Fault diagnosis method and system

Abstract

The present invention relates to a fault diagnosis method and system, which are characterized by comprising: obtaining a vibration signal of a device under test; inputting the vibration signal of the device under test into a fault diagnosis model that integrates a trained one-dimensional dual-channel convolution and a bidirectional long short-term memory neural network to obtain a fault diagnosis classification result of the device under test, which can be widely used in the field of advanced manufacturing control technology.

Description

Technical Field

[0001] The present invention relates to the field of advanced manufacturing control technology, and in particular to a fault diagnosis method and system. Background Art

[0002] The precision, complexity and intelligence of the equipment used in the industrial production process are becoming increasingly higher. If these devices are not well managed, once a failure occurs, it will cause a series of negative impacts such as downtime and even endanger personal safety. Therefore, fault diagnosis of equipment has great practical guiding significance.

[0003] With the continuous development of artificial intelligence and industrial Internet of Things, the use of deep learning methods to achieve fault diagnosis has become the current development trend. At present, the most popular fault diagnosis method is based on signal analysis. This method uses certain transformations (empirical decomposition model, short-time Fourier transform or wavelet decomposition, etc.) to transform time series signals into frequency domain signals. However, this method is relatively time-consuming and laborious to extract features. Although the method of manually extracting features has achieved good results in the application of fault diagnosis, its diagnostic performance mostly depends on the effect of manually extracted features, and does not give full play to the unique advantage of deep learning's self-learning.

[0004] In addition, the current traditional fault diagnosis methods generally have the disadvantage of complex network structure, which increases the burden on computers, and their diagnostic accuracy will be significantly reduced as the number of fault types increases. Summary of the invention

[0005] In view of the above problems, an object of the present invention is to provide a simple and efficient fault diagnosis method and system that can avoid manual participation in feature extraction.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, a fault diagnosis method is provided, comprising:

[0008] Obtain the vibration signal of the device under test;

[0009] The vibration signal of the equipment under test is input into the fault diagnosis model that integrates the trained one-dimensional dual-channel convolution and bidirectional long short-term memory neural network to obtain the fault diagnosis classification result of the equipment under test.

[0010] Furthermore, the construction process of the fault diagnosis model integrating the one-dimensional dual-channel convolution and the bidirectional long short-term memory neural network is as follows:

[0011] Collect several historical fault vibration signals of the equipment to be tested, and classify and process the several historical fault vibration signals to obtain several different fault types;

[0012] Pre-process the collected historical fault vibration signals;

[0013] Down-sampling the pre-processed historical fault vibration signal;

[0014] The downsampled historical fault vibration signal dataset is divided into a training set and a test set;

[0015] The constructed fault diagnosis model is trained and tested using training sets and test sets, and a fault diagnosis model integrating the trained one-dimensional dual-channel convolution and bidirectional long short-term memory neural network is obtained.

[0016] Furthermore, the preprocessing of the collected historical fault vibration signals includes:

[0017] Define a function for obtaining historical fault vibration signals to integrate historical fault vibration signals of different fault types in different folders into one data set;

[0018] According to the fault type of the historical fault vibration signal, a label function of the historical fault vibration signal is defined.

[0019] Furthermore, the fault diagnosis model integrating the one-dimensional dual-channel convolution and the bidirectional long short-term memory neural network includes a first convolution neural network, a second convolution neural network, a feature fusion module, a first bidirectional long short-term memory network, a second bidirectional long short-term memory network, a batch normalization layer and a fully connected layer;

[0020] The first convolutional neural network and the second convolutional neural network are both used to extract local spatial features of the downsampled fault vibration signal;

[0021] The feature fusion module is used to fuse the local spatial features extracted by the first convolutional neural network and the second convolutional neural network to obtain spatial features of different scales;

[0022] The first bidirectional long short-term memory network and the second bidirectional long short-term memory network are both used to extract temporal features according to the fused spatial features;

[0023] The BN layer is used to perform batch normalization on the extracted time series features;

[0024] The fully connected layer is used to output the fault type according to the batch normalization result.

