A hoist cross bearing fault diagnosis method and system

Abstract

The application discloses a hoist cross bearing fault diagnosis method and system, and relates to the technical field of bearing fault diagnosis. The method comprises the following steps: step one: a multi-dimensional space-time cooperative neural network is built, which comprises a domain-invariant feature extraction module, a multi-dimensional space-time feature fusion module and a bearing fault classification module; wherein the domain-invariant feature extraction module comprises a multi-scale convolution layer and a cooperative normalization layer, and the method proceeds to step two; the application innovatively proposes a multi-dimensional space-time cooperative neural network model, a cooperative normalization method is designed on the basis of a traditional multi-scale convolution neural network model, and the normalization of batch data and single sample data is mixed to reduce the covariance offset and overfitting phenomenon; meanwhile, the multi-dimensional space-time feature fusion module adaptively captures the information interaction of local cross channels and the importance of different features in the learning space, and enhances the extraction ability of the model to the domain-invariant features of bearing faults.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bearing fault diagnosis, and in particular to a hoist cross bearing fault diagnosis method and system. BACKGROUND

[0002] Mine hoist systems are the core link in mine operations, responsible for transporting ore, waste rock, personnel, and equipment from underground to the surface. These systems often operate under extreme working conditions, such as high loads, harsh weather, and complex geological conditions, making their key components, especially the main shaft bearings at both ends of the hoist, not only bear the full torque of the hoist, but also face great wear and failure risks as important parts of winding hoist steel wire ropes. The failure of the main shaft bearing not only causes the hoist to stop, but also may cause serious safety accidents, so it is crucial to effectively diagnose and predict the maintenance of the main shaft bearing of the mine hoist.

[0003] Currently, the bearing fault diagnosis of the main shaft of the mine hoist still faces a series of challenges:

[0004] 1. Multiple operating conditions: Mine hoist systems operate under different speed conditions, load requirements, and operating modes, resulting in varying operating conditions of the bearings;

[0005] 2. Difficulty in data collection: In mine hoist systems, due to the harsh environment and complex equipment layout, it is very difficult to achieve comprehensive monitoring of the bearing state and collect data;

[0006] 3. Model generalization ability requirement: Existing fault diagnosis models often rely on multi-source domain and labeled data information interaction. When the source domain information is reduced or only one source domain is available, how to extract key domain-invariant features in samples to achieve efficient generalization learning of the model still faces great challenges;

[0007] 4. Complexity of feature extraction: The fault features of the hoist bearing may be different at different time scales and frequencies, and traditional feature extraction methods are difficult to fully capture these complex general features. SUMMARY

[0008] The purpose of the present application is to solve the problems mentioned in the background art, and to provide a hoist cross bearing fault diagnosis method and system.

[0009] The purpose of the present application can be achieved by the following technical solutions:

[0010] In a first aspect, the present application provides a hoist cross bearing fault diagnosis method, comprising the following steps:

[0011] Step one: build a multi-dimensional space-time collaborative neural network, including a domain-invariant feature extraction module, a multi-dimensional space-time feature fusion module and a bearing fault classification module; wherein the domain-invariant feature extraction module includes a multi-scale convolution layer and a collaborative normalization layer, and step two is entered;

[0012] Step two: collect vibration signal data under different working conditions, and obtain source domain and target domain data sets after preprocessing; establish a bearing database for each working condition in the source domain, and divide the bearing database into a training set and a test set, and step three is entered:

[0013] Step three: build a multi-scale convolution layer, and input the initial feature signal X into the multi-scale convolution layer to obtain multi-scale convolution features X1, X2, …, X5 of X, and step four is entered;

[0014] Step four: build a collaborative normalization layer, input the multi-scale convolution features X1, X2, …, X5 into the collaborative normalization layer to obtain the normalized results Z1, Z2, …, Z5 of the features, and step five is entered;

[0015] Step five: build a multi-dimensional space-time feature fusion module, input the normalized results Z1, Z2, …, Z5 into the multi-dimensional space-time feature fusion module to realize adaptive extraction of important features in the channel and space, complete feature extraction of Layer1 layer, and then sequentially pass through layer2, Layer3, … Layer5 layer based on steps three to five to obtain multi-scale features MF1, MF2, …, MF5, and step six is entered;

[0016] Step six: cascade the multi-scale features MF1, MF2, …, MF5 along the channel to confirm the final fusion feature MF, and build a bearing fault classification module, input the fusion feature MF into the fault classification module to realize bearing fault classification in the source domain, and step seven is entered;

[0017] Step seven, based on steps three to six, repeatedly train the overall network model to obtain a trained overall network model, and input the test set of the target domain into the trained overall network model to obtain the test set accuracy.

[0018] As a preferred embodiment of the application, in step two, vibration signal data under different working conditions is collected, and source domain and target domain data sets are obtained after preprocessing; a bearing database for each working condition in the source domain is established, and the bearing database is divided into a training set and a test set, specifically:

[0019] Based on the published PU bearing data set, bearing data sets under multiple different working conditions are collected, m sample data are taken for each working condition, sample databases in the source domain and the target domain are established, and the sample databases are divided into a training set and a test set, and the sample ratio of the training set and the test set is 8:2.

