WO2022262072A1

WO2022262072A1 - Method and system for training target model on the basis of modification of sample signal

Info

Publication number: WO2022262072A1
Application number: PCT/CN2021/108156
Authority: WO
Inventors: 郭春林; 郭尔富
Original assignee: 华北电力大学; 北京大地纵横科技有限公司
Priority date: 2021-06-16
Filing date: 2021-07-23
Publication date: 2022-12-22
Also published as: CN113537288A; CN113537288B

Abstract

A method and system for training a target model on the basis of modification of a sample signal. The method comprises: selecting, from a plurality of models, a target model which needs to be trained, and obtaining a configuration file for the target model (101); determining an object device related to the target model on the basis of the configuration file for the target model, and determining a plurality of sample signals associated with the object device (102); when the object device operates normally, performing signal acquisition or signal simulation on at least two sample signals in the plurality of sample signals so as to obtain a normal sample set comprising the at least two sample signals (103); modifying at least one sample signal in the normal sample set according to a preset modification mode to obtain an abnormal sample set corresponding to the normal sample set (104); and training the target model on the basis of the normal sample set, the abnormal sample set, and a preset training algorithm to obtain a trained target model (105).

Description

Method and system for training target model based on modification of sample signal

technical field

The present invention relates to the technical field of artificial intelligence, and more specifically, to a method and system for training a target model based on modification of sample signals.

Background technique

At present, with the development of artificial intelligence technology, a large number of machine learning algorithms continue to emerge. Machine learning algorithms, especially deep learning, have achieved great success in recent years, and data is the key factor that makes machine learning possible. Technicians can use simple algorithms to implement machine learning, but without good data the algorithms cannot be optimized.

It can be seen that in the model training based on machine learning, the data quality of the sample signal/sample data affects the training effect of the model. However, in actual situations, the number or rate of failures of some types of equipment during operation is relatively low, so the sample signal/sample data of this type of equipment in normal operation has a relatively large amount of data, while abnormal operation or failure The data size of the sample signal/sample data is small. In this case, abnormal sample signals/sample data with a small amount of data cannot meet the requirements of model training or testing.

Contents of the invention

In order to solve the above problems in the prior art, the present application proposes a model training method and system based on signal modification. The signal modification-based model training method and system of the present application are applicable to various predetermined algorithms, including machine learning algorithms, fitting algorithms, and the like.

According to one aspect of the present invention, there is provided a method for training a target model based on modification of a sample signal, the method comprising:

Select a target model to be trained from multiple models, and obtain a configuration file of the target model;

determining an object device involved in the object model based on the configuration file of the object model, and determining various sample signals associated with the object device;

When the subject device is operating normally, perform signal acquisition or signal simulation on at least two sample signals among the various sample signals, so as to obtain a normal sample set including the at least two sample signals;

Modifying at least one sample signal in the normal sample set according to a preset modification method to obtain an abnormal sample set corresponding to the normal sample set;

The target model is trained based on the normal sample set, the abnormal sample set and the preset training algorithm, so as to obtain the trained target model.

The various sample signals include: vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness.

At least one sample signal among the at least two sample signals is a vibration/acoustic emission signal collected by a sensor close to the device casing.

At least one of the at least two sample signals is a sound signal collected outside the device.

According to still another aspect of the present invention, a method for testing a target model based on modification of a sample signal is provided, the method comprising:

Determine the target model that needs to be tested and obtain the configuration file of the target model;

When the subject device is running normally, perform signal acquisition or signal simulation on at least two sample signals among the various sample signals, so as to obtain a normal sample set including the at least two sample signals;

The target model is tested based on the normal sample set and the abnormal sample set, so as to determine the performance index of the target model based on the test results.

According to still another aspect of the present invention, a system for training a target model based on modification of sample signals is provided, the system comprising:

a selection device, configured to select a target model to be trained from multiple models, and obtain a configuration file of the target model;

determining means, configured to determine the target device involved in the target model based on the configuration file of the target model, and determine various sample signals associated with the target device;

An acquisition device, configured to perform signal acquisition or signal simulation on at least two sample signals among various sample signals when the subject device is operating normally, so as to acquire a normal sample set including the at least two sample signals;

A modifying device, configured to modify at least one sample signal in the normal sample set according to a preset modification method, so as to obtain an abnormal sample set corresponding to the normal sample set;

The training device is used to train the target model based on the normal sample set, the abnormal sample set and a preset training algorithm, so as to obtain the trained target model.