[0025] Furthermore, the fault diagnosis model integrating the one-dimensional dual-channel convolution and the bidirectional long short-term memory neural network also includes a Dropout layer for discarding certain connection points during neural network training to prevent overfitting.

[0026] Further, the first convolutional neural network is provided with three one-dimensional convolutional layers and one one-dimensional maximum pooling layer, and the second convolutional neural network is provided with six one-dimensional convolutional layers and two one-dimensional maximum pooling layers;

[0027] The one-dimensional convolution layer of the first convolutional neural network uses a relatively large convolution kernel to extract the global spatial features of the downsampled fault vibration signal;

[0028] The one-dimensional maximum pooling layer of the first convolutional neural network is used to reduce the dimension of the global spatial features extracted by the one-dimensional convolutional layer of the first convolutional neural network, and extract the corresponding local spatial features;

[0029] The one-dimensional convolution layer of the second convolutional neural network uses a relatively small convolution kernel to extract the global spatial features of the downsampled fault vibration signal;

[0030] The one-dimensional maximum pooling layer of the second convolutional neural network is used to reduce the dimension of the global spatial features extracted by the one-dimensional convolutional layer of the second convolutional neural network, and extract the corresponding local spatial features.

[0031] Furthermore, the calculation formula for the side length of the feature graph of the first convolutional neural network and the second convolutional neural network is:

[0032] or

[0033] Among them, W out Represents the edge length of the output feature map; W in Represents the side length of the input feature map; S represents the convolution step size; K represents the convolution kernel length, and the calculation results are rounded up.

[0034] In a second aspect, a fault diagnosis system is provided, comprising:

[0035] A data acquisition module, used to acquire the vibration signal of the device under test;

[0036] The fault diagnosis classification module is used to input the vibration signal of the equipment under test into the fault diagnosis model that integrates the trained one-dimensional dual-channel convolution and bidirectional long short-term memory neural network to obtain the fault diagnosis classification result of the equipment under test.

[0037] According to a third aspect, a processing device is provided, comprising computer program instructions, wherein the computer program instructions are used to implement the steps corresponding to the above-mentioned fault diagnosis method when executed by the processing device.

[0038] In a fourth aspect, a computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions are used to implement the steps corresponding to the above-mentioned fault diagnosis method when executed by a processor.

[0039] The present invention adopts the above technical solution, which has the following advantages:

[0040] 1. The present invention aims at simplicity and high efficiency, can automatically extract data, and avoid the influence of manual participation in feature extraction. The present invention uses an end-to-end fault diagnosis mode to construct a fault diagnosis model that integrates dual-channel convolution and bidirectional long short-term memory neural network. The original vibration signal is used as the input of the fault diagnosis model, and the dual-channel convolutional neural network is used as a feature extractor and the bidirectional long short-term memory neural network is used as a classifier. The constructed model has the ability to express features in time and space. It can not only utilize the information with time series characteristics collected by the sensor, but also can mine deeper information. Compared with the traditional fault diagnosis model, the training time is shorter and the effect is more stable.

[0041] 2. The fault diagnosis model constructed by the present invention can get rid of the step of manually extracting features required by traditional fault diagnosis, and does not require signal processing, which can reduce human errors.

[0042] 3. The present invention reduces the sample size by downsampling the data, thereby reducing the burden on the training network of the computing device without affecting the accuracy. The accuracy of fault diagnosis for four categories can reach 100%, and the accuracy of fault diagnosis for ten categories can reach 99.8%. Compared with traditional fault diagnosis, the training convergence speed of the fault diagnosis model using the present invention is faster and more accurate.

[0043] In summary, the present invention can be widely applied in the field of advanced manufacturing control technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Throughout the accompanying drawings, the same reference numerals are used to represent the same components. In the accompanying drawings:

[0045] Figure 1 is a structural diagram of a fault diagnosis model provided by an embodiment of the present invention;

[0046] Figure 2 It is a schematic diagram of the accuracy of four-category fault diagnosis of a device under test provided by an embodiment of the present invention;

[0047] Figure 3 It is a schematic diagram of loss values ​​for four-category fault diagnosis of a device under test provided by an embodiment of the present invention;