[0020] As a preferred embodiment of the present application, the multi-scale convolution layer comprises 5 convolution layers with different kernel sizes and 1 wide convolution layer with a kernel size of 64, specifically:

[0021] On the PU dataset, the initial input feature signal X is passed through the multi-scale convolution layer to obtain the multi-scale convolution features X1, X2, …, X5 of X, i.e. the convolution features under different kernel sizes:

[0022] X1 = conv1(conv(X));

[0023] X2 = conv2(conv(X));

[0024] X3 = conv3(conv(X));

[0025] X4 = conv4(conv(X));

[0026] X5 = conv5(conv(X));

[0027] where X is the input feature signal, conv is the wide convolution layer, and conv1, conv2, conv3, conv4, and conv5 are convolution layers with different scales.

[0028] As a preferred embodiment of the present application, a cooperative normalization layer is built, and the multi-scale convolution features X1, X2, …, X5 obtained in step 3 are input into the cooperative normalization layer to obtain the normalized results Z1, Z2, …, Z5 of the features. Taking the multi-scale convolution feature X1 as an example, the specific steps are as follows:

[0029] Step 4-1, each channel is divided into two sub-channels C1 and C2, which is achieved by:

[0030] X1 c1 = X1[:, 0:C / / 2, :, :]

[0031] X1 c2 = X1[:, C / / 2:c, :, :]

[0032] Step 4-2, in channel C1, the data is divided into several groups and the BN operation is performed, and the parameters of the data are updated by group; Specifically:

[0033]

[0034] where, represents the mean value obtained by performing batch normalization in channel C1, denotes the variance value obtained by performing batch normalization in the channel C1, denotes the scale transformation and offset of the result of implementing batch normalization in the channel C1.

[0035] Step 4-3, in the channel C2, Solve the mean and standard deviation of each sample in the H, W dimension based on the IN strategy, and keep the N, C dimension, that is, only calculate the mean and standard deviation within the channel, specifically:

[0036]

[0037]

[0038] wherein, denotes the mean value obtained by performing batch normalization in the channel C2, denotes the variance value obtained by performing batch normalization in the channel C1, denotes the scale transformation and offset of the result of implementing instance normalization in the channel C2; I denotes instance normalization.

[0039] Step 4-4, recombine the normalized sub-channel results to obtain the final output Z:

[0040] Z1=[Y B，1 , Y I，1 , Y B，2 , Y I，2 ,......, Y B，C / / 2 , Y I，C / / 2 ];

[0041] wherein, BN is batch normalization, IN is instance normalization, is the feature value of each sample in the N, H and W dimensions in the channel C1, is the feature value of each sample in the H, W dimensions in the channel C2, γc and βc are learnable parameters, ε is a constant, and [.] denotes splicing along the channel dimension.

[0042] As a preferred embodiment of the application, a multi-dimensional spatio-temporal feature fusion module is built, the normalized results Z1, Z2, …, Z5 are input into the multi-dimensional spatio-temporal feature fusion module, adaptive extraction of important features in the channel and space is realized, feature extraction of the Layer1 layer is completed, then multi-scale features MF1, MF2, …, MF5 are obtained based on steps three to five through layer2, Layer3, …, Layer5 layers in turn, and taking the normalized result Z1 as an example, the input of the multi-dimensional spatio-temporal feature fusion module is specifically:

[0043] Step 5-1, for a given input feature Z1, respectively perform the maximum pooling and average pooling operations to obtain two features Im and Ia, then perform feature concatenation on Im and Ia, and then input to the channel attention branch and the spatial attention branch respectively;

[0044] Im = max(Z1);

[0045]

[0046] Sc and Ss are the results of feature concatenation on Im and Ia;

[0047] Step 5-2, in the channel attention branch, input the feature Sc into the channel weight generation unit, determine the convolution kernel size of two one-dimensional convolution layers as k1 and k2 through input channel number self-adaption, then input the feature Sc into the above two one-dimensional convolution layers respectively, self-adaptively extract the deep features of Sc, and finally pass through the Softmax layer to obtain the attention weights Wc1 and Wc2 in the channel branch;

[0048] Wc1 = σ [Conv1(Sc, max(1, [(γ1×C+b1) / 2]))];

[0049] Wc2 = σ [Conv2(Sc, max(1, [(γ2×C+b2) / 2]))];

[0050] Wherein, γ and b are hyperparameters, [.] represents an odd integer obtained by rounding up, Conv1 and Conv2 are one-dimensional 1*1 convolution, and σ is a Softmax activation function;

[0051] Step 5-3, in the spatial attention branch, input the feature Ss into the spatial weight generation unit, and the feature Ss respectively passes through two 7*7 convolution layers to obtain the initial features S s1 and S s1' in the spatial branch, then realize the dimensionality of the features through the linear layer FC1, the Relu activation function and the linear layer FC2, and obtain the important features S s2 and S s2' of the sample in the spatial branch, and finally obtain the attention weights Ws1 and Ws2 in the spatial branch through the Softmax activation function;

[0052] Ws1 = σ (FC2(η(FC1(Conv3(Ss))))) ;

[0053] Ws2 = σ (FC2(η(FC1(Conv4(Ss))))) ;

[0054] Wherein, Conv3 and Conv4 are 7*7 convolution;

[0055] Step 5-4, aggregate the channel weight and the spatial weight to determine the important part of the feature between the channel and the space, specifically:

[0056] MF1 1 = (Wc1+Ws1) x Im+ (Wc2+Ws2) x Ia;

[0057] complete the feature extraction of the Layer1 layer to obtain the feature MF1 1 , and then the feature MF1 1 based on steps 3 to 5, sequentially pass through layer2, Layer3, …… Layer5 layers to obtain multi-scale features MF1.

