WO2022037068A1

WO2022037068A1 - Method for diagnosis of fault in machine tool bearing

Info

Publication number: WO2022037068A1
Application number: PCT/CN2021/083476
Authority: WO
Inventors: 唐敦兵; 刘长春; 朱海华; 聂庆玮; 王立平; 宋家烨; 张毅
Original assignee: 南京航空航天大学
Priority date: 2021-03-22
Filing date: 2021-03-29
Publication date: 2022-02-24
Also published as: CN113128561A

Abstract

The invention relates to the technical field of intelligent manufacturing. Provided is a method for diagnosis of a fault in a machine tool bearing, comprising: establishing a digital twin workshop, and loading a machining task to a machine tool in the digital twin workshop, wherein the digital twin workshop corresponds to a real workshop; acquiring fault data of the machine tool during a process of the machine tool executing the machining task, and generating fault diagnosis models of a first type according to the acquired fault data; performing variable working condition transfer learning training on the generated fault diagnosis models, and obtaining fault diagnosis models of a second type, wherein the fault diagnosis models of the second type include a model for diagnosis of a fault in a machine tool bearing in the real workshop; and using the fault diagnosis models of the second type to detect a fault in the machine tool bearing in the real workshop. The diagnosis method is applicable to monitoring of a machine tool workshop.

Description

A method for fault diagnosis of machine tool bearings

technical field

The invention relates to the technical field of intelligent manufacturing, in particular to a fault diagnosis method for a machine tool bearing.

Background technique

As a typical manufacturing equipment, a machine tool is called the industrial mother machine. Failures in machine tools, if not corrected in a timely manner, can result in reduced accuracy and affect productivity and yield. Therefore, in intelligent manufacturing workshops, fault diagnosis methods are crucial for machine tools with bearings as key components.

In the current data-driven fault diagnosis model, machine learning algorithms and classifiers with signal processing methods are commonly used. The data-driven method can directly identify the fault from the bearing vibration signal collected by the sensor, so there is no need to understand the internal bearing of the bearing. structure.

In practical applications, the data-driven fault diagnosis scheme has achieved certain achievements and effects. But there are also some drawbacks, such as: traditional data-driven methods with machine learning algorithms strictly require that training and testing data must be under the same working conditions and have the same distribution and feature space. It is therefore not suitable for real-world working conditions that often change over time, making data difficult to obtain. At the same time, for these machine learning methods, it is first necessary to use enough training data to train the fault diagnosis model; then, the test data under the same working conditions is used to test the performance of the model; but the working conditions of the workshop machine tool bearings are not in the real world. may remain the same; as the bearing operating time increases, the fault diameter becomes larger and the load cannot be the same all the time. Moreover, the larger the scale and the more links of the automated production line, the more prominent this problem will be. This requires frequent collection of fault data, but in actual operation, the collection of fault data will also cause damage to actual production to a certain extent, such as causing equipment downtime, disruption of workshop production plans, etc. This kind of damage may not Serious but still affect the production accuracy. It can be seen that the traditional method is not suitable for working conditions that change with time, and there are great limitations.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a method for diagnosing machine tool bearing faults, which can alleviate the problem of damage to actual machine tool bearings in order to collect fault data.

To achieve the above object, the embodiments of the present invention adopt the following technical solutions:

Establish a digital twin workshop, and load the processing tasks of the machine tool in the digital twin workshop, wherein the digital twin workshop corresponds to the real workshop; collect the fault data of the machine tool during the operation of the processing task of the machine tool, and according to the collected fault data A first type of fault diagnosis model is generated, wherein the first type of fault diagnosis model includes: a fault diagnosis model of a machine tool bearing in a digital twin workshop; the generated fault diagnosis model is subjected to variable working condition transfer learning training, and the first type of fault diagnosis model is obtained. Two-type fault diagnosis model, wherein the second-type fault diagnosis model includes: a model for fault diagnosis of machine tool bearings in the actual workshop; using the second-type fault diagnosis model to detect machine tool bearings in the actual workshop failure.