[0048] Figure 4It is a schematic diagram of a confusion matrix for four-category fault diagnosis of a device under test provided by an embodiment of the present invention;

[0049] Figure 5 It is a schematic diagram of the accuracy of diagnosing ten types of faults of a device under test provided by an embodiment of the present invention;

[0050] Figure 6 It is a schematic diagram of loss values ​​for diagnosing ten types of faults of a device under test provided by an embodiment of the present invention;

[0051] Figure 7 It is a schematic diagram of a confusion matrix for diagnosing ten types of faults of a device under test provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0053] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0054] Although the terms first, second, third, etc. can be used in the text to describe multiple elements, components, regions, layers and / or sections, these elements, components, regions, layers and / or sections should not be limited by these terms. These terms can only be used to distinguish an element, component, region, layer or section from another region, layer or section. Unless the context clearly indicates, terms such as "first", "second" and other numerical terms do not imply order or sequence when used in the text. Therefore, the first element, component, region, layer or section discussed below can be referred to as the second element, component, region, layer or section without departing from the teaching of the example embodiments.

[0055] In view of the problem that traditional fault diagnosis methods are difficult to adaptively extract effective features from fault vibration signals and do not fully utilize the timing characteristics of fault data, the fault diagnosis method and system provided by the embodiment of the present invention use the original vibration signal of the device under test as the input of the fault diagnosis model. The fault diagnosis model uses two convolutional neural networks with different kernel sizes and different depths to automatically extract different signal features from the original vibration signal, and then performs feature fusion. According to the learned features, a bidirectional long short-term memory neural network is used to identify the fault type. The present invention can overcome the problems of traditional feature extraction methods relying on expert experience and insufficient information utilization, and can automatically extract features and diagnose fault data without any preprocessing and operation of the input features, making equipment fault diagnosis simpler and more efficient.

[0056] Example 1

[0057] This embodiment provides a fault diagnosis method, including the following steps:

[0058] 1) Collect several historical fault vibration signals of the equipment under test to classify various fault types, and build a fault diagnosis model that integrates one-dimensional dual-channel convolution and bidirectional long short-term memory neural network, specifically:

[0059] 1.1) Collect several historical fault vibration signals of the equipment under test, and classify and process the several historical fault vibration signals to obtain several different fault types.

[0060] Specifically, historical fault vibration signals are collected using sensors.

[0061] Specifically, the fault type can be classified according to the specific requirements of the specific equipment, which will not be elaborated here.

[0062] Specifically, the fault vibration signals after classification processing are respectively set in corresponding folders.

[0063] 1.2) Preprocess the collected historical fault vibration signals:

[0064] 1.2.1) Define a function for obtaining historical fault vibration signals to integrate fault vibration signals of different fault types in different folders into one data set in preparation for input into model training.

[0065] Specifically, the location and number of historical fault vibration signals in each folder are determined, the number of historical fault vibration signals in each folder, for example, 460,800, is obtained through a for loop function, and the fault vibration data is stored in a shape of 225 rows and 2048 columns.

[0066] 1.2.2) According to the fault type of the historical fault vibration signal, define the label function of the historical fault vibration signal.

[0067] Specifically, the sample length of the historical fault vibration signal is set to 2048, the number of samples is set to 225, and classification labels are created.

[0068] For example: if the fault type is healthy, the classification label is 0; if the fault type is inner raceway fault, the classification label is 1; if the fault type is rolling element fault, the classification label is 2; and if the fault type is outer raceway fault, the classification label is 3.

[0069] 1.3) Down-sample the pre-processed historical fault vibration signal.

[0070] Specifically, the preprocessed historical fault vibration signal may be downsampled by 2, 4 or 8 times according to actual needs.

[0071] For example, the historical fault vibration signal with a sample length of 2048 is downsampled by 4 times, and the position index is performed every 4 points, and the fault sample length is changed to 512 to reduce the computational burden.

[0072] 1.4) Divide the downsampled historical fault vibration signal dataset into a training set and a test set.

[0073] Specifically, the downsampled historical fault vibration signal data set is divided into a ratio of 4:1.

