Diagnostic device

Abstract

The present invention provides a diagnostic device that utilizes pre-processed measurement data, ML values, and interpretation values. By using all of these values / data, the phenomena, events, and behaviors of the process can be analyzed, so that many aspects can be taken into account.

Description

TECHNICAL FIELD

[0001] The present invention relates to diagnosis of a process and control thereof. The process is for example a water treatment plant, a paper machine, etc. BACKGROUND

[0002] Today, machine learning algorithms are used with systems that analyze and estimate the behavior of a process, such as a paper machine or water treatment. Processes are often multivariate, and thus they are difficult to follow or understand. Machine learning provides the ability to learn automatically for the system and provides the ability to improve from experience without explicit programming. Thus, a computer system uses machine learning (ML) utility algorithms and statistical models to perform a specific task or tasks without using explicit instructions. There are several ML algorithms. Here only some of them are mentioned: linear regression, logistic regression, K-means, feed-forward neural networks, etc.

[0003] The inference of the results of the ML algorithms is often difficult to explain, especially from a complex process. Therefore, explanation values are used to help the user to explain the results. So, the explanation values are used to explain how the ML algorithm arrived at a specific result, but also to categorize the way the process works. The explanation values are obtained by using for example SHAP (Shapley additive explanations) values, LIME method or DeepLIFT method.

[0004] Figure 1 An example of a known control arrangement is shown, where a process 1 is driven by an actuator 2, which is controlled by a controller 3. Measurement values 4 are taken from the process, which are used as feedback data for the controller. The controller compares the measurement values with one or more setpoint values 5 and forms a control command for the actuator.

[0005] The measurements 4 can also be used for other purposes, in which case it is convenient to pre-process 6 the measurement data before actual use. The pre-processing can include for example data merging, aligning time format, modifying metadata, data validation, etc. In Figure 1 In the example, machine learning (ML) 7 is used to extract information and patterns from large data sets. The matching learning algorithm is typically based on a statistical model, where the computer can perform a specific task without exact instructions, but instead relies on recognizing patterns. The recognized patterns can be obtained by building a mathematical model based on a training data set. Prediction (simulation) and pattern recognition can be made by inputting new data into the mathematical model.

[0006] Because it is difficult to see from the output (prediction / simulation) of the ML what has happened in the process, the Figure 1The SHAP values in the embodiments of the application are used to trace how the ML predictions link back to the input variables. For each prediction, a rating number is calculated for each input variable to indicate the contribution of that variable to the final prediction. These rating numbers can be seen as explanation values indicating the importance of the input values at a given point in time.

[0007] The explanation values are used to verify how the ML algorithm and the ML model work 9. This can be done more easily from the explanation values than from the ML predictions. Thus, if the ML model does not work properly, it can be changed.

[0008] The ML values are used in the prediction unit 10 for the behavior of the prediction process. The predictions can be used to provide recommendations 12 to the process 1. The predictions can also be used to suggest corrections 11 to change the set point 5 of the controller 3.

[0009] Although ML values are used, there is no device that it can also exploit other data in an automatic way. SUMMARY

[0010] The object of the application is to provide a diagnostic device that exploits pre-processed measurement data, ML values and explanation values. By using all these values / data, it is possible to analyze the phenomena, events and behavior of the process so that many aspects can be taken into account. This can be done automatically. The object is achieved in the manner described in the independent claim. The dependent claims specify different embodiments of the application.