[0058] As a preferred embodiment of the present application, in step six, the cascade of multi-scale features MF1, MF2, ……, MF5 along the channel is realized, the final fusion feature MF is confirmed, and the bearing fault classifier is built, the fusion feature MF obtained in step six is input into the fault classifier to realize the fault classification in the source domain, specifically:

[0059] MF = Concate [MF1, MF2, MF3, MF4, MF5];

[0060] OutPut = σ (Fc (MF));

[0061] Combined with the cross-entropy loss function to realize the fault classification of the bearing:

[0062] L = - (ylog (OutPut) + (1-y) log (1-OutPut));

[0063] Wherein, Fc is the full connection layer, σ is the Softmax activation function, Output is the output classification result of the model, L is the classification loss, and y is the fault label in the source domain.

[0064] As a preferred embodiment of the present application, in step seven, the overall network model is repeatedly trained based on steps three to six to obtain the trained overall network model, and the test set of the target domain is input into the trained overall network model to obtain the test set accuracy and feature visualization result, specifically:

[0065] Step 7-1, select the learning rate, optimizer, training batch size and training round number of the overall network model;

[0066] Step 7-2, input the training data set obtained in step two into the set bearing fault diagnosis model to obtain the trained cross-bearing fault diagnosis model;

[0067] Step 7-3, inputting test data under different working conditions in the target domain into the trained cross-bearing fault diagnosis model to realize cross-working condition fault diagnosis of the bearing.

[0068] In a second aspect, the application provides a hoist cross-bearing fault diagnosis system, comprising a domain-invariant feature extraction module, a multi-dimensional spatio-temporal feature fusion module and a bearing fault classification module.

[0069] The domain-invariant feature extraction module is composed of a multi-scale convolution layer and a collaborative normalization layer, and is used to realize adaptive extraction and fusion of important features of input samples; the multi-scale convolution layer is composed of a small convolution layer with different sizes of convolution kernels and a wide convolution layer, and is used to obtain feature relationships between different scales of samples and long-term relationship features of all input samples; the collaborative normalization layer simultaneously learns statistical characteristics between channels and independent characteristics between instances, and reduces the covariance shift and overfitting phenomenon through mixed normalization of batch data and single sample data;

[0070] The multi-dimensional spatio-temporal feature fusion module innovatively replaces the traditional (maximum / average) pooling module, and adaptively captures local cross-channel information interaction; a spatial weight generation unit is designed in the spatial branch thereof to adaptively learn the importance of different features in the space.

[0071] The bearing fault classification module is composed of a full connection layer and a Softmax activation function, and is used to input the fused convolution features into a bearing fault classifier to output a classification result of bearing faults.

[0072] Compared with the prior art, the application has the following beneficial effects:

[0073] 1. The application uses only data of a single source domain to extract general domain-invariant features, and can realize intelligent fault diagnosis of bearings across working conditions without relying on any information of a target domain.

[0074] 2. The application combines the advantages of general feature extraction of the multi-dimensional spatio-temporal feature fusion module and the advantages of reducing sample internal covariance shift of the collaborative normalization on the basis of the traditional multi-scale convolution neural network model, and the average accuracy of the method reaches 98.74% in six different cross-working condition (cross-speed, cross-load) tasks of the PU bearing data set, which significantly improves the accuracy and model generalization ability in cross-working condition bearing fault diagnosis compared with some mainstream domain generalization methods. BRIEF DESCRIPTION OF DRAWINGS

[0075] In order to facilitate understanding of those skilled in the art, the application will be further described below with reference to the drawings.

[0076] Figure 1 The flowchart of the method;

[0077] Figure 2 A framework diagram of a multi-dimensional space-time collaborative neural network model;

[0078] Figure 3 A schematic diagram of collaborative normalization;

[0079] Figure 4 A framework diagram of a multi-dimensional space-time feature fusion module. DETAILED DESCRIPTION

[0080] The technical solutions of the present application will be described clearly and completely below in conjunction with embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0081] Please refer to Figures 1-4 The present application provides a hoist cross-bearing fault diagnosis method, which comprises the following steps:

[0082] Step 1, build a network of a cross-bearing fault diagnosis model, including a domain-invariant feature extraction module, a multi-dimensional space-time feature fusion module and a bearing fault classification module; the domain-invariant feature extraction module includes a multi-scale convolution layer and a collaborative normalization layer.

[0083] Step 2, collect vibration signal data under different working conditions, and after a pretreatment step, obtain source domain and target domain data sets respectively, establish a bearing database for each working condition in the source domain, and divide the database into a training set and a test set, the number ratio of bearings in the training set and the test set is 8:2, and the specific steps are as follows:

[0084] Based on the disclosed bearing data set, collect bearing data sets under multiple different working conditions, take n sample data for each working condition, establish image databases in the source domain and the target domain, and divide the databases into a training set and a test set, the sample ratio of the training set and the test set is 8:2.

