Method for generating an output signal from a virtual sensor of an industrial machine having reliability / reliability check

Abstract

According to an innovative method (3000), an output signal is generated from a virtual sensor of an industrial machine, the output signal corresponding to a sequence of values of output variables of a software model, the output variables being related to the measurement or evaluation of parameters of the industrial machine, the method being executed during machine operation time, a) receiving (3400) at least one input sequence of values of at least one input variable of a software model related to at least one physical quantity of the industrial machine, b) predicting (3500) a sequence of output values of the output variables via the software model based on the at least one input sequence, c) calculating (3600) for one or more or all of the values of the predicted output sequence, corresponding flags and / or values regarding the reliability and / or statistical evaluation of the predicted values, d) based on the corresponding flags and / or values, deciding (3700) whether to discard none or one or more of the values of the predicted output sequence, and c) using (3800) only the non-discarded values of the predicted output sequence as the output signal.

Description

Technical Field

[0001] The subject matter disclosed herein relates to a method for generating an output signal from a virtual sensor of an industrial machine that takes into account the inaccuracy of the virtual sensor. In particular, one or more predicted values from a predictor within the virtual sensor can be discarded if they are considered unreliable. Further, the subject matter disclosed herein relates to a system for evaluating a virtual sensor and an industrial machine.

Background Art

[0002] Virtual sensors are becoming increasingly popular in the field of industrial control, particularly in the oil and gas industry.

[0003] As is known, a virtual sensor is typically a software that can provide a signal related to a physical quantity, such as temperature or pressure, within an industrial system when it is not possible or desirable to directly measure this physical quantity via a physical sensor. The virtual sensor predicts the value of a variable of interest based on the values of one or more other variables directly measured within or on the industrial system.

[0004] Virtual sensors are often based on artificial intelligence and may require, for example, preliminary training by machine learning.

[0005] The accuracy of a virtual sensor typically depends on its preliminary training and the data used for training.

[0006] Ideally, the virtual sensor should be trained in combination with exactly the same industrial machinery that is intended to receive virtual measurements during operation. In other words, the "domain" used for training the virtual sensor should also be used for the operation or normal operation of the virtual sensor. Using different domains is already known. For example, the paper by L. Strazzera, V. Gori, and G. Veneri, titled "DANNTe: a case study of a turbo-machinery sensor virtualization under domain shift" and published in 2021 in relation to the Conference on Neural Information Processing Systems, describes a turbomachinery application consisting of constructing a virtual sensor using data collected from a machine during the so-called "source domain" (winter time) and applying it to data collected from the same machine during the so-called "target domain" (summer time).

[0007] The paper by Hossein Shahabadi Farahani et al., titled "A Novel Method for Designing Transferable Soft Sensors and its Application" on arxiv.org of Cornell University, discloses domain adversarial training for "transferring" a "soft sensor" from one plant (corresponding to the "source domain") to another plant (corresponding to the "target domain"). According to this paper, when the "soft sensor" is properly and sufficiently "transferred" through domain adversarial training, its output is used in the new plant and the generated output is considered to be always correct.

[0008] The patent application published as US Patent Application Publication No. 2021 / 110147 (A1) discloses systems and methods specific to image processing, particularly human detection in different scenes, i.e., different "domains". This document teaches the use of domain adversarial learning. According to this document, if the confidence score for an image is low, i.e., if the prediction provided by the system regarding the scene is not sufficiently reliable, the system requests that a human annotate the image rather than automatically annotating it. Further, the system requires annotated data belonging to the target domain from the start. In fact, this document discloses semi - automatic domain adaptation.

[0009] It is desirable for the "target domain" to be more different from the "source domain". In particular, for example, it is desirable to train a virtual sensor on a "prototype machine" within the premises of a machine manufacturer and use the trained virtual sensor with a "fleet machine", i.e., a machine installed and operating within the premises of a user, typically a client of the machine manufacturer. The more different the domains are, the higher the likelihood of inaccuracy. Thus, the applicant has come to consider the risk of inaccuracy, particularly the risk of inaccuracy resulting from the difference between the "source domain" and the "target domain".

