Fault diagnosis method and device based on HACAM-WDCNN

Abstract

The invention provides a fault diagnosis method based on HACAM-WDCNN, and provides a rotating machinery fault diagnosis method based on HACAM-WDCNN for solving the problems of difficulty in construction of a multi-channel input mutual relation of operation data, local feature deficiency and the like. According to the method, the local features are extracted by using two-layer point-by-point convolution before global feature attention distribution, so that the model is lightweight as much as possible, and meanwhile, the SKnet and the ECA are combined, so that the problem that the convolution kernel size and the channel number of the original ECA model are difficult to optimally match is solved; the thought of combination of local and global features improves the operation of attention ignoring of local features and rough global compression of the features in the previous channel. And meanwhile, by using the cavity convolution and the point-by-point convolution, the lightweight is realized to the maximum extent, the model parameters are reduced, and the training time is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a fault diagnosis method and device based on HACAM-WDCNN. Background Art

[0002] With the breakthrough development of modern information technologies such as the new generation of communication technology, cloud computing, big data, artificial intelligence, and virtual reality, mechanical fault diagnosis has entered the "big data" era, and intelligent algorithms based on deep learning and other methods have achieved good results in the field of fault diagnosis. However, previous studies have mostly focused on fault diagnosis of single sensor signals, lacking the effective use of multi-source operating data information.

[0003] Many key components of equipment cannot guarantee long-term healthy operation due to their complex and harsh working conditions. They often shut down for maintenance, which causes incalculable economic and property losses and even serious consequences such as life safety. Therefore, in order to ensure the safety of equipment operation, it is of great significance to quickly and accurately diagnose the faults of key components.

[0004] Traditional channel attention models channel importance by reducing and then increasing the dimension of the features of global average pooling. This method makes it difficult to construct the relationship between channels and does not take into account the loss of local features caused by the global pooling operation. Summary of the invention

[0005] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a fault diagnosis method and device based on HACAM-WDCNN.

[0006] According to a first aspect, the present invention provides a fault diagnosis method based on HACAM-WDCNN, the method comprising:

[0007] Obtain historical operation data when the equipment fails, and preprocess the historical operation data to obtain preprocessed data;

[0008] Constructing a local attention module and a global attention module, combining the local attention module and the global attention module into a HACAM module using a residual connection method, and embedding the HACAM module into a specific convolution block to construct a HACAM-WDCNN diagnostic model;

[0009] Inputting the preprocessed data into the HACAM-WDCNN diagnostic model for model training and testing to obtain a trained diagnostic model;

[0010] The real-time operation data of the equipment is obtained, and the real-time operation data of the equipment is input into the trained diagnosis model to obtain the real-time diagnosis result of the equipment fault.

[0011] Preferably, the specific steps of constructing the HACAM-WDCNN diagnostic model are: using point-by-point convolution to extract local features to construct a local attention module; extracting global features in parallel through a global maximum pooling layer and a global average pooling layer, and extracting the interaction between the two through an improved SKnet to construct a global attention module; using a residual connection method to connect the local attention module and the global attention module to their respective original data, respectively, and combine them into a HACAM module; embedding the HACAM module into a specific convolution block of WDCNN to construct a HACAM-WDCNN diagnostic model.

[0012] Preferably, the HACAM-WDCNN diagnostic model includes: a convolution block, a HACAM module, a Dropout layer, and a fully connected block; the Dropout layer is located between the convolution block and the HACAM module, the HACAM module is embedded in the middle layer of the WDCNN, and the fully connected block is connected to the last convolution block.

[0013] Preferably, the specific steps of extracting global features in parallel through the global maximum pooling layer and the global average pooling layer include: based on the ECA attention mechanism, adding a global maximum pooling layer to the Squeeze part; based on the SKnet attention mechanism, the global pooling layer uses an improved SKnet with shared parameters to extract global features in parallel; based on the above adjustment results, determining the influence of the features extracted by the global maximum pooling layer and the features extracted by the global average pooling layer on fault diagnosis.

