Multi-source cross-domain gear transmission fault diagnosis method and system

Abstract

The invention discloses a multi-source cross-domain gear transmission fault diagnosis method and system, and the method comprises the steps: carrying out the collection and processing of multi-channel acceleration vibration signals of a gear box under different fault types, and obtaining multi-source cross-domain data; performing symmetric point mode processing on the multi-source cross-domain data to obtain image sample data; according to an independence criterion, inputting the image sample data into a pre-constructed initial diagnosis model for network pre-training processing to obtain a pre-trained model, the initial diagnosis model comprising a deep residual sparse encoder and a decoder; performing classification replacement and model fine tuning processing on the pre-training model to obtain a target diagnosis model; and inputting the to-be-diagnosed fault data into the target diagnosis model for fault diagnosis processing to obtain a fault identification result. The gear transmission fault diagnosis method and device can improve comprehensiveness and accuracy of gear transmission fault diagnosis, and can be widely applied to the technical field of fault diagnosis.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a multi-source cross-domain gear transmission fault diagnosis method and system. Background Art

[0002] In modern industrial production, gears are a very important transmission device, and their working condition directly determines whether the mechanical equipment can operate normally. Since gears often run for a long time in a complex environment with high speed, heavy load and high temperature, tooth wear or breakage often occurs, leading to failures. In related technologies, some progress has been made in industrial diagnosis problems by using models to autonomously establish a mapping relationship between gear operation data and its health status, but they are always constrained by the consistency requirements of the state space of the source domain and the target domain. Since a machine may only have one or two faults from the beginning of production to the end of its life, the fault type data that can be collected is very limited, and it is usually difficult to cover all fault types in the target domain, affecting the versatility and accuracy of fault diagnosis. In summary, the technical problems existing in related technologies need to be improved. Summary of the invention

[0003] The main purpose of the embodiments of the present application is to propose a multi-source and cross-domain gear transmission fault diagnosis method and system, which can improve the versatility and accuracy of fault diagnosis.

[0004] To achieve the above-mentioned purpose, an embodiment of the present application provides a multi-source cross-domain gear transmission fault diagnosis method, the method comprising:

[0005] Collect and process the multi-channel acceleration vibration signals of the gearbox under different fault types to obtain multi-source cross-domain data;

[0006] Performing symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data;

[0007] Inputting the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model, wherein the initial diagnosis model includes a deep residual sparse encoder and a decoder;

[0008] Performing classification replacement and model fine-tuning processing on the pre-trained model to obtain a target diagnosis model;

[0009] The fault data to be diagnosed is input into the target diagnosis model for fault diagnosis processing to obtain a fault identification result.

[0010] In some embodiments, performing symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data includes the following steps:

[0011] Get preset parameters;

[0012] Performing image conversion processing on the multi-source cross-domain data according to the preset parameters to obtain a converted image;

[0013] Performing image similarity calculation processing on the converted image according to the perceptual hash algorithm, and constructing an optimization function according to the calculation result;

[0014] Solving the optimization function according to the sparrow search algorithm, and performing strategy update processing on the preset parameters according to the solution result to obtain the target parameters;

[0015] Image conversion processing is performed on the multi-source cross-domain data according to the target parameters to obtain the image sample data.

[0016] In some embodiments, performing image similarity calculation processing on the converted image according to the perceptual hash algorithm and constructing an optimization function according to the calculation result includes the following steps:

[0017] Performing classification and extraction processing on the conversion image to obtain a sample set;

[0018] Performing similarity calculation on every two images of the sample set according to the perceptual hash algorithm to obtain a calculation result set;

[0019] The image similarities of the same type in the calculation result set are added together, and the addition results of different types are subtracted from each other to obtain the optimization function.

[0020] In some embodiments, inputting the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model includes the following steps:

[0021] Performing a working condition label adding process on the image sample data according to the independence criterion to obtain an input image;

[0022] Inputting the input image into the initial diagnosis model, and outputting an output image;

[0023] Performing model loss processing on the input image and the output image according to a model loss function to obtain a loss value;

[0024] The network parameters of the initial diagnosis model are updated according to the loss value to obtain the pre-trained model.

[0025] In some embodiments, performing model loss processing on the input image and the output image according to the model loss function to obtain a loss value comprises the following steps:

[0026] Performing image difference and square calculation processing on the input image and the output image to obtain an image mean square error;

[0027] Performing Hilbert-Schmidt independence calculation processing on the working condition label of the input image and the hidden features extracted by the encoder in the initial diagnosis model to obtain an independence value;

[0028] Performing sparsity constraint processing on the initial diagnostic model to obtain a relative entropy value;

[0029] The image mean square error, the independence value and the relative entropy value are added according to the model loss function, and the addition result is added to the parameter regularization calculation result to obtain the loss value.

[0030] In some embodiments, before inputting the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion, the method further includes pre-building the initial diagnosis model, specifically including the following steps:

[0031] Stacking multiple convolutional layers to obtain the deep residual sparse encoder;

[0032] A decoder is constructed according to the deep residual sparse encoder, wherein the structure of the decoder is symmetrical to the structure of the deep residual sparse encoder;

[0033] The deep residual sparse encoder and the decoder are connected through a fully connected layer to construct the initial diagnosis model.

[0034] In some embodiments, performing classification replacement and model fine-tuning on the pre-trained model to obtain a target diagnosis model includes the following steps:

[0035] Replacing the decoder of the pre-trained model with a classifier to obtain a replacement model;

[0036] Partially annotated data is obtained, and supervised learning is performed on the replacement model according to the partially annotated data to obtain the target diagnosis model.

