model selection and generation system includes an input unit and a processing unit. The input unit is configured to receive a plurality of input value trajectories including an operation input value trajectory and a simulation input value trajectory. The processing unit is configured to implement a simulator for an industrial process. The processing unit is configured to generate a plurality of industrial process behavior data. The industrial process behavior data is generated for a plurality of input value trajectories, and the generation of the industrial process behavior data for the plurality of input value trajectories includes the use of the simulator. The processing unit is configured to implement a plurality of machine learning algorithms for modeling an industrial process. The processing unit is configured to process first behavior data among the plurality of behavior data using a first machine learning algorithm among the plurality of machine learning algorithms to determine a first modeled result of the first machine learning algorithm. The processing unit is configured to determine to train the first machine learning algorithm using the first behavior data or to implement a second machine learning algorithm among the plurality of machine learning algorithms, and the determination includes a comparison between the first modeled result of the first machine learning algorithm and a performance condition.

[0075] In this way, the best machine learning algorithm used for process monitoring can be selected and trained.

[0076] In one example, the performance condition includes one or more of a target accuracy, a target false detection rate, and a target detection omission rate.

[0077] According to one example, the processing unit is configured to process second behavior data among the plurality of behavior data using the first machine learning algorithm to determine a second modeled result of the first machine learning algorithm. The processing unit is configured to determine to train the first machine learning algorithm using the second behavior data or to implement the second machine learning algorithm, and the determination includes a comparison between the second modeled result of the first machine learning algorithm and the performance condition.

[0078] According to one example, the processing unit is configured to process the first behavior data using a second machine learning algorithm to determine a first modeled result of the second machine learning algorithm. The processing unit is configured to determine to train the second machine learning algorithm using the first behavior data or to implement a third machine learning algorithm among a plurality of machine learning algorithms, and the determination includes a comparison between the first modeled result of the second machine learning algorithm and a performance condition.

[0079] According to one example, the processing unit is configured to select at least some of a plurality of input value trajectories.

[0080] In one example, the selection of at least some of the plurality of input value trajectories includes utilizing a determined sensitivity of a trained machine learning algorithm with respect to at least a portion of the plurality of behavior data.

[0081] According to one example, the determination to train an existing machine learning algorithm or implement a new machine learning algorithm using the behavior data includes determining the sensitivity of the existing trained machine learning algorithm with respect to at least a portion of the plurality of behavior data.

[0082] According to one example, the processing unit is configured to determine to stop training an existing machine learning algorithm, and the determination includes determining the sensitivity of the trained machine learning algorithm with respect to at least a portion of the plurality of behavior data.

[0083] According to one example, the determination of the sensitivity of the trained machine learning algorithm with respect to at least a portion of the plurality of behavior data includes an analysis of the loss function of the trained machine learning algorithm with respect to at least a portion of the plurality of behavior data.

[0084] According to one example, the first machine learning algorithm is the simplest machine learning algorithm among a plurality of machine learning algorithms.

[0085] According to one example, the second machine learning algorithm is the second simplest machine learning algorithm among a plurality of machine learning algorithms.

[0086] According to one example, the third machine learning algorithm is the third simplest machine learning algorithm among a plurality of machine learning algorithms.

[0087] Industrial process model generation method An example of an industrial process model generation method is receiving a plurality of input value trajectories including operation input value trajectories and simulation input value trajectories related to an industrial process; implementing a simulator for the industrial process by a processing unit; generating a plurality of industrial process behavior data by the processing unit, the industrial process behavior data being generated for a plurality of input value trajectories, and the generation of the industrial process behavior data for the plurality of input value trajectories including utilization of the simulator; implementing a machine learning algorithm for modeling the industrial process by the processing unit, the processing unit being configured to train the machine learning algorithm; processing, by the processing unit, first behavior data among the plurality of behavior data with the machine learning algorithm so as to determine a first modeled result; determining, by the processing unit, whether to train or not to train the machine learning algorithm using the first behavior data, the determination including a comparison between the first modeled result and a performance condition; processing, by the processing unit, second behavior data among the plurality of behavior data with the machine learning algorithm so as to determine a second modeled result; The processing unit determines whether to train or not to train a machine learning algorithm using the second behavior data, or whether to further train or not to further train the machine learning algorithm using the second behavior data, and the determination includes a comparison between the second modeled result and the performance condition.