[0014] Preferably, the formula for connecting the local attention module and the global attention module to their respective original data using the residual connection method is:

[0015]

[0016] in, Represents the corresponding product of elements; F is the input feature, F∈R C×L , where R is a real number used to reflect the dimension of the parameter, C is the number of channels of the feature, and L is the length of the feature map; M L is the local attention coefficient, M L ∈R C×1 ;M G is the global attention coefficient, M G ∈R C×1 , where R is a real number used to reflect the dimension of the parameter, and C is the number of feature channels.

[0017] Preferably, the specific steps of inputting the preprocessed data into the HACAM-WDCNN diagnostic model for model training and testing are: obtaining a one-dimensional time series signal sample in the preprocessed data, and performing equal-length interception on the one-dimensional time series signal sample to obtain one-dimensional data. The one-dimensional data is set as a training set, a validation set, and a test set according to a preset ratio; the training set is used to train the HACAM-WDCNN diagnostic model, and the validation set is used to validate the trained diagnostic model, and finally the test set is used to test the trained diagnostic model.

[0018] Preferably, the preprocessed data is input into the HACAM-WDCNN diagnostic model for model training and testing. After the trained diagnostic model is obtained, it also includes: using the trained diagnostic model to diagnose faults under different working conditions to verify the robustness and noise resistance of the model.

[0019] Preferably, the specific steps for verifying the robustness and noise resistance of the model are: inputting the test data under different working conditions into the trained diagnostic model for experimental testing, and verifying the robustness according to the test results; adding white noise of different powers to the test data under the same working conditions, conducting experimental testing with strong noise data under variable working conditions, and verifying the noise resistance according to the test results.

[0020] According to the second aspect, the present invention provides a fault diagnosis device based on HACAM-WDCNN, which is used to implement the fault diagnosis method based on HACAM-WDCNN of the first aspect, including: a data acquisition unit, used to acquire historical operation data when the equipment fails, and preprocess the historical operation data to obtain preprocessed data; a model building unit, used to construct a local attention module and a global attention module, and use the residual connection method to combine the local attention module and the global attention module into a HACAM module, and embed the HACAM module into a specific convolution block to construct a HACAM-WDCNN diagnosis model; a model training unit, used to input the preprocessed data into the HACAM-WDCNN diagnosis model for model training and testing to obtain a trained diagnosis model; a real-time diagnosis unit, used to acquire real-time operation data of the equipment, and input the real-time operation data of the equipment into the trained diagnosis model to obtain real-time diagnosis results of equipment failures.

[0021] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has the following advantages:

[0022] The present invention aims to solve the problems of difficulty in constructing the relationship between multi-channel inputs of operating data and missing local features, and proposes a rotating machinery fault diagnosis method based on HACAM-WDCNN. The present invention uses two layers of point-by-point convolution to extract local features before allocating global feature attention, making the model as lightweight as possible. At the same time, SKnet is combined with ECA to solve the problem that the convolution kernel size of the original ECA model is difficult to optimally match with the number of channels. The idea of ​​combining local and global features improves the previous channel attention that ignores local features and roughly compresses the features globally. At the same time, the use of dilated convolution and point-by-point convolution achieves lightweightness to the greatest extent, reduces model parameters, and shortens training time. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 is a flow chart of a fault diagnosis method based on HACAM-WDCNN according to an embodiment of the present invention;

[0025] Figure 2 is a schematic diagram of an original ECA process according to an embodiment of the present invention;

[0026] Figure 3 is a schematic diagram of an original SKnet according to an embodiment of the present invention;

[0027] Figure 4 is a schematic diagram of a local attention module network structure according to an embodiment of the present invention;

[0028] Figure 5 is a schematic diagram of implementation details of the global attention module of an embodiment of the present invention;

[0029] Figure 6 is a schematic diagram of an improved SKnet according to an embodiment of the present invention;

[0030] Figure 7 is a schematic diagram of a residual connection process according to an embodiment of the present invention;

[0031] Figure 8 is a schematic diagram of the HACAM-WDCNN network structure according to an embodiment of the present invention;

[0032] Fig. 9 is a structural block diagram of a HACAM set according to an embodiment of the present invention;

[0033] Fig.10 is a schematic diagram of a test bench according to an embodiment of the present invention;

[0034] Fig.11 It is a schematic diagram of an experimental fault diagnosis process of an embodiment of the present invention;

