Rolling bearing fault diagnosis method and system based on convolutional neural network, terminal and storage medium

Abstract

The invention relates to a rolling bearing fault diagnosis method and system based on a convolutional neural network, a terminal and a storage medium, and the method comprises the steps: obtaining a fault feature coefficient data set, obtaining a fault feature frequency total data set of a rolling bearing through the combination of a pre-constructed rotating speed data set of the rolling bearing, and obtaining a fault feature coefficient data set of the rolling bearing; determining a plurality of representative fault characteristic frequencies, and constructing a characteristic frequency network model based on a convolutional neural network; acquiring a vibration data set, performing data preprocessing on the vibration data set, and training the characteristic frequency network model by using a preprocessing result to obtain a fault diagnosis model; and acquiring real-time vibration data of the rolling bearing and inputting the real-time vibration data into the fault diagnosis model to obtain a fault diagnosis result of the rolling bearing. Tests show that the scheme of the invention not only shows better performance in coping with a bearing data set of the Kaisizi University and a bearing data set of the Peldbon University, but also meets business requirements in an actual production environment.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a rolling bearing fault diagnosis method, system, terminal and storage medium based on convolutional neural network. Background Art

[0002] Rolling bearings are precision mechanical parts widely used in various types of mechanical equipment. Their main function is to reduce friction losses. In daily production processes, rolling bearing failure is also one of the most common causes of mechanical failure. For heavy industries such as wind power, metallurgy, and coal, maintaining long-term stability and high efficiency of the production system is of paramount importance. Therefore, in these industries, condition monitoring and fault diagnosis of rolling bearings are of great significance in avoiding accidents, reducing machinery maintenance costs, and improving production efficiency.

[0003] At present, the fault diagnosis solutions for rolling bearings mainly include the following types.

[0004] Traditional bearing fault diagnosis methods dominated by manual experience are mainly based on experience, knowledge base and manual judgment, or based on models or expert systems. These methods have high requirements for on-site experience and professional knowledge, are highly dependent on specific personnel, equipment or systems, and have poor versatility.

[0005] A relatively new bearing fault diagnosis method dominated by artificial intelligence cannot be realized without a large amount of data. However, in the actual operation of mechanical equipment, the bearing operating environment is highly complex, and the collected data often contains noise and sensitive features present distributions of different scales. Based on this, two representative bearing fault diagnosis methods are proposed:

[0006] The existing solution 1 proposes to design the convolution kernel size and step size of the fault feature extraction layer based on the fault feature frequency of the rolling bearing. When the bearing model and operating speed are known, this solution can indeed improve the accuracy of fault diagnosis. Since its network structure design is highly dependent on the bearing model and speed data, although this solution can more easily find sensitive features, it is often difficult to obtain these parameters as the basis for diagnosis in practical applications, and the possibility of the solution being widely used is very low.

[0007] The existing solution 2 proposes a feature extraction network composed of a multi-layer small core network and a single-layer large core network in parallel to achieve multi-scale fault feature extraction. This solution can improve the network's ability to detect fault feature frequencies. However, in this solution, the network structure is designed based on experimental experience, the fault feature frequency is not considered, and the design of only two convolution kernels, one large and one small, can hardly be called multi-scale; in addition, the LSTM module in the solution also increases the computational complexity of training and reasoning.

[0008] Finally, it should be pointed out that, whether it is the existing solution 1 or the existing solution 2, the training and testing of both are based on laboratory data sets. The data in such data sets often have less noise and more obvious fault characteristics than the data obtained through sampling in the actual environment. Therefore, it is basically difficult for both to achieve the expected excellent results in practical applications.

[0009] In view of the various defects existing in the above-mentioned prior art, how to propose a new rolling bearing fault diagnosis solution that can be effectively applied in actual industrial production scenarios has become an urgent problem to be solved by technicians in this field. Summary of the invention

[0010] In order to meet the business requirements in actual industrial production scenarios and better avoid the interference of noise on fault diagnosis, the present application provides a rolling bearing fault diagnosis method, system, terminal and storage medium based on convolutional neural network.