[0074] 1.5) Construct a fault diagnosis model that integrates one-dimensional dual-channel convolution and bidirectional long short-term memory neural network:

[0075] Specifically, Figure 1 As shown, the fault diagnosis model includes a first convolutional neural network, a second convolutional neural network, a feature fusion module, a first bidirectional long short-term memory network, a second bidirectional long short-term memory network, a batch normalization (BN) layer, a Dropout (random discard) layer and a fully connected layer, wherein the first convolutional neural network is provided with three one-dimensional convolutional layers and one-dimensional maximum pooling layer, and the second convolutional neural network is provided with six one-dimensional convolutional layers and two one-dimensional maximum pooling layers, and two convolutional neural networks with different channels, different kernels and different depths are used to extract different features.

[0076] The one-dimensional convolution layer of the first convolutional neural network uses a relatively large convolution kernel to extract the global spatial features of the downsampled fault vibration signal. The larger the convolution kernel used, the larger the receptive field, the more information seen, and the better the global spatial features obtained. Therefore, the one-dimensional convolution layer in the first convolutional neural network uses a relatively large convolution kernel.

[0077] The one-dimensional maximum pooling layer of the first convolutional neural network is used to reduce the dimension of the global spatial features extracted by the one-dimensional convolutional layer of the first convolutional neural network and extract the corresponding local spatial features, which can effectively reduce the parameter matrix, thereby reducing the number of calculations, speeding up the calculation speed, and preventing overfitting to a certain extent.

[0078] The one-dimensional convolution layer of the second convolutional neural network uses a relatively small convolution kernel to extract the global spatial features of the downsampled fault vibration signal. Using a smaller convolution kernel will use more filters and more activation functions, and the extracted global spatial features will be more complex, allowing the convolutional neural network to learn a more discriminative mapping function.

[0079] The one-dimensional maximum pooling layer of the second convolutional neural network is used to reduce the dimension of the global spatial features extracted by the one-dimensional convolutional layer of the second convolutional neural network, and extract the corresponding local spatial features.

[0080] The feature fusion module is used to fuse the local spatial features extracted by the first convolutional neural network and the second convolutional neural network to obtain spatial features of different scales to fully mine data information.

[0081] The first bidirectional long short-term memory network and the second bidirectional long short-term memory network are both used to extract temporal features based on the fused spatial features. The first bidirectional long short-term memory network and the second bidirectional long short-term memory network have the same functions, and one more layer is set to achieve better classification results.

[0082] The batch normalization layer is used to batch normalize the extracted time series features so that the data conforms to a distribution with a mean of 0 and a standard deviation of 1.

[0083] The Dropout layer is used to discard certain connection points during neural network training to prevent overfitting after training.

[0084] The fully connected layer is used to adopt the Softmax activation function and output the fault type according to the batch normalization result.

[0085] Specifically, since the first convolutional neural network and the second convolutional neural network do not utilize the characteristic that the collected data has time series characteristics, the first bidirectional long short-term memory neural network and the second bidirectional long short-term memory network are provided, which can mine the time series information in the one-dimensional signal and realize more accurate fault diagnosis. More specifically, the first bidirectional long short-term memory neural network and the second bidirectional long short-term memory network classify the entire input based on the fused features of the input, divide the hidden layer into two independent positive and negative hidden layers, and the output layer has both past and future information. Using information from both ends of the sequence to estimate the output will make the result more accurate.

[0086] Specifically, since two CNN channels, the first convolutional neural network and the second convolutional neural network, are provided, the two channels need to be fused and input into the long short-term memory neural network. Therefore, to ensure that the output feature map sizes of the two channels are consistent, full zero padding can be used, where the feature map side length calculation formula is:

[0087] With padding = 'SAME':

[0088]

[0089] With padding = 'VALID':

[0090]

[0091] Among them, W out Represents the edge length of the output feature map; W in represents the side length of the input feature map; S represents the convolution step size; K represents the convolution kernel length. The calculation results are rounded up.

[0092] 1.6) Input the training set into the constructed fault diagnosis model for training, and determine the weight, loss and accuracy of the fault diagnosis model based on the Adam optimization algorithm (an optimization algorithm), the MSE loss function (a loss function) and the accuracy evaluation index. Since the training of the fault diagnosis model is a method disclosed in the prior art, the specific process will not be described in detail here.