[0011] The inventive diagnostic device for a multivariable process comprises a data processing module 6 for processing measurement data of the multivariable process and performing pre-processed measurement data 6A. The device further comprises a machine learning module 7 for performing machine learning values 7A from the pre-processed measurement data 6A. The diagnostic device further comprises an explanation value module 8 for forming explanation values 8A from the machine learning values 7A and a deviation calculation module 14. The deviation calculation module is arranged to calculate deviations 8D between the explanation values 8A and normal explanation values 8N, deviations 7D between the machine learning values 7A and normal machine learning values 7N, and deviations 6D between the pre-processed measurement data 6A and normal pre-processed measurement data 6N. The diagnostic device further comprises at least one estimator 15, each estimator being arranged to track a specific disturbance condition of the multivariable process with said deviations 6D, 7D, 8D and to form an estimate 33 of the severity of the disturbance condition. BRIEF DESCRIPTION OF DRAWINGS

[0012] In the following, the application is described in more detail by referring to the attached drawings, in which

[0013] Figure 1 Fig. illustrates an example of a prior art device,

[0014] Figure 2 Fig. illustrates an example of a diagnostic device according to the application,

[0015] Figure 3 Fig. illustrates an example of an estimator according to the present application,

[0016] Figure 4 Fig. illustrates another example of an estimator according to the present application,

[0017] Figure 5 Fig. illustrates an example of LE or fuzzy mapping, and

[0018] Figure 6 Fig. illustrates another example of LE or fuzzy mapping. DETAILED DESCRIPTION

[0019] Figure 2 Fig. illustrates an example of an inventive diagnostic device for a multivariate process 1. The process can comprise a plurality of processes, so as a whole it can be a combination of processes running together. It comprises a data processing module 6 for processing measurement data of the multivariate process and performing pre-processed measurement data 6A. The device further comprises a machine learning module 7 for performing machine learning values 7A from the pre-processed measurement data 6A. The diagnostic device further comprises an interpretation value module 8 for forming interpretation values 8A from the machine learning values 7A, and a deviation calculation module 14. The deviation calculation module is arranged for calculating a deviation 8D between the interpretation values 8A and normal interpretation values 8N, a deviation 7D between the machine learning values 7A and normal machine learning values 7N, and a deviation 6D between the pre-processed measurement data 6A and normal pre-processed measurement data 6N. The deviation calculation module can have several modules to make the calculations, e.g. a module for calculating the deviation 8D between the interpretation values 8A and normal interpretation values 8N. The deviation calculation module 14 can also be a distributed module with independent modules making the calculations.

[0020] The diagnostic device further comprises at least one estimator 15, each estimator being arranged to track a specific disturbance or a specific quality condition of the multivariable process with said bias 6D, 7D, 8D and to form an estimate 33 of the severity of the disturbance condition. For example, in papermaking, one estimator can be arranged to track retention of fine particles, while another estimator can track sizing performance. The output 15A of each estimator 15 can be used as is or together with the output of the other estimators to provide recommendations and / or guidance commands 16, such as commands to change the set points of the controller 3, recommendations to change the raw material's recommended treatment 1, recommendations to improve washing, recommendations to optimize retention, quality indicators indicating the health of the process or sub-process, etc. The recommendations can differ from process to process. The output 15A of each estimator can be used alone or together with the output of the other estimators for controlling, optimizing or troubleshooting the multivariable process. The control and / or optimization can comprise controlling one or more of the dosing amount of chemicals, the dosing point of chemicals, the dosing interval of chemicals, the selection of the type of chemicals used in the process, and the process conditions, such as pH, temperature, flow rate of the process flow.

[0021] The machine-learned explanation values and the machine-learned normal explanation values are, for example, SHAP values, values from the LIME method, values from the DeepLIFT method, or any other possible explanation values.

[0022] The LIME method explains individual model predictions, which are based on a local approximation of the model around a given prediction. LIME refers to the simplified input x as an interpretable input. The mapping x = hx(x) converts the binary vector of interpretable inputs to the original input space. Different types of hx mappings are used for different input spaces.

[0023] DeepLIFT is a recursive prediction explanation method. It assigns to each input xi the value CAXiAy, which represents the effect of setting that input to the reference value instead of its original value. This means that the DeepLIFT mapping x = hx(x) converts binary values to the original input, where 1 means that the input takes its original value and 0 means that the reference value is taken. The reference value represents a typical uninformative background value for the feature.