SUMMARY OF THE INVENTION

[0010] According to a first aspect, the subject matter disclosed herein relates to a computer - implemented method for generating an output signal from a virtual sensor of an industrial machine based on a software model having input variables related to physical quantities of the industrial machine and output variables related to the measurement or evaluation of parameters of the industrial machine. During machine operation time, values of the output variables are predicted, and one or more of them can be discarded based on calculated flags and / or values related to, for example, the reliability and / or confidence level of the prediction.

[0011] According to a second aspect, the subject matter disclosed herein relates to a machine learning artificial intelligence virtual sensor for an industrial machine, and the output of the virtual sensor is related to the measurement or evaluation of parameters of the industrial machine. The virtual sensor includes a feature extractor, a domain classifier, a predictor, and a discarder, and the output of the predictor corresponds to the output of the virtual sensor. During training time, the extractor, classifier, and predictor are active, and the discarder is inactive. During machine operation time, the extractor, classifier, predictor, and discarder are all active.

[0012] According to a third aspect, the subject matter disclosed herein relates to a computer-implemented system for evaluating an industrial machine. This computer-implemented system includes at least one virtual sensor of an industrial machine configured to output only predicted values having a specific reliability and / or a specific confidence level.

Brief Description of the Drawings

[0013] Many of the disclosed embodiments of the present invention, and the attendant advantages thereof, will be better understood and more fully appreciated when considered in connection with the accompanying drawings. Referring to the following detailed description of the invention for the purpose of carrying out the invention, a more complete understanding will be readily obtained.

Figure 1

Figure 2

Figure 3

Detailed Description of the Invention

[0014] A virtual sensor is typically one piece of software that can provide values of variables within an industrial system, such as temperature or pressure values that are impossible or undesirable to measure directly through physical sensors. Generally, the output of a virtual sensor can be related to the measurement or evaluation of parameters of an industrial machine. The values first calculated and then provided by a virtual sensor can be affected by (e.g., calculation) errors. An innovative virtual sensor can itself determine whether the value calculated by the virtual sensor is incorrect or overly incorrect and discard it. In the case of an innovative virtual sensor, the error may be due to the fact that the virtual sensor is not fully trained on a production machine but is at least partially trained on a prototype machine.

[0015] Note that since a virtual sensor is generally first trained and then used, the expression "during machine operation time" is used herein to distinguish it from "during training time". However, the expression "during machine operation time" does not exclude the possibility that there may be a delay (fixed or variable) (short, e.g., 1 second, or long, e.g., 1 year) between the time when some data is collected from one or more physical sensors after training is completed and the time when such collected data is processed by an innovative virtual sensor to generate an output signal. It should not be excluded that an innovative virtual sensor may have a first training period, a subsequent first operation period, a subsequent second training period, a subsequent second operation period, etc. Furthermore, it should not be excluded that the operation period may at least partially overlap with the training period, in which case the innovative virtual sensor learns during operation.

[0016] Background knowledge for easier understanding of this specification can be found in the papers of L. Strazzera, V. Gori, and G. Veneri, titled "DANNTe: a case study of a turbo-machinery sensor virtualization under domain shift" and published in 2021 in relation to the Conference on Neural Information Processing Systems, and the paper of Y. Ganin et al., titled "Domain-Adversarial Training of Neural Networks" and published in 2016 by the Journal of Machine Learning Research.

[0017] FIG. 1 shows a schematic block diagram of an embodiment of an innovative virtual sensor 100. The virtual sensor 100 includes a feature extractor 20, a predictor 40, a domain classifier 60, and a rejector 80. During operation of the virtual sensor 100, the feature extractor 20 receives input variable values, extracts feature values from the input variable values, and provides the feature values to the predictor 40 and the domain classifier 60. During operation of the virtual sensor 100, the predictor 40 makes a prediction using the received feature values and outputs a predicted output variable value, and the domain classifier 60 receives the feature values, estimates the domain of the received input variable values based on the received feature values, and calculates a reliability flag corresponding to the predicted output variable value output by the predictor 40.