[0037] To achieve the above objectives, another aspect of the embodiments of the present application provides a multi-source cross-domain gear transmission fault diagnosis system, the system comprising:

[0038] The first module is used to collect and process multi-channel acceleration vibration signals of the gearbox under different fault types to obtain multi-source cross-domain data;

[0039] The second module is used to perform symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data;

[0040] The third module is used to input the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model, wherein the initial diagnosis model includes a deep residual sparse encoder and a decoder;

[0041] The fourth module is used to perform classification replacement and model fine-tuning processing on the pre-trained model to obtain a target diagnosis model;

[0042] The fifth module is used to input the fault data to be diagnosed into the target diagnosis model for fault diagnosis processing to obtain a fault identification result.

[0043] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the above-mentioned method when executing the computer program.

[0044] To achieve the above objective, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0045] The embodiments of the present application include at least the following beneficial effects: the present application provides a multi-source cross-domain gear transmission fault diagnosis method and system, which acquires multi-source cross-domain data by collecting and processing multi-channel acceleration vibration signals of the gearbox under different fault types, and processes the multi-source cross-domain data by symmetrical point pattern to obtain image sample data, which can fuse multi-channel sensor signals, so that the original time domain signal can be visualized in a graphical manner. In addition, the scheme inputs the image sample data into the pre-constructed initial diagnosis model for network pre-training according to the independence criterion to obtain a pre-trained model, the initial diagnosis model includes a deep residual sparse encoder and a decoder, and a sparse autoencoder is formed by the deep residual sparse encoder and the decoder, which can reduce the information redundancy problem when the number of hidden layer nodes in the autoencoder network is too large by using a sparse autoencoder, and accelerate the training speed of the neural network. In addition, the scheme obtains a target diagnosis model by performing classification replacement and model fine-tuning on the pre-trained model, inputs the fault data to be diagnosed into the target diagnosis model for fault diagnosis processing, and obtains a fault identification result, which can improve the generalization and accuracy of the fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of a multi-source cross-domain gear transmission fault diagnosis method provided by an embodiment of the present application;

[0047] Figure 2is a schematic diagram of fusing multi-channel information provided by an embodiment of the present application;

[0048] Figure 3 is a schematic diagram of a network structure of an initial diagnosis model provided in an embodiment of the present application;

[0049] Figure 4 is a training flow chart of an initial diagnosis model provided in an embodiment of the present application;

[0050] Figure 5 It is a structural schematic diagram of a multi-source cross-domain gear transmission fault diagnosis system provided by an embodiment of the present application;

[0051] Figure 6 It is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are only examples of systems and methods consistent with some aspects of the embodiments of the present application as detailed in the attached claims.

[0053] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiment of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".

[0054] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0056] Before describing the embodiments of the present application in detail, some nouns and terms involved in the embodiments of the present application are first described. The nouns and terms involved in the embodiments of the present application are subject to the following explanations.

[0057] 1) Deep learning is a branch of machine learning that uses multi-layered neural networks (called deep neural networks) to simulate the complex decision-making capabilities of the human brain. Some form of deep learning powers most of the artificial intelligence (AI) applications in our lives today.

[0058] 2) Symmetric Dot Pattern (Symmetrized Dot Pattern or Symmetric Dot Pattern) image, also known as SDP image, is an image representation method based on polar coordinate system. It realizes signal visualization by converting the original signal (such as time series data) into a mirror-symmetrical snowflake image.

[0059] In modern industrial production, gears are a very important transmission device, and their working condition directly determines whether mechanical equipment can operate normally. Since gears often run for a long time in a complex environment with high speed, heavy load and high temperature, tooth wear or breakage often occurs, leading to failure. Once a gear fails, machinery and equipment will not operate normally, which leads to reduced production efficiency or serious accidents. Therefore, if the operating condition of a gear can be diagnosed correctly and quickly, it can help various types of mechanical equipment to operate safely and efficiently. Traditional methods based on manual analysis, experience and prior knowledge cannot meet the requirements of timely and accurate diagnosis. In this context, data-driven fault diagnosis methods have been widely studied and applied. This method can use the model to autonomously establish a mapping relationship between gear operation data and its health status without requiring researchers to have complex physical modeling or diagnostic expertise.

[0060] Deep learning methods have been more widely used in the field of fault diagnosis due to their ability to learn high-level abstract features from complex data, such as convolutional neural networks (CNN), autoencoders (AE), and generative adversarial networks (GAN). These methods have higher diagnostic accuracy than those early classical models with shallow architectures and weak generalization performance, such as vector machines and decision trees.

[0061] However, many early studies were based on an assumption that both training data and test data come from the same distribution. However, in actual industrial production, the working environment and operating conditions of mechanical equipment often change due to different process requirements, and the different speeds and voltages used when the gears are working will affect their operating data. This means that the data used for training and testing come from different distributions, a phenomenon known as domain shift. When there is a domain shift between the training data (source domain) and the test data (target domain), the traditional supervised diagnosis model will deteriorate dramatically.

[0062] To address this distribution difference problem, current articles widely use domain adaptation (DA) methods in transfer learning. This method suppresses the distribution difference of transferable features by learning domain-invariant representations from a common feature space with different distributions. Some common DA cross-domain fault diagnosis strategies are maximum mean difference (MMD), maximum classifier difference (MCD), and generative adversarial network (GAN). Under various working conditions, the fault diagnosis effect of deep domain adaptation models is better than that of conventional deep models.