[0011] After testing, the solution of this application not only showed better performance in dealing with the Case Western Reserve University bearing dataset and the Paderborn University bearing dataset, but also met the business requirements in the actual production environment.

[0012] In a first aspect, the present application provides a rolling bearing fault diagnosis method based on a convolutional neural network, which adopts the technical solution described below.

[0013] A rolling bearing fault diagnosis method based on convolutional neural network comprises the following steps:

[0014] Acquire a fault characteristic coefficient data set of a rolling bearing under various fault states, combine the fault characteristic coefficient data set with a pre-constructed speed data set of the rolling bearing to obtain a total fault characteristic frequency data set of the rolling bearing, determine a plurality of representative fault characteristic frequencies from the total fault characteristic frequency data set, and use the plurality of fault characteristic frequencies to construct a characteristic frequency network model based on a convolutional neural network;

[0015] Acquire vibration data of the rolling bearing in a normal state and in various fault states, summarize and form a vibration data set, perform data preprocessing on the vibration data set, and use the vibration data set after the data preprocessing to train the characteristic frequency network model to obtain a fault diagnosis model;

[0016] Real-time vibration data of the rolling bearing is acquired, and the real-time vibration data is input into the fault diagnosis model to obtain a fault diagnosis result of the rolling bearing.

[0017] By adopting the above technical scheme, a multi-scale rolling bearing fault diagnosis network is proposed, which has achieved excellent diagnostic performance in different theoretical data sets and practical applications. At the same time, the network construction and model training process of this scheme is based on a large amount of data collected in actual production environments, which further enhances the generalization and robustness of the fault diagnosis model finally obtained, and the subsequent application prospects are very broad.

[0018] Preferably, the method of acquiring a fault characteristic coefficient data set of a rolling bearing under various fault states, combining the fault characteristic coefficient data set with a pre-constructed speed data set of the rolling bearing to obtain a total fault characteristic frequency data set of the rolling bearing specifically comprises the following steps:

[0019] Divide the speed of the rolling bearing into intervals to obtain a plurality of speed intervals, select a specific number of speed values ​​from each speed interval, and aggregate them to form a speed data set RS of the rolling bearing;

[0020] Retrieving the fault characteristic coefficient data sets of rolling bearings under four fault states from the bearing model library, including the outer ring fault characteristic coefficient data set D1, the inner ring fault characteristic coefficient data set D2, the rolling element fault characteristic coefficient data set D3, and the cage fault characteristic coefficient data set D4;

[0021] The total data set of rolling bearing fault characteristic frequency is obtained by calculation, and the calculation formula is:

[0022] F fault_total ={f ij |f ij =d i × j ,d i ∈(D1∪D2∪D3∪D4),r j ∈RS}

[0023] Among them, d i represents the i-th fault characteristic coefficient in the union of four fault characteristic coefficient data sets, r j Represents the jth speed in the speed data set.

[0024] Preferably, the determining of a plurality of representative fault characteristic frequencies from the total fault characteristic frequency data set specifically comprises the following steps:

[0025] The change of the sum of square errors of the total data set of the fault characteristic frequency with the K value is calculated to obtain the elbow diagram, and the K value is determined by the elbow diagram. The K value is the number of fault characteristic frequencies used to construct the characteristic frequency network model. The calculation formula is:

[0026]

[0027] Among them, x_ft i Represents the dataset F fault_total The i-th data point in represents the jth cluster center, C j represents the cluster in the jth cluster;

[0028] The total data set of fault characteristic frequencies is clustered using the k-means++ algorithm to obtain K cluster centers, each of which corresponds to a representative fault characteristic frequency.

[0029] By adopting the above technical solution, the machine learning method is used to perform a cluster analysis on the fault frequencies of the four most typical rolling bearing faults, namely the outer ring, inner ring, rolling element, and cage, and then the representative fault characteristic frequency of the rolling bearing is obtained; combined with the speed range of the operating equipment, the subsequent construction of the characteristic frequency network model and the fault diagnosis of the rolling bearing can be completed, and the entire operation process has extremely high reusability and reuse value.