The method for diagnosing machine tool bearing faults provided by the embodiments of the present invention proposes a method of integrating digital twins and fault diagnosis, and uses digital twins to simulate the state changes of bearings during actual machine tool processing to collect data of machine tool bearings in the digital twin space, so as to Training a model for fault diagnosis of machine tool bearings in virtual space alleviates the problem of damage to actual machine tool bearings in order to collect fault data. Among them, deep transfer learning is used to transfer the fault diagnosis model of machine tool bearings in the digital twin space to the machine tool bearings in the real workshop, and the model is retrained with a small amount of data of the machine tool bearings in the real workshop, and the fault diagnosis model of the machine tool bearings in the real workshop can be obtained. . It can avoid the time cost of collecting the full life cycle data of the bearing in the real workshop, and avoid the occurrence of equipment downtime and workshop production plan disruption due to the need to collect fault data.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a flowchart of a machine tool bearing fault diagnosis based on digital twin and deep transfer learning provided by an embodiment of the present invention;

Fig. 2 is the corresponding relation diagram of the digital twin workshop and the real workshop according to the embodiment of the present invention;

3 is a framework diagram of an improved deep residual learning algorithm provided by an embodiment of the present invention;

Fig. 4 is a dropout network architecture diagram provided by an embodiment of the present invention;

5 is a schematic diagram of a process visualization system based on a digital twin provided by an embodiment of the present invention;

6 is a schematic diagram of a machine tool bearing fault diagnosis system provided by an embodiment of the present invention;

FIG. 7 is a schematic diagram of a method flow according to an embodiment of the present invention.

detailed description

In order to make those skilled in the art better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Hereinafter, embodiments of the present invention will be described in detail, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but not to be construed as a limitation of the present invention. It will be understood by those skilled in the art that the singular forms "a", "an", "the" and "the" as used herein can include the plural forms as well, unless expressly stated otherwise. It should be further understood that the word "comprising" used in the description of the present invention refers to the presence of stated features, integers, steps, operations, elements and/or components, but does not exclude the presence or addition of one or more other features, Integers, steps, operations, elements, components and/or groups thereof. It will be understood that when we refer to an element as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Also, "connected" or "coupled" as used herein may include wireless connections or couplings. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items. It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as herein, are not to be taken in an idealized or overly formal sense. explain.

An embodiment of the present invention provides a fault diagnosis method for a machine tool bearing, as shown in FIG. 7 , including:

S1. Establish a digital twin workshop, and load the processing tasks of the machine tool in the digital twin workshop.

Among them, the digital twin workshop corresponds to the real workshop.

S2. Collect the fault data of the machine tool during the operation of the machining task of the machine tool, and generate a first-type fault diagnosis model according to the collected fault data.

Wherein, the first type of fault diagnosis model includes: a fault diagnosis model of a machine tool bearing in a digital twin workshop.

S3 , performing transfer learning training on the generated fault diagnosis model under variable working conditions, and obtaining a second type of fault diagnosis model.

Wherein, the second type of fault diagnosis model includes: a model for fault diagnosis of machine tool bearings in the real workshop.

S4. Use the second type of fault diagnosis model to detect the fault of the machine tool bearing in the real workshop.

As shown in FIG. 1 , in this embodiment, the machine tool bearing fault data is simulated and generated in the digital twin workshop, and a machine tool bearing fault diagnosis model in the digital twin space is established. According to a small amount of machine tool bearing fault data collected in the real workshop, the The machine tool bearing fault diagnosis model in the digital twin space is subjected to variable working condition transfer learning training, so as to generate a model that can adapt to the machine tool bearing fault diagnosis in the real workshop. The fault diagnosis model of the machine tool bearing under the current working condition can monitor the health status of the bearing in the virtual space and the real space through the assistance of the digital twin. In the specific implementation, technicians can design a digital twin model of workshop machine tools and internal components including bearings according to the actual situation of the workshop where the machine tool bearings are located. The digital twin model can collect workshop machine tool data in real time and reflect the production of workshop machine tools. situation. In the established digital twin workshop, the processing tasks of the machine tool are simulated and run, the bearing fault data of the machine tool during the simulation operation is collected, and the simulation data is pre-trained by deep learning, and the bearing fault diagnosis model in the digital twin workshop is obtained. The machine tool bearing fault diagnosis model in the digital twin space is subjected to variable working condition transfer learning training to generate a model that can adapt to the machine tool bearing fault diagnosis in the real workshop.