[0024] SHAP (SHapley Additive exPlanation) explanation values attribute to each feature the change in the expected model prediction when that feature is adjusted. These values explain how, if we did not know any of the features of the current output f(x), we would obtain the expected value E[f(z)] that would be predicted from the base value. The order in which the features are added to the expected value matters. However, this has already been taken into account in the SHAP values.

[0025] Figure 2 It is also shown (also with Figure 1Process 1, driven by actuator 2 (which is controlled by controller 3), is similar to the process 2. Measured values ​​4 are obtained from the process, and the controller compares the measured values ​​with one or more setpoint values ​​and generates control commands 3A for actuator 2.

[0026] As already described, measurement 4 can also be used for other purposes and can be preprocessed 6. For example, preprocessing may include data merging, aligning time formats, modifying metadata, data validation, etc. Figure 2 In this example, Machine Learning 7 is used to extract information and patterns from large datasets. Recognized patterns can be obtained by building mathematical models based on training datasets. Predictions (simulations) and pattern recognition can be performed by feeding new data into the mathematical model.

[0027] The explanatory value 8, like the SHAP value, is typically used to track how the 9ML value links back to the input variables. For each prediction, a rating number is calculated for each input variable to indicate that variable's contribution to the final prediction. These rating numbers are explanatory values ​​representing the importance of the input value at a given point in time.

[0028] It can be noted that the deviation / error between the normal explained value and the current ML prediction / estimated explained value is calculated, as well as the deviation between the normal ML value and the ML value, and the normal (preprocessed) measurement data and the preprocessed measurement data. The normal explained value can be a repository value found from the good running cycle of the processing. Therefore, the normal explained value 8N of machine learning, the normal machine learning value 7N, and the normal preprocessed measurement data 6N are values / data 13A that have been derived from the good running cycle of the processing. For example, the normal values ​​can be derived as simple values ​​or intermediate values ​​of these good periods. The normal operation of the processing occurs during the period when the processing or combined processing runs well. Therefore, for all data (preprocessed, ML prediction, and ML explained values), normal (optimal) values ​​can be given or estimated (from the stored values). Therefore, a library of normal historical values ​​can exist, where the processing has been determined to run optimally.

[0029] Therefore, during operations where single or combined processing is not optimally performed, discrepancies, deviations, or errors are detected from measured values, ML values, and interpreted values. This is detected as a deviation from the normal value. Differences from the normal values ​​6N, 7N, and 8N, such as 6D, 7D, and 8D (see...) Figure 3 This is used as input to estimator 15. Although the deviation calculation module 14 is shown as a separate module, it may also be part of estimator 15. Typically, deviation is related to error. The magnitude of the error indicates how much the setpoint value needs to be changed or how much the setpoint value should be changed.

[0030] Figure 3 An example of estimator 15 is shown, which uses bias / error 6D, 7D, and 8D. Figure 3The example of FIG. 6A shows three error values for three variables, but as shown, more variables and error values can be used if desired. Thus, at least one error / bias value 6D for the measurement data 6A, at least one error / bias value 7D for the ML value 7A, and at least one error / bias value 8D for the interpretation value 8A can be used in the estimator of the present application.

[0031] The estimator 15 includes at least one P module 17, 17A, 17C and I module 18, 18A, 18C or D module 19, 19A, or any combination of these modules. As previously mentioned, the bias is the data input into the module. The estimator also includes input mapping modules 20, 21, 22, 20A, 21A, 22A, 20C, 21C for each output 23, 24, 25, 23A, 24A, 25A, 23C, 24C of the module. In addition, the estimator includes a summation module 26 for summing the outputs 27, 28, 29, 27A, 28A, 29A, 27C, 28C of the input mapping modules 20, 21, 22, 20A, 21A, 22A, 20C, 21C and an output scaling module 30 for scaling the output 31 of the summation module. In addition, the estimator includes an output mapping module 32 to provide a normalized output 33. The normalized output is the estimate value, as mentioned above, which is used for recommendations, etc.