[0085] Step 3, the multi-scale convolution layer contains 5 convolution layers with different convolution kernel sizes and 1 wide convolution layer with a convolution kernel size of 64, and the specific steps are as follows:

[0086] On the PU data set, the initial input feature signal X passes through the multi-scale convolution layer to obtain the multi-scale convolution features X1, X2…, X5 of X, i.e. the convolution features under different convolution kernel sizes:

[0087] X1 = conv1(conv(X)) (1)

[0088] X2 = conv2(conv(X)) (2)

[0089] X3 = conv3(conv(X)) (3)

[0090] X4 = conv4(conv(X)) (4)

[0091] X5 = conv5(conv(X)) (5)

[0092] wherein X is an input feature signal, conv is a wide convolution layer, conv1, conv2, conv3, conv4, and conv5 are convolution layers of different scales.

[0093] Step 4, build a collaborative normalization layer, input the multi-scale convolution features X1, X2, …, X5 obtained in step 3 into the collaborative normalization layer to obtain the normalized results Z1, Z2, …, Z5 of the features. Taking the multi-scale convolution feature X1 as an example, the specific process is as follows:

[0094] Step 4-1, split each channel into two sub-channels C1 and C2. This can be achieved in the following way:

[0095] X1 c1 = X1[:, 0:C / / 2, :, :] (6)

[0096] X1 c2 = X1[:, C / / 2:c, :, :] (7)

[0097] Step 4-2, in channel C1, The data is divided into several groups and the BN operation is performed, and the parameters of the data are updated by group. A group of data determines the direction of the gradient at this time, thereby reducing the randomness of the descent. The specific process is as follows:

[0098]

[0099] Step 4-3, in channel C2, Based on the IN strategy, the mean and standard deviation of the data in the H and W dimensions of each sample are solved, and the N and C dimensions are retained, that is, only the mean and standard deviation within the channel are calculated. The specific process is as follows:

[0100]

[0101] Step 4-4, recombine the normalized sub-channel results to form the final output Z:

[0102] Z1 = [Y B，1 , Y I，1 , Y B，2 , Y I，2 , …, Y B，C / / 2 , Y I，C / / 2 ] (14)

[0103] where BN is batch normalization, IN is instance normalization, is the eigenvalue of each sample in channel C1 in N, H and W dimensions, is the eigenvalue of each sample in channel C2 in H and W dimensions, γc and βc are learnable parameters, and ε is a very small constant, which is 10 -5 or 10 -8 , [.] represents splicing along the channel dimension.

[0104] Step 5, build a multi-dimensional spatio-temporal feature fusion module, input the normalized results Z1, Z2, …, Z5 into the multi-dimensional spatio-temporal feature fusion module, realize adaptive extraction of important features in channels and space, and obtain multi-scale features MF1, MF2, …, MF5. Taking feature Z1 as an example, the input of the multi-dimensional spatio-temporal feature fusion module is as follows:

[0105] Step 5-1, for a given input feature Z1, two features Im and Ia are obtained after maximum pooling and average pooling operations are performed respectively, then Im and Ia are input into the channel attention branch and the spatial attention branch after feature splicing.

[0106] Im = max (Z1) (15)

[0107]

[0108] where Sc represents the input of the channel attention branch, and Ss represents the input of the spatial attention branch.

[0109] Step 5-2, in the channel attention branch, the feature Sc is input into the channel weight generation unit. First, two one-dimensional convolution layers with adaptive input channel number are determined, the convolution kernel size is k1 and k2, then the feature Sc is input into the above two one-dimensional convolution layers respectively, the deep features of Sc are adaptively extracted, and finally the attention weights Wc1 and Wc2 in the channel branch are obtained through the Softmax layer. The specific process is as follows:

[0110] Wc1 = σ [Conv1 (Sc, max (1, [(γ1×C+b1) / 2]))] (18)

[0111] Wc2 = σ [Conv2 (Sc, max (1, [(γ2×C+b2) / 2]))] (19)

[0112] where γ and b are hyperparameters, usually set as γ = 2 and b = 1. [.] represents the odd number of upward rounding, Conv1 and Conv2 are one-dimensional 1*1 convolution, and σ is the Softmax activation function.

[0113] Step 5-3, in the spatial attention branch, similarly, the feature Ss is input into the spatial weight generation unit. First, the feature Ss passes through two 7*7 convolution layers respectively to obtain the initial features Ss1 and Ss2 in the spatial branch. s1' Then, the dimension of the feature is increased and decreased through the linear layer FC1, the Relu activation function and the linear layer FC2 to obtain the important features Ss2 and Ss2 of the sample in the spatial branch. s2' Finally, the attention weights Ws1 and Ws2 in the spatial branch are obtained through the Softmax activation function. Specifically as follows:

[0114] Ws1=σ(FC2(η(FC1(Conv3(Ss)))))(20)

[0115] Ws2=σ(FC2(η(FC1(Conv4(Ss)))))(21)

[0116] Where, Conv3, Conv4 is 7*7 convolution.

[0117] Step 5-4, the channel weight and the spatial weight are summarized to determine the important part of the feature between the channel and the space. Specifically as follows:

[0118] MF1 1 =(Wc1+Ws1)×Im+(Wc2+Ws2)×Ia(22)

[0119] The feature extraction of Layer1 layer is completed to obtain the feature MF1 1 Then the feature MF1 1 Based on steps 3 to 5, the multi-scale features MF1 are obtained by sequentially passing through layer2, Layer3, …, Layer5 layers.

[0120] Where, Output represents the fused feature. Since the sum of the feature weights of Im and Ia is equal to 1, the useful part between the features Im and Ia is retained, and the useless part is discarded, thereby realizing effective feature fusion.