In this embodiment, the machine tool bearing fault data is generated by simulation in the digital twin workshop, and the machine tool bearing fault diagnosis model in the digital twin space is established. The fault diagnosis model performs transfer learning training for variable working conditions, thereby generating a model that can adapt to the fault diagnosis of machine tool bearings in real workshops. The fault diagnosis model of the machine tool bearing can monitor the health status of the bearing in the virtual space and the real space through the assistance of the digital twin.

Further, before establishing the digital twin workshop, it also includes:

S01. Acquire and collect point cloud data of the real workshop through a three-dimensional laser scanner installed in the real workshop.

The objects scanned by the three-dimensional laser scanner at least include: machine tools installed in the real workshop.

S02, using the point cloud data of the real workshop to establish a digital twin model.

Wherein, the objects represented by the digital twin model include at least: the overall structure and internal components of the machine tool installed in the real workshop, and the internal components include: the bearing of the machine tool.

Specifically in S02, the point cloud data of the real workshop is used to establish a digital twin model, including:

S021. Read the real-time running data of the machine tool equipment in the real workshop through the OPC UA communication framework.

S022. Store the real-time running data in a database and use it as the source data of the Unity data-driven engine.

Wherein, the Unity data-driven engine is used to drive the digital twin model. The data connection described in this embodiment can be understood as: using the three-dimensional laser scanning technology to build a digital twin model of the CNC equipment in the workshop, reading the real-time operating data of the machine tool equipment through the OPC UA communication framework, and converting the format into the database as a guide The source data of the Unity data-driven engine uses the Unity data-driven engine to drive the digital twin model, thereby realizing the synchronous movement of the physical model and the digital twin model, that is, data connectivity.

In the existing scheme to obtain fault data, taking the machine tool bearing failure as an example, it is necessary to carry out accelerated damage experiments on the machine tool bearing. However, in the solution of this embodiment, it is only necessary to perform model migration after training with the simulation data in the digital twin model. When the model is migrated, a small amount of fault data of the actual machine tool needs to be used for training, and then the model training of the actual machine tool can be completed. The real-time perception of data in this embodiment refers to the data collected by the sensor when the machine tool is working normally, including a large amount of normal data and a small amount of fault data, and there is no need to damage the machine tool bearing in order to obtain a large amount of machine tool fault data.

In this embodiment, a 3D laser scanner can be used to collect point cloud data from the real workshop, and preprocessing, registration, and splicing of point clouds can be performed to establish a digital twin model of workshop machine tools and internal components including bearings. The digital twin model realizes data connection with the actual machine tools in the real workshop. During the operation of the digital twin workshop, all the data of the actual machine tools in the real workshop will be sensed in real time and transmitted to the digital twin workshop. The digital twin model can collect data in real time and display it in real time The status of the actual machine tools in the workshop, through the real-time interaction between the actual machine tools in the workshop and the digital twin model, the two can grasp the dynamic changes of each other in time and respond in real time, and the production process is continuously optimized.

Specifically in S2, the failure data of the machine tool during the operation of the machining task of the machine tool is collected, and the first type of failure diagnosis model is generated according to the collected failure data, including:

S21. Import the machining tasks of the machine tool into the digital twin workshop for simulation, and collect simulation fault data as source domain data.

S22. After preprocessing the obtained source domain data, import a deep residual learning algorithm for training, and when the training accuracy rate reaches 100%, obtain the first type of fault diagnosis model.

The training data used in the transfer learning training in this embodiment is divided into "source domain data" and "target domain data", both of which are commonly used names in the field of machine learning. The "simulated fault data" simulated in this embodiment is used for training with the source domain data of the transfer learning; while the data actually collected by the machine tool is used for training with the target domain data of the transfer learning. In practical applications, the preprocessing of source domain data includes: using mean interpolation to deal with missing values and data normalization. For example, as shown in Figure 2, a large number of order processing tasks are simulated and run in the digital twin workshop, so as to collect a large number of simulated fault data of the machine tools in the digital twin workshop, so that enough source domain data can be obtained. Then, the collected source domain data, that is, a large number of simulated fault data of machine tools in the digital twin workshop, are preprocessed, and then imported into the improved deep residual learning algorithm for training. After the number of training times reaches 630, the training accuracy rate reaches 100 %, so as to draw the bearing fault diagnosis model in the digital twin workshop. Then, the fault diagnosis results are given through the model of machine tool bearing fault diagnosis in the digital twin workshop: normal data, inner ring fault, rolling element fault or center outer ring fault.