[0093] 1.7) Determine whether the training classification results of the training set are accurate. If they are accurate, save the fault diagnosis model and input the test set into the model for fault classification; otherwise, enter step 1.6) and continue training until the training classification results are accurate to obtain the final trained fault diagnosis model.

[0094] Specifically, whether the classification result of the training set is accurate can be determined based on the accuracy rate and pre-set requirements.

[0095] 2) Obtain the vibration signal of the device under test and input it into the trained fault diagnosis model to obtain the fault classification result of the device under test.

[0096] The fault diagnosis method of the present invention is described in detail below through specific embodiments. This embodiment is verified and analyzed using a rolling bearing fault diagnosis data set published by Case Western Reserve University in the United States. This data set is recognized in the field of fault diagnosis and has a certain authority for experimental example verification:

[0097] 1) Collect historical fault vibration signals of several rolling bearings to be tested, classify various fault types, and build a four-classification fault diagnosis model that integrates one-dimensional dual-channel convolution and bidirectional long short-term memory neural network:

[0098] 1.1) Several historical fault vibration signals of the rolling bearing to be tested are collected by sensors, and various fault types are classified and processed to obtain four different fault types, including health, inner raceway fault, rolling element fault and outer raceway fault.

[0099] 1.2) Preprocess the collected historical fault vibration signals.

[0100] Set the folder location of the historical fault vibration signal, define the function of obtaining the historical fault vibration signal (determine the location and number of files in the folder, obtain the number of historical fault vibration signals in each folder through the for loop function, for example, 460800, and store the fault vibration data in the shape of 225 rows and 2048 columns), define the label function of the fault vibration data according to the fault type of the historical fault vibration signal, and the corresponding relationship between the specific fault condition and the classification label is shown in Table 1 below:

[0101] Table 1: Correspondence between fault conditions and classification labels

[0102]

[0103] 1.3) Downsample the preprocessed historical fault vibration signal by 4 times.

[0104] 1.4) Divide the historical fault vibration signal dataset after 4 times downsampling into a training set and a test set.

[0105] 1.5) Input the training set into the constructed fault diagnosis model for training, and determine the weight, loss and accuracy of the fault diagnosis model based on the Adam optimization algorithm, MSE loss function and accuracy evaluation index.

[0106] 1.6) Determine whether the classification results of the training set are accurate. If they are accurate, save the fault diagnosis model and input the test set into the model for fault classification. Otherwise, continue training until the classification results are accurate, and obtain the final trained fault diagnosis model that integrates one-dimensional dual-channel convolution and bidirectional long short-term memory neural network.

[0107] 2) Obtain the vibration signal of the rolling bearing to be tested, pre-process it, and input it into the trained fault diagnosis model to obtain the fault diagnosis classification result of the equipment to be tested.

[0108] like Figure 2 and 3 , which is a schematic diagram of the accuracy and loss value of the fault diagnosis classification results obtained by the method of the present invention. It can be seen that with the increase of iterations, the accuracy can reach 100%.

[0109] like Figure 4 As shown, it is a schematic diagram of the confusion matrix of the fault diagnosis classification results obtained by the method of the present invention, wherein the accuracy of the evaluation model corresponding to each category can further analyze the test results of the fault diagnosis model in the present invention.

[0110] When used, the fault diagnosis model of the present invention can reduce the computational burden of the computing device, and each iteration takes only 1 to 2 seconds, which is a speed that cannot be achieved by traditional fault diagnosis methods.

[0111] In addition, the fault diagnosis model in the present invention is also applicable to ten categories of fault diagnosis, and its steps are roughly the same as those of the four categories, the only difference being that its fault types are divided into 10 categories.

[0112] like Figure 5 , Figure 6 and Figure 7 As shown, they are respectively the accuracy, loss value and confusion matrix diagram of the fault diagnosis classification results obtained by the method of the present invention. It can be seen that the accuracy can still reach 99.8%, proving that the fault diagnosis model of the present invention has good performance.