[0035] Fig.12 It is a structural schematic diagram of a fault diagnosis device based on HACAM-WDCNN according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] The present invention provides a fault diagnosis method based on HACAM-WDCNN, which selects representative rolling bearings in rotating machinery as the diagnosis object, takes a rotor data set collected by a rotor fault test bench designed and manufactured by a certain company as an example for experiment, and simulates different faults encountered by the rotating machinery by adding counterweights to small holes at different positions. The data set includes operation data collected by two X-axis acceleration sensors, two Y-axis acceleration sensors, a current sensor, a voltage sensor, a torque sensor, and a pressure sensor, and constructs a HACAM-WDCNN diagnosis model to accurately predict the location and type of rolling bearing faults, thereby achieving the purpose of efficient and accurate diagnosis of rotating machinery faults.

[0038] According to an embodiment of the present invention, a fault diagnosis method embodiment based on HACAM-WDCNN is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0039] In this embodiment, a fault diagnosis method based on HACAM-WDCNN is provided, which can be used for the above-mentioned mobile terminals, such as mobile phones, tablet computers, etc. Figure 1 is a flow chart of a fault diagnosis method based on HACAM-WDCNN according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0040] Step S1: Acquire historical operation data when the equipment fails, and preprocess the historical operation data to obtain preprocessed data.

[0041] In this embodiment, the operating data of the equipment when a fault occurs is collected, including vibration, current, voltage, etc., and the operating data is standardized to complete the preliminary data preprocessing. In actual engineering applications, the widely used interception length is 1024 or 2048. In this embodiment, 1024 samples are selected as the unit length to intercept the original vibration signal, and the standardization formula is:

[0042]

[0043] Where x is the input sample, z is the sample value after standardization, u is the mean of the 1024 sample values, and SD is the tolerance of the 1024 sample values.

[0044] In this embodiment, data preprocessing includes but is not limited to data outlier processing, missing value supplementation and data standardization, which can be specifically performed by replacing data outliers with the median, filling missing values ​​with the mean, and then performing data standardization. The present invention is not limited to this, and it is sufficient to achieve data standardization.

[0045] Step S2: construct a local attention module and a global attention module, use the residual connection method to combine the local attention module and the global attention module into a HACAM module, and embed the HACAM module into a specific convolution block to construct a HACAM-WDCNN diagnostic model.

[0046] In this embodiment, Figure 2 The figure shows the original ECA process. First, the feature map of shape (B, C, L) is globally averaged pooled to obtain a feature map of shape (B, C, 1), where B represents the number of samples in a batch, C represents the number of input channels, and L is the length of the time series. Then, the feature channel dimension and the time dimension are converted to obtain a feature map of shape (B, 1, C). Finally, a fixed-size convolution kernel is used to perform a convolution operation on the last dimension, i.e., the current channel dimension, to construct the relationship between channels.

[0047] like Figure 3 As shown in the figure, the original SKnet uses convolutions of multiple branches with different sizes to extract the features of the input vector, adds the convolution results together, and then generates the corresponding weights of different branches through operations such as flat pooling and fully connected layers.

[0048] Optionally, the specific steps of constructing the HACAM-WDCNN diagnostic model are: using point-by-point convolution to extract local features to construct a local attention module; extracting global features in parallel through a global maximum pooling layer and a global average pooling layer, and extracting the interaction between the two through an improved SKnet; using a residual connection method to connect the local attention module and the global attention module to their respective original data, and combining them into a HACAM module; embedding the HACAM module into a specific convolution block of WDCNN to construct a HACAM-WDCNN diagnostic model.

[0049] In this embodiment, Figure 4 As shown in the figure, point-by-point convolution is used to extract local features and build a local attention module. The implementation details are as follows:

[0050]

[0051] In the formula, F1 C Represents the convolution kernel size of 1, and the number is and two layers of convolution of C; σ is the Sigmoid function; δ is the Relu function.