Further, the deep residual learning algorithm is imported for training as described in S221, including:

S2211. Use a dropout network structure to preprocess the fault data, wherein the fault data includes vibration data of the machine tool.

S2212. After completing the preprocessing, perform parameter structure setting on the fault diagnosis model.

S2213. Load the set parameters and train the fault diagnosis model.

In the preferred solution, in S2212, the parameter structure is set for the fault diagnosis model, including: setting the training algebra to 30 generations, the number of training times to 180 times, and the learning rate to 0.001, and after the training is completed, the set The parameter structure is stored.

S2213, in the process of training the fault diagnosis model, including:

Load the improved deep residual learning algorithm for training, wherein the improved deep residual learning algorithm includes:

y' ^(l) = r ^(l) ×y ^(l) (4)

Among them, in the process of training, each neuron is switched off probabilistically, f represents the activation function, and y represents the output value. z represents the value after the summation of neurons, i represents the number of neurons, r represents the probability of neuron shutdown (0 or 1), w represents the weight, b _i represents the bias of the ith neuron, w _i represents the weight of the ith neuron, _zi represents the sum of the weight of the ith neuron and the bias of the ith neuron, _yi represents the input signal, y' represents the input signal after dropout processing, z _i ' represents the value after the summation of neurons after dropout processing, and l represents the number of layers of the identity mapping layer of the residual neural network.

For example, the present embodiment may adopt the improved deep residual learning algorithm framework as shown in FIG. 3, and the training process of the improved deep residual learning algorithm includes:

During training, it is first necessary to standardize all the collected vibration data, and use the original data stitching method for image transformation. Specifically, in the present invention, the dropout network structure is used to directly utilize the original data, and training and prediction can be performed by simple processing, and the whole process does not require any time-frequency domain conversion and other signal processing techniques for the signal.

Secondly, after the data preprocessing is completed, the parameter structure of the model is set, the training algebra is set to 30 generations, the number of training times is set to 180 times, and the learning rate is 0.001. The parameters of the entire network after training can be saved as .m in Matlab file, as long as it is trained once, the trained weight file can be loaded for prediction in the same scene.

The saved weight file can be updated dynamically, that is, re-training is performed on the current training parameters, which potentially increases the training algebra of the model. When new faults are found, the model can be dynamically trained to make the model more and more perfect.

Finally, when the model reaches the learning rate of 0.001 or the maximum training times of 180 times, the model training is completed. Since the model training is all data with labels, the label is the fault type of the bearing. In this way, as long as the model is trained well, there is no need to rely on From the operator's experience, the output of the model is the failure type of the machine tool bearing. Specifically, the deep network structure of the diagnosis model, as shown in FIG. 4 is the dropout network architecture diagram provided by the embodiment of the present invention, which can better represent the complex nonlinear relationship between the bearing vibration signal and the bearing state.

In the improved deep residual learning algorithm, the calculation process of the implementation of the dropout layer adopts the calculation process shown in the above formulas (1) to (6). The training phase requires probabilistic shutdowns at each neuron. Equation (3) uses the Bernolli function. The function of the Bernoulli function is to generate a random vector of 0 and 1 of a certain length according to the probability. In addition, f represents the activation function. y stands for output. z represents the value after the summation of the neurons, which becomes the output through the activation function.

Specifically, in S3, the generated fault diagnosis model is subjected to variable working condition transfer learning training, and a second type of fault diagnosis model is obtained, including:

S31. Collect target domain data from the real workshop.

S32. Use the target domain data to perform sample migration training, and generate a second type of fault diagnosis model under variable working conditions.

Data preprocessing is performed on the target domain data collected from the real workshop. A small amount of target domain data is used for transfer learning training. The transfer training in the present invention selects sample transfer training, and the output label of the corresponding neural network will also be adjusted and modified accordingly according to the label of the target domain data. Continue to use sample transfer training to generate models that can adapt to fault diagnosis of machine tool bearings in real workshops under variable working conditions. The fault diagnosis results are given through the model of machine tool bearing fault diagnosis in the real workshop: normal data, inner ring fault, rolling element fault, center outer ring fault, orthogonal outer ring fault or positive outer ring fault. It should be noted that the various data types mentioned in this embodiment include: point cloud data collected from the real workshop; real-time operation data of machine tools in the real workshop (the real-time operation data includes normal operation data and Fault data); the fault data of the machine tool, the types of fault data can be various, such as the vibration data of the machine tool, especially the vibration data of the machine tool bearing. In practical applications, as shown in Figure 1, corresponding labels can be set for different faults; the training data of transfer learning is divided into "source domain data" and "target domain data", among which, the simulated "simulated fault data" is trained on the source domain data used for transfer learning. The actual data collected by the machine tool is used for training on the target domain data of transfer learning. The main idea of this embodiment is to perform transfer learning training by using a small amount of target domain data, that is, a small amount of target domain data collected, that is, fault data collected by an actual machine tool.