[0088] In one example, determining whether to train or not to train a machine learning algorithm using the first behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0089] In one example, determining whether to train or not to train a machine learning algorithm using the second behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0090] In one example, determining whether to further train or not to further train a machine learning algorithm using the second behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0091] In one example, determining the sensitivity of the trained machine learning algorithm to the plurality of behavior data includes analyzing the loss function of the machine learning algorithm trained with respect to at least a portion of the plurality of behavior data.

[0092] In one example, the method includes the processing unit determining to stop training the machine learning algorithm, and the determination includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data.

[0093] In one example, the method includes the processing unit selecting at least some of the plurality of input value trajectories.

[0094] In one example, selecting at least some of a plurality of input value trajectories includes using the determined sensitivity of a trained machine learning algorithm for at least a portion of a plurality of behavior data.

[0095] Industrial Process Model Selection and Generation Method An example of an industrial process model selection and generation method is receiving a plurality of input value trajectories including operation input value trajectories and simulation input value trajectories related to an industrial process, implementing a simulator for the industrial process by a processing unit, generating a plurality of industrial process behavior data by a processing unit, the industrial process behavior data being generated for a plurality of input value trajectories, and the generation of the industrial process behavior data for the plurality of input value trajectories including the use of the simulator, implementing a first machine learning algorithm among a plurality of machine learning algorithms for modeling the industrial process by a processing unit, processing first behavior data among a plurality of behavior data using the first machine learning algorithm among the plurality of machine learning algorithms by a processing unit so as to determine a first modeled result of the first machine learning algorithm, including training the first machine learning algorithm using the first behavior data by a processing unit, or determining to implement a second machine learning algorithm among the plurality of machine learning algorithms, the determination including a comparison between the first modeled result of the first machine learning algorithm and a performance condition.

[0096] In one example, the method includes processing, by a processing unit, second behavior data of a plurality of behavior data with a first machine learning algorithm to determine second modeled results of the first machine learning algorithm, and determining, by the processing unit, whether to train the first machine learning algorithm or implement a second machine learning algorithm using the second behavior data, the determination including a comparison of the second modeled results of the first machine learning algorithm with performance conditions.

[0097] In one example, the method includes processing, by a processing unit, first behavior data with a second machine learning algorithm to determine first modeled results of the second machine learning algorithm, and determining, by the processing unit, whether to train the second machine learning algorithm or implement a third machine learning algorithm of a plurality of machine learning algorithms using the first behavior data, the determination including a comparison of the first modeled results of the second machine learning algorithm with performance conditions.

[0098] In one instance, the method includes selecting, by a processing unit, at least some of a plurality of input value trajectories.

[0099] In one example, selecting at least some of a plurality of input value trajectories includes using a determined sensitivity of a trained machine learning algorithm for at least a portion of a plurality of behavior data.

[0100] In one example, determining whether to train an existing machine learning algorithm or implement a new machine learning algorithm using behavior data includes determining a sensitivity of an existing trained machine learning algorithm for at least a portion of a plurality of behavior data.

[0101] In one example, the processing unit is configured to determine to stop training an existing machine learning algorithm, and the determination includes determining the sensitivity of the trained machine learning algorithm to at least a portion of a plurality of behavior data.

[0102] In one example, determining the sensitivity of the trained machine learning algorithm to at least a portion of a plurality of behavior data includes analyzing the loss function of the trained machine learning algorithm with respect to at least a portion of the plurality of behavior data.

[0103] The industrial process training data generation system, the industrial process monitoring system, the method for generating industrial process training data, and the industrial process monitoring method are described in more detail with respect to specific detailed embodiments, and FIGS. 1 to 2 are referred to again.

[0104] FIG. 1 shows an overview of an integrated workflow for model training and training data generation using a high-fidelity simulator.

[0105] First, there are a number of predefined operator input value trajectories and disturbance trajectories, for example, a set of step changes on the relevant setpoint of the process, or the activation or deactivation of an actuator failure such as a valve failure. For simplicity, the operator input value trajectories and disturbance trajectories can be collectively referred to as input value trajectories. Such step change experiments are well known from system identification. Using the input value trajectories and disturbance trajectories, labels can be created together with the simulation data. Data from the introduction of disturbances in the simulation is labeled as an anomaly or a fault. In the case of a system identification task, the data after a setpoint change (response) functions as label information for the machine learning process.

