WO2023097774A1

WO2023097774A1 - Method and system for generating fault data of industrial robot, terminal, and storage medium

Info

Publication number: WO2023097774A1
Application number: PCT/CN2021/137927
Authority: WO
Inventors: 郭媛君; 杨之乐; 安钊; 刘祥飞; 冯伟; 王尧
Original assignee: 深圳先进技术研究院
Priority date: 2021-11-30
Filing date: 2021-12-14
Publication date: 2023-06-08
Also published as: CN114326655A

Abstract

The present application relates to a method and system for generating fault data of an industrial robot, a terminal, and a storage medium. The method comprises: extracting real fault data of an industrial robot, labeling category tags for the real fault data according to fault categories, and using the category tags as condition information and generating a real fault data set together by the condition information and the real fault data; inputting the real fault data set into a conditional adversarial generative network for training to obtain a trained generative adversarial model; and generating different categories of fault data of the industrial robot according to the trained generative adversarial model. Embodiments of the present application improve the quality of fault data generation, expand fault data of a core part of an industrial robot, and facilitate the improvement of the efficiency of industrial robot operation state monitoring and the accuracy of system fault diagnosis.

Description

Industrial robot fault data generation method, system, terminal and storage medium

technical field

The application belongs to the technical field of mechanical engineering, and in particular relates to a method, system, terminal and storage medium for generating fault data of an industrial robot.

Background technique

In recent years, with the support of technologies such as neural networks and machine vision, industrial robots have developed towards a highly anthropomorphic and intelligent direction. Because industrial robots can produce efficiently around the clock, ensuring the safety of the entire product production system of an enterprise and keeping industrial robots in an efficient working state has also received widespread attention from enterprises and researchers. For enterprises, once the industrial robot system fails, it will cause the production of the entire production line to stagnate. If the faulty robot is not repaired in time, the robot failure may turn into a huge production accident, and even pose a threat to the life safety of the company's staff. Since robot failure will cause unpredictable consequences, it is particularly important to carry out research on industrial robot fault diagnosis systems and reduce the human and material resources consumed by enterprises in dealing with industrial robot faults.

Fault diagnosis technology studies the response of changes in machine or unit operating status in diagnostic information. However, most of the fault data obtained from industrial robots belong to the data under normal working conditions, and there are only a small amount of fault precursor data. The effective and available fault data of industrial robots is very scarce, which will greatly affect the training of deep neural networks. As a result, the obtained industrial robot fault diagnosis model cannot be practically applied due to weak generalization ability and insufficient expression ability.

Most of the existing fault data generation technologies use transfer learning or conventional generative adversarial networks for data generation. The disadvantage of transfer learning is that it is usually only suitable for processing limited small data sets, and the "knowledge" in other fields is not necessarily feasible in a specific field. Conventional generative adversarial networks use the generator and the discriminator to learn against each other for training, and there is a large distribution difference between the generated fault data and the actual fault data.

Contents of the invention

The present application provides a method, system, terminal and storage medium for generating fault data of industrial robots, aiming to solve one of the above-mentioned technical problems in the prior art at least to a certain extent.

In order to solve the above problems, the application provides the following technical solutions:

A method for generating fault data of an industrial robot, comprising:

Extracting the real fault data of the industrial robot, marking the real fault data with a category label according to the fault category, and using the category label as condition information together with the real fault data to generate a real fault data set;

The real fault data set is input into the conditional confrontation generation network for training, and the trained generation confrontation model is obtained;

Generate different types of industrial robot fault data according to the trained generation confrontation model.

The technical solution adopted in the embodiment of the present application further includes: the generating the real fault data set together with the real fault data using the category label as condition information includes:

Carrying out time-frequency domain feature extraction on the real fault data, and forming a real fault data set according to the extracted features;

classify the real fault data into fault categories, and put a category label on each real fault data in the real fault data set;

Perform conditional processing on the category labels, and use the conditional processed category labels as condition information and the real fault data to form a new real fault data set.

The technical solution adopted in the embodiment of the present application also includes: the training of the real fault data set input condition against the generation network includes:

The real fault data and condition information are input into the generator of the confrontation generating network for training, and the simulated fault data of different categories are generated by the generator;

The simulated fault data and condition information are input into a discriminator for training, and the discriminator outputs a discrimination result.