According to another aspect of the present invention, there is provided a system for testing a target model based on modification of a sample signal, the system comprising:

Obtaining means for determining a target model to be tested and obtaining a configuration file of the target model;

An acquisition device, configured to perform signal acquisition or signal simulation on at least two sample signals among multiple sample signals when the subject device is operating normally, so as to acquire a normal sample set including the at least two sample signals;

The testing device is used for testing the target model based on the normal sample set and the abnormal sample set, so as to determine the performance index of the target model based on the test result.

According to another aspect of the present invention, a model training method based on signal modification is provided. Firstly, a large number of normal samples and abnormal samples are generated, and then a model is trained using the samples and a set algorithm for diagnosing the state of the target device, distinguish or recognize. in,

The normal samples are generated by collecting or simulating two or more signals under normal operation of the target device. (The types of sample signals include but are not limited to vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image, brightness, and can also be a combination of various signals)

The abnormal samples are generated by modifying certain signals in normal samples.

Wherein modifying a certain signal in the normal sample is to modify a certain signal according to the amplitude, the square of the amplitude, the energy value, the square of the energy value or their ratio.

Among them, during the training process, the amount of signal modification in abnormal samples is also reduced to obtain a higher-precision model.

According to still another aspect of the present invention, a model evaluation method based on signal modification is provided, and the model is used for diagnosing, judging or identifying the state of the target device. in:

Generate normal samples by collecting or simulating two or more signals under normal operation of the target device.

Anomalous samples are generated by modifying a certain signal in normal samples.

The model is tested with the normal samples and abnormal samples, and then the performance of the model is evaluated according to the test results.

Wherein modifying a certain signal in the normal sample is modifying the signal according to the amplitude, the square of the amplitude, the energy value, the square of the energy value or their ratio.

Among them, the accuracy of the model is evaluated by reducing the amount of signal modification in abnormal samples during the training process.

The technical solution of the present invention can effectively and accurately obtain abnormal samples or negative samples, and combine the obtained abnormal samples or negative samples with normal samples or positive samples to train or test the model. Through the model training and testing technology of the technical solution of the present invention, a model with better recognition accuracy can be obtained. The model detection effect obtained by the method or system of the present invention is good, the technical difficulty and cost are relatively low, and it can be widely used in the scheme of identifying with various devices.

Description of drawings

A more complete understanding of the exemplary embodiments of the present invention can be had by referring to the following drawings:

1 is a flowchart of a method for training a target model based on modification of a sample signal according to an embodiment of the present invention;

Fig. 2 is the flow chart of the method for testing target model based on the modification of sample signal according to the embodiment of the present invention;

3 is a schematic structural diagram of a system for training a target model based on modification of sample signals according to an embodiment of the present invention;

Fig. 4 is a schematic structural diagram of a system for testing a target model based on modification of a sample signal according to an embodiment of the present invention.

detailed description

FIG. 1 is a flowchart of a method 100 for training a target model based on modification of sample signals according to an embodiment of the present invention. Method 100 starts at step 101 .

In step 101, a target model to be trained is selected from multiple models, and a configuration file of the target model is obtained. In the actual scene of industrial production or equipment operation, various types and/or various sizes of equipment are widely used in various locations, production links, monitoring links, etc. For this reason, if it is necessary to determine the operating status of any type of equipment, or obtain parameters of any type of equipment, etc., it is necessary to obtain the model or equipment model of each different type of equipment. Generally, the model of each different type of equipment or the equipment model can be used to determine the operating state of the equipment, obtain the operating parameters of the equipment, and so on. For this reason, when it is necessary to train or test a model of a specific device, it is necessary to select a target model to be trained from models associated with multiple different devices or device models.

Typically, each model or device model has a configuration file, and the configuration file is used to describe various properties of the model or device model. The various attributes are, for example: input parameters, output parameters, model type, model effect, model accuracy, device type, device name, device identifier, and the like. To this end, the target model has various attributes, and for example, the device type, device name, device identifier, etc. of the target model can be determined via the configuration file of the target model.

In step 102, an object device involved in the object model is determined based on the configuration file of the object model, and various sample signals associated with the object device are determined. As mentioned above, the device type, device name, device identifier, etc. of the target model can be determined through the configuration file of the target model. And further, the target device involved in the target model may be determined by parsing the configuration file of the target model. The target device may be any type of device. In addition, the configuration file of the target model may also include information of various sample signals associated with the target device. Alternatively, after the target device involved in the target model is determined, the device identifier or device name of the target device can be used to search in the sample signal information library to obtain information on various sample signals associated with the target device .

The multiple sample signals include one or more of the following: vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness. In general, one or more sample signals of vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness can be used to characterize, train, test, describe target device. It should be understood that this application is only described by taking vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness as examples, and those skilled in the art should understand Note that this application can use any reasonable sample signal. In practical scenarios, various types of sensors can be used to acquire any one of vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image, and brightness.