[0063] In engineering applications, fault data may come from various working conditions. However, these methods only use data from one source domain, so their performance is limited. In recent years, multi-source domain adaptation methods have been proposed to address the limitations of single-domain adaptation methods. In the related art, a domain adaptation method for learning potential diagnostic knowledge from multiple rotating machines is proposed. This method is based on adversarial DA and uses multiple classifiers and a discriminator to learn information from multiple source domains. In the related art, a pre-trained stacked autoencoder can also be used to extract multi-source domain features from different working conditions, and a domain discriminator is used to align each source domain with the target domain. Experimental results show that the multi-source domain adaptation method has better diagnostic performance than the single-source domain adaptation method.

[0064] Multi-source domain adaptation has achieved better experimental results than single-source domain adaptation because it takes into account prior knowledge under more working conditions. However, a major limitation of this method is that it requires multiple source domains with fully labeled categories to provide training data, and it is very difficult or even impossible to obtain sample data of all fault types under multiple working conditions. Therefore, the related technology proposes a global multi-source domain adaptation model GlocalNet, which fully integrates multi-source information while additionally considering the impact of category differences in each domain, reducing the category requirements for training data.

[0065] Although the above methods have made some progress in industrial diagnosis problems, they are always constrained by the state space consistency requirements of the source domain and the target domain. Since a machine may only have one or two faults from the beginning of production to the end of its life, the fault type data that can be collected is very limited, and it is usually difficult to cover all fault types in the target domain. In addition, many literatures only use 1-D frequency domain data for fault diagnosis, which will lose some effective information in the data and limit the diagnostic performance of the model.

[0066] In view of this, a multi-source cross-domain gear transmission fault diagnosis is provided in an embodiment of the present application. The scheme detects the stage of the target application. The scheme obtains multi-source cross-domain data by collecting and processing the multi-channel acceleration vibration signals of the gearbox under different fault types, and performs symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data, which can fuse multi-channel sensor signals and improve the visualization ability of the original time domain signal in a graphical manner. In addition, the scheme inputs the image sample data into the pre-constructed initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model. The initial diagnosis model includes a deep residual sparse encoder and a decoder. The deep residual sparse encoder and the decoder constitute a sparse autoencoder. The sparse autoencoder can reduce the information redundancy problem when the number of hidden layer nodes in the autoencoder network is too large by using a sparse autoencoder, and accelerates the training speed of the neural network. In addition, the scheme obtains the target diagnosis model by performing classification replacement and model fine-tuning processing on the pre-trained model, and inputs the fault data to be diagnosed into the target diagnosis model for fault diagnosis processing to obtain the fault identification result, which can improve the generalization and accuracy of the fault diagnosis model.

[0067] A multi-source cross-domain gear transmission fault diagnosis method provided in an embodiment of the present application relates to the field of fault diagnosis technology. A multi-source cross-domain gear transmission fault diagnosis method provided in an embodiment of the present application can be applied to a terminal, can also be applied to a server, and can also be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and a car terminal, etc., but is not limited to this; the server side can be configured as an independent physical server, or it can be configured as a server cluster or distributed system composed of multiple physical servers, and can also be configured to provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms and other basic cloud computing services. The server can also be a node server in a blockchain network; the software can be an application that implements a multi-source cross-domain gear transmission fault diagnosis method, etc., but is not limited to the above forms.

[0068] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0069] Figure 1 is an optional flow chart of a multi-source cross-domain gear transmission fault diagnosis method provided in an embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S105.

[0070] Step S101, collecting and processing multi-channel acceleration vibration signals of the gearbox under different fault types to obtain multi-source cross-domain data;

[0071] Step S102, performing symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data;

[0072] Step S103, inputting the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model, wherein the initial diagnosis model includes a deep residual sparse encoder and a decoder;

[0073] Step S104, performing classification replacement and model fine-tuning processing on the pre-trained model to obtain a target diagnosis model;

[0074] Step S105 , inputting the fault data to be diagnosed into the target diagnosis model for fault diagnosis processing to obtain a fault identification result.

[0075] Steps S101 to S105 shown in the embodiment of the present application are to collect multi-channel (x-axis, y-axis, etc.) acceleration vibration signals of gearboxes with different fault types at various speeds, and to organize and summarize the data to obtain multi-source cross-domain data. The multi-source cross-domain data is then subjected to symmetric point pattern processing to obtain image sample data, which is a symmetric point pattern image (SDP image) obtained by converting the acceleration vibration signal. The image sample data is then input into a pre-constructed initial diagnostic model for network pre-training processing according to the independence criterion, which is a constraint condition based on the Hilbert-Schmidt Independence Criterion (HSIC), a statistical method for measuring the independence between two variables, and is used as a constraint condition in multiple fields to optimize the performance of a model or algorithm. The initial diagnostic model includes a deep residual sparse encoder and a decoder, wherein the deep residual sparse encoder and the decoder are symmetrical structures. The pre-trained model is subjected to classification replacement and model fine-tuning. The decoder of the pre-trained model is replaced with a classifier, and the model is fine-tuned through partial labeled data to obtain the target diagnosis model. Finally, the collected vibration signal of the unknown fault gearbox is pre-processed and then input into the trained model to obtain the gearbox fault identification result.

[0076] In some embodiments, performing symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data includes the following steps:

[0077] Get preset parameters;

[0078] Performing image conversion processing on the multi-source cross-domain data according to the preset parameters to obtain a converted image;

[0079] Performing image similarity calculation processing on the converted image according to the perceptual hash algorithm, and constructing an optimization function according to the calculation result;

[0080] Solving the optimization function according to the sparrow search algorithm, and performing strategy update processing on the preset parameters according to the solution result to obtain the target parameters;

[0081] Image conversion processing is performed on the multi-source cross-domain data according to the target parameters to obtain the image sample data.