The technical solution adopted in the embodiment of the present application also includes: the training of the real fault data set into the conditional adversarial generation network further includes:

Using the stochastic gradient descent method to update the discriminator, and judge whether the number of training times of the discriminator reaches k times based on the hyperparameter k, if it does not reach k times, then train the discriminator again; if it reaches k Second-rate,

The generator is updated by stochastic gradient descent method, and the generator is retrained.

The technical solution adopted in the embodiment of the present application also includes: the update of the discriminator by using the stochastic gradient descent method is specifically:

The discriminator is updated by adding stochastic gradients.

The technical solution adopted in the embodiment of the present application also includes: the update of the generator using the stochastic gradient descent method is specifically:

The generator is updated by subtracting the stochastic gradient.

The technical solution adopted in the embodiment of the present application also includes: the loss function of the conditional confrontation generation network is:

Among them, D represents the discriminator, G represents the generator, y represents the conditional information, G(z|y) represents the conditional information and the noise signal is input into the generator G together, and D(G(z|y))) represents the generator The simulated fault data generated by G is then output to the discriminator D to determine the authenticity, and D(x|y) represents the real fault data x and the condition information y are input to the discriminator D to determine the authenticity;

Represents the probability that the data comes from real fault data after being judged by the discriminator D,

Represents the probability that the data comes from the simulated fault data generated by the generator G after being judged by the discriminator D; log is logarithm.

Another technical solution adopted in the embodiment of the present application is: an industrial robot fault data generation system, including:

Data extraction module: used to extract the real fault data of the industrial robot, mark the real fault data with a category label according to the fault category, and use the category label as condition information together with the real fault data to generate a real fault data set;

Model training module: used to input the real fault data set into the conditional confrontation generation network for training to obtain a trained generation confrontation model;

Data generation module: used to generate different types of industrial robot fault data according to the trained generation confrontation model.

Another technical solution adopted by the embodiment of the present application is: a terminal, the terminal includes a processor and a memory coupled to the processor, wherein,

The memory stores program instructions for realizing the method for generating fault data of an industrial robot;

The processor is configured to execute the program instructions stored in the memory to control the generation of industrial robot fault data.

Another technical solution adopted in the embodiment of the present application is: a storage medium storing program instructions executable by a processor, and the program instructions are used to execute the method for generating fault data of an industrial robot.

Compared with the prior art, the beneficial effect of the embodiment of the present application lies in that the industrial robot fault data generation method, system, terminal and storage medium of the embodiment of the present application mark different types of real fault data with category labels, and use the category labels as The condition information is input into the condition generation confrontation network, so as to control the condition generation confrontation network to generate different types of fault data of the core components of industrial robots, so that the generated data can be controlled, the quality of fault data generation is improved, and the faults of the core components of industrial robots are expanded. The data helps to improve the efficiency of monitoring the operating status of industrial robots and the accuracy of system fault diagnosis. At the same time, the embodiment of the present application introduces the hyperparameter k, so that the training speed of the discriminator is accelerated, and the training efficiency of the conditional generative adversarial network is higher.

Description of drawings

Fig. 1 is the flow chart of the industrial robot fault data generating method of the embodiment of the present application;

Fig. 2 is the structural representation of the industrial robot fault data generating system of the embodiment of the present application;

FIG. 3 is a schematic diagram of a terminal structure in an embodiment of the present application;

FIG. 4 is a schematic structural diagram of a storage medium according to an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, not to limit the present application.

In view of the deficiencies in the prior art, the industrial robot fault data generation method of the embodiment of the present application uses a conditional confrontation generation network to generate fault data. This method adopts the idea of "game theory" and generates a small amount of The real fault data and fault category labels of the conditional generative confrontation network are used for confrontation training, and finally a powerful generative confrontation model is obtained, and then the fault data similar to the real fault data structure distribution of industrial robots is generated by the generative confrontation model, which solves the problem of industrial robots. The problem of insufficient fault data improves the generalization ability and expression ability of the industrial robot fault diagnosis model, and at the same time enhances the practicability of the industrial robot fault diagnosis model.