In step 103, when the target device is running normally, signal acquisition or signal simulation is performed on at least two sample signals among various sample signals, so as to obtain a normal sample set including the at least two sample signals.

The normal sample set is one of the data collected by sensors such as vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness when the target device is in normal operation. A sample signal composed of one or more sample signals or data. In the present application, in order to make the model accuracy of the trained or tested target model higher, the present application performs signal acquisition or signal simulation on at least two sample signals among various sample signals, so as to obtain the A collection of normal samples for a sample signal.

For example, the at least two sample signals are vibration signal, sound signal and voltage signal, and the normal sample set includes a plurality of samples arranged in order of sample sampling time, wherein each sample includes vibration signal, sound signal and voltage signal , and each sample has a sampling time. That is, each sample in the set of normal samples is a signal group or signal set having a sample time and including each sample signal at that sample time. When storing the sample data, the normal sample set can include at least two sample subsets, and each sample subset is a vibration sample signal subset, an acoustic emission sample signal subset, a sound sample signal subset or an electric field strength sample signal subset, etc. It should be understood that the division of signal subsets is only for data storage or data display. In practice, each sample includes each of at least two sample signals. Alternatively, a plurality of sample signal groups are included in the normal sample set, and each sample signal group includes a single vibration sample signal, a single acoustic emission sample signal and a single sound sample signal, for example, each sample signal group is <vibration sample signal, acoustic sample signal Transmit sample signal, sound sample signal>. It should be appreciated that each sample signal group can be considered as a sample in the set of normal samples.

For example, at least one sample signal among the at least two sample signals is a vibration/acoustic emission signal collected by a sensor closely attached to a device housing of the target device. Alternatively, at least one sample signal among the at least two sample signals is a sound signal collected outside the device of the subject device. In an actual situation, the sensor may be placed close to the casing of the device or the target device, placed outside the device or the target device, or placed inside the device or the target device.

In step 104, at least one sample signal in the normal sample set is modified according to a preset modification manner, so as to obtain an abnormal sample set corresponding to the normal sample set. In actual situations, equipment with a stable operating state has fewer failures or a lower failure rate in actual operation, so the sample signal/sample data volume of this type of equipment is relatively large for normal operation, while abnormal operation Or the data amount of the sample signal/sample data at the time of failure is small. In this case, it is often difficult to obtain enough abnormal sample signals. To this end, the present application modifies at least one sample signal in the normal sample set according to a preset modification manner.

The preset modification methods include: increasing or decreasing the corresponding modification amount for the amplitude, the square of the amplitude, the energy value or the square of the energy value, so that the amplitude, the square of the amplitude, or the energy value of the corresponding modification amount can be increased or decreased Or the value or value range of the square of the energy value changes. For example, increase or decrease the corresponding modifier k for the amplitude sample signal in the normal sample set of the target device; increase or decrease the corresponding modifier b for the amplitude square sample signal in the normal sample set of the target device; Increase or decrease the corresponding modifier c for the energy value sample signal in the normal sample set of the target device or increase or decrease the corresponding modifier d for the energy value square sample signal in the normal sample set of the target device, so that the normal sample set of the target device The amplitude sample signal, the amplitude square sample signal, the energy value sample signal or the energy value square sample signal in becomes the sample signal when the abnormal operation or faulty operation occurs. Thus, the abnormal sample set corresponding to the normal sample set is formed from the amplitude sample signal, the amplitude square sample signal, the energy value sample signal or the energy value square sample signal when the target device is abnormally or malfunctioning.

The preset modification methods also include: increasing or decreasing the corresponding ratio for the amplitude, amplitude square, energy value or energy value square, so that the corresponding ratio of amplitude, amplitude square, energy The value or value interval of the value or the square of the energy value changes. For example, the amplitude sample signal in the normal sample set of the target device increases or decreases by the corresponding ratio e, the amplitude square sample signal in the normal sample set of the target device increases or decreases by the corresponding rate f, and the target device The energy value sample signal in the normal sample set of the target device increases or decreases by the corresponding ratio g, or the energy value square sample signal in the normal sample set of the target device increases or decreases by the corresponding rate h, so that the normal sample set of the target device The amplitude sample signal, the amplitude square sample signal, the energy value sample signal or the energy value square sample signal in becomes the sample signal when the abnormal operation or faulty operation occurs. Thus, the abnormal sample set corresponding to the normal sample set is formed from the amplitude sample signal, the amplitude square sample signal, the energy value sample signal or the energy value square sample signal when the target device is abnormally or malfunctioning.