[0052] In this embodiment, Figure 5 As shown, the implementation details of the global attention module are:

[0053]

[0054] X GMP =Max(X1,X2,...,X L )

[0055] X' GAP =RS(X GAP )

[0056] X' GMP =RS(X GMP )

[0057] In the formula, X i is the i-th element of the input feature X in the time dimension, L is the length of each sample time series, RS is the feature reshaping operation, X GAP is the global average feature after global average pooling, X GMP is the global maximum feature after global maximum pooling. X' GAP ,X' GMP They represent the global average feature and the global maximum feature after the feature reshaping operation. The time series length of this embodiment is 1024.

[0058] Optionally, the specific steps of extracting global features in parallel through a global maximum pooling layer and a global average pooling layer include: based on the ECA attention mechanism, adding a global maximum pooling layer to the Squeeze part; based on the SKnet attention mechanism, the global pooling layer uses an improved SKnet with shared parameters to extract global features in parallel; based on the above adjustment results, determining the impact of the features extracted by the global maximum pooling layer and the features extracted by the global average pooling layer on fault diagnosis.

[0059] In this embodiment, the improved ECA attention mechanism adds a global maximum pooling layer in the Squeeze part to extract features in parallel with the original global average pooling layer.

[0060] In this embodiment, the improved SKnet is different from the previous SKnet in the multi-channel input usage scenario. Using SKnet to extract the ECA channel interaction relationship only requires a single-channel convolution layer, and the dimensionality reduction operation is not conducive to the extraction of the interaction relationship. Therefore, the dimensionality reduction operation in the original Sknet is abandoned, and the interaction relationship between the two is extracted through a single-channel convolution layer.

[0061] In this embodiment, Figure 6 As shown in Figure 1, the maximum pooling layer and the average pooling layer extract the multi-scale channel relationship through the shared parameter SKnet, and the shared parameter operation establishes a connection between the two branch features. The extracted feature dimension is converted and input into the improved SKnet with shared parameter operation, which can be regarded as a special case that the number of input feature channels of SKnet is 1 and the extracted features are no longer time series features but channel relationships. The shared parameter operation is the global average feature X MAX and X GAP The global maximum feature uses the same improved SKnet for feature extraction, and the operation of shared parameter operation takes into account the mixed information of the original features. Afterwards, since the number of SKnet input feature channels is 1, the dimension reduction operation in SKnet is discarded. Considering that the number of fault diagnosis channels is generally between 32 and 128, an improved SKnet with two branches and convolution kernel sizes of 3 and 5 is selected. Finally, the features output by the improved SKnet are added, and the global attention weight coefficient is generated using sigmoid.

[0062] Taking the global average pooling feature as input, the improved SKnet can be described as follows: In this embodiment, convolution layers with convolution kernel sizes of 3 and 5 are used to extract features respectively and the results are added element by element. The specific formula is:

[0063]

[0064] In the formula, It indicates a convolutional layer with a kernel size of 5 and a kernel number of 1. It indicates a convolutional layer with a kernel size of 3 and a kernel number of 1. It means element-by-element addition; Indicated by Extracted deep features; Indicated by The extracted deep features; U represents the fused features.

[0065] In this embodiment, an average pooling operation is performed in the channel dimension, a fully connected layer is used to generate weight coefficients, and Softmax is used to control the coefficients between 0 and 1, and the weights add up to 1. The specific implementation details are:

[0066]

[0067] a+b=1

[0068] Where, X i is the i-th element of the fusion feature U in the channel dimension, a and b are and Attention weight. Note that V MAX Calculation process and V GAP Exactly the same, no need to repeat the description here.

[0069] In this embodiment, M G =σ(RS(δ(V GAP )+δ(V MAX ))), where σ is the Sigmoid function, δ is the Relu function, V MAX , V GAP It is the feature map after the improved ECA maximum pooling operation is processed by SKnet with shared parameters, M G is the global attention coefficient.

[0070] In this embodiment, Figure 7 As shown, the local attention module and the global attention module are connected to their respective original data using the residual connection method, that is, the weight recalibration is realized by combining the global and local attention using the residual connection idea. The specific implementation details are as follows:

[0071]

[0072]

[0073] in, Represents the corresponding product of elements; F is the input feature, F∈R C×L , where R is a real number used to reflect the dimension of the parameter, C is the number of channels of the feature, and L is the length of the feature map; M L is the local attention coefficient, ML ∈R C×1 ;M G is the global attention coefficient, M G ∈R C×1 , where R is a real number used to reflect the dimension of the parameter, and C is the number of feature channels.