Specifically, please refer to FIG. 1 , which is a flowchart of a method for generating fault data of an industrial robot according to an embodiment of the present application. The industrial robot fault data generating method of the embodiment of the present application comprises the following steps:

S10: Extract the real fault data of the core components of the industrial robot, classify the real fault data, label each real fault data according to the classification results, and use the category labels as condition information together with the real fault data to generate a real fault data set;

In this step, the generation process of the real fault data set specifically includes the following steps:

S11: Perform time-frequency domain feature extraction on the extracted real fault data, and form a real fault data set according to the extracted features;

S12: classify the real fault data into fault categories, and put a category label on each real fault data in the real fault data set;

S13: Carry out conditional processing on the category label, and use the category label as condition information together with the real fault data to form a new real fault data set;

Among them, the conditional processing method of the category label is as follows: digitize the category label, for example: name the first type of fault data as 1, the second type of fault data as 2, and so on, so that the real fault data and The label information after conditional processing is input into the conditional confrontation generation network together.

S20: Input the real fault data set into the conditional confrontation generation network for training, and use the generator of the conditional confrontation generation network to form a joint hidden layer representation of the input real fault data and condition information, and generate different types of simulated fault data;

In this step, the generation adversarial loss of the conditional adversarial generation network is a loss function, and the distribution error between the simulated fault data generated by the generator and the marked real fault data is measured through the loss function, and the generator and the discriminator are optimized. The loss function is defined as:

Among them, D represents the discriminator, G represents the generator, y represents the conditional information (category label), G(z|y) represents the conditional information and the noise signal is input into the generator G together, D(G(z|y)) ) means that the simulated fault data generated by the generator G is output to the discriminator D to determine the authenticity.

Represents the probability that the data comes from the simulated fault data generated by the random noise generator G after being judged by the discriminator D. In the embodiment of the present application, the maximum and minimum strategy is selected for network training, so that the ability of the discriminator D to distinguish authenticity becomes stronger and stronger, and the generation quality of the generator G is more and more similar to the real fault data x. The logarithm log is added for the convenience of calculation.

S30: Input the simulated fault data and condition information generated by the generator into the discriminator for training, and output the discrimination result through the discriminator;

S40: Updating the discriminator by using the stochastic gradient descent method through the optimizer;

In this step, the optimizer updates the discriminator by adding stochastic gradients.

S50: Based on the hyperparameter k, it is judged whether the number of training times of the discriminator reaches k times, if not, re-execute S30; if it reaches k times, execute S60;

S60: Update the generator by using the stochastic gradient descent method through the optimizer, and re-execute S20;

In the embodiment of this application, the optimizer updates the generator by subtracting the stochastic gradient. In the embodiment of the present application, by introducing the hyperparameter k, first train the discriminator k times, and then train the generator once, until the distribution difference between the simulated fault data generated by the generator and the real fault data gradually decreases. The difference in real data makes the distribution difference between the generated fault data output by the network and the real fault data gradually decrease. Updating the discriminator by accelerating training makes the conditional adversarial generative network converge faster.

S70: cyclically execute S20 to S60, until the number of training times of the conditional confrontation generation network reaches the preset number of times, and obtain the trained generation confrontation model;

S80: Input the random noise and the conditionalized category labels into the trained generative confrontation model, and generate different categories of fault data similar to the real fault data structure distribution of the industrial robot through the generative confrontation model.

Based on the above, the industrial robot fault data generation method of the embodiment of the present application labels different categories of real fault data, and inputs the category labels as condition information into the condition generation adversarial network, so as to control the condition generation adversarial network to generate different categories The fault data of core components of industrial robots makes the generated data controllable, improves the quality of fault data generation, expands the fault data of core components of industrial robots, and helps to improve the efficiency of monitoring the operating status of industrial robots and the accuracy of system fault diagnosis . At the same time, the embodiment of the present application introduces the hyperparameter k, so that the training speed of the discriminator is accelerated, and the training efficiency of the conditional generative adversarial network is higher.

Please refer to FIG. 2 , which is a schematic structural diagram of an industrial robot fault data generation system according to an embodiment of the present application. The industrial robot fault data generation system 40 of the embodiment of the present application includes:

Data extraction module 41: used to extract the real fault data of the industrial robot, mark the real fault data with category labels according to the fault category, and use the category labels as condition information together with the real fault data to generate a real fault data set;

Model training module 42: used to input the real fault data set into the conditional confrontation generation network for training, and obtain the trained generation confrontation model;

Data generating module 43: used to generate different types of industrial robot fault data according to the trained generative confrontation model.

Please refer to FIG. 3 , which is a schematic diagram of a terminal structure in an embodiment of the present application. The terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51 .

The memory 52 stores program instructions for realizing the above-mentioned method for generating fault data of an industrial robot.

The processor 51 is used to execute the program instructions stored in the memory 52 to control the generation of industrial robot fault data.