[0106] These input value trajectories are used to control the process in a number of simulation runs. One simulation run captures the system behavior and response for one input value trajectory. At the end of these simulation runs, an initial training dataset has been created.

[0107] In the next step, a machine learning algorithm (where the algorithm can mean two or more algorithms operating together) is trained by being presented with training data that changes the parameters of the underlying model based on a loss function (it is established that the machine learning algorithm can be based on deep learning networks, linear regression, random forests, SVM, etc.). At the end of this step, a first model has been trained. During this step, the training dataset can be split into a training dataset and a test (holdout) dataset. Further, the training dataset can be repeatedly split into training and validation datasets (cross-validation).

[0108] In the next step, the system extracts the sensitivity of the machine learning algorithm over the entire space of possible inputs to the algorithm. One way is to analyze the loss function of the model (a measure of the model prediction error) in both the training set sample or the validation set sample. However, it is known how this can be done in other ways.

[0109] Based on additional constraints (e.g., minimum and maximum values of setpoints, or system trip thresholds), the most beneficial new inputs are generated. When using existing samples within the training or validation dataset, new inputs can be generated and input value trajectories can be created in the vicinity of the original data points.

[0110] The new setpoint trajectory and disturbance trajectory (input value trajectory) are used as inputs for a new simulation run. These runs generate new data points. First, the existing model is tested at the new data points. If the performance is good enough (e.g., based on a target accuracy, or a target false detection / missed detection rate), the process ends and the created machine learning model can be used online.

[0111] If the model is not good enough, the new data points are added to the training dataset and the machine learning model is retrained to improve it.

[0112] In a variation of the above process, more than one machine learning model can be trained, or a group of machine learning algorithms or models can be trained. The most beneficial points are a mix of the most beneficial points across all models. A model that continues to produce significantly worse results than other models (e.g., based on a statistical test) can be excluded from the process.

[0113] In another variation of the process described above, the training dataset also includes historical data from the actual plant for initial training.

[0114] Figure 2 shows an external process to the process depicted in Figure 1. Here, the search process starts with a very simple machine learning model (e.g., logistic regression or an artificial neural network with a few layers). When the model performance saturates at an insufficient accuracy (or some other performance metric), or starts to decline while more beneficial data is presented, the search process changes to a more complex model, either by a predefined list of machine learning models, or by adding additional degrees of freedom to the model architecture, e.g., by adding layers. Also, a combination of the two methods is possible, first increasing the complexity of a given class of deep learning network, and then changing to a more complex network architecture.

[0115] As detailed above, a workflow has been developed that determines new simulation inputs using the performance of a machine learning model or individual samples within a training dataset or validation dataset, thereby resulting in simulation runs that are highly beneficial to the machine learning model. New data is generated by the simulation system. An already trained algorithm is tested against new, never-before-seen data. If the performance of the algorithm is sufficient, the process stops. If the performance is not sufficient, new data is added to the training data, the machine learning model is retrained, and the process continues.

[0116] To further illustrate an industrial process model generation system, an industrial process model selection and generation system, and related methods, reference is made to FIGS. 3 through 5.

[0117] FIG. 3 shows a detailed example of an existing simulation workflow. Simulations of industrial processes take two types of inputs: [1] inputs that an operator of the process would make, and [2] inputs that are only possible in the simulation and are actually outside the control of the operator, such as equipment failures, raw material quality, external temperature, etc.

[0118] To generate data for training a machine learning model, trajectories of these inputs need to be defined to control the simulation. The trajectories define, for example, "in 5 minutes, the operator opens the valve" and "in 20 minutes, the valve starts to leak 50% of the flow".

[0119] The output of the process simulator is all the values that are normally available in a process control system (setpoints, sensor measurements, actuator values such as valve positions).

[0120] This method is used, for example, to train a machine learning algorithm to (1) detect process anomalies, (2) detect device failures, (3) predict the future behavior of a process, and (4) select the best possible next controller output.

[0121] The output of the training can supply new previously unseen data and is, for example, a model that can perform one of tasks (1) through (4). The model is connected to an actual industrial process (which generates data points of the same type as the simulator version), performs the task, and generates the corresponding output.