[0113] Example 2

[0114] This embodiment provides a fault diagnosis system, including:

[0115] The data acquisition module is used to acquire the vibration signal of the device under test.

[0116] The fault diagnosis classification module is used to input the vibration signal of the equipment under test into the fault diagnosis model that integrates the trained one-dimensional dual-channel convolution and bidirectional long short-term memory neural network to obtain the fault diagnosis classification result of the equipment under test.

[0117] In a preferred embodiment, the construction process of the fault diagnosis model integrating one-dimensional dual-channel convolution and bidirectional long short-term memory neural network is as follows:

[0118] A number of historical fault vibration signals of the device under test are collected, and the historical fault vibration signals are classified and processed to obtain a number of different fault types.

[0119] Preprocess the collected historical fault vibration signals.

[0120] The preprocessed historical fault vibration signal is downsampled.

[0121] The downsampled historical fault vibration signal dataset is divided into a training set and a test set.

[0122] The constructed fault diagnosis model is trained and tested using training sets and test sets, and a fault diagnosis model integrating the trained one-dimensional dual-channel convolution and bidirectional long short-term memory neural network is obtained.

[0123] Example 3

[0124] This embodiment provides a processing device corresponding to the fault diagnosis method provided in this embodiment 1. The processing device can be a processing device used for a client, such as a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the method of embodiment 1.

[0125] The processing device includes a processor, a memory, a communication interface and a bus, and the processor, the memory and the communication interface are connected through the bus to complete mutual communication. The memory stores a computer program that can be run on the processing device, and the processing device executes the fault diagnosis method provided in this embodiment 1 when running the computer program.

[0126] In some implementations, the memory may be a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0127] In some other implementations, the processor may be a central processing unit (CPU), a digital signal processor (DSP), or other general-purpose processors of various types, which are not limited herein.

[0128] Example 4

[0129] This embodiment provides a computer program product corresponding to the fault diagnosis method provided in this embodiment 1. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the fault diagnosis method described in this embodiment 1 are loaded.

[0130] Computer readable storage media can be tangible devices that hold and store instructions used by instruction execution devices. Computer readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any combination thereof.

[0131] The above embodiments are only used to illustrate the present invention, wherein the structure, connection mode and manufacturing process of each component may be changed. Any equivalent transformations and improvements based on the technical solution of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. A fault diagnosis method, characterized in that: include: Obtain the vibration signal of the device under test; The vibration signal of the equipment under test is input into the fault diagnosis model that integrates the trained one-dimensional dual-channel convolution and bidirectional long short-term memory neural network to obtain the fault diagnosis classification result of the equipment under test; The construction process of the fault diagnosis model integrating the one-dimensional dual-channel convolution and the bidirectional long short-term memory neural network is as follows: Collect several historical fault vibration signals of the equipment to be tested, and classify and process the several historical fault vibration signals to obtain several different fault types; Pre-process the collected historical fault vibration signals; Down-sampling the pre-processed historical fault vibration signal; The downsampled historical fault vibration signal dataset is divided into a training set and a test set; The constructed fault diagnosis model is trained and tested using training sets and test sets, and a fault diagnosis model integrating the trained one-dimensional dual-channel convolution and bidirectional long short-term memory neural network is obtained. The fault diagnosis model integrating the one-dimensional dual-channel convolution and the bidirectional long short-term memory neural network comprises a first convolution neural network, a second convolution neural network, a feature fusion module, a first bidirectional long short-term memory network, a second bidirectional long short-term memory network, a batch normalization layer and a fully connected layer; The first convolutional neural network and the second convolutional neural network are both used to extract local spatial features of the downsampled fault vibration signal; The feature fusion module is used to fuse the local spatial features extracted by the first convolutional neural network and the second convolutional neural network to obtain spatial features of different scales; The first bidirectional long short-term memory network and the second bidirectional long short-term memory network are both used to extract temporal features according to the fused spatial features; The batch normalization layer is used to perform batch normalization on the extracted time series features; The fully connected layer is used to output the fault type according to the batch normalization result; The first convolutional neural network is provided with three one-dimensional convolutional layers and one one-dimensional maximum pooling layer, and the second convolutional neural network is provided with six one-dimensional convolutional layers and two one-dimensional maximum pooling layers; The one-dimensional convolution layer of the first convolutional neural network uses a relatively large convolution kernel to extract the global spatial features of the downsampled fault vibration signal; The one-dimensional maximum pooling layer of the first convolutional neural network is used to reduce the dimension of the global spatial features extracted by the one-dimensional convolutional layer of the first convolutional neural network, and extract the corresponding local spatial features; The one-dimensional convolution layer of the second convolutional neural network uses a relatively small convolution kernel to extract the global spatial features of the downsampled fault vibration signal; The one-dimensional maximum pooling layer of the second convolutional neural network is used to reduce the dimension of the global spatial features extracted by the one-dimensional convolutional layer of the second convolutional neural network, and extract the corresponding local spatial features.