[0082] In the embodiment of the present application, the collected vibration signal can be divided into s ij , where i represents the i-th working condition, j represents the j-th segment of data information, and the division formula is:

[0083] n i =floor(s i / k);

[0084] Where n i represents the number of samples divided into the i-th working condition, s i represents the vibration signal of the i-th working condition collected, k represents the fixed length of the segmentation, and the floor function represents the rounding function. Then a set of preset parameters θ0, ξ0, τ0 are obtained, and the preset parameters can be set according to the actual situation. ij The signal is denoted as d, d i is the i-th sampling point of the d signal, and d i+τ is the τth sampling point of signal d after the ith sampling point, and uses the symmetrical point mode to transform d in the time domain into the SDP image in the spatial domain. The specific transformation formula is:

[0085]

[0086] In the formula, r(i) represents the polar coordinate radius of the space domain corresponding to the point, d max and d min They represent the maximum and minimum values ​​of the time domain signal d respectively. The spatial domain polar coordinate angle of the signal is determined using the following formula:

[0087]

[0088] Where θ is the rotation angle of the mirror symmetry plane (θ = 360m / n, n is the number of mirror symmetry planes), ξ is the gain of the plotting angle (ξ ≤ θ), and τ is the time interval factor (0 ≤ τ ≤ 10). Figure 2 , Figure 2 A schematic diagram of fusing multi-channel information provided in an embodiment of the present application, Figure 2 201 represents channel 1, that is, the x-axis vibration signal, 202 represents channel 2, that is, the y-axis vibration signal, and the whole is the SDP image obtained by conversion.

[0089] Then, the image similarity of the converted image is calculated according to the perceptual hash algorithm, and the optimization function is constructed according to the calculation results. The optimization function is solved by the sparrow search algorithm (SSA). By setting the sparrow population size to n, the discoverer ratio k1, the follower ratio k2, the scout ratio k3, the warning value R2, the safety value ST and other parameters, the positions of their respective parameter spaces X = {θ0, ξ0, τ0} are updated according to the strategy:

[0090]

[0091]

[0092] In the formula, It represents the jth position coordinate of sparrow i at iteration t. This coordinate represents the position of the sparrow in multidimensional space and is used to represent its current state or position; iter max represents the maximum number of iterations; Q represents a random perturbation factor, which is usually used to introduce randomness and increase the diversity of solutions, thereby avoiding falling into the local optimum; L represents a guiding factor, which is usually used to guide the search direction of individuals in strategy adjustment; α is a control factor, which is used to adjust the amplitude of position update during the exploration process to make the search have a certain stability; A, A + represents the auxiliary factor, which is used to balance the speed and direction control during position adjustment; Indicates the position of the sparrow with the best fitness in the current population, which is used to guide the sparrows towards the optimal solution and improve the accuracy of the solution; represents the position of the sparrow with the worst fitness in the current population, which is used to adjust the position of the follower or scout to prevent them from continuing to move in an unfavorable direction; β represents the scaling factor, which is used to amplify or reduce the effect of certain updates, so that the algorithm is adaptable to different situations; K represents the control parameter, which is used to affect the position adjustment under the safety value strategy; ε represents the minimum constant, which is used to prevent zero from appearing in the denominator to ensure the stability of the calculation. If the optimization function value corresponding to the sparrow parameter space is not updated within a certain number of rounds or reaches the maximum number of iterations, the algorithm is exited and the corresponding optimal target value g(X best ) max With the parameter value X found above best ={θ best ,ξ best ,τ best}, all s ij The vibration signal is converted into an SDP spatial domain image as the subsequent input.

[0093] One of the above technical solutions has the following advantages or beneficial effects: the embodiment of the present application uses a perceptual hash function to construct an evaluation index, extracts multiple samples of each type of fault signal, calculates the similarity of SDP images between samples of the same type of fault, calculates the similarity of SDP images between samples of different types of fault, and takes the difference between the two as the objective function, and maximizes the similarity of fault images of the same type and minimizes the similarity of fault images of different types, that is, maximizes the objective function as the optimization target, and optimizes the parameters of SDP within the definition domain using the sparrow search algorithm, so as to find the SDP parameters that make the same type of fault images most similar and the different types of fault images least similar, and finally uses this group of parameters to convert the divided vibration signal data into SDP image samples in the spatial domain. The embodiment of the present application can fuse multi-channel sensor signals by adopting a symmetrical point pattern algorithm, so that the graphic differences directly reflect some changes in signal amplitude and frequency, and can clearly map complex time series in the form of scattered points in a polar coordinate diagram, so that the original time domain signal can be displayed in a graphical manner, thereby improving the visualization capability.

[0094] In some embodiments, performing image similarity calculation processing on the converted image according to the perceptual hash algorithm and constructing an optimization function according to the calculation result includes the following steps:

[0095] Performing classification and extraction processing on the conversion image to obtain a sample set;

[0096] Performing similarity calculation on every two images of the sample set according to the perceptual hash algorithm to obtain a calculation result set;

[0097] The image similarities of the same type in the calculation result set are added together, and the addition results of different types are subtracted from each other to obtain the optimization function.