[0122] In this flow, a human defines the input trajectory, examines the results of the ML algorithm (the performance of the model), and determines which input value trajectory to use in the next iteration.

[0123] Figure 4 shows a detailed example of the development made by the inventors. Essentially, the new development is to analyze the model generated by the machine learning algorithm to find out what new data brings the best improvement to the generated model.

[0124] If the model is good enough to be used in the actual process, no new data is generated and the model is used. In this flow, the system or method tests the model, analyzes the model, and determines which new input value trajectory is used in the next iteration.

[0125] Figure 5 shows a detailed example of further development made by the inventor. The extension is related to the introduction of another loop in addition to the repeated data generation and training.

[0126] After the model acceptance check, the improvement checker tests whether the model is still improving by adding new data to the dataset. This can be done by directly analyzing the sensitivity of the model to the inputs (if the sensitivity is low across all inputs, the model cannot be further improved), or simply by tracking whether the model has seen improvements beyond a threshold in the last n versions. Performance is measured by some performance metric such as the F1 score (for classification) or RMSE (root mean square error) (for regression).

[0127] If the model is no longer improving (and not improving based on the output of the sensitivity layer), the model manager maintains a list of algorithms that start with the simplest ML algorithms (e.g., linear regression (for regression) or logistic regression (for classification)) and move towards more complex algorithms (e.g., support vector machines and deep learning artificial neural networks). Each time the ML model stops improving, the next more complex model is selected, training is restarted, always starting with all the training data available at that point.

[0128] The complexity of an artificial neural network is determined by the complexity of its architecture, measured, for example, by the number of hidden layers and the number of nodes within the hidden layers.

[0129] In this flow, the system or method still tests the model, analyzes the model, and determines which new input value trajectories are used in the next iteration. Additionally, the system or method determines whether the currently used machine learning algorithm can still be improved and achieve the required performance, or whether a more complex algorithm should be used.

[0130] Note: The reason for not initially using the most complex algorithm is that it "overfits" the training data (basically creating a lookup table for the training data) and cannot generalize to data outside the training data set.

[0131] In FIGS. 3 to 5, it is shown that different computers, or processing units of such computers, can perform different functions. This can be a fact, but if necessary, a single processing unit can also perform all different functions.

[0132] Sensitivity analysis Regarding analyzing the sensitivity of a machine learning algorithm or model, the following are three examples related to this.

[0133] One example of analyzing the sensitivity of a machine learning model is to select n industrial process behavior data for which the machine learning model is most different from the actual values (this difference is also called the prediction error). To generate an input value trajectory to assist in improving the model, the n input value trajectories used to generate the n industrial process behaviors with the largest prediction errors are changed, for example, by randomly changing the initial state of the simulation or by randomly changing some of the input values within the input value trajectory.

[0134] In another example, another machine learning model algorithm is trained to predict the prediction error of the first machine algorithm using the initial state and other characteristics of the input value trajectory as predictor variables. This machine learning model can be, for example, a decision tree, and the decision tree returns a sequence of decisions on which input value trajectory values result in insufficient predictions for the current machine learning model. Using this information, it is possible to accurately generate new input value trajectories with these contents, for example, a specific initial state for simulation, a specific type of operator input (such as setpoint change, manual actuator change), and the parameters of these inputs (such as initial setpoint and new setpoint), or failure types.

[0135] In another example specific to artificial neural networks, new high-information data points are generated as follows: An optimization algorithm (such as a genetic algorithm, reinforcement learning algorithm, or Bayesian optimization algorithm) generates an input value trajectory for use in generating behavior data using a process simulation system. For each behavior data, the derivative (i.e., sensitivity) of the neural network prediction (model response) with respect to the model parameters can be evaluated without having to undergo the entire training and validation cycles. This means that this gradient information can be evaluated for a large number of candidate points in a short time. The resulting points are "high-information" points in the following sense: During subsequent training, a gradient descent algorithm takes large steps towards a better solution when the prediction error is high, i.e., when the model can be improved. During subsequent training, a gradient descent algorithm stays near its current solution when the prediction error is low, i.e., when the model generalizes well to the newly discovered input trajectory.