[0018] According to the embodiment of FIG. 1, the input variable values may come from one or more physical sensors of a prototype machine installed and operating in a test environment, i.e., the "first domain" (so-called "source domain"), or from one or more physical sensors of an industrial machine installed and operating in a production environment, i.e., the "second domain" (so-called "target domain"), but it should be noted that they do not come from both machines simultaneously. In fact, especially (but not only) during the training period, a set of input variable values may be provided to virtual sensors grouped into so-called "batches", and the batches advantageously include data from the "source domain" and the "target domain", even if the data from the "source domain" and the "target domain" are not actually processed simultaneously but only during the same period. Thus, the domain classifier 60 can understand (more precisely, estimate) whether the input variable values received by the virtual sensors at a certain time come from the "first domain" or the "second domain", and determine the "domain loss". When a data batch is used during the training period, the output from the domain classifier may be an average value derived from all the data in the batch, e.g., the average domain loss. Usually, the input variable values come from the "first domain" and the "second domain" during the training of the virtual sensors and then only from the "second domain". When a data batch is used, a single batch may include data from only the "first domain", data from only the "second domain", or data from both the "first domain" and the "second domain". As already explained, there may be a delay between the time when data is detected by the physical sensors and the time when the data is processed by the virtual sensors. Generally, the number of domains may be three or more, e.g., there may be two or more prototype machines used for training.

[0019] As will be described later, the domain classifier 60 can also calculate a reliability flag corresponding to the predicted output variable value output by the predictor 40. In other words, according to the embodiment of FIG. 1, the domain classifier 60 can also understand (more precisely, estimate) whether the output variable value output by the predictor 40 at a certain time is reliable. Generally, since the output variable value sometimes changes, the reliability flag sometimes changes.

[0020] According to the embodiment of FIG. 1, during the operation of the virtual sensor 100, the discarder 80 receives the predicted output variable value output from the predictor 40 and the corresponding reliability flag from the domain classifier 60, discards the values considered unreliable based on the calculation / estimation of the domain classifier 60, and outputs only the reliable values.

[0021] As shown in FIG. 2, the innovative virtual sensor 100 may be a component of an innovative evaluation system 1000, i.e., a computer-implemented system capable of evaluating an industrial machine (or equivalently, an industrial system or industrial plant) typically in a production state. In the embodiment of FIG. 2, some inputs of the prior art evaluation system 300 are combined with some physical sensors 200 of the industrial machine, and some other inputs of the system 300 are combined with the output of the virtual sensor 100. The input of the virtual sensor 100 is combined with some physical sensors 200 of the industrial machine. Note that the physical sensors 200 of the industrial machine are external to the innovative evaluation system 1000 (see the slanted dashed line).

[0022] An embodiment of an innovative method for generating an output signal from a virtual sensor of an industrial machine is shown in FIG. 3. This method (as well as embodiments of other methods) is a so-called "computer-implemented method". Note that the arrows in this flowchart do not represent the flow of data, but only the flow of control. In particular, the data generated by the activity of block 3800 is not intended to be provided as an input to the activity of block 3400. Flowchart 3000 of FIG. 3 has a BEGIN block 3100 and several subsequent blocks. Blocks 3200 and 3300 (described in detail later) correspond to method steps that are executed before the machine operation time, i.e., during the training time of the innovative virtual sensor, while blocks 3400, 3500, 3600, 3700, and 3800 (described in detail later) correspond to method steps that are executed during the machine operation time, i.e., after the training time of the innovative virtual sensor. In flowchart 3000 of FIG. 3, there is no end block as if the innovative virtual sensor never stops operating. However, it should be understood that the innovative virtual sensor can be stopped at any time by an "external command" that can be issued by an operator and / or one piece of software. In flowchart 3000 of FIG. 3, blocks 3400, 3500, 3600, 3700, and 3800 form a loop and are repeatedly executed cyclically. Depending on a particular embodiment, the iteration may be performed for each new set of input variable values, e.g., for any new set of values provided as input to the innovative virtual sensor at a certain time. For example, if the virtual sensor has only one input variable, the iteration may be performed at any new value of the input variable provided to the sensor input.