[0121] Step 6, the cascade of multi-scale features MF1, MF2, …, MF5 along the channel is realized to confirm the final fused feature MF, and a bearing fault classifier is built. The fused feature MF obtained in step 6 is input into the fault classifier to realize the fault classification in the source domain. Specifically as follows:

[0122] MF=Concate[MF1,MF2,MF3,MF4,MF5](23)

[0123] OutPut=σ(Fc(MF))(24)

[0124] The cross-entropy loss function is combined to realize the fault classification of the bearing:

[0125] L = -(ylog(OutPut) + (1-y)log(1-OutPut)) (25)

[0126] Wherein, Fc is a full connection layer, sigma is a Softmax activation function, Output is an output classification result of the model, L is a classification loss, and y is a fault label in the source domain.

[0127] In step 7, the overall network model is repeatedly trained based on steps 3-6 to obtain a trained overall network model, and the test set of the target domain is input into the trained overall network model to obtain the test set accuracy and feature visualization result. Specifically as follows:

[0128] Step 7-1, select the learning rate, optimizer, training batch size and training round number of the overall network model;

[0129] Step 7-2, input the training data set obtained in step 2 into the bearing fault diagnosis model set, to obtain a trained cross-bearing fault diagnosis model;

[0130] Step 7-3, input the test data under different working conditions in the target domain into the trained cross-bearing fault diagnosis model to realize the cross-working condition fault diagnosis of the bearing.

[0131] An implementation system of a cross-bearing fault diagnosis method of an elevator, comprising a domain-invariant feature extraction module, a multi-dimensional spatio-temporal feature fusion module and a bearing fault classification module; the domain-invariant feature extraction module comprises a multi-scale convolution layer and a collaborative normalization layer, and specifically as follows:

[0132] The multi-scale convolution layer is composed of small convolution layers with different sizes of convolution kernels and a wide convolution layer, and obtains the feature relationship between different scales of samples and the long-term features of all input samples;

[0133] The collaborative normalization layer simultaneously learns the statistical characteristics between channels and the independent characteristics between instances, and reduces the covariance shift and overfitting phenomenon through mixed normalization of batch data and single sample data;

[0134] The multi-dimensional spatio-temporal feature fusion module designs a channel weight generation unit in its channel dimension to adaptively capture the local cross-channel information interaction; a spatial weight generation unit is designed in its spatial branch to adaptively learn the importance of different features in the space;

[0135] The bearing fault classifier is composed of a full connection layer and a Softmax activation function, the fused convolution features are input into the bearing fault classifier, and the classification result of the bearing fault is output.

[0136] Embodiment 1

[0137] The application is a hoist cross bearing fault diagnosis method, which specifically applies to the following steps:

[0138] Step 1, build a multi-dimensional space-time collaborative neural network, including a domain-invariant feature extraction module, a multi-dimensional space-time feature fusion module and a bearing fault classification module; the domain-invariant feature extraction module includes a multi-scale convolution layer and a collaborative normalization layer;

[0139] Step 2, collect vibration signal data under different working conditions, and after preprocessing, obtain source domain and target domain data sets respectively, establish a bearing database for each working condition, and divide the database into a training set and a test set, the number of bearings in the training set and the test set is in the proportion of 8:2, as follows:

[0140] Based on the published PU bearing data set, collect bearing data sets under 4 different working conditions, take 200 and 50 sample data for each working condition respectively, establish sample databases in the source domain and the target domain, and divide the databases into a training set and a test set, the sample ratio of the training set and the test set is 8:2.

[0141] Step 3, the multi-scale convolution layer contains 5 convolution layers with different kernel sizes and 1 wide convolution layer with a kernel size of 64, as follows:

[0142] On the PU data set, the initial input feature signal X passes through the multi-scale convolution layer to obtain the multi-scale convolution features X1, X2…, X5 of X, that is, the convolution features under different kernel sizes:

[0143] X1=conv1(conv(X)) (1)

[0144] X2=conv2(conv(X)) (2)

[0145] X3=conv3(conv(X)) (3)

[0146] X4=conv4(conv(X)) (4)

[0147] X5=conv5(conv(X)) (5)

[0148] Wherein, X is the input feature signal, co is the wide convolution layer, conv1, conv2, conv3, conv4, and conv5 are convolution layers with different scales;

[0149] Step 4, build a collaborative normalization layer, input the multi-scale convolutional features X1, X2…, X5 obtained in step 3 into the collaborative normalization layer to obtain the normalized results Z1, Z2…, Z5 of the features. Taking the multi-scale convolutional feature X1 as an example, the input of the collaborative normalization layer is as follows:

[0150] Step 4-1, split each channel into two sub-channels C1 and C2. The specific process is as follows:

[0151] X1 c1 = X1[:, 0:C / / 2, :, :](6)

[0152] X1 c2 = X1[:, C / / 2:c, :, :](7)

[0153] Step 4-2, in channel C1, divide the data into several groups and perform BN operation, update the parameters of the data by group, and one group of data determines the direction of this gradient, thereby reducing the randomness of descent. The specific process is as follows:

[0154]

[0155]

[0156] Step 4-3, in channel C2, based on the IN strategy to solve the mean and standard deviation of each sample in the H, W dimensions, and keep the N, C dimensions, that is, only calculate the mean and standard deviation within the channel. The specific process is as follows:

[0157]

[0158] Step 4-4, recombine the normalized sub-channel results to obtain the final output Z:

[0159] Z1= [Y B，1 , Y I，1 , Y B，2 , Y I，2 , …, Y B,C / / 2 , Y I,C / / 2 ](14)

[0160] wherein, BN is batch normalization, IN is instance normalization, is the feature value of each sample in the N, H and W dimensions in channel C1, is the feature value of each sample in the H, W dimensions in channel C2, γc and βc are learnable parameters, and ε is a very small constant, which is 10 -5 or 10 -8 , [.] represents concatenation along the channel dimension.