[0030] Preferably, the use of a plurality of the fault characteristic frequencies to construct a characteristic frequency network model based on a convolutional neural network specifically comprises the following steps:

[0031] Construct a characteristic frequency network model based on a convolutional neural network, wherein the characteristic frequency network model includes a characteristic frequency network and a main network, wherein:

[0032] The characteristic frequency network includes K parallel convolution modules, and the value of each cluster center corresponds to the convolution kernel size and convolution step size of a one-dimensional convolution layer in a convolution module in the characteristic frequency network.

[0033] The main network includes 5 serial convolution modules, the convolution kernel size of the one-dimensional convolution layer in the first convolution module in the main network is 5 and the convolution step size is 1, and the convolution kernel size of the one-dimensional convolution layer in the remaining convolution modules in the main network is 3 and the convolution step size is 1.

[0034] Each of the convolution modules is composed of a serial one-dimensional convolution layer, a batch normalization layer, an activation function layer, and a maximum pooling layer;

[0035] A flattened connection layer and a fully connected layer are configured for the feature frequency network model, the outputs of the feature frequency network and the main network are both used as inputs of the flattened connection layer, the output of the flattened connection layer is used as input of the fully connected layer, and the output of the fully connected layer is the output of the feature frequency network model.

[0036] By adopting the above technical solution, the specific architecture of the characteristic frequency network model in the solution is refined and limited, which also provides a technical basis for obtaining the final fault diagnosis model through training in the subsequent operation process.

[0037] Preferably, the step of acquiring vibration data of the rolling bearing in a normal state and in various fault states, aggregating the data to form a vibration data set, and performing data preprocessing on the vibration data set specifically comprises the following steps:

[0038] Acquire vibration data of rolling bearings in normal states and various fault states, summarize and form a vibration data set, divide the vibration data set according to a preset division ratio to obtain a training set, a verification set and a test set, wherein the training set and the verification set are used in the training phase of the characteristic frequency network model, and the test set can be used in the reasoning phase of the characteristic frequency network model;

[0039] The vibration data set is subjected to data preprocessing, wherein the data preprocessing at least includes input data length processing, data noise reduction, data enhancement, and data standardization or normalization, to obtain the vibration data set subjected to the data preprocessing.

[0040] By adopting the above technical solution, the vibration data set used as the training data of the characteristic frequency network model is optimized, which ensures the accuracy and effectiveness of the data, improves the stability and convergence speed of the model training, and increases the generalization of the characteristic frequency network model after training.

[0041] Preferably, the vibration data set that has undergone the data preprocessing is used to train the characteristic frequency network model to obtain a fault diagnosis model, which specifically includes the following steps:

[0042] Determining the training set, the validation set, and the test set from the vibration data set that has undergone the data preprocessing, batch processing the training set according to a preset total number of training rounds and a batch size for each round, and inputting the training set batch by batch into the characteristic frequency network model;

[0043] For each batch of the training sets input into the characteristic frequency network model, forward propagation, loss calculation, back propagation and data recording are performed in sequence, and this step is repeated until the total number of training rounds is reached and the data processing of all the training sets is completed;

[0044] The characteristic frequency network model is evaluated using the validation set, and when the performance indicators in the evaluation results meet the preset standards, the weight file of the current characteristic frequency network model is recorded and the current characteristic frequency network model is used as a fault diagnosis model.

[0045] By adopting the above technical scheme, the training process of the characteristic frequency network model in the scheme is refined and limited, and the weight file recorded after the training can also be directly applied to other untrained characteristic frequency network models to quickly obtain a complete fault diagnosis model, which undoubtedly greatly broadens the application scenarios and scope of application of this scheme.

[0046] Preferably, the step of acquiring the real-time vibration data of the rolling bearing, inputting the real-time vibration data into the fault diagnosis model, and obtaining the fault diagnosis result of the rolling bearing specifically comprises the following steps:

[0047] Acquire real-time vibration data of the rolling bearing, and perform data preprocessing on the real-time vibration data, wherein the data preprocessing at least includes input data length processing, data noise reduction, data enhancement, and data standardization or normalization;

[0048] The fault diagnosis model is loaded, and the real-time vibration data that has undergone the data preprocessing is input into the fault diagnosis model to obtain a fault diagnosis result, wherein the fault diagnosis result at least includes whether there is a fault prompt, the fault type, and the fault location.