[0032] The P, I, and D modules 17, 17A, 17C, 18, 18A, 18C, 19, 19A and combinations thereof PI, PD, ID, and PID are known, but the bias / error of the interpretation value or ML value has not been used as an input before. The P module 17, 17A, 17C has a weighting coefficient which is multiplied with the input error value. The I module includes an integrator unit 118, 118A, 118C which integrates the input error value for a particular period. The integrated input error value is multiplied with a second weighting coefficient 180, 180A, 180C. The D-module includes a differentiator unit 119, 119A which forms the derivative of the error value during a particular period. The derivative is multiplied with a third weighting coefficient 190, 190A. As can be seen, all P, I, and D modules and combinations thereof have a weighting coefficient unit. These units can have the same weighting coefficient or different weighting coefficients. The weighting coefficients can weight the importance of the proportional (P), integral (I), and derivative (D) parts of the error value and also adjust or fine tune the estimate by increasing or decreasing the contribution of each single input calculation.

[0033] It is not always necessary to have all P, I, and D modules, but as mentioned before, if they are really used, they can be in the estimator. Figure 3 In the embodiment of FIG. 6A, the P, I, and D modules together provide a PID calculation for the interpretation error value 8D and the ML error value 7D and a PI calculation for the measurement data error 6D.

[0034] Thus, the estimator according to the application comprises at least one module arranged to handle the deviation 8D between the interpreted value 8A and the normal interpreted value 8N, at least one module arranged to handle the deviation 7D between the machine learned value 7A and the normal machine learned value 7N, and at least one module arranged to handle the deviation 6D between the pre-processed measurement data 6A and the normal pre-processed measurement data 6N. The number of inputs (deviations) used by the estimator can also be different. For example, the estimator can use only one deviation of the measurement data, four deviations of the four different ML values, and two deviations of the two different interpreted values.

[0035] Figure 4 Another possible example is shown, where the D module is not needed, so the estimator of this example has a PI calculation. As mentioned above, the estimator can have only those modules needed for the P, I, D, PI, PD, ID or PID calculation of the implementation of the set point controller. It is also worth mentioning that the estimator can make different calculations for different error values. For example, Figure 3 The implementation of the error value 8D can be modified to another scheme, where a PID calculation is made for the error value 8D and a P calculation is made for the other error value 7D (i.e. the I module 18A and the D module 19A have been deleted).

[0036] As mentioned above, the set point estimator also comprises input mapping modules 20, 21, 22, 20A, 21A, 22A, 20C, 21C for each output 23, 24, 25, 23A, 24A, 25A, 23C, 24C of the P, I and D modules. See Figure 3 The input mapping transforms the result of each output of the P, I or D module to a value between -2 and 2. This can be seen as a normalization of the value. The input mapping is formed by a linguistic equation (LE) or fuzzy logic. By using the input mapping it is easy to take non-linearities into account. The tuning of the estimator is also relatively smooth, because the properties of the handling are taken into account in the input mapping. The mapping modules of the estimator can use any mapping curve separately. For example, in Figure 3 the modules 20 and 20A can be formed by LE or one module 20 has been formed by LE and the other module 20A by fuzzy logic.

[0037] Figure 5 An example of a mapping curve 50 is shown, which is formed by a linguistic equation or fuzzy logic. X is the input variable, which is transformed to the output variable Y. The maximum and minimum values of X and Y are determined. A linear formula (such as Y = ax + b) determines that the Y value between X occurs between the maximum and minimum values. If X is greater than the maximum X value, Y is the maximum Y. If X is smaller than the minimum X value, Y is the minimum Y.

[0038] A mapping curve can also be a curve other than a linear curve. It can be another curve that matches the processed features better. Figure 6 Two other possible instances of the mapping curve are shown. The solid line describes the piecewise linear mapping curve 60, and the dashed line describes the S-curve mapping 61. Other curves are also possible. Therefore, refer to... Figure 3 The mapping module can use any mapping curve independently. For example, modules 20 and 20A can have the same mapping curve, such as a linear curve, or different curves, such as different linear curves, or piecewise linear curves and S-curves.