In addition, the preset modification method also includes: increasing or decreasing a corresponding steady-state modification amount for the amplitude, the square of the amplitude, the energy value or the square of the energy value. As mentioned above, magnitude, magnitude squared, energy value or energy value squared may have respective steady state modifiers.

Wherein is the amplitude, the square of the amplitude, the energy value or the square of the energy value, increasing or decreasing the corresponding steady-state modifier includes: the steady-state modifier of the amplitude, the square of the amplitude, the energy value or the square of the energy value in the statistical normal sample set △Y corresponds to the maximum value Ymax and minimum value Ymin of the parameter; set the adjustment target coefficient a, where 0<a<1; count the value Ysignal of the corresponding parameter of the steady-state modification amount △Y of the sample to be modified;

Calculate the steady-state modifier △Y=Ymin+a×(Ymax-Ymin)-Ysignal,

For amplitude, amplitude squared, energy value or energy value squared, increase or decrease the corresponding steady-state modifier ΔY.

The adjustment target coefficient a is obtained by adding a fixed value and a random value generated by a probability model, and satisfies 0<a<1. For example, a=m1+m2, where m1 is a fixed value and m2 is a random value generated by a probability model.

In step 105, the target model is trained based on the normal sample set, the abnormal sample set and a preset training algorithm, so as to obtain a trained target model. Wherein, the preset training algorithm may be any reasonable training algorithm in the field of artificial intelligence, deep learning, and machine learning algorithm. After obtaining the trained target model, input the actual signal generated by the target device in actual operation into the trained target model, so as to use the trained target model to diagnose, judge or identify the operating state of the target device, Therefore, it is determined that the target device is in a normal state, an abnormal state or any other state. That is, the target model can diagnose, judge or identify the operating state of the target device according to the collected sample signal, so as to determine whether the target device is in a normal state, an abnormal state or any other state.

Inputting the pre-stored first test sample set into the trained target model, so that the trained target model can diagnose, judge or identify the result state of the target device, and obtain a verification state corresponding to the pre-stored first test sample set. Wherein the verification state may correspond to each sample signal group or sample subset of the first test sample set respectively.

Determine the degree of difference between the verification state and the result state, and when the degree of difference is less than or equal to a predetermined threshold, determine that the trained target model meets the requirements. Since the result state of the object equipment corresponding to each sample signal group or sample subset in the first test sample set is known or predetermined, the difference between the verification state and the result state can be used as the accurate target model for determining basis of degree. For example, when a specific set of test sample signals is input to the trained target model, the obtained verification state is that the ratio of the normal state to the abnormal state of the object equipment is 95:5, while the known or predetermined result state of the object The ratio of normal state to abnormal state of equipment is 99:1. It can be seen that the degree of difference is 4%=(99-95)/(99+1). The predetermined threshold is a preset difference threshold, and the predetermined threshold or the difference threshold may be considered as the minimum requirement for the accuracy of diagnosis, discrimination or identification of the target model. For example, when the predetermined threshold is 5%, the degree of difference is 4%, and the degree of difference is less than the predetermined threshold, then it is determined that the trained target model meets the requirements. When the predetermined threshold is 2%, the degree of difference is 4%, and if the degree of difference is greater than the predetermined threshold, it is determined that the trained target model does not meet the requirements.

Alternatively, after obtaining the normal sample set including the at least two sample signals, and obtaining the abnormal sample set corresponding to the normal sample set, the normal sample set is divided into a first normal sample subset and a second normal sample set Subsets, dividing the set of abnormal samples into a first subset of abnormal samples and a second subset of abnormal samples. The first quantity ratio of the sample signals (sample signals of normal samples or normal sample signals) in the first normal sample subset and the second normal sample subset is any reasonable ratio such as 3:7, 5:5, 6:4, etc. Wherein the second quantity ratio of the sample signals (sample signals of abnormal samples or abnormal sample signals) in the first abnormal sample subset and the second abnormal sample subset is any reasonable ratio such as 3:7, 5:5, 6:4, etc. Wherein, the first quantity ratio and the second quantity ratio may be equal or unequal.

The target model is trained based on the first normal sample set, the first abnormal sample set and a preset training algorithm, so as to obtain a trained target model. Among them, the preset training algorithm can be any reasonable training algorithm in the field of artificial intelligence, deep learning, and machine learning algorithm.

Preferably, the second set of normal samples and the second sub-set of abnormal samples are used to form a second test sample set, and the second test sample set is input into the trained target model, so that the trained target model can diagnose, distinguish or identify the target device The result state is to obtain the verification state corresponding to the pre-stored second test sample set; and determine the degree of difference between the verification state and the result state. When the degree of difference is less than or equal to a predetermined threshold, it is determined that the trained target model meets the requirements; when the degree of difference is greater than the predetermined threshold, it is determined that the trained target model does not meet the requirements.