[0136] Therefore, an input trajectory with a maximized sensitivity of the model response is either one that causes the current model to be improved or one that validates the current model. In a special case, when the input (predictor variable) to the artificial neural network contains only information that is part of the input value trajectory and the output of the process simulation (process behavior data) is part of the output (predicted value), the steps of the process simulation can be avoided and the derivative (model response) of the neural network prediction with respect to the model parameters can be used to directly evaluate a large number of candidate input value trajectories. In this case, the time to evaluate the candidate input trajectories is even shorter.

[0137] Although the present invention has been illustrated and described in detail in the drawings and the foregoing description, such illustration and description should be considered as illustrative or exemplary and not restrictive. The present invention is not limited to the disclosed embodiments. Other variations of the disclosed embodiments can be understood and achieved by those skilled in the art in practicing the invention claimed in the claims, considering the drawings, the disclosure, and the appended claims. The inventions described in the original claims of this application are set forth below. [1] An industrial process model generation system, comprising: an input unit; a processing unit, the input unit is configured to receive a plurality of input value trajectories including operational input value trajectories and simulation input value trajectories associated with an industrial process; the processing unit is configured to implement a simulator of the industrial process; the processing unit is configured to generate a plurality of industrial process behavior data, the industrial process behavior data being generated for at least some of a plurality of input value trajectories, and generating the industrial process behavior data for at least some of the plurality of input value trajectories includes utilizing the simulator; the processing unit is configured to implement a machine learning algorithm to model the industrial process; the processing unit is configured to train the machine learning algorithm; the processing unit is configured to process first behavioral data of the plurality of behavioral data with the machine learning algorithm to determine a first modeled outcome; the processing unit is configured to determine whether to train or not train the machine learning algorithm using the first behavioral data, the determination including a comparison of the first modeled results to a performance condition; the processing unit is configured to process second behavioral data of the plurality of behavioral data with the machine learning algorithm to determine a second modeled outcome; The processing unit is configured to determine whether to train or not train the machine learning algorithm using the second behavioral data, or whether to further train or not train the machine learning algorithm using the second behavioral data, the determination including a comparison of the second modeled results to the performance conditions. [2] The system of [1], wherein the plurality of input value trajectories include one or more of process data, temperature data, pressure data, flow rate data, level data, voltage data, current data, power data, actuator data, valve data, sensor data, and controller data. [3] The system described in [1] or [2], wherein the decision to train or not train the machine learning algorithm using the first behavioral data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavioral data. [4] The system described in any one of [1] to [3], wherein the decision of whether to train or not train the machine learning algorithm using the second behavioral data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavioral data. [5] The system described in any one of [1] to [4], wherein the decision to further train or not further train the machine learning algorithm using the second behavioral data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavioral data. [6] The system described in [4] or [5], wherein determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavioral data includes analyzing a loss function of the trained machine learning algorithm with respect to at least a portion of the plurality of behavioral data. [7] The system described in any one of [1] to [6], wherein the processing unit is configured to decide to stop training the machine learning algorithm, the decision including determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavioral data. [8] The system of any one of [1] to [7], wherein the processing unit is configured to select at least some of the plurality of input value trajectories. [9] An industrial process model selection and generation system, comprising: an input unit; a processing unit, the input unit is configured to receive a plurality of input value trajectories, including operational input value trajectories and simulation input value trajectories; the processing unit is configured to implement a simulator of the industrial process; The processing unit is configured to generate a plurality of industrial process behavior data, the industrial process behavior data being generated for at least some of the plurality of input value trajectories, and the generation of the industrial process behavior data for at least some of the plurality of input value trajectories includes utilization of the simulator. The processing unit is configured to implement a plurality of machine learning algorithms for modeling the industrial process. The processing unit is configured to process first behavior data among the plurality of behavior data using a first machine learning algorithm among the plurality of machine learning algorithms to determine a first modeled result of the first machine learning algorithm. The processing unit is configured to determine to train the first machine learning algorithm using the first behavior data or to implement a second machine learning algorithm among the plurality of machine learning algorithms, the determination including a comparison between the first modeled result of the first machine learning algorithm and a performance condition, the system.

[10] The processing unit is configured to process second behavior data among the plurality of behavior data using the first machine learning algorithm to determine a second modeled result of the first machine learning algorithm, and the processing unit is configured to determine to train the first machine learning algorithm using the second behavior data or to implement the second machine learning algorithm, the determination including a comparison between the second modeled result of the first machine learning algorithm and the performance condition, the system according to [9].