[0098] In the embodiment of the present application, k samples are extracted from all n working conditions respectively, and converted into SDP images to obtain a sample set (I 11 ,I 12 …I 1k …I nk ), by perceiving and calculating the image similarity between the two images, a set of calculation results is obtained. Among them, the perceptual hash algorithm is used to calculate the similarity of the generated SDP image, and the specific calculation formula is:

[0099]

[0100] Where D i,jrepresents the coefficient of the image I in the i-th row and j-th column after discrete cosine transform (DCT); M is the median of the retained DCT coefficients; n is the size of the retained DCT coefficient matrix. After the calculation result set is obtained, the similarities of images belonging to the same working condition are added, and the similarities of images under different working conditions are subtracted to obtain the final optimization function g(x). The specific formula of the optimization function is:

[0101]

[0102] Wherein, I represents the SDP image.

[0103] One of the above technical solutions has the following advantages or beneficial effects: the embodiment of the present application calculates the similarity of images through a perceptual hash algorithm and establishes a corresponding optimization function. It can obtain better SDP parameters by solving the optimization function, thereby converting vibration signal data into SDP image samples in the spatial domain, thereby improving the image conversion effect.

[0104] In some embodiments, inputting the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model includes the following steps:

[0105] Performing a working condition label adding process on the image sample data according to the independence criterion to obtain an input image;

[0106] Inputting the input image into the initial diagnosis model, and outputting an output image;

[0107] Performing model loss processing on the input image and the output image according to a model loss function to obtain a loss value;

[0108] The network parameters of the initial diagnosis model are updated according to the loss value to obtain the pre-trained model.

[0109] In the embodiment of the present application, in order to make the model learn domain invariant features, it is necessary to introduce the HSIC (Hilbert-Schmidt independence criterion) independence criterion to reduce the independence between the hidden features learned by the model and the working condition labels of the image sample data. Among them, the working condition label is a one-dimensional vector, d∈R k , if the sample comes from the i-th condition, then d i =1, otherwise d i=0. After adding the working condition label for each sample, the HSIC constraint can be added to the loss of the model. By minimizing the HSIC constraint, the hidden features learned by the model are minimized in correlation with the working condition to which the sample belongs. Specifically, since the relevant technical activities require multiple source domains with fully labeled categories to provide training data, that is, it is necessary to collect working data of gears under various working conditions, such as "healthy", "worn", and "damaged" fault types. In order to enable the autoencoder to extract common features under different working conditions as much as possible, thereby minimizing the differences in data distribution in different domains in the features, the embodiment of the present application adopts the HSIC independence criterion to constrain the feature extraction process of the encoder: first, use one-hot encoding to obtain the working condition label corresponding to the sample: if the sample comes from the i-th working condition, then d i =1, otherwise d i = 0. The final working condition label is expressed as d = [d1, d2, ... d n ], where n represents the number of working conditions. Let Z = [z1, z2, … z m ] is the extracted one-dimensional hidden feature, and the following formula and function are used to perform feature mapping on d and z to obtain the kernel matrix K d ,K z :

[0110]

[0111] Using the kernel matrix K d ,K z The biased estimate of HSIC is calculated as:

[0112]

[0113] Where, P = {(d 1 ,z 1 ),...,(d n ,z n )} represents the corresponding combination of elements in sets d and Z. and They are respectively composed of the kernel function k d and k Z The two reproducing kernel Hilbert spaces constructed, K d ,K z are the kernel matrices corresponding to d and Z respectively, is the center matrix. The HSIC penalty term finally added can be expressed as:

[0114]

[0115] Finally, the network parameters of the initial diagnosis model are updated according to the loss value to obtain a pre-trained model.

[0116] One of the above technical solutions has the following advantages or beneficial effects: By minimizing HSIC, the features learned by the model will have the greatest independence from the domain label. At the same time, multiple training data with fully labeled categories are no longer required during training, which greatly reduces the requirements for training data.

[0117] In some embodiments, performing model loss processing on the input image and the output image according to the model loss function to obtain a loss value comprises the following steps:

[0118] Performing image difference and square calculation processing on the input image and the output image to obtain an image mean square error;

[0119] Performing Hilbert-Schmidt independence calculation processing on the working condition label of the input image and the hidden features extracted by the encoder in the initial diagnosis model to obtain an independence value;

[0120] Performing sparsity constraint processing on the initial diagnostic model to obtain a relative entropy value;

[0121] The image mean square error, the independence value and the relative entropy value are added according to the model loss function, and the addition result is added to the parameter regularization calculation result to obtain the loss value.

[0122] In the embodiment of the present application, the model loss function The formula is as follows:

[0123]

[0124] Where θ={W,b,W ′ ,b ′} are the model parameters of DRSAE, is the reconstructed sample of the input SDP image sample X, Z is the extracted one-dimensional hidden feature, d is the domain label of the dataset X, K d and K Zare the kernel matrices corresponding to the hidden feature Z and the domain label d, respectively. β and η are penalty coefficients, which are used to control the importance weights of the sparsity constraint term and the HSIC constraint term, respectively. The model loss function of the embodiment of the present application calculates the image difference between the input image and the output image, combines the sparsity constraint, and additionally adds the L2 parameter regularization, and finally calculates the model loss value. Specifically, the embodiment of the present application obtains the output value of the network through the forward propagation process, and then in the back propagation process, the Adam optimization algorithm is used to update the network parameters with the goal of minimizing the mean square error loss function and the HSIC constraint between the input image and the output image. The training adopts a cyclic iterative method, and the data is sent to the network for iterative training until the network reaches the expected optimization target or the maximum number of iterations.

[0125] One of the above technical solutions has the following advantages or beneficial effects: the embodiment of the present application calculates the loss of the model by combining the mean square error loss function and the HSIC constraint, and can adopt the HSIC independence criterion. It no longer requires multiple fully labeled source domain data, which greatly reduces the requirements for training data, has better practical value, and improves the versatility of the initial diagnostic model.