Preferably, it also includes, by reducing the modification amount, obtaining an adjusted abnormal sample set, and training the target model based on the normal sample set, the adjusted abnormal sample set and a preset training algorithm, so as to obtain the trained high-precision target model. For example, after increasing or decreasing the corresponding modification amount for the amplitude, amplitude square, energy value or energy value square, so that after increasing or decreasing the magnitude of the corresponding modification amount, by reducing the modification amount, the obtained Adjusted set of outlier samples.

It also includes obtaining an adjusted abnormal sample set by reducing the ratio, and training the target model based on the normal sample set, the adjusted abnormal sample set and a preset training algorithm, thereby obtaining a trained high-precision target Model. For example, after increasing or decreasing the corresponding ratio for the amplitude, the square of the amplitude, the energy value or the square of the energy value, it further includes: reducing the ratio to obtain an adjusted abnormal sample set.

FIG. 2 is a flowchart of a method 200 for testing a target model based on modification of sample signals according to an embodiment of the present invention. Method 200 starts at step 201 .

In step 201, a target model to be tested is determined and a configuration file of the target model is obtained. In the actual scene of industrial production or equipment operation, various types and/or various sizes of equipment are widely used in various locations, production links, monitoring links, etc. For this reason, if it is necessary to determine the operating status of any type of equipment, or obtain parameters of any type of equipment, etc., it is necessary to obtain the model or equipment model of each different type of equipment. Generally, the model of each different type of equipment or the equipment model can be used to determine the operating state of the equipment, obtain the operating parameters of the equipment, and so on. To this end, before testing, determine the target model that needs to be tested.

Typically, each model or device model has a configuration file, and the configuration file is used to describe various properties of the model or device model. Various attributes are, for example: input parameters, output parameters, model type, model action, model accuracy, device type, device name, device identifier, etc. To this end, the target model has various attributes, and for example, the device type, device name, device identifier, etc. of the target model can be determined via the configuration file of the target model.

In step 202, an object device involved in the object model is determined based on the configuration file of the object model, and various sample signals associated with the object device are determined. As mentioned above, the device type, device name, device identifier, etc. of the target model can be determined through the configuration file of the target model. And further, the target device involved in the target model may be determined by parsing the configuration file of the target model. The target device may be any type of device. In addition, the configuration file of the target model may also include information of various sample signals associated with the target device. Alternatively, after the target device involved in the target model is determined, the device identifier or device name of the target device can be used to search in the sample signal information library to obtain information on various sample signals associated with the target device .

The various sample signals include: vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness. In general, one or more sample signals of vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness can be used to characterize, train, test, describe target device. It should be understood that this application is only described by taking vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness as examples, and those skilled in the art should understand Note that this application can use any reasonable sample signal. In practical scenarios, various types of sensors can be used to acquire any one of vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image, and brightness.

In step 203, when the target device is running normally, signal acquisition or signal simulation is performed on at least two sample signals among the various sample signals, so as to obtain a normal sample set including the at least two sample signals.

For example, at least one sample signal among the at least two sample signals is a vibration/acoustic emission signal collected by a sensor close to the device casing. Alternatively, at least one of the at least two sample signals is a sound signal collected outside the device. In an actual situation, the sensor may be placed close to the casing of the device or the target device, placed outside the device or the target device, or placed inside the device or the target device.

In step 204, at least one sample signal in the normal sample set is modified according to a preset modification manner, so as to obtain an abnormal sample set corresponding to the normal sample set. In actual situations, equipment with a stable operating state has fewer failures or a lower failure rate in actual operation, so the sample signal/sample data volume of this type of equipment is relatively large for normal operation, while abnormal operation Or the data amount of the sample signal/sample data at the time of failure is small. In this case, it is often difficult to obtain enough abnormal sample signals. To this end, the present application modifies at least one sample signal in the normal sample set according to a preset modification manner.

The preset modification methods include: increasing or decreasing the corresponding ratio for the amplitude, the square of the amplitude, the energy value or the square of the energy value, so that the amplitude, the square of the amplitude, or the energy value of the corresponding ratio can be increased or decreased Or the value or value range of the square of the energy value changes. For example, the amplitude sample signal in the normal sample set of the target device increases or decreases by the corresponding ratio e, the amplitude square sample signal in the normal sample set of the target device increases or decreases by the corresponding rate f, and the target device The energy value sample signal in the normal sample set of the target device increases or decreases by the corresponding ratio g, or the energy value square sample signal in the normal sample set of the target device increases or decreases by the corresponding rate h, so that the normal sample set of the target device The amplitude sample signal, the amplitude square sample signal, the energy value sample signal or the energy value square sample signal in becomes the sample signal when the abnormal operation or faulty operation occurs. Thus, the abnormal sample set corresponding to the normal sample set is formed from the amplitude sample signal, the amplitude square sample signal, the energy value sample signal or the energy value square sample signal when the target device is abnormally or malfunctioning.