[11] The processing unit is configured to process the first behavior data using the second machine learning algorithm to determine a first modeled result of the second machine learning algorithm, and the processing unit is configured to determine to train the second machine learning algorithm using the first behavior data or to implement a third machine learning algorithm among the plurality of machine learning algorithms, the determination including a comparison between the first modeled result of the second machine learning algorithm and the performance condition, the system according to [9].

[12] The determination of whether to train an existing machine learning algorithm or implement a new machine learning algorithm using the behavior data includes determining the sensitivity of the existing machine learning algorithm to at least a portion of the plurality of behavior data, the system according to any one of [9] to

[11] .

[13] The processing unit is configured to determine to stop training an existing machine learning algorithm, the determination including determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data, the system according to any one of [9] to

[12] .

[14] The determination of the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of behavior data includes analyzing the loss function of the trained machine learning algorithm regarding at least a portion of the plurality of behavior data, the system according to

[12] or

[13] .

[15] The processing unit is configured to select at least some of the plurality of input value trajectories, the system according to any one of [9] to

[14] .

[16] An industrial process model generation method, Receiving a plurality of input value trajectories including operation input value trajectories and simulation input value trajectories related to an industrial process; Implementing a simulator of the industrial process by a processing unit; Generating a plurality of industrial process behavior data by the processing unit, the industrial process behavior data being generated for at least some of the plurality of input value trajectories, and the generation of the industrial process behavior data for at least some of the plurality of input value trajectories including utilization of the simulator; Implementing a machine learning algorithm for modeling the industrial process by the processing unit, the processing unit being configured to train the machine learning algorithm; Processing, by the processing unit, first behavior data among the plurality of behavior data with the machine learning algorithm so as to determine a first modeled result; determining, by the processing unit, whether to train or not train the machine learning algorithm using the first behavioral data, the determining including comparing the first modeled results to a performance condition; processing, by the processing unit, second behavioral data of the plurality of behavioral data with the machine learning algorithm to determine a second modeled outcome; determining, by the processing unit, whether to train or not train the machine learning algorithm using the second behavioral data, or whether to further train or not train the machine learning algorithm using the second behavioral data; The method, wherein the determining includes comparing the second modeled results to the performance conditions.

[17] A method for industrial process model selection and generation, comprising: receiving a plurality of input value trajectories, including operational input value trajectories and simulation input value trajectories associated with an industrial process; implementing a simulator of said industrial process by a processing unit; generating, by the processing unit, a plurality of industrial process behavior data, the industrial process behavior data being generated for at least some of the plurality of input value trajectories, wherein generating the industrial process behavior data for at least some of the plurality of input value trajectories includes utilizing the simulator; implementing, by the processing unit, a first machine learning algorithm of a plurality of machine learning algorithms that model the industrial process; processing, by the processing unit, first behavioral data of the plurality of behavioral data with the first machine learning algorithm of the plurality of machine learning algorithms to determine a first modeled outcome of the first machine learning algorithm; and determining, by the processing unit, to use the first behavioral data to train the first machine learning algorithm or to implement a second machine learning algorithm of the plurality of machine learning algorithms, wherein the determining includes comparing a first modeled result of the first machine learning algorithm to a performance condition.