[0126] In some embodiments, before inputting the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion, the method further includes pre-building the initial diagnosis model, specifically including the following steps:

[0127] Stacking multiple convolutional layers to obtain the deep residual sparse encoder;

[0128] A decoder is constructed according to the deep residual sparse encoder, wherein the structure of the decoder is symmetrical to the structure of the deep residual sparse encoder;

[0129] The deep residual sparse encoder and the decoder are connected through a fully connected layer to construct the initial diagnosis model.

[0130] In the embodiment of the present application, in order to obtain the most representative features of the SDP image, the embodiment of the present application uses an autoencoder AE to construct an initial diagnosis model, extracts the features of the image by encoding and decoding the above SDP image, and reconstructs the image using the acquired features. Figure 3 In the encoder part, multiple 2D convolutional layers are stacked to extract important features from the image. The weights of the convolutional layer are a set of two-dimensional filters. During the forward pass, each filter is convolved on the input feature map to generate the output feature map of the convolutional layer. Specifically, the convolution process can be described as:

[0131]

[0132] Among them, X[l-1] represents the input from the l-1th layer, W[l] represents the weight matrix of the lth convolutional layer, b[l] represents the bias matrix of the lth convolutional layer, and the i and c in the subscripts indicate that the i-th channel of the input is used and the result of the convolution calculation belongs to the c-th channel of the output.

[0133] The design of the decoder is completely symmetrical with the structure of the encoder. The deep residual sparse encoder and decoder are then connected through a fully connected layer, and the hidden features are enlarged back to the original size of the input image. At this time, the transposed convolution layer is used for upsampling, which is essentially similar to the content in the forward convolution expression. The difference is that normal convolution uses the stride parameter as the moving step size of the convolution kernel to reduce the size of the output, while transposed convolution uses stride as the expansion step size to expand the size of the output.

[0134] In order to avoid the problem of gradient vanishing caused by stacking multiple convolutional layers, a residual structure is introduced here, which is defined as:

[0135] out = F(x) + x;

[0136] In the formula, x represents the input of the current layer, F(x) can be expressed as the output of the current layer, and the output of the current layer after the residual structure is out.

[0137] In order to avoid the negative impact of process data information loss and feature information redundancy on the performance of the autoencoder network, the sparse autoencoder (SAE) adds sparsity restrictions to the hidden layer nodes based on the ordinary autoencoder. In order to add sparsity restrictions to the nodes, the average activation value of the layer nodes must first be defined as:

[0138]

[0139] In the formula, α h (X) represents the activation value of the hth node in this layer. The method of adding sparsity constraints is to manually set an expected activation level ρ close to 0, and then add constraints requiring the average activation value of the hidden layer to be Penalize those hidden layers whose average activation value is far from ρ. It can be expressed as:

[0140]

[0141] In the formula, the KL divergence (relative entropy) is calculated. If the value is very low, it means that the activation degree of the network layer is low. By adding a sparsity constraint to the autoencoder, the optimization objective function of the sparse autoencoder is obtained as follows:

[0142]

[0143] Among them, β is the sparsity constraint coefficient.

[0144] One of the above technical solutions has the following advantages or beneficial effects: The embodiment of the present application adopts a sparse autoencoder model, which not only solves the information redundancy problem when the number of hidden layer nodes of the autoencoder network is too large, but also speeds up the training speed of the neural network, while avoiding the overfitting problem to a certain extent.

[0145] In some embodiments, performing classification replacement and model fine-tuning on the pre-trained model to obtain a target diagnosis model includes the following steps:

[0146] Replacing the decoder of the pre-trained model with a classifier to obtain a replacement model;

[0147] Partially annotated data is obtained, and supervised learning is performed on the replacement model according to the partially annotated data to obtain the target diagnosis model.

[0148] In an embodiment of the present application, the decoder part of the pre-trained model is replaced with a softmax classifier, and fine-tuned using partially labeled gear fault data to achieve the prediction of gear fault types.

[0149] The formula of the classifier is as follows:

[0150]

[0151] The fault diagnosis effect of this model mainly depends on supervised learning using labeled data. In view of the problem that traditional domain adaptation models usually require one or more source domain data to have complete labels, and in actual industrial scenarios, source domain data is difficult to cover all fault types in the target domain, the model optimizes the loss function in the fine-tuning process and allows the use of multiple partially labeled source domain data. By utilizing the complementary information between these source domain data, accurate diagnosis of all gear fault types in the target domain can be achieved.

[0152] Assume that the gear annotation data under k′ working conditions are used in the training, and t1, t2, …, t k′ By combining these partially labeled data, the fine-tuning loss function of the replacement model can be defined as:

[0153]

[0154] Among them, L i represents the cross entropy loss on the i-th domain, n i,j represents the number of samples belonging to the jth fault type on the i-th domain, y m represents the fault type label of the mth sample, fj (x m ,θ) represents the probability that the model predicts that the mth sample is the jth fault type.

[0155] The fault diagnosis effect of this model mainly depends on supervised learning using labeled data. In view of the problem that traditional domain adaptation models usually require one or more source domain data to have complete labels, and in actual industrial scenarios, it is difficult for source domain data to cover all fault types in the target domain, the model optimizes the loss function in the fine-tuning process and allows the use of multiple partially labeled source domain data. By utilizing the complementary information between these source domain data, accurate diagnosis of all gear fault types in the target domain can be achieved.