Wherein is the amplitude, the square of the amplitude, the energy value or the square of the energy value, increasing or decreasing the corresponding steady-state modifier includes:

Statistically calculate the maximum value Ymax and minimum value Ymin of the corresponding parameters of the amplitude, amplitude square, energy value, or steady-state modifier △Y of the energy value square in the normal sample set;

Set the adjustment target coefficient a, where 0<a<1;

Count the value Ysignal of the corresponding parameter of the steady-state modifier △Y of the sample to be modified;

Calculate the steady-state modifier △Y=Ymin+a×(Ymax-Ymin)-Ysignal,

In step 205, the target model is tested based on the normal sample set and the abnormal sample set, so as to determine the performance index of the target model based on the test results. The target model can diagnose, judge or identify the operating state of the target device according to the collected sample signals, so as to determine whether the target device is in a normal state, an abnormal state or any other state. The performance index of the target model may include the diagnostic accuracy, discrimination accuracy or recognition accuracy of the target model.

The target model is tested based on the normal sample set and the abnormal sample set, so as to determine the performance index of the target model based on the test results. Diagnose, judge or identify the result status of the target device, and obtain the verification status corresponding to the normal sample set and/or the abnormal sample set. Determining the correct ratio of the result status based on the verification status, determining the test result based on the correct ratio and determining the performance index of the target model based on the test result. For example, after the normal sample set and the abnormal sample set are respectively or sequentially input into the trained target model, the trained target model diagnoses, discriminates or identifies the result status of the target device 100 times, and determines the 100 result status diagnosis based on the verification status , the number of correct discrimination or recognition is 99 times, then the correct ratio of determining the result state based on the verification state is 99/100=99%, then the test result is 99%.

According to one embodiment, when the test result is greater than or equal to 97%, it is determined that the performance index of the target model is high precision, and when the test result is less than 97% and greater than or equal to 90%, it is determined that the performance index of the target model is medium precision, and When the test result is less than 90%, it is determined that the performance index of the target model is low precision.

Preferably, it also includes obtaining an adjusted abnormal sample set by reducing the ratio, and testing the target model based on the normal sample set and the adjusted abnormal sample set, so as to determine the diagnostic accuracy, discrimination accuracy or recognition accuracy of the target model precision. For example, after increasing or decreasing the corresponding ratio for the amplitude, the square of the amplitude, the energy value or the square of the energy value, it further includes: reducing the ratio to obtain an adjusted abnormal sample set.

Preferably, it also includes, by reducing the modification amount, obtaining an adjusted abnormal sample set, and testing the target model based on the normal sample set and the adjusted abnormal sample set, so as to determine the diagnostic accuracy, discrimination accuracy or recognition accuracy. For example, after increasing or decreasing the corresponding modification amount for the amplitude, amplitude square, energy value or energy value square, so that after increasing or decreasing the magnitude of the corresponding modification amount, by reducing the modification amount, the obtained Adjusted set of outlier samples.

FIG. 3 is a schematic structural diagram of a system 300 for training a target model based on modification of sample signals according to an embodiment of the present invention. The system 300 includes: selection means 301 , determination means 302 , acquisition means 303 , modification means 304 , training means 305 and recognition means 306 .

The selecting means 301 is configured to select a target model to be trained from multiple models, and obtain a configuration file of the target model. In the actual scene of industrial production or equipment operation, various types and/or various sizes of equipment are widely used in various locations, production links, monitoring links, etc. For this reason, if it is necessary to determine the operating status of any type of equipment, or obtain parameters of any type of equipment, etc., it is necessary to obtain the model or equipment model of each different type of equipment. Generally, the model of each different type of equipment or the equipment model can be used to determine the operating state of the equipment, obtain the operating parameters of the equipment, and so on. For this reason, when it is necessary to train or test a model of a specific device, it is necessary to select a target model to be trained from models associated with multiple different devices or device models.

The determining unit 302 is configured to determine an object device involved in the object model based on the configuration file of the object model, and determine various sample signals associated with the object device. As mentioned above, the device type, device name, device identifier, etc. of the target model can be determined through the configuration file of the target model. And further, the target device involved in the target model may be determined by parsing the configuration file of the target model. The target device may be any type of device. In addition, the configuration file of the target model may also include information of various sample signals associated with the target device. Alternatively, after the target device involved in the target model is determined, the device identifier or device name of the target device can be used to search in the sample signal information library to obtain information on various sample signals associated with the target device .