[0074] Optionally, the HACAM-WDCNN diagnostic model includes: a convolution block, a HACAM module, a Dropout layer, and a fully connected block; the Dropout layer is located between the convolution block and the HACAM module, the HACAM module is embedded in the middle layer of the WDCNN, and the fully connected block is connected to the last convolution block.

[0075] In this embodiment, Figure 8 As shown in the figure, HACAM-WDCNN includes five different convolution blocks arranged in sequence. Each convolution block consists of a convolution layer, a batch normalization layer, a maximum pooling layer, and a Relu activation function layer. The convolution kernel size of the first convolution layer is 64, which matches the input with a time step of 1024. The convolution kernel sizes of the remaining convolution layers are all 3. The number of convolution kernels in the convolution layers is 16, 32, 64, 64, and 64, respectively. The HACAM module is added after the second, third, and fourth convolution blocks, and the Dropout layer is located after these three specific convolution blocks and before HACAM, forming a HACAM set, as shown in the figure. Fig. 9 As shown in Figure 2, the fully connected block consists of two fully connected layers and a Relu activation function layer.

[0076] Step S3: Input the preprocessed data into the HACAM-WDCNN diagnostic model for model training and testing to obtain a trained diagnostic model.

[0077] Optionally, obtain one-dimensional time series signal samples in the preprocessed data, and perform equal-length truncation on the one-dimensional time series signal samples to obtain one-dimensional data; set the one-dimensional data as a training set, a validation set, and a test set according to a preset ratio; use the training set to train the HACAM-WDCNN diagnostic model, use the validation set to verify the trained diagnostic model, and finally use the test set to test the trained diagnostic model.

[0078] In this embodiment, each one-dimensional time series signal sample after preprocessing is cut into equal lengths, and the one-dimensional time series signal samples are set as training set, verification set and test set according to a preset ratio of 8:1:1; the initial HACAM-WDCNN diagnostic model is trained with the training set, and verified with the verification set, and finally the model is tested with the preprocessed test set.

[0079] In this embodiment, the cross entropy loss function is used to test the model. The cross entropy loss function used in the present invention is:

[0080]

[0081] In the formula, E is the objective function, n is the number of samples, and y k is the true value of the kth category, t k is the predicted value of the kth category.

[0082] In this embodiment, a gradient descent algorithm is used to train the fault diagnosis model. The gradient descent algorithm can be expressed as:

[0083]

[0084] In the formula, θ is the model parameter vector to be optimized, α is the learning rate, is the gradient of the loss function J(θ) with respect to the parameter θ.

[0085] The gradient calculation formula is:

[0086]

[0087] In the formula, Represents the function J on the i-th element θ i The partial derivative of .

[0088] Step S4: Acquire real-time operation data of the equipment, input the real-time operation data of the equipment into the trained diagnosis model, and obtain real-time diagnosis results of equipment faults.

[0089] Optionally, the preprocessed data is input into the HACAM-WDCNN diagnostic model for model training and testing. After the trained diagnostic model is obtained, it also includes: using the trained diagnostic model to diagnose faults under different working conditions to verify the robustness and noise resistance of the model.

[0090] Optionally, test data under different working conditions are input into the trained diagnostic model for experimental testing, and the robustness is verified based on the test results. White noise of different powers is added to the test data under the same working condition, and experimental testing is performed using strong noise data under variable working conditions, and noise resistance is verified based on the test results.

[0091] In this embodiment, generalization experiments and noise resistance experiments are carried out respectively. In the generalization experiment, the training model and the test model are tested using data sets of different working conditions. In the noise resistance experiment, Gaussian white noise of different powers is added to the data set, and the model is tested using strong noise data of variable working conditions.

[0092] In this embodiment, the generalization experiment and the noise resistance experiment are cross-tested. For example, the model is trained with a data set of a working condition with a rotation speed of 500r / min as the training set, and the model is tested with a data set of a working condition with a rotation speed of 1000r / min. In the second part, Gaussian white noise is added, and the model is tested with strong noise data of variable working conditions. For example, Gaussian white noise of -4db, -2db, 0db, and 2db is added to the data set respectively, and a cross-test is performed. The robustness and noise resistance of the model proposed in the present invention are tested through these two parts.