[0023] According to the method of FIG. 3 (and embodiments of other methods), the input signal of the innovative virtual sensor corresponds to a sequence of values of the input variables of the software model, and the output signal of the innovative virtual sensor corresponds to a sequence of values of the output variables of the software model. Generally, there are a number of input variables, and thus a corresponding number of input signals. Generally, there may be two or more output variables, and thus two or more output signals, which means that the innovative virtual sensor can "virtually" measure two or more variables in the industrial system simultaneously. The input variables are related to the physical quantities of the industrial system, and the output variables are related to the measurement or evaluation of the parameters of the industrial system (which may be physical quantities).

[0024] The method of FIG. 3 (and embodiments of other methods) includes at least the following steps performed during machine operation time: a) receiving at least one input sequence of values of the input variables of the software model (block 3400 in FIG. 3); b) predicting an output sequence of values of the output variables through the software model based on the at least one input sequence (block 3500 in FIG. 3); c) calculating corresponding flags and / or values regarding the reliability and / or statistical evaluation of the predicted values for one or more or all of the values of the predicted output sequence (block 3600 in FIG. 3); d) discarding none or one or more of the values of the predicted output sequence based on the corresponding flags and / or values (the number of values discarded depends on the values of the flags and / or values) (block 3700 in FIG. 3), and e) using only the non-discarded values of the predicted output sequence as the output signal (block 3800 in FIG. 3).

[0025] Step b) follows step a) after a while. Depending on the embodiment, this may be a very short time, for example, even in microseconds if this delay is only due to the processing time of the microprocessor, or a very long time, for example, several days or weeks or months if the delay is due to the storage of data previously collected from industrial machines. In other words, the delay may depend on where the value is received from.

[0026] Note that in step b), typically, for each input value in the input sequence, there is an output value in the output sequence.

[0027] Furthermore, note that the calculations in step c) are not necessarily performed for each of the values of the predicted output sequence.

[0028] The software model of this method is preferably an artificial intelligence software model, particularly a machine learning artificial intelligence software model.

[0029] Step c) basically offers two possibilities, and these possibilities may be alternative or even used simultaneously. According to the first possibility that is easier to implement, a prediction reliability flag is calculated, for example, for each of the predicted values in step b). In particular, according to this first possibility, if the corresponding predicted value is considered reliable, the flag is "true". According to the second possibility, a statistical evaluation value is calculated, for example, for each of the predicted values in step b). In particular, according to this second possibility, the value is a confidence level and corresponds to the probability that the predicted value is correct.

[0030] Regarding step e), further details are provided below. Note that the output of step b), i.e., the output before the discard in step d), may also have useful purposes, especially during the training time.

[0031] The method of Figure 3 (and embodiments of other methods) is preferably preceded by the following preliminary steps before the machine operation time, namely - At least, it includes the step of training a software model on a source domain different from the target domain corresponding to the machine operation time (blocks 3200 and 3300 in FIG. 3).

[0032] More preferably, the software model is trained on both the source domain and the target domain. In FIG. 3, blocks 3200 and 3300 are generally parallel because training with source domain data and training with target domain data are completely or partially overlapping as it can be considered that such two training activities are simultaneous. As already explained, for example, there may be a sequence of training data batches including both source domain data and target domain data, or, for example, a sequence in which a training data batch including only source domain data and a training data batch including only target domain data alternate.

[0033] The above preliminary step is the following step, that is - It may include the step of training a software model on data collected from at least a prototype machine different from the industrial machine (block 3200 in FIG. 3) (in this case, the values of the input variables and output variables from the prototype should be used).

[0034] It should be noted that typically, a "prototype machine" includes a much larger number of physical sensors, for example, more than 10 times as many as a "fleet machine".

[0035] The above preliminary step is the following step, that is - It may include the step of training a software model on data collected from the industrial machine (block 3300 in FIG. 3) (in this case, only the values of the input variables from the industrial machine should be used).