[0156] See also Figure 4 , the model pre-training and fine-tuning process is as follows Figure 4 As shown, 401 represents source domain 1, 402 represents source domain 2, 403 represents source domain k, k is a positive integer, 404 represents samples in multiple source domains, 405 represents domain features, and 406 represents hidden features. It can be seen that the model can simultaneously use fault information in multiple domains to improve its prediction accuracy and robustness. In unsupervised training, HSIC values ​​are calculated for multiple source domains and their corresponding domain labels. Through HSIC constraints, the model can learn domain-independent data features. In supervised training, partial annotation information from multiple domains is used to improve the accuracy of cross-domain diagnosis.

[0157] One of the above technical solutions has the following advantages or beneficial effects: when training the model, the embodiment of the present application does not require any source domain to contain annotations of all fault types. In practice, it is very convenient to obtain partial fault type data of each working condition, thereby improving the versatility of the model.

[0158] The following is a detailed description of the embodiments of the present invention with reference to specific application examples:

[0159] The data set used in the embodiment of the present application can be a public data set from a relevant scientific research unit, such as the "WT-Planetary gearbox dataset" data set published by Beijing Jiaotong University. In the public data set, the test bench consists of a motor, a planetary gearbox, a fixed-axis gearbox and a loading device, with four planetary gears rotating around a sun gear. The five health conditions of the sun gear are divided into five conditions: gear with a broken tooth, gear wear, crack in the root, missing one tooth, and healthy condition. The vibration data is collected by the SMIC CA-YD-1181 accelerometer, and the speed pulse is captured by the encoder. The sampling rate of all channels is set to 48kHz. In the early data processing stage, the parameters of the sparrow search algorithm are initialized as follows: the dimension of the reference data dims = 3; the warning value ST = 0.6; the proportion of discoverers PD = 0.7; the proportion of sparrows aware of danger SD = 0.2; the maximum number of iterations = 1000.

[0160] In the encoder part, the embodiment of the present application chooses to stack multiple 2D convolutional layers to extract important features in the image. With its unique advantage of translation invariance, the convolutional layer has become a common tool for capturing local features in images. The weights of the convolutional layer are a set of two-dimensional filters. During the forward pass, each filter is convolved on the input feature map to generate the output feature map of the convolutional layer. In the hyperparameter setting of the encoder structure, the embodiment of the present application refers to some commonly used and effective parameter selections and image sizes and network structures, and selects the ReLU activation function, the convolution kernel size is 3x3, the step size is 2, and the zero padding mode is set to "same". In the process of layer-by-layer convolution, the size of the feature matrix is ​​reduced to one-half after each layer, and the number of channels is doubled. This convolution method is widely used in various CNNs. At the same time, the embodiment of the present application refers to the network structure of ResNet, and the residual structure is introduced to alleviate the problem of gradient disappearance of multi-layer convolutional layers without bringing a lot of additional computational overhead. At the end of the encoder, the embodiment of the present application uses a global average pooling layer to compress the two-dimensional features extracted by the convolution layer into one dimension, and further reduces the feature length through two fully connected layers.

[0161] The design of the decoder is completely symmetrical with the structure of the encoder. After passing through a similar fully connected layer, the hidden features need to be enlarged back to the original size of the input image. At this time, a transposed convolution layer is needed for upsampling, which is essentially similar to the content in the forward convolution expression. The difference is that normal convolution uses the stride parameter as the moving step size of the convolution kernel to reduce the size of the output, while transposed convolution uses stride as the expansion step size to expand the size of the output.

[0162] After completing the pre-training, the embodiment of the present application replaces the decoder part of the model with a softmax classifier, and fine-tunes it with labeled data, so that the gear fault type can be predicted. The diagnostic effect of the model mainly depends on supervised learning using labeled data. Many previous domain adaptation models require one or more fully labeled source domain data, but in practice, source domain data is often difficult to cover all fault types in the target domain. During the training process, the learning rate is set to 0.001, the training epoch is 50, and the batch size is 64. In addition, in order to improve the training effect, the embodiment of the present application introduces a cosine annealing strategy during training. For the two penalty coefficients β and η in the method, the embodiment of the present application sets β to 0.5 and η to 1.

[0163] See also Figure 5 The embodiment of the present application further provides a multi-source cross-domain gear transmission fault diagnosis system, which can implement the multi-source cross-domain gear transmission fault diagnosis method mentioned above, and the system includes:

[0164] The first module 501 is used to collect and process multi-channel acceleration vibration signals of the gearbox under different fault types to obtain multi-source cross-domain data;

[0165] The second module 502 is used to perform symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data;

[0166] The third module 503 is used to input the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model, wherein the initial diagnosis model includes a deep residual sparse encoder and a decoder;

[0167] The fourth module 504 is used to perform classification replacement and model fine-tuning processing on the pre-trained model to obtain a target diagnosis model;

[0168] The fifth module 505 is used to input the fault data to be diagnosed into the target diagnosis model for fault diagnosis processing to obtain a fault identification result.

[0169] It can be understood that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0170] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned multi-source cross-domain gear transmission fault diagnosis method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.

[0171] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0172] See also Figure 6 , Figure 6 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0173] The processor 601 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0174] The memory 602 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 602 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 602, and the processor 601 calls and executes the multi-source cross-domain gear transmission fault diagnosis method of the embodiment of the present application;

[0175] Input / output interface 603, used to implement information input and output;

[0176] Communication interface 604, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0177] Bus 605 , which transmits information between various components of the device (e.g., processor 601 , memory 602 , input / output interface 603 , and communication interface 604 );

[0178] The processor 601 , the memory 602 , the input / output interface 603 and the communication interface 604 are connected to each other in communication within the device via a bus 605 .