[0049] By adopting the above technical scheme, the application of the fault diagnosis model and the acquisition of fault diagnosis results in the scheme are refined and limited. The data preprocessing process of real-time vibration data also realizes the screening and filtering of input information for the fault diagnosis model in actual application scenarios, thereby maximizing the accuracy and reliability of the final fault diagnosis results.

[0050] In a second aspect, the present application provides a rolling bearing fault diagnosis system based on a convolutional neural network, which adopts the technical solution described below.

[0051] A rolling bearing fault diagnosis system based on convolutional neural network includes the following modules:

[0052] A characteristic frequency network model construction module is configured to obtain a fault characteristic coefficient data set of a rolling bearing under various fault states, combine the fault characteristic coefficient data set with a pre-constructed speed data set of the rolling bearing to obtain a total fault characteristic frequency data set of the rolling bearing, determine a plurality of representative fault characteristic frequencies from the total fault characteristic frequency data set, and use the plurality of fault characteristic frequencies to construct a characteristic frequency network model based on a convolutional neural network;

[0053] A fault diagnosis model generation module is configured to obtain vibration data of the rolling bearing in a normal state and in a variety of fault states, summarize to form a vibration data set, perform data preprocessing on the vibration data set, and use the vibration data set after the data preprocessing to train the characteristic frequency network model to obtain a fault diagnosis model;

[0054] The rolling bearing fault diagnosis module is configured to obtain real-time vibration data of the rolling bearing, input the real-time vibration data into the fault diagnosis model, and obtain a fault diagnosis result of the rolling bearing.

[0055] By adopting the above technical scheme, the aforementioned rolling bearing fault diagnosis method based on convolutional neural network is implemented at the hardware level. It can be seen that the fault diagnosis model constructed by the scheme is relatively clear and easy to operate at the hardware level, with a high degree of reproducibility. It can be obtained by combining the existing deep learning and convolutional neural network technologies. The overall implementation cost of the scheme is low, which is conducive to large-scale promotion and application.

[0056] In a third aspect, the present application provides a smart terminal, which adopts the following technical solution:

[0057] An intelligent terminal includes a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the rolling bearing fault diagnosis method based on convolutional neural network as described above.

[0058] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0059] A computer-readable storage medium, wherein at least one instruction, at least one program, code set or instruction set is stored in the readable storage medium, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the rolling bearing fault diagnosis method based on a convolutional neural network as described above.

[0060] In summary, the present application includes at least one of the following beneficial technical effects:

[0061] 1. The solution of the present application uses a machine learning method to perform a cluster analysis on the fault frequencies of the outer ring, inner ring, rolling element, and retainer of the rolling bearing, and then obtains the representative fault characteristic frequency of the rolling bearing; combined with the speed range of the operating equipment, the subsequent construction of the characteristic frequency network model and the fault diagnosis of the rolling bearing can be completed, and the entire operation process has extremely high reusability and reuse value.

[0062] 2. The solution of this application proposes a multi-scale rolling bearing fault diagnosis network, so that the overall solution not only shows better performance in dealing with the Case Western Reserve University bearing dataset and the Paderborn University bearing dataset, but also meets the business requirements in the actual production environment.

[0063] 3. The solution of this application realizes data-driven deep learning. The network construction and model training process in the solution are based on a large amount of data collected in actual production environments, which further enhances the generalization and robustness of the fault diagnosis model finally obtained, and the subsequent application prospects are very broad. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] 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, they can also refer to these drawings to obtain other drawings without paying creative work.