[0093] The present invention utilizes SKnet to selectively apply convolution kernels of different scales, thereby capturing this characteristic of multi-scale features and realizing multi-scale construction of channel relationships. On the basis of the original ECA global average pooling, a global maximum pooling branch is added, and the information of the two branches is comprehensively considered using SKnet with shared parameters. A local attention module is built to extract local features, thereby building a HACAM-WDCNN diagnostic model, solving the problem that the original ECA model convolution kernel size and the number of channels are difficult to optimally match. The idea of ​​combining local and global features improves the previous channel attention that ignores local features and roughly compresses the features globally. At the same time, the use of void convolution and point-by-point convolution achieves lightweight reduction of model parameters and shortens training time to the greatest extent.

[0094] In order to further verify the effectiveness and feasibility of the method of the present invention, the present invention conducts experiments on a rotor fault data set collected by a rotor fault test bench designed and manufactured by a company. Fig.10 As shown in the figure, different faults are simulated by adding counterweights to small holes at different positions. The data set includes operating data collected by two X-axis accelerometers, two Y-axis accelerometers, current sensors, voltage sensors, torque sensors, and pressure sensors. The sensor sampling frequency is 8K, with 250,000 sampling points. The ten common health states of the rotor are trained and tested, and the experimental data are shown in Table 1:

[0095] Table 1

[0096]

[0097] like Fig.11As shown in the figure, in this experiment, the data set of simulated faults in the fault experiment is regarded as historical data. After outlier processing and missing value supplementation, the historical data is standardized to obtain standard preprocessed data. The preprocessed data is then divided into a validation set, a training set and a test set according to a predetermined data division standard. The validation set is used to verify the model, adjust the model parameters and evaluate the model; the training set is used for model training, and the data in the training set is optimized using the cross entropy loss function. When the optimization conditions are not met, the data is gradient descent adjusted, and then the adjusted data is optimized again using the cross entropy loss function until the optimization conditions are met, and the optimized model is output; the test set is used to test the performance of the model after optimizing the parameters using the validation set and the training set, and obtain the rolling bearing classification results. Finally, the corresponding fault results are output, and each fault result is labeled to obtain the ten common health states in Table 1.

[0098] In this embodiment, a fault diagnosis device based on HACAM-WDCNN is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0099] This embodiment provides a fault diagnosis device based on HACAM-WDCNN, such as Fig.12 As shown, including:

[0100] A data acquisition unit, used to acquire historical operation data when the equipment fails, and pre-process the historical operation data to obtain pre-processed data;

[0101] A model building unit, used for constructing a local attention module and a global attention module, combining the local attention module and the global attention module into a HACAM module by using a residual connection method, and embedding the HACAM module into a specific convolution block to construct a HACAM-WDCNN diagnostic model;

[0102] A model training unit, used for inputting the preprocessed data into the HACAM-WDCNN diagnostic model for model training and testing to obtain a trained diagnostic model;

[0103] The real-time diagnosis unit is used to obtain the real-time operation data of the equipment, input the real-time operation data of the equipment into the trained diagnosis model, and obtain the real-time diagnosis result of the equipment fault.

[0104] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0105] The fault diagnosis device based on HACAM-WDCNN in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0106] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A fault diagnosis method based on HACAM-WDCNN, characterized in that: The method comprises: Obtain historical operation data when the equipment fails, and preprocess the historical operation data to obtain preprocessed data; Constructing a local attention module and a global attention module, combining the local attention module and the global attention module into a HACAM module using a residual connection method, and embedding the HACAM module into a specific convolution block to construct a HACAM-WDCNN diagnostic model; Inputting the preprocessed data into the HACAM-WDCNN diagnostic model for model training and testing to obtain a trained diagnostic model; The real-time operation data of the equipment is obtained, and the real-time operation data of the equipment is input into the trained diagnosis model to obtain the real-time diagnosis result of the equipment fault.