[0161] Step 5, build a multi-dimensional spatio-temporal feature fusion module, input the normalized results Z1, Z2…, Z5 into the multi-dimensional spatio-temporal feature fusion module to realize adaptive extraction of important features in the channel and space, and obtain multi-scale features MF1, MF2…MF5. Taking the feature Z1 as an example, the input of the multi-dimensional spatio-temporal feature fusion module is as follows:

[0162] Step 5-1, for a given input feature Z1, two features Im and Ia are obtained after maximum pooling and average pooling operations are performed respectively, then the features Im and Ia are input into the channel attention branch and the spatial attention branch after feature splicing.

[0163] Im = max(Z1) (15)

[0164]

[0165] Wherein, Sc represents the input of the channel attention branch, and Ss represents the input of the spatial attention branch.

[0166] Step 5-2, in the channel attention branch, the feature Sc is input into the channel weight generation unit. First, two one-dimensional convolution layers with convolution kernel sizes of k1 and k2 are adaptively determined by inputting the channel number, then the feature Sc is input into the above two one-dimensional convolution layers respectively, the deep features of Sc are adaptively extracted, and finally the attention weights Wc1 and Wc2 in the channel branch are obtained through the Softmax layer. The specific process is as follows:

[0167] Wc1 = σ [Conv1(Sc, max(1, [(γ1×C+b1) / 2]))] (18)

[0168] Wc2 = σ [Conv2(Sc, max(1, [(γ2×C+b2) / 2]))] (19)

[0169] Wherein, γ and b are hyperparameters, usually set as γ = 2 and b = 1. [.] represents an odd number of upward rounding, Conv1 and Conv2 are one-dimensional 1*1 convolution, and σ is a Softmax activation function.

[0170] Step 5-3, in the spatial attention branch, the feature Ss is input into the spatial weight generation unit in the same way. First, the feature Ss is input into two 7*7 convolution layers to obtain the initial features Ss1 and Ss2 in the spatial branch. s1' Then, the features are upgraded and downgraded through the linear layer FC1, the Relu activation function and the linear layer FC2 to obtain the important features Ss2 and Ss3 of the samples in the spatial branch. s2', and finally the attention weights Ws1 and Ws2 in the spatial branch are obtained through the Softmax activation function. Specifically as follows:

[0171] Ws1=σ(FC2(η(FC1(Conv3(Ss)))))(20)

[0172] Ws2=σ(FC2(η(FC1(Conv4(Ss)))))(21)

[0173] where Conv3 and Conv4 are 7*7 convolutions.

[0174] Step 5-4, aggregate the channel weights and spatial weights to determine the important part of the features between channels and spaces, as follows:

[0175] MF1=(Wc1+Ws1)×Im+(Wc2+Ws2)×Ia(22)

[0176] Complete the feature extraction of Layer1 layer to obtain the feature MF1 1 Then the feature MF1 1 Based on steps 3 to 5, sequentially pass through layer2, Layer3, …, Layer5 layers to obtain multi-scale features MF1.

[0177] where Output represents the fused features. Since the sum of the feature weights of Im and Ia is equal to 1, the useful part between features Im and Ia is retained, and the useless part is discarded, thereby achieving effective feature fusion.

[0178] Step 6, implement the concatenation of multi-scale features MF1, MF2, …, MF5 along the channel, confirm the final fused feature MF, and build a bearing fault classifier. The fused feature MF obtained in step 6 is input into the fault classifier to realize fault classification in the source domain, as follows:

[0179] MF=Concate[MF1,MF2,MF3,MF4,MF5](23)

[0180] OutPut=σ(Fc(MF))(24)

[0181] Combine the cross-entropy loss function to realize the fault classification of the bearing:

[0182] L=-(ylog(OutPut)+(1-y)log(1-OutPut))(25)

[0183] where Fc is a fully connected layer, σ is a Softmax activation function, Output is the output classification result of the model, L is the classification loss, and y is the fault label in the source domain.

[0184] In step 7, the overall network model is repeatedly trained based on steps 3-6 to obtain a trained overall network model, and a test set of the target domain is input into the trained overall network model to obtain a test set accuracy and feature visualization result. Specifically as follows:

[0185] Step 7-1, the learning rate, optimizer, training batch size and training number of rounds of the overall network model are selected;

[0186] Step 7-2, input the training data set obtained in step 2 into the bearing fault diagnosis model set, to obtain a trained cross-bearing fault diagnosis model;

[0187] Step 7-3, input the test data under different working conditions in the target domain into the trained cross-bearing fault diagnosis model to realize cross-condition fault diagnosis of bearings.