Claims

1. An industrial process model generation system, comprising: - an input unit; and - a processing unit, wherein the input unit is configured to receive a plurality of input value trajectories including actual input data and / or simulated input data related to an industrial process, and the input value trajectories are used to control the industrial process in a number of simulation runs so as to generate an initial training data set, wherein the processing unit is configured to implement a simulator of the industrial process, wherein the processing unit is configured to implement a machine learning algorithm for modeling the industrial process, wherein the processing unit is configured to train the machine learning algorithm including using the initial training data set, wherein the processing unit is configured to generate a plurality of industrial process behavior data by using the simulator, the industrial process behavior data being generated for at least some of the plurality of input value trajectories, the processing unit is configured to select at least some of the plurality of input value trajectories, and at least some of the selection of the plurality of input value trajectories includes using a determined sensitivity of the machine learning algorithm to at least a part of the plurality of industrial process behavior data, and the industrial process behavior data includes one or more of process data, actuator data, and set values, wherein the processing unit is configured to train the machine learning algorithm in the following step-by-step process, as part of the step-by-step process, the processing unit is configured to process first behavior data among the plurality of industrial process behavior data by using the machine learning algorithm to determine a first modeled result, as part of the step-by-step process, the processing unit is configured to determine whether to train or not to train the machine learning algorithm by using the first behavior data, and the determination includes a comparison between the first modeled result and performance conditions, as part of the step-by-step process, the processing unit is configured to process second behavior data among the plurality of industrial process behavior data by using the machine learning algorithm to determine a second modeled result, As part of the step - by - step process, after determining whether to train or not to train the machine learning algorithm using the first behavior data, the processing unit is configured to determine whether to train or not to train the machine learning algorithm using the second behavior data, the determination including a comparison of the second modeled result and the performance condition, As part of the step - by - step process, the processing unit is configured to continue training the machine learning algorithm with behavior data and related model results until the machine learning algorithm is no longer significantly changed by training. When the training is stopped, a final machine learning algorithm is provided, The final trained machine learning algorithm is configured to supply new data for performing one or more of: (1) detecting process anomalies, (2) detecting device failures, (3) predicting future behavior of the process, (4) selecting the best possible next controller output, a system.

2. The system according to claim 1, wherein the plurality of input value trajectories includes one or more of process data, temperature data, pressure data, flow rate data, level data, voltage data, current data, power data, actuator data, valve data, sensor data, controller data.

3. The determination of whether to train or not to train the machine learning algorithm using the first behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of industrial process behavior data, the system according to claim 1 or 2.

4. The determination of whether to train or not to train the machine learning algorithm using the second behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of industrial process behavior data, the system according to any one of claims 1 to 3.

5. The determination of whether to further train or not further train the machine learning algorithm using the second behavior data includes determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of industrial process behavior data, the system according to any one of claims 1 to 4.

6. The determination of the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of industrial process behavior data includes analyzing the loss function of the trained machine learning algorithm regarding at least a portion of the plurality of industrial process behavior data, the system according to claim 4 or 5.

7. The processing unit is configured to determine to stop training the machine learning algorithm, the determination including determining the sensitivity of the trained machine learning algorithm to at least a portion of the plurality of industrial process behavior data, the system according to any one of claims 1 to 6.

8. An industrial process model generation method, comprising: Receiving a plurality of input value trajectories including actual input data and / or simulated input data related to an industrial process, the input value trajectories being used to control the industrial process in a number of simulation runs so as to generate an initial training data set; Implementing, by a processing unit, a simulator of the industrial process; Implementing, by the processing unit, a machine learning algorithm for modeling the industrial process; Training, by the processing unit, the machine learning algorithm including utilization of the initial training data set; Generating, by the processing unit, a plurality of industrial process behavior data, the industrial process behavior data being generated for at least some of the plurality of input value trajectories, the processing unit selecting at least some of the plurality of input value trajectories, at least some of the selection of the plurality of input value trajectories including utilization of the determined sensitivity of the machine learning algorithm to at least a portion of the plurality of industrial process behavior data, the industrial process behavior data including one or more of process data, actuator data, and set values. The processing unit is configured to train the machine learning algorithm in the following step - by - step process: As part of the step - by - step process, the processing unit processes first behavior data among a plurality of industrial process behavior data with the machine learning algorithm so as to determine a first modeled result; As part of the step - by - step process, the processing unit determines whether to train or not to train the machine learning algorithm using the first behavior data, and the determination includes a comparison between the first modeled result and a performance condition; As part of the step - by - step process, the processing unit processes second behavior data among the plurality of industrial process behavior data with the machine learning algorithm so as to determine a second modeled result; As part of the step - by - step process, after determining whether to train or not to train the machine learning algorithm using the first behavior data, the processing unit determines whether to train or not to train the machine learning algorithm using the second behavior data; The determination includes a comparison between the second modeled result and the performance condition; As part of the step - by - step process, the processing unit is configured to continue training the machine learning algorithm with behavior data and related model results until the machine learning algorithm is no longer significantly changed by training. When the training is stopped, a final machine learning algorithm is provided; The final trained machine learning algorithm is configured to supply new data for performing one or more of (1) detecting process anomalies, (2) detecting device failures, (3) predicting future behavior of the process, and (4) selecting the best possible next controller output.

Images