2. A fault diagnosis method according to claim 1, characterized in that: The preprocessing of the collected historical fault vibration signals includes: Define a function for obtaining historical fault vibration signals to integrate historical fault vibration signals of different fault types in different folders into one data set; According to the fault type of the historical fault vibration signal, a label function of the historical fault vibration signal is defined.

3. A fault diagnosis method as claimed in claim 1, characterized in that: The fault diagnosis model integrating the one-dimensional dual-channel convolution and the bidirectional long short-term memory neural network also includes a Dropout layer, which is used to discard certain connection points during neural network training to prevent overfitting.

4. A fault diagnosis method as claimed in claim 1, characterized in that: The calculation formula for the side length of the feature graphs of the first convolutional neural network and the second convolutional neural network is: or in, Represents the edge length of the output feature map; Represents the edge length of the input feature map; Represents the convolution step size; Indicates the convolution kernel length, and the calculation results are rounded up.

5. A fault diagnosis system, characterized in that: include: A data acquisition module, used to acquire the vibration signal of the device under test; A fault diagnosis classification module is used to input the vibration signal of the equipment under test into a fault diagnosis model that integrates the trained one-dimensional dual-channel convolution and bidirectional long short-term memory neural network to obtain the fault diagnosis classification result of the equipment under test; The fault diagnosis model integrating the one-dimensional dual-channel convolution and the bidirectional long short-term memory neural network comprises a first convolution neural network, a second convolution neural network, a feature fusion module, a first bidirectional long short-term memory network, a second bidirectional long short-term memory network, a batch normalization layer and a fully connected layer; The first convolutional neural network and the second convolutional neural network are both used to extract local spatial features of the downsampled fault vibration signal; The feature fusion module is used to fuse the local spatial features extracted by the first convolutional neural network and the second convolutional neural network to obtain spatial features of different scales; The first bidirectional long short-term memory network and the second bidirectional long short-term memory network are both used to extract temporal features according to the fused spatial features; The batch normalization layer is used to perform batch normalization on the extracted time series features; The fully connected layer is used to output the fault type according to the batch normalization result; The first convolutional neural network is provided with three one-dimensional convolutional layers and one one-dimensional maximum pooling layer, and the second convolutional neural network is provided with six one-dimensional convolutional layers and two one-dimensional maximum pooling layers; The one-dimensional convolution layer of the first convolutional neural network uses a relatively large convolution kernel to extract the global spatial features of the downsampled fault vibration signal; The one-dimensional maximum pooling layer of the first convolutional neural network is used to reduce the dimension of the global spatial features extracted by the one-dimensional convolutional layer of the first convolutional neural network, and extract the corresponding local spatial features; The one-dimensional convolution layer of the second convolutional neural network uses a relatively small convolution kernel to extract the global spatial features of the downsampled fault vibration signal; The one-dimensional maximum pooling layer of the second convolutional neural network is used to reduce the dimension of the global spatial features extracted by the one-dimensional convolutional layer of the second convolutional neural network, and extract the corresponding local spatial features.

6. A processing device, characterized in that: The method comprises computer program instructions, wherein the computer program instructions are used to implement the steps corresponding to the fault diagnosis method according to any one of claims 1 to 4 when executed by a processing device.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, wherein the computer program instructions are used to implement the steps corresponding to the fault diagnosis method according to any one of claims 1 to 4 when executed by a processor.

Images