[0179] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned multi-source cross-domain gear transmission fault diagnosis method is implemented.

[0180] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0181] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0182] The embodiment of the present application provides a multi-source cross-domain gear transmission fault diagnosis method and system. The scheme acquires and processes the multi-channel acceleration vibration signals of the gearbox under different fault types to obtain multi-source cross-domain data, performs symmetrical point pattern processing on the multi-source cross-domain data, obtains image sample data, and can fuse multi-channel sensor signals, so that the original time domain signal can be visualized in a graphical manner. In addition, the scheme inputs the image sample data into the pre-constructed initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model. The initial diagnosis model includes a deep residual sparse encoder and a decoder. The deep residual sparse encoder and the decoder form a sparse autoencoder. The sparse autoencoder can reduce the information redundancy problem when the number of hidden layer nodes in the autoencoder network is too large by using a sparse autoencoder, and accelerates the training speed of the neural network. In addition, the scheme obtains a target diagnosis model by performing classification replacement and model fine-tuning processing on the pre-trained model, inputs the fault data to be diagnosed into the target diagnosis model for fault diagnosis processing, and obtains a fault identification result, which can improve the generalization and accuracy of the fault diagnosis model.

[0183] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0184] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0185] The system embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0186] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0187] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0188] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0189] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0190] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0191] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0192] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0193] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A multi-source cross-domain gear transmission fault diagnosis method, characterized in that: The method comprises the following steps: Collect and process the multi-channel acceleration vibration signals of the gearbox under different fault types to obtain multi-source cross-domain data; Performing symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data; Inputting the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model, wherein the initial diagnosis model includes a deep residual sparse encoder and a decoder; Performing classification replacement and model fine-tuning processing on the pre-trained model to obtain a target diagnosis model; The fault data to be diagnosed is input into the target diagnosis model for fault diagnosis processing to obtain a fault identification result.

2. The method according to claim 1, characterized in that The step of performing symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data comprises the following steps: Get preset parameters; Performing image conversion processing on the multi-source cross-domain data according to the preset parameters to obtain a converted image; Performing image similarity calculation processing on the converted image according to the perceptual hash algorithm, and constructing an optimization function according to the calculation result; Solving the optimization function according to the sparrow search algorithm, and performing strategy update processing on the preset parameters according to the solution result to obtain the target parameters; Image conversion processing is performed on the multi-source cross-domain data according to the target parameters to obtain the image sample data.

3. The method according to claim 2, characterized in that The process of performing image similarity calculation on the converted image according to the perceptual hash algorithm and constructing an optimization function according to the calculation result comprises the following steps: Performing classification and extraction processing on the conversion image to obtain a sample set; Performing similarity calculation on every two images of the sample set according to the perceptual hash algorithm to obtain a calculation result set; The image similarities of the same type in the calculation result set are added together, and the addition results of different types are subtracted from each other to obtain the optimization function.

4. The method according to claim 1, characterized in that The step of inputting the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model comprises the following steps: Performing a working condition label adding process on the image sample data according to the independence criterion to obtain an input image; Inputting the input image into the initial diagnosis model, and outputting an output image; Performing model loss processing on the input image and the output image according to a model loss function to obtain a loss value; The network parameters of the initial diagnosis model are updated according to the loss value to obtain the pre-trained model.

5. The method according to claim 4, characterized in that The step of performing model loss processing on the input image and the output image according to the model loss function to obtain a loss value comprises the following steps: Performing image difference and square calculation processing on the input image and the output image to obtain an image mean square error; Performing Hilbert-Schmidt independence calculation processing on the working condition label of the input image and the hidden features extracted by the encoder in the initial diagnosis model to obtain an independence value; Performing sparsity constraint processing on the initial diagnostic model to obtain a relative entropy value; The image mean square error, the independence value and the relative entropy value are added according to the model loss function, and the addition result is added to the parameter regularization calculation result to obtain the loss value.

6. The method according to claim 1, characterized in that Before inputting the image sample data into the pre-built initial diagnosis model for network pre-training processing according to the independence criterion, the method further includes pre-building the initial diagnosis model, which specifically includes the following steps: Stacking multiple convolutional layers to obtain the deep residual sparse encoder; A decoder is constructed according to the deep residual sparse encoder, wherein the structure of the decoder is symmetrical to the structure of the deep residual sparse encoder; The deep residual sparse encoder and the decoder are connected through a fully connected layer to construct the initial diagnosis model.

7. The method according to any one of claims 1 to 6, characterized in that: The step of performing classification replacement and model fine-tuning on the pre-trained model to obtain a target diagnosis model includes the following steps: Replacing the decoder of the pre-trained model with a classifier to obtain a replacement model; Partially annotated data is obtained, and supervised learning is performed on the replacement model according to the partially annotated data to obtain the target diagnosis model.

8. A multi-source cross-domain gear transmission fault diagnosis system, characterized in that: The system comprises: The first module is used to collect and process multi-channel acceleration vibration signals of the gearbox under different fault types to obtain multi-source cross-domain data; The second module is used to perform symmetrical point pattern processing on the multi-source cross-domain data to obtain image sample data; The third module is used to input the image sample data into a pre-built initial diagnosis model for network pre-training processing according to the independence criterion to obtain a pre-trained model, wherein the initial diagnosis model includes a deep residual sparse encoder and a decoder; The fourth module is used to perform classification replacement and model fine-tuning processing on the pre-trained model to obtain a target diagnosis model; The fifth module is used to input the fault data to be diagnosed into the target diagnosis model for fault diagnosis processing to obtain a fault identification result.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images