[0036] It should be noted that the training of the virtual sensor can be based on more prototype machine data than industrial machine data. In other words, advantageously, the training of the virtual sensor can be possible without requiring long activities using industrial machines during production. In fact, the loss weighting mechanism can be arranged in place to compensate for different amounts. For example, the domain loss during training for processing industrial machine data can have a higher weight than the training domain loss when processing prototype machine data if the amount of training prototype machine data is larger than the amount of training prototype machine data.

[0037] According to the method of FIG. 3 (and embodiments of other methods), the software model is typically implemented via a neural network. Advantageously, the training is domain adversarial training of the neural network. In particular, the software model is realized through a first neural network configured to embody a feature extractor, a second neural network configured to embody a domain classifier, and a third neural network configured to embody a predictor, and the training is domain adversarial training of the entire neural network which is a set of the first, second, and third neural networks. More specifically, the first neural network (i.e., the "feature extractor") is trained on both the source domain and the target domain, the second neural network (i.e., the "domain classifier") is trained on both the source domain and the target domain, and the third neural network (i.e., the "predictor") is trained only on the source domain.

[0038] The calculation in step c) (of the flag and / or value) may be performed by the domain classifier based on at least one input sequence of values, for example, providing a domain loss flag. Preferably, such a calculation is performed through the same single software model used for prediction.

[0039] The verification of the software model can be performed by considering another variable that is strictly correlated with the output variable, even without measuring the output variable (corresponding to the output of the virtual sensor) on the industrial machine.

[0040] The above method may be implemented through a machine learning artificial intelligence virtual sensor for an industrial machine (or equivalently, an industrial system or industrial plant), such as the sensor 100 shown in FIG. 1 for example.

[0041] An innovative virtual sensor generally - A feature extractor (labeled 20 in the embodiment of FIG. 1) having one or more variable inputs (labeled 22 in the embodiment of FIG. 1) and one or more feature outputs (labeled 24 in the embodiment of FIG. 1), wherein the one or more variable inputs are configured to be coupled to corresponding one or more physical sensors of the industrial machine, and the feature extractor is configured to output feature values at the one or more feature outputs, the feature extractor; - A predictor (labeled 40 in the embodiment of FIG. 1) having one or more feature inputs (labeled 42 in the embodiment of FIG. 1) and at least one variable output (labeled 44 in the embodiment of FIG. 1), wherein the feature inputs of the predictor are coupled to the feature outputs of the feature extractor, and the predictor is configured to output a sequence of variable prediction values at the at least one variable output, the predictor; - A domain classifier (labeled 60 in the embodiment of FIG. 1) having one or more feature inputs (labeled 62 in the embodiment of FIG. 1) and a domain loss output (labeled 64 in the embodiment of FIG. 1), wherein the feature inputs of the domain classifier are coupled to the feature outputs of the feature extractor, and the domain classifier is configured to output a sequence of domain loss flags at the domain loss output, the domain classifier; - An eliminator (labeled 80 in the embodiment of FIG. 1) having a loss input (labeled 83 in the embodiment of FIG. 1), the loss input being coupled to the domain loss output of a domain classifier, the eliminator being configured to signal internally and / or externally a discard action (which may be only the possibility of a discard action) for a variable prediction value when the corresponding domain loss value exceeds a predetermined threshold, such signaling being performed via a signal (not shown in FIG. 1) inside the eliminator that may also be provided to a specific output of the eliminator (not shown in FIG. 1), comprising the eliminator, According to a particular embodiment, the output of the virtual sensor (labeled 100 in the embodiment of FIG. 1) may correspond to the output of the eliminator (labeled 80 in the embodiment of FIG. 1), or a combination of the output of the predictor (labeled 40 in the embodiment of FIG. 1) and the output of the eliminator (labeled 80 in the embodiment of FIG. 1). The virtual sensor (labeled 100 in the embodiment of FIG. 1) is - During training time, the feature extractor, classifier, and predictor are active and the eliminator is inactive. - During machine operation time, the feature extractor, classifier, predictor, and eliminator are active.

[0042] Basically, assume that when the domain classifier determines that there is a domain loss, that is, when it cannot correctly estimate the domain, the output value predicted by the predictor should be considered untrustworthy. If the domain classifier can also provide a specific confidence level for the domain, for example, the corresponding confidence level (not necessarily the same) may be applicable to the output value predicted by the predictor.