[0188] The method proposed in the application uses python programming language and pytorch framework to build a network framework on an Intel(R) Core(TM) i3-7100 CPU@3.90GHz host for related experiments. The learning rate of the trained network is set to 1x e -4 , the optimization strategy is Momentum, the Adam optimization algorithm is used, the momentum coefficient weight decay is 0.005, the batch-size is set to 64, and the learning rate decay factor is 0.2.

[0189] In order to show the superior performance of the algorithm of the application, the application selects some popular bearing cross-condition fault diagnosis algorithms based on CNN in recent years as comparison models, and takes the PU data set of Paderborn University in Germany as an example. The comparison experimental results are shown in Table 1.

[0190] Table 1 Comparison of different methods on PU bearing database

[0191]

[0192] From the experimental results, it can be seen that the average accuracy of the method of the application reaches 98.74%, which is higher than that of the other five mainstream domain generalization models. Compared with other methods, the generalization ability of the method of the application is better, the cross-condition learning result is more stable, and the method has stronger deep feature extraction and global feature capture ability compared with other network models, especially in the generalization learning of working conditions 0 to 1, the accuracy reaches 99.81%.

[0193] The preferred embodiments of the application disclosed above are only to facilitate the elucidation of the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments. Obviously, many modifications and variations can be made in light of the teachings above. The description is chosen and described in order to provide the best illustration of the application and its practical application to those skilled in the art and to enable those skilled in the art to best utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. A hoist cross bearing fault diagnostic method, characterized by, The method comprises the following steps: Step one: build a multi-dimensional space-time collaborative neural network, including a domain-invariant feature extraction module, a multi-dimensional space-time feature fusion module and a bearing fault classification module; wherein the domain-invariant feature extraction module comprises a multi-scale convolution layer and a collaborative normalization layer, and step two is entered; Step two: collect vibration signal data under different working conditions, and obtain source domain and target domain data sets after preprocessing; establish a bearing database for each working condition in the source domain, and divide the bearing database into a training set and a test set, and step three is entered; Step three: build a multi-scale convolution layer, and input the initial feature signal X into the multi-scale convolution layer to obtain the multi-scale convolution features X1, X2, …, X5 of X, and step four is entered; Step four: build a collaborative normalization layer, input the multi-scale convolution features X1, X2, …, X5 into the collaborative normalization layer to obtain the normalized results Z1, Z2, …, Z5 of the features, and step five is entered; Step five: build a multi-dimensional space-time feature fusion module, input the normalized results Z1, Z2, …, Z5 into the multi-dimensional space-time feature fusion module to realize adaptive extraction of important features in the channel and space, complete feature extraction of Layer1, and then sequentially pass through layer2, Layer3, …, Layer5 based on steps three to five to obtain multi-scale features MF1, MF2, …, MF5, and step six is entered; Step six: realize the cascade of multi-scale features MF1, MF2, …, MF5 along the channel, confirm the final fusion feature MF, build a bearing fault classification module, input the fusion feature MF into the fault classification module, realize bearing fault classification in the source domain, and step seven is entered; Step seven: repeatedly train the overall network model based on steps three to six to obtain the trained overall network model, and input the test set of the target domain into the trained overall network model to obtain the test set accuracy.

2. The method according to claim 1, wherein, In step two, the vibration signal data under different working conditions are collected, and the source domain and target domain data sets are obtained after preprocessing; a bearing database for each working condition in the source domain is established, and the bearing database is divided into a training set and a test set, which is specifically: Based on the published PU bearing data set, a bearing data set under multiple different working conditions is collected, m sample data are taken for each working condition, a sample database in the source domain and the target domain is established, and the sample database is divided into a training set and a test set.

3. The method of claim 1, wherein, In step three, the multi-scale convolution layer comprises five convolution layers with different kernel sizes and one wide convolution layer with a kernel size of 64, which is specifically: On the PU data set, the initial input feature signal X is input into the multi-scale convolution layer to obtain the multi-scale convolution features X1, X2, …, X5 of X, i.e. convolution features under different kernel sizes: X1=conv1(conv(X)); X2=conv2(conv(X)); X3=conv3(conv(X)); X4=conv4(conv(X)); X5=conv5(conv(X)); Wherein X is the input feature signal, conv is a wide convolution layer, conv1, conv2, conv3, conv4, and conv5 are convolution layers of different scales.

4. The method of claim 1, wherein, In step four, a collaborative normalization layer is built, and the multi-scale convolution features X1, X2, …, X5 obtained in step three are input into the collaborative normalization layer to obtain the normalized results Z1, Z2, …, Z5 of the features. Taking the multi-scale convolution feature X1 as an input of the collaborative normalization layer as an example, the specific process is as follows: In step 4-1, each channel is split into two sub-channels C1 and C2, which is achieved by the following: X1 c1 = X1 [ :, 0 : C / / 2, :, : ] ; X1 c2 = X1[ :, C / / 2: c, :, : ] ; Step 4-2, in sub-channel C1, The data is divided into several groups and the BN operation is performed, and the parameters of the data are updated by group; specifically: ； ； ； Step 4-3, in sub-channel C2, Based on the IN policy, the mean and standard deviation of each sample in the H, W dimensions are solved, and the N, C dimensions are retained, that is, only the mean and standard deviation within the channel are averaged, and the specific is: ； ； ； In step 4-4, the normalized sub-channel results are recombined to obtain the final output Z1: ； where BN is batch normalization, IN is instance normalization, is the feature value of each sample in the sub-channel C1 in the N, H and W dimensions, is the feature value of each sample in the sub-channel C2 in the H and W dimensions, are all learnable parameters, and ε is a constant, and [.] represents the concatenation along the channel dimension.