Further in S32, the sample transfer training includes:

Set the label of the source domain data to 0 and the label of the target domain data to 1. Divide the modeling sample into a first part and a second part, use the first part for modeling and predict the first fault diagnosis model through the second part, and then obtain the established model according to the prediction result Discrimination for source domain data and target domain data. That is, for the fault diagnosis model (the first fault diagnosis model) established by the simulated data, transfer learning is performed through the discrimination between the source domain data and the target domain data, so that the parameters in the first fault diagnosis model are optimized to suit the actual situation. A second type of fault diagnosis model for machine tool faults. After that, normalize the prediction results with the discrimination degree greater than the preset degree, and then bring the sample weight parameters of the second fault diagnosis model for training. Specifically, in order to improve the accuracy of the second fault diagnosis model, the discrimination The prediction results whose degree is greater than the preset degree are normalized and then brought into the second fault diagnosis model for model parameter optimization training. Wherein, the data volume of the first part is larger than that of the second part. For example, as shown in Figure 5:

This embodiment can be implemented as a process visualization system based on a digital twin, and the specific process of the transfer learning in S32 is as follows:

S321: Define the source domain data and target domain data collected by the machine tool bearing, the data label of the source domain is 0, and the data label of the target domain is 1;

S322: Carry out cross-validation modeling for the model trained in S22. In the given modeling samples, take out most of the samples for modeling, leave a small part of the samples for prediction with the newly established model, and find this small part The prediction error of the sample, record their sum of squares, in order to judge the discrimination of the model between the target domain and the source domain;

S323: If the sum of the squares of the prediction errors of the small part of the samples is too small, it means that the model trained this time has a high degree of discrimination. After the prediction results are normalized, the sample weight parameters of the model are brought into the model for training;

S334: If the sum of the squares of the prediction errors of this small part of the samples is too large, it means that the discrimination degree of the model trained this time needs to be improved. There are two situations that cause the discrimination degree to be low: if the wrongly classified samples appear in the source domain, then It is considered that the difference with the target domain is large, and the weight reduction process is performed; if the wrongly classified samples appear in the target domain, it is considered that the model learning is not sufficient, and the learning should be strengthened;

S335: After the prediction error of the model is corrected, the result of the model fault diagnosis will be visually displayed in the process visualization system based on the digital twin.

In the machine tool bearing fault diagnosis system shown in Figure 6, the function button area in the figure can complete the corresponding functions in turn. After the "source domain data preprocessing" button is pressed, the machine tool bearing data in the digital twin workshop will be preprocessed; after the "train network" button is pressed, the machine tool bearing fault diagnosis model in the digital twin workshop will be trained; After the "target domain data preprocessing" button is pressed, the machine tool bearing data in the actual workshop will be preprocessed; after the "migration training data" button is pressed, the machine tool bearing fault diagnosis model in the actual workshop will be trained.

In this embodiment, a fault diagnosis method for machine tool bearings based on digital twin and deep transfer learning is proposed. The digital twin is used to simulate the state change of the bearing during the actual machine tool machining process to collect the data of the machine tool bearing in the digital twin space for training. The model for fault diagnosis of machine tool bearings in virtual space avoids damage to actual machine tool bearings in order to collect fault data. Using deep transfer learning, the model of machine tool bearing fault diagnosis in the digital twin space is transferred to the machine tool bearing in the real workshop, and the model is retrained with a small amount of data of the machine tool bearing in the real workshop, and the fault diagnosis model of the machine tool bearing in the real workshop can be obtained. The method proposed in the present invention can avoid consuming a lot of time cost by collecting the full life cycle data of the actual workshop bearing, and avoid the occurrence of equipment downtime and workshop production plan disruption due to the need to collect fault data.

Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, please refer to the partial descriptions of the method embodiments. The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical scope disclosed by the present invention can easily think of changes or substitutions. All should be included within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

A method for diagnosing machine tool bearing faults, comprising:

Establish a digital twin workshop, and load the processing tasks of the machine tool in the digital twin workshop, wherein the digital twin workshop corresponds to the real workshop;

Collect the fault data of the machine tool during the operation of the machining task of the machine tool, and generate a first type of fault diagnosis model according to the collected fault data, wherein the first type of fault diagnosis model includes: the machine tool bearing in the digital twin workshop fault diagnosis model;

The generated fault diagnosis model is subjected to variable working condition transfer learning training, and a second type of fault diagnosis model is obtained, wherein the second type of fault diagnosis model includes: a model for fault diagnosis of machine tool bearings in the actual workshop;

The failure of the machine tool bearing in the real workshop is detected by using the second type of fault diagnosis model.
The method according to claim 1, characterized in that, before establishing the digital twin workshop, further comprising:

Acquire and collect point cloud data of the real workshop through a three-dimensional laser scanner installed in the real workshop, wherein the objects scanned by the three-dimensional laser scanner at least include: machine tools installed in the real workshop;

Use the point cloud data of the real workshop to establish a digital twin model, wherein the objects represented by the digital twin model include at least: the overall structure and internal components of the machine tool installed in the real workshop, the internal components Including: Bearings for machine tools.
The method according to claim 2, wherein, establishing a digital twin model using the point cloud data of the real workshop includes:

Read the real-time operating data of the machine tool equipment in the real workshop through the OPC UA communication framework;

The real-time operating data is stored in a database and used as source data for a Unity data-driven engine, which is used to drive the digital twin model.
The method according to claim 1, wherein the collecting the fault data of the machine tool during the operation of the machining task of the machine tool, and generating the first type of fault diagnosis model according to the collected fault data, comprising:

Import the processing tasks of the machine tool into the digital twin workshop for simulation, and collect simulation fault data as source domain data;

After preprocessing the obtained source domain data, a deep residual learning algorithm is imported for training, and when the training accuracy rate reaches 100%, the first type of fault diagnosis model is obtained.
The method according to claim 4, wherein the importing a deep residual learning algorithm for training comprises:

The fault data is preprocessed by using a dropout network structure, wherein the fault data includes vibration data of the machine tool;

After completing the preprocessing, set the parameter structure of the fault diagnosis model;

Load the set parameters and train the fault diagnosis model.
The method according to claim 5, characterized in that, in the process of training the fault diagnosis model, comprising:

Load the improved deep residual learning algorithm for training, wherein the improved deep residual learning algorithm includes:

y' (l) = r (l) ×y (l) (4)

Among them, in the process of training, each neuron is shut down probabilistically, f represents the activation function, y represents the output value, z represents the value after the summation of neurons, i represents the number of neurons, and r represents the number of neurons. The probability of neuron shutting down, w represents the weight, bi represents the bias of the ith neuron, wi represents the weight of the ith neuron, zi represents the weight of the ith neuron and the ith neuron , y i represents the input signal, y' represents the input signal after dropout processing, zi ' represents the value after the summation of neurons after dropout processing, and l represents the identity mapping of the residual neural network The number of layers.
The method according to claim 5, wherein the setting of the parameter structure for the fault diagnosis model comprises:

The training algebra is set to 30 generations, the training times is set to 180 times, the learning rate is 0.001, and the set parameter structure is stored after the training is completed.
The method according to claim 4, characterized in that, performing the transfer learning training on the generated fault diagnosis model under variable working conditions to obtain the second type of fault diagnosis model, comprising:

collecting target domain data from the real workshop;

Use the target domain data to perform sample transfer training, and generate a second type of fault diagnosis model under variable working conditions.
The method according to claim 8, wherein the sample transfer training comprises:

Set the label of the source domain data to 0 and the label of the target domain data to 1;

Divide the modeling sample into a first part and a second part, use the first part for modeling and predict the first fault diagnosis model through the second part, and then obtain the established model according to the prediction result For the degree of discrimination between the source domain data and the target domain data, the data volume of the first part is larger than that of the second part;

The prediction results with the discrimination degree greater than the preset degree are normalized, and then the sample weight parameters of the second fault diagnosis model are brought into the training.