[0039] The outputs 27, 28, 29, 27A, 28A, 29A, 27C, and 28C of the input mapping modules 20, 21, 22, 20A, 21A, 22A, 20C, and 21C are summed in the summing module 26. Therefore, all deviation / error values ​​are taken into account. Then, the output scaling module 30 scales the sum output 31, and the output mapping module 32 normalizes the scaled sum to provide a normalized output 33, which is the estimator output.

[0040] Furthermore, the output of one estimator can be used as input to another estimator along with any combination of measurements, ML predictions, and performance values ​​(such as SHAP), which provides cascading connections between estimators.

[0041] The method of the present invention for generating an estimate of the severity of interfering or quality conditions in a multivariate treatment utilizes the diagnostic apparatus described herein for generating an estimate of the severity of interfering or quality conditions. The method uses the estimate of the severity of the interfering or quality conditions to provide recommendations and / or instructions for controlling and / or optimizing the multivariate treatment. Control and / or optimization may include control over one or more of the following: the amount of chemical added, the point of chemical addition, the interval of chemical addition, the selection of the type of chemical used in the treatment, treatment conditions such as pH, temperature, the flow rate of the treatment stream, and treatment stream delays, such as pulp, pulping, or water flow delays in treatment equipment such as towers, tanks, pulpers, pools, or other treatment facilities.

[0042] The method of the present invention can control industrial processes, such as multivariate processes, including pulp processing, papermaking, paperboard manufacturing or tissue paper manufacturing processes, industrial water or wastewater treatment processes, raw water treatment processes, water reuse processes, municipal water or wastewater treatment processes, sludge treatment processes, mining processes, oil recovery processes, or any other industrial processes.

[0043] As mentioned above, the application provides an automatic way to provide estimators for an analytical process 1. The process can be, for example, a water treatment process or a paper making process. The process can be an industrial process, such as: pulp treatment, paper making, board or tissue making process, industrial water or waste water treatment process, raw water treatment process, water re-use process, municipal water or waste water treatment process, sludge treatment process, mining process, oil recovery process or any other industrial process. The process is typically a multivariable process, and thus a large number of measurements are taken. In order to understand how the ML algorithm arrives at the predicted values, an explained value is formed to evaluate the input parameters. There are also normal measurement data, ML values and explained values that indicate that the process is running well, and a bias / error value of the values / data can be formed and used for analytical purposes.

[0044] The device of the application can be located in the same place as the tracked process. However, it is also possible that it is located elsewhere, which makes it possible to track the process remotely. For example, the measurement data 4 is sent to the diagnostic device of the application via a communication network, which processes the measurement data and sends the estimator output, which can be used to adjust the process. The output of the estimator can be sent to the owner of the process, to the maintenance center of the process or to any desired destination.

[0045] From the above it is evident that the application is not limited to the embodiments described herein, but can be implemented with many other different embodiments within the scope of the independent claims.

Claims

1. A diagnostic device for controlling and / or optimizing a multivariate industrial process, the device having a data processing module (6) for processing measurement data of the multivariate process and performing pre-processed measurement data (6A) and a machine learning module (7) for performing machine learning values (7A) from the pre-processed measurement data (6A), characterized in that, The diagnostic device comprises an interpreted value module (8) for forming interpreted values (8A) from the machine learned values (7A), and a deviation calculation module (14) for calculating deviations (8D) between the interpreted values (8A) and normal interpreted values (8N), deviations (7D) between the machine learned values (7A) and normal machine learned values (7N), and deviations (6D) between the pre-processed measured data (6A) and normal pre-processed measured data (6N), The diagnostic device further comprises at least one estimator (15) each arranged to track a particular disturbance condition or quality condition of the multivariable process using deviations (8D) between the interpreted values (8A) and normal interpreted values (8N), deviations (7D) between the machine learned values (7A) and normal machine learned values (7N), and deviations (6D) between the pre-processed measured data (6A) and normal pre-processed measured data (6N), and to form an estimate of the severity of the disturbance condition or the quality condition, the output of the estimator being used for controlling and / or optimizing the multivariable industrial process.