5. The method of claim 1, wherein, In step five, a multi-dimensional spatio-temporal feature fusion module is built, and the normalized results Z1, Z2, …, Z5 are input into the multi-dimensional spatio-temporal feature fusion module to realize adaptive extraction of important features in the channel and space, complete the feature extraction of Layer1 layer, and then sequentially pass through layer2, Layer3, … Layer5 layer based on steps three to five to obtain multi-scale features MF1, MF2, …, MF5. Taking the normalized result Z1 as an input of the multi-dimensional spatio-temporal feature fusion module as an example, the specific process is as follows: In step 5-1, for a given input feature Z1, two features Im and Ia are obtained after maximum pooling and average pooling operations, respectively. Then, Im and Ia are input into the channel attention branch and the spatial attention branch after feature concatenation; Im = max(Z1); ； ； In step 5-2, in the channel attention branch, the feature Sc is input into the channel weight generation unit, and two one-dimensional convolution layers with convolution kernel sizes of k1 and k2 are adaptively determined based on the input channel number. Then, the feature Sc is input into the above two one-dimensional convolution layers to adaptively extract the deep features of the feature Sc. Finally, the attention weights Wc1 and Wc2 in the channel branch are obtained through the Softmax layer. ； ； where γ and b are hyperparameters, denotes the ceiling of an odd number, Conv1, Conv2 is one-dimensional 1*1 convolution, and σ is a Softmax activation function. Step 5-3, in the spatial attention branch, input the feature Ss into the spatial weight generation unit, and the feature Ss respectively passes through two 7*7 convolution layers to obtain the initial feature in the spatial branch and Then, the dimension of the feature is increased and reduced through the linear layer FC1, the Relu activation function and the linear layer FC2 to obtain the important feature of the sample in the spatial branch and Finally, the attention weights Ws1 and Ws2 in the spatial branch are obtained through the Softmax activation function. Ws1 = σ(FC2(η(FC1(Conv3(Ss))))) ; Ws2 = σ(FC2(η(FC1(Conv4(Ss))))) ; Wherein, Conv3 and Conv4 are 7*7 convolutions. In step 5-4, the channel weight and the spatial weight are summarized to determine the important part of the features between the channel and the space, which is specifically as follows: MF1 1 = (Wc1+ Ws1) x Im+ (Wc2+ Ws2) x Ia; The feature extraction of the Layer1 layer is completed, and the feature MF1 is obtained 1 Then the feature MF1 1 Based on steps three to five, the multi-scale features MF1 are obtained by sequentially passing through the layer2, Layer3,..., Layer5 layers.

6. The hoist cross bearing fault diagnostic method of claim 1, wherein, In step six, the multi-scale features MF1, MF2, …, MF5 are cascaded along the channel to confirm the final fusion feature MF, and a bearing fault classifier is built. The fusion feature MF obtained in step six is input into the fault classifier to realize fault classification in the source domain, which is specifically as follows: MF = Concate[MF1, MF2, MF3, MF4, MF5]; OutPut = σ(Fc(MF)); The cross-entropy loss function is combined to realize the fault classification of the bearing: L = -(ylog(OutPut) + (1-y)log(1-OutPut)); Wherein, Fc is a full connection layer, sigma is a Softmax activation function, Output is an output classification result of the model, L is a classification loss, and y is a fault label in the source domain.

7. The hoist cross bearing fault diagnostic method of claim 1, wherein, In step seven, the overall network model is repeatedly trained based on steps three to six to obtain a trained overall network model, and a test set of the target domain is input into the trained overall network model to obtain a test set accuracy and a feature visualization result, specifically as follows: Step 7-1, selecting a learning rate, an optimizer, a training batch size and a training number of rounds of the overall network model; Step 7-2, inputting the training data set obtained in step two into the bearing fault diagnosis model to obtain a trained cross-bearing fault diagnosis model; Step 7-3, inputting test data under different working conditions in the target domain into the trained cross-bearing fault diagnosis model to realize cross-condition fault diagnosis of bearings.

8. A hoist cross bearing fault diagnostic system characterized by The application is applied to a cross-bearing fault diagnosis method of the hoist according to any one of claims 1-7, comprising a domain-invariant feature extraction module, a multi-dimensional space-time feature fusion module and a bearing fault classification module. The domain-invariant feature extraction module is composed of a multi-scale convolution layer and a collaborative normalization layer, and is used for adaptive extraction and fusion of important features of input samples; the multi-scale convolution layer is composed of a small convolution layer with different convolution kernel sizes and a wide convolution layer, and is used for obtaining feature relationships between different scales of samples and long-term relationship features of all input samples; The collaborative normalization layer simultaneously learns statistical characteristics between channels and independent characteristics between instances, and reduces the covariance shift and overfitting phenomenon through mixed normalization of batch data and single sample data; The multi-dimensional space-time feature fusion module innovatively replaces the traditional pooling module, and designs a channel weight generation unit in the channel dimension to adaptively capture local cross-channel information interaction; A space weight generation unit is designed in the spatial branch to adaptively learn the importance of different features in the space; The bearing fault classification module is composed of a full connection layer and a Softmax activation function, and is used for inputting the fused convolution features into a bearing fault classifier to output a classification result of bearing faults.

Images