[0043] In the embodiment of FIG. 1 (as well as other sensor embodiments), it should be noted that the predictor 40 and the classifier 60 have specific outputs used during the training time. In particular, output 48 corresponds to output 44, output 68 corresponds to output 64, and output 66 is additional and will be described below.

[0044] The sensor 100 of FIG. 1 (as well as other sensor embodiments) includes a domain classifier 60, and the domain classifier 60 also has a domain label output 66, and is configured to output in output 66 a sequence of domain labels (i.e., the domain estimated from the input variable values at consecutive times) mainly used during the training time.

[0045] In the embodiment of FIG. 1 (as well as other sensor forms), the feature extractor 20 may comprise a first neural network or be a first neural network, the domain classifier 60 may comprise a second neural network or be a second neural network, and the predictor 40 may comprise a third neural network or be a third neural network.

[0046] In the embodiment of FIG. 1 (as well as other sensor embodiments), the rejector 80 is more advantageously configured to reject the variable prediction value in at least one variable output 44 of the predictor 40 based on the corresponding domain loss value.

[0047] In the embodiment of FIG. 1 (as well as other sensor embodiments), the rejector 80 is further configured to receive the variable prediction value from at least one variable output 44 of the predictor 40 (at input 82) and output the non-rejected variable prediction value (at output 84).

[0048] According to some sensor embodiments, the rejector is configured to store the non-rejected variable prediction value.

[0049] According to some sensor embodiments, the discarder is configured to store information regarding the discard action. For example, all predicted values may be output and / or stored by the discarder together with their reliability flags and / or corresponding values regarding, for example, their confidence levels, such that the decision of whether and how to use the values predicted by the innovative virtual sensor is delegated to other devices or systems.

[0050] As already mentioned, the innovative virtual sensor may be a component of an innovative computer-implemented system for performing evaluations on industrial machines (or equivalently, industrial systems or industrial plants) such as those shown in FIG. 2, for example. Generally, such an innovative evaluation system can include several virtual sensors, some or all of which are innovative virtual sensors. Generally, such an innovative evaluation system can be configured to be coupled to several physical sensors, as shown in FIG. 2, for example.

[0051] The innovative computer-implemented evaluation system can be configured to execute a computational analysis computer program and process data from at least one or more innovative virtual sensors (labeled 100 in the embodiment of FIG. 2) of the industrial machine to be evaluated, optionally from one or more non-innovative virtual sensors (not shown in the embodiment of FIG. 2) of the industrial machine to be evaluated, and / or from one or more physical sensors (labeled 200 in the embodiment of FIG. 2) of the industrial machine to be evaluated.

[0052] FIG. 2 shows a physical sensor 200 of an industrial machine having a corresponding output 210. A first set 220 of the output 210 is coupled to an input 102 of the innovative virtual sensor 100 (corresponding to the input 22 of the feature extractor 20 in FIG. 1), and a second set 230 of the output 210 is coupled to a first set of inputs 330 of the “conventional” evaluation system 300. An output 104 of the innovative virtual sensor 100 (which may correspond to the output 84 of the eliminator 80 in FIG. 1) is coupled to a second set of inputs 320 of the “conventional” evaluation system 300. The “conventional” evaluation system 300 has an output 340 for providing an evaluation result based on the data at its inputs 320 and 330.

[0053] Note that an evaluation system including an innovative evaluation system may be designed to perform evaluations on two or more industrial machines (or equivalently, an industrial system or an industrial plant).

Claims

1. A computer-implemented method (3000) for generating an output signal from a virtual sensor of an industrial machine, wherein the output signal corresponds to a sequence of values of an output variable of a software model, the output variable being related to the measurement or evaluation of a parameter of the industrial machine, the method comprising the following steps executed during machine operation time, namely: a) receiving (3400) at least one input sequence of values of at least one input variable of the software model, the at least one input variable being related to at least one physical quantity of the industrial machine; b) predicting (3500) an output sequence of values of the output variable via the software model based on the at least one input sequence; c) calculating (3600) corresponding flags and / or values regarding the reliability and / or statistical evaluation of the predicted values for one or more or all of the values of the predicted output sequence; d) discarding (3700) the values of the predicted output sequence based on the corresponding flags and / or values; e) using (3800) only the non-discarded values of the predicted output sequence as the output signal, a computer-implemented method (3000).