2. The diagnostic device of claim 1, wherein The machine learned interpreted values and the machine learned normal interpreted values are SHAP values, values from the LIME method, values from the DeepLIFT method or any other possible interpreted values.

3. The diagnostic device of claim 2, wherein, The machine learned normal interpreted values (8N), normal machine learned values (7N) and normal pre-processed measured data (6N) are values / data (13A) that have been derived from a good operating period of the process.

4. The diagnostic device of claim 3, wherein The estimator comprises at least one P-module (17, 17A, 17C), I-module (18, 18A, 18C) or D-module (19, 19A, 19C), or any combination of these modules, at least one module arranged to process deviations (8D) between the interpreted values (8A) and normal interpreted values (8N), at least one module arranged to process deviations (7D) between the machine learned values (7A) and normal machine learned values (7N), at least one module arranged to process deviations (6D) between the pre-processed measured data (6A) and normal pre-processed measured data (6N).

5. The diagnostic device of claim 4, wherein, The estimator further comprises an input mapping module (20, 21, 22, 20A, 21A, 22A, 20C, 21C) connected to the output (23, 24, 25, 23A, 24A, 25A, 23C, 24C) of the P-module (17, 17A, 17C), I-module (18, 18A, 18C) or D-module (19, 19A, 19C), or any combination of these modules, a summation module (26) summing the output (27, 28, 29, 27A, 28A, 29A, 27C, 28C) of the input mapping module, an output scaling module (30) scaling the output (31) of the summation module, an output mapping module (32) for providing a normalized output (33) as estimator output.

6. The diagnostic device of claim 5, wherein, The input mapping module (20, 21, 22, 20A, 21A, 22A, 20C, 21C) and the output mapping module (32) are formed by linguistic equations or fuzzy logic.

7. The diagnostic device of claim 6, wherein The mapping curves of the input mapping module (20, 21, 22, 20A, 21A, 22A, 20C, 21C) and the output mapping module (32) provide linear curves, piecewise linear curves, S-curves and / or other curve forms.

8. The diagnostic device according to any one of claims 1 to 7, characterized in that The diagnostic device comprises at least one deviation calculation module (14) to provide a deviation (8D) between the machine-learned interpretation value (8A) and the machine-learned normal interpretation value (8N), a deviation (7D) between the machine-learned value (7A) and the normal machine-learned value (7N), and / or a deviation (6D) between the pre-processed measurement data (6A) and the normal pre-processed measurement data (6N).

9. The diagnostic device of claim 8, wherein, The deviation calculation module (14) is part of the estimator (15).

10. The diagnostic device of claim 8, wherein, The deviation calculation module (14) is a module separate from the estimator (15).

11. The diagnostic device of claim 1, wherein The estimate of one estimator is an input to the other estimator for use by the other estimator.

12. A method for forming an estimate of the severity of an interference condition or a quality condition in a multivariate process, characterized by, The diagnostic device according to any one of claims 1 to 11 is used to form an estimate of the severity of an interference condition or a quality condition.

13. The method of claim 12, wherein, The estimate of the severity of the interference condition or the quality condition is used to provide a recommendation and / or a guidance command in the multivariable process to control and / or optimize the multivariable process.

14. The method according to claim 12 or 13, characterized in that, The multivariable process is a pulp process, a papermaking, a paperboard or tissue making process, an industrial water or wastewater treatment process, a raw water treatment process, a water re-use treatment, a municipal water or wastewater treatment process, a sludge treatment process, a mining process or a petroleum recovery process.

15. The method of claim 13, wherein, The control and / or optimization includes control of one or more of the following: dosage of chemicals, dosage point of chemicals, dosage interval of chemicals, selection of type of chemicals used in the process, process conditions, process flow delays.

Images