The obtaining means 303 is configured to perform signal collection or signal simulation on at least two sample signals among the various sample signals when the target device is running normally, so as to obtain a normal sample set including the at least two sample signals.

The modifying means 304 is configured to modify at least one sample signal in the normal sample set according to a preset modification method, so as to obtain an abnormal sample set corresponding to the normal sample set. In actual situations, equipment with a stable operating state has fewer failures or a lower failure rate in actual operation, so the sample signal/sample data volume of this type of equipment is relatively large for normal operation, while abnormal operation Or the data amount of the sample signal/sample data at the time of failure is small. In this case, it is often difficult to obtain enough abnormal sample signals. To this end, the present application modifies at least one sample signal in the normal sample set according to a preset modification manner.

Calculate the steady-state modifier △Y=Ymin+a×(Ymax-Ymin)-Ysignal,

The training device 305 is configured to train the target model based on the normal sample set, the abnormal sample set and a preset training algorithm, so as to obtain a trained target model. Wherein, the preset training algorithm may be any reasonable training algorithm in the field of artificial intelligence, deep learning, and machine learning algorithm. After obtaining the trained target model, input the actual signal generated by the target device in actual operation into the trained target model, so as to use the trained target model to diagnose, judge or identify the operating state of the target device, Therefore, it is determined that the target device is in a normal state, an abnormal state or any other state. That is, the target model can diagnose, judge or identify the operating state of the target device according to the collected sample signals, so as to determine whether the target device is in a normal state, an abnormal state or any other state.

The identification means 306 is configured to input the actual signal generated by the target equipment during actual operation into the trained target model after obtaining the trained target model, so as to use the trained target model to analyze the operating state of the target device Diagnose, judge or identify, so as to determine whether the target device is in a normal state or an abnormal state.

FIG. 4 is a schematic structural diagram of a system 400 for testing a target model based on modification of sample signals according to an embodiment of the present invention. The system 400 includes: obtaining means 401 , determining means 402 , obtaining means 403 , modifying means 404 , and testing means 405 .

Obtaining means 401, configured to determine a target model that needs to be tested and obtain a configuration file of the target model. In the actual scene of industrial production or equipment operation, various types and/or various sizes of equipment are widely used in various locations, production links, monitoring links, etc. For this reason, if it is necessary to determine the operating status of any type of equipment, or obtain parameters of any type of equipment, etc., it is necessary to obtain the model or equipment model of each different type of equipment. In general, a model of each different type of equipment or equipment model can be used to determine the operating status of the equipment, obtain the operating parameters of the equipment, and so on. To this end, before testing, determine the target model that needs to be tested.

The determining unit 402 is configured to determine an object device involved in the object model based on the configuration file of the object model, and determine various sample signals associated with the object device. As mentioned above, the device type, device name, device identifier, etc. of the target model can be determined through the configuration file of the target model. And further, the target device involved in the target model may be determined by parsing the configuration file of the target model. The target device may be any type of device. In addition, the configuration file of the target model may also include information of various sample signals associated with the target device. Alternatively, after the target device involved in the target model is determined, the device identifier or device name of the target device can be used to search in the sample signal information library to obtain information on various sample signals associated with the target device .

The obtaining means 403 is configured to perform signal collection or signal simulation on at least two sample signals among the various sample signals when the target device is running normally, so as to obtain a normal sample set including the at least two sample signals.

The normal sample set is one of the data collected by sensors such as vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness when the target device is in normal operation. A sample signal composed of one or more sample signals or data. In the present application, in order to make the model accuracy of the trained or tested target model higher, the present application performs signal acquisition or signal simulation on at least two sample signals among various sample signals, so as to obtain the Normal sample collection of sample signals.

The modification means 404 is configured to modify at least one sample signal in the normal sample set according to a preset modification method, so as to obtain an abnormal sample set corresponding to the normal sample set.

In actual situations, equipment with a stable operating state has fewer failures or a lower failure rate in actual operation, so the sample signal/sample data volume of this type of equipment is relatively large for normal operation, while abnormal operation Or the data amount of the sample signal/sample data at the time of failure is small. In this case, it is often difficult to obtain enough abnormal sample signals. To this end, the present application modifies at least one sample signal in the normal sample set according to a preset modification manner.