2. The fault diagnosis method based on HACAM-WDCNN according to claim 1, characterized in that: The specific steps of constructing the HACAM-WDCNN diagnostic model are: Use point-by-point convolution to extract local features and build a local attention module; The global features are extracted in parallel through the global maximum pooling layer and the global average pooling layer, and the interaction between the two is extracted by improving SKnet to build a global attention module; The local attention module and the global attention module are respectively connected to their respective original data using a residual connection method to form a HACAM module; The HACAM module is embedded into the specific convolution block of WDCNN to construct the HACAM-WDCNN diagnostic model.

3. The fault diagnosis method based on HACAM-WDCNN according to claim 2 is characterized in that: The HACAM-WDCNN diagnostic model includes: a convolution block, a HACAM module, a Dropout layer, and a fully connected block; The Dropout layer is located between the convolutional block and the HACAM module, the HACAM module is embedded in the middle layer of the WDCNN, and the fully connected block is connected to the last convolutional block.

4. The fault diagnosis method based on HACAM-WDCNN according to claim 2 is characterized in that: The specific steps of extracting global features in parallel through the global maximum pooling layer and the global average pooling layer include: Based on the ECA attention mechanism, a global maximum pooling layer is added to the Squeeze part; Based on the SKnet attention mechanism, the global pooling layer uses an improved SKnet with shared parameters to extract global features in parallel; Based on the above adjustment results, the influence of the features extracted by the global maximum pooling layer and the features extracted by the global average pooling layer on fault diagnosis is determined.

5. The fault diagnosis method based on HACAM-WDCNN according to claim 2 is characterized in that: The formula for connecting the local attention module and the global attention module to their respective original data using the residual connection method is: in, Represents the corresponding product of elements; F is the input feature, F∈R C×L , where R is a real number used to reflect the dimension of the parameter, C is the number of channels of the feature, and L is the length of the feature map; M L is the local attention coefficient, M L ∈R C×1 ;M G is the global attention coefficient, M G ∈R C×1 , where R is a real number used to reflect the dimension of the parameter, and C is the number of feature channels.

6. The fault diagnosis method based on HACAM-WDCNN according to claim 1, characterized in that: The specific steps of inputting the pre-processed data into the HACAM-WDCNN diagnostic model for model training and testing are: Acquire one-dimensional time series signal samples in the preprocessed data, and perform equal-length interception on the one-dimensional time series signal samples to obtain one-dimensional data; The one-dimensional data is set as a training set, a validation set, and a test set according to a preset ratio; The training set is used to train the HACAM-WDCNN diagnostic model, the validation set is used to validate the trained diagnostic model, and finally the test set is used to test the trained diagnostic model.

7. The fault diagnosis method based on HACAM-WDCNN according to claim 1, characterized in that: After the pre-processed data is input into the HACAM-WDCNN diagnostic model for model training and testing to obtain a trained diagnostic model, the method further includes: The trained diagnostic model is used to diagnose faults under different working conditions to verify the robustness and noise resistance of the model.

8. The fault diagnosis method based on HACAM-WDCNN according to claim 7 is characterized in that: The specific steps for verifying the robustness and noise resistance of the model are: Input the test data under different working conditions into the trained diagnostic model for experimental testing, and verify the robustness based on the test results; White noise of different powers is added to the test data under the same working conditions, and experimental tests are carried out using strong noise data under variable working conditions. The noise resistance is verified based on the test results.

9. A fault diagnosis device based on HACAM-WDCNN, used to implement the fault diagnosis method based on HACAM-WDCNN as described in claim 1, characterized in that: include: A data acquisition unit, used to acquire historical operation data when the equipment fails, and pre-process the historical operation data to obtain pre-processed data; A model building unit, used for constructing a local attention module and a global attention module, combining the local attention module and the global attention module into a HACAM module by using a residual connection method, and embedding the HACAM module into a specific convolution block to construct a HACAM-WDCNN diagnostic model; A model training unit, used for inputting the preprocessed data into the HACAM-WDCNN diagnostic model for model training and testing to obtain a trained diagnostic model; The real-time diagnosis unit is used to obtain the real-time operation data of the equipment, input the real-time operation data of the equipment into the trained diagnosis model, and obtain the real-time diagnosis result of the equipment fault.

Images