2. The method further comprises the following preliminary steps executed before machine operation time, namely - training (3200, 3300) the software model at least on a source domain different from the target domain corresponding to machine operation time, the method (3000) according to claim 1.

3. The software model is implemented via a neural network (20, 40, 60), the training being domain adversarial training of the neural network (20, 40, 60), the method (3000) according to claim 2.

4. The preliminary steps comprise - training (3200) the software model with respect to at least data collected from a prototype machine different from the industrial machine, the method (3000) according to claim 2.

5. The preliminary steps comprise - training (3300) the software model also with respect to data collected from the industrial machine, the method according to claim 4.

6. The method according to claim 2, wherein in step c), the flag and / or value is preferably calculated via a domain classifier (60) configured to output a domain loss flag and / or value based on the at least one input sequence via the software model.

7. A machine learning artificial intelligence virtual sensor (100) for an industrial machine, wherein the output of the virtual sensor is related to the measurement or evaluation of parameters of the industrial machine, and the virtual sensor comprises: - A feature extractor (20) having one or more variable inputs (22) and one or more feature outputs (24), wherein the one or more variable inputs (22) are configured to be coupled to corresponding one or more physical sensors (200) of the industrial machine, and the feature extractor (20) is configured to output feature values at the one or more feature outputs (24), the feature extractor (20); - A domain classifier (60) having one or more feature inputs (62) and a domain loss output (64), wherein the feature inputs (62) are coupled to the feature outputs (24), and the domain classifier (60) is configured to output a sequence of domain loss flags at the domain loss output (64), the domain classifier (60); - A predictor (40) having one or more feature inputs (42) and at least one variable output (44), wherein the feature inputs (42) are coupled to the feature outputs (24), and the predictor (40) is configured to output a sequence of variable prediction values at the at least one variable output (44), the predictor (40); - A rejector (80) having a loss input (83), wherein the loss input (83) is coupled to the domain loss output (64), and the rejector (80) is configured to signal a rejection action for the variable prediction value when the corresponding domain loss value exceeds a predetermined threshold, the rejector (80), comprising: The output of the virtual sensor (100) corresponds to the output of the rejector (80), or a combination of the output of the predictor (40) and the output of the rejector (80); The virtual sensor (100) is: - During the training time, the feature extractor (20), the classifier (60) and the predictor (40) are active, and the rejector (80) is inactive. - A machine learning artificial intelligence virtual sensor (100) for industrial machines, configured such that the feature extractor (20), the classifier (60), the predictor (40), and the discard unit (80) are active during the machine operation time.

8. The feature extractor (20) comprises a first neural network, The domain classifier (60) comprises a second neural network, The virtual sensor (100) according to claim 7, wherein the predictor (40) comprises a third neural network.

9. The virtual sensor according to claim 7, wherein the discard unit (80) is further configured to discard the variable prediction value in the at least one variable output (44) based on the corresponding domain loss value.

10. The virtual sensor according to claim 9, wherein the discard unit (80) is further configured to receive a variable prediction value from the at least one variable output (44) (82) and output a variable prediction value that has not been discarded (84).

11. The virtual sensor according to claim 9, wherein the discard unit (80) is further configured to store the variable prediction value that has not been discarded.

12. The virtual sensor according to claim 11, wherein the discard unit (80) is further configured to store information regarding the discard action.

13. A computer-implemented system (1000) for evaluating an industrial machine, - The computer-implemented system (1000) comprising at least one virtual sensor (100) of the industrial machine according to any one of claims 7 to 12.

14. The computer-implemented system (1000) according to claim 13, wherein the system is configured to execute a computational analysis computer program and process data from at least the virtual sensor (100) of the industrial machine and optionally data from one or more physical sensors (200) of the industrial machine.

Images