The maximum value Ymax and the minimum value Ymin of the parameters corresponding to the steady-state modifier △Y of the amplitude, amplitude square, energy value or energy value square in the normal sample set are counted;

Set the adjustment target coefficient a, where 0<a<1;

Calculate the steady-state modifier △Y=Ymin+a×(Ymax-Ymin)-Ysignal,

The testing device 405 is configured to test the target model based on the normal sample set and the abnormal sample set, so as to determine the performance index of the target model based on the test results. The target model can diagnose, judge or identify the operating state of the target device according to the collected sample signals, so as to determine whether the target device is in a normal state, an abnormal state or any other state. The performance index of the target model may include the diagnostic accuracy, discrimination accuracy or recognition accuracy of the target model.

The target model is tested based on the normal sample set and the abnormal sample set, so as to determine the performance index of the target model based on the test results. Diagnose, judge or identify the result status of the target device, and obtain the verification status corresponding to the normal sample set and/or the abnormal sample set. Determining the correct ratio of the result status based on the verification status, determining the test result based on the correct ratio and determining the performance index of the target model based on the test result. For example, after the normal sample set and the abnormal sample set are respectively or sequentially input into the trained target model, the trained target model diagnoses, discriminates or identifies the result status of the target device 100 times, and determines the 100 result status diagnosis based on the verification status , the number of correct discrimination or recognition is 99 times, then the correct ratio of the result state determined based on the verification state is 99/100=99%, then the test result is 99%.

Claims

A method for training a target model based on modification of a sample signal, the method comprising:

Select a target model to be trained from multiple models, and obtain a configuration file of the target model;

determining an object device involved in the object model based on the configuration file of the object model, and determining various sample signals associated with the object device;

When the subject device is running normally, perform signal acquisition or signal simulation on at least two sample signals among the various sample signals, so as to obtain a normal sample set including the at least two sample signals;

Modifying at least one sample signal in the normal sample set according to a preset modification method to obtain an abnormal sample set corresponding to the normal sample set;

The target model is trained based on the normal sample set, the abnormal sample set and the preset training algorithm, so as to obtain the trained target model.
The method according to claim 1, the plurality of sample signals include: vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness.
According to the method according to claim 1, at least one sample signal among the at least two sample signals is a vibration/acoustic emission signal collected by a sensor close to the device casing.
The method of claim 1, wherein at least one of the at least two sample signals is a sound signal collected externally to the device.
A method for testing a target model based on modification of a sample signal, the method comprising:

Determine the target model that needs to be tested and obtain the configuration file of the target model;

determining an object device involved in the object model based on the configuration file of the object model, and determining various sample signals associated with the object device;

When the subject device is running normally, perform signal acquisition or signal simulation on at least two sample signals among the various sample signals, so as to obtain a normal sample set including the at least two sample signals;

Modifying at least one sample signal in the normal sample set according to a preset modification method to obtain an abnormal sample set corresponding to the normal sample set;

The target model is tested based on the normal sample set and the abnormal sample set, so as to determine the performance index of the target model based on the test results.
A system for training a target model based on modification of a sample signal, the system comprising:

a selection device, configured to select a target model to be trained from multiple models, and obtain a configuration file of the target model;

determining means, configured to determine the target device involved in the target model based on the configuration file of the target model, and determine various sample signals associated with the target device;

An acquisition device, configured to perform signal acquisition or signal simulation on at least two sample signals among multiple sample signals when the subject device is operating normally, so as to acquire a normal sample set including the at least two sample signals;

A modifying device, configured to modify at least one sample signal in the normal sample set according to a preset modification method, so as to obtain an abnormal sample set corresponding to the normal sample set;

The training device is used to train the target model based on the normal sample set, the abnormal sample set and a preset training algorithm, so as to obtain the trained target model.
The system according to claim 5, the plurality of sample signals include: vibration, sound, velocity, displacement, stress, pressure, voltage, current, power, electric field strength, magnetic field strength, temperature, image and brightness.
The system according to claim 5, at least one sample signal among the at least two sample signals is a vibration/acoustic emission signal collected by a sensor close to the device casing.
The system of claim 5, at least one of the at least two sample signals is a sound signal collected externally to the device.
A system for testing a target model based on modification of a sample signal, the system comprising:

Obtaining means for determining a target model to be tested and obtaining a configuration file of the target model;

determining means, configured to determine the target device involved in the target model based on the configuration file of the target model, and determine various sample signals associated with the target device;

An acquisition device, configured to perform signal acquisition or signal simulation on at least two sample signals among multiple sample signals when the subject device is operating normally, so as to acquire a normal sample set including the at least two sample signals;

A modifying device, configured to modify at least one sample signal in the normal sample set according to a preset modification method, so as to obtain an abnormal sample set corresponding to the normal sample set;

The testing device is used for testing the target model based on the normal sample set and the abnormal sample set, so as to determine the performance index of the target model based on the test result.