Wherein, the processor 51 may also be referred to as a CPU (Central Processing Unit, central processing unit). The processor 51 may be an integrated circuit chip with signal processing capability. The processor 51 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components . A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.

Please refer to FIG. 4 , which is a schematic structural diagram of a storage medium according to an embodiment of the present application. The storage medium of the embodiment of the present application stores a program file 61 capable of realizing all the above-mentioned methods, wherein the program file 61 can be stored in the above-mentioned storage medium in the form of a software product, and includes several instructions to make a computer device (which can It is a personal computer, a server, or a network device, etc.) or a processor (processor) that executes all or part of the steps of the methods in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. , or terminal devices such as computers, servers, mobile phones, and tablets.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in this application may be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to the embodiments shown in the present application, but is to be accorded the widest scope consistent with the principles and novel features disclosed in the present application.

Claims

A method for generating fault data of an industrial robot, comprising:

Extracting the real fault data of the industrial robot, marking the real fault data with a category label according to the fault category, and using the category label as condition information together with the real fault data to generate a real fault data set;

The real fault data set is input into the conditional confrontation generation network for training, and the trained generation confrontation model is obtained;

Generate different types of industrial robot fault data according to the trained generation confrontation model.
The method for generating fault data of an industrial robot according to claim 1, wherein said generating a real fault data set with said category label as condition information together with said real fault data comprises:

Carrying out time-frequency domain feature extraction on the real fault data, and forming a real fault data set according to the extracted features;

classify the real fault data into fault categories, and put a category label on each real fault data in the real fault data set;

Perform conditional processing on the category labels, and use the conditional processed category labels as condition information and the real fault data to form a new real fault data set.
The method for generating fault data of an industrial robot according to claim 2, wherein said inputting said real fault data set into a conditional confrontation generation network for training comprises:

The real fault data and condition information are input into the generator of the confrontation generating network for training, and the simulated fault data of different categories are generated by the generator;

The simulated fault data and condition information are input into a discriminator for training, and the discriminator outputs a discrimination result.
The method for generating fault data of an industrial robot according to claim 3, wherein said inputting said real fault data set into a conditional confrontation generation network for training further comprises:

Using the stochastic gradient descent method to update the discriminator, and judge whether the number of training times of the discriminator reaches k times based on the hyperparameter k, if it does not reach k times, then train the discriminator again; if it reaches k Second-rate,

The generator is updated by stochastic gradient descent method, and the generator is retrained.
The method for generating fault data of an industrial robot according to claim 4, wherein the updating of the discriminator by using the stochastic gradient descent method is specifically:

The discriminator is updated by adding stochastic gradients.
The method for generating fault data of an industrial robot according to claim 4, wherein the updating of the generator by using the stochastic gradient descent method is specifically:

The generator is updated by subtracting the stochastic gradient.
The method for generating fault data of an industrial robot according to any one of claims 1 to 6, wherein the loss function of the conditional confrontation generation network is:

Among them, D represents the discriminator, G represents the generator, y represents the conditional information, G(z|y) represents the conditional information and the noise signal is input into the generator G together, and D(G(z|y))) represents the generator The simulated fault data generated by G is then output to the discriminator D to determine the authenticity, and D(x|y) represents the real fault data x and the condition information y are input to the discriminator D to determine the authenticity;
Represents the probability that the data comes from real fault data after being judged by the discriminator D,
Represents the probability that the data comes from the simulated fault data generated by the generator G after being judged by the discriminator D; log is logarithm.
An industrial robot fault data generating system is characterized in that it comprises:

Data extraction module: used to extract the real fault data of the industrial robot, mark the real fault data with a category label according to the fault category, and use the category label as condition information together with the real fault data to generate a real fault data set;

Model training module: used to input the real fault data set into the conditional confrontation generation network for training to obtain a trained generation confrontation model;

Data generation module: used to generate different types of industrial robot fault data according to the trained generation confrontation model.
A terminal, characterized in that the terminal includes a processor and a memory coupled to the processor, wherein,

The memory is stored with program instructions for realizing the method for generating industrial robot fault data according to any one of claims 1-7;

The processor is configured to execute the program instructions stored in the memory to control the generation of industrial robot fault data.
A storage medium, characterized in that it stores program instructions executable by a processor, and the program instructions are used to execute the method for generating fault data of an industrial robot according to any one of claims 1 to 7.