[0065] Figure 1 It is a flow chart of a rolling bearing fault diagnosis method based on a convolutional neural network according to an embodiment of the present application;

[0066] Figure 2 Yes Figure 1 A schematic diagram of data flow in step S1 of a rolling bearing fault diagnosis method based on a convolutional neural network is shown;

[0067] Figure 3 It is an elbow diagram used in a rolling bearing fault diagnosis method based on a convolutional neural network in an embodiment of the present application;

[0068] Figure 4 It is a structural schematic diagram of a characteristic frequency network model constructed by a rolling bearing fault diagnosis method based on a convolutional neural network in an embodiment of the present application;

[0069] Figure 5 Yes Figure 1 A data flow diagram of step S3 in a rolling bearing fault diagnosis method based on a convolutional neural network is shown;

[0070] Figure 6 It is a schematic diagram of the module architecture of a rolling bearing fault diagnosis system based on a convolutional neural network according to an embodiment of the present application. DETAILED DESCRIPTION

[0071] The present application provides a rolling bearing fault diagnosis method, system, terminal and storage medium based on convolutional neural network. In order to make the purpose, technical solution and advantages of the present application clearer, the implementation mode of the present application will be further described below.

[0072] The following is a detailed description of a rolling bearing fault diagnosis method, system, terminal and storage medium based on a convolutional neural network of the present application in conjunction with the drawings in the specification.

[0073] A rolling bearing fault diagnosis method based on a convolutional neural network in this embodiment is as follows: Figure 1 As shown, the following steps are included:

[0074] S1. Obtain a fault characteristic coefficient data set of a rolling bearing under various fault conditions, combine the fault characteristic coefficient data set with a pre-constructed rotational speed data set of the rolling bearing to obtain a total fault characteristic frequency data set of the rolling bearing, determine a plurality of representative fault characteristic frequencies from the total fault characteristic frequency data set, and use the plurality of fault characteristic frequencies to construct a characteristic frequency network model based on a convolutional neural network.

[0075] First of all, it should be made clear that, generally speaking, rolling bearing failures mainly include the following four types, namely outer ring failure, inner ring failure, rolling element failure, and cage failure. When a rolling bearing failure occurs, analysts can analyze the corresponding fault characteristic frequency from the data. The calculation formula is,

[0076] F fault =F cf *RPM

[0077] Among them, F fault is the fault characteristic frequency, F cf is the fault characteristic coefficient, and RPM is the rotation speed (revolutions per minute).

[0078] Based on the above formula, Figure 2 As shown, step S1 specifically includes the following process:

[0079] S111, dividing the rotation speed of the rolling bearing into intervals to obtain a plurality of rotation speed intervals, selecting a specific number of rotation speed values ​​from each of the rotation speed intervals, and summarizing them to form a rotation speed data set RS of the rolling bearing.

[0080] In actual operation, the speed of rolling bearings can be divided into five intervals according to the frequency of my country's power grid and the common number of motor poles of asynchronous motors and synchronous motors. Three common values ​​are taken in each interval to form the speed data set RS.

[0081] For example, the speed of the rotating magnetic field of an 8-pole motor (pole pair number is 4) is 750 rpm (the actual speed is slightly lower than the speed of the rotating magnetic field). The unit can be divided into intervals of 100, and five intervals can be selected: [500,600], [600,700], [700,800], [800,900], and [900,1000]. Three common values ​​can be selected from each interval. For example, the values ​​in the interval [700,800] can be average interval values ​​such as 700, 750, and 800, or empirical values ​​such as 700, 750, and 780.

[0082] S112. Retrieve the fault characteristic coefficient data sets of the rolling bearing under four fault states from the bearing model library, including the outer ring fault characteristic coefficient data set D1, the inner ring fault characteristic coefficient data set D2, the rolling element fault characteristic coefficient data set D3, and the cage fault characteristic coefficient data set D4.

[0083] It should be noted that the data in a single fault characteristic coefficient data set only represents the data of the rolling bearing under one fault state, so any fault characteristic coefficient data set may contain data of more than one type of rolling bearing. For this reason, the fault diagnosis model finally obtained in this embodiment can be applied to the fault diagnosis analysis of multiple types of faulty bearings.

[0084] S113. The total data set of rolling bearing fault characteristic frequency is obtained by calculation. The calculation formula is:

[0085] F fault_total ={f ij |f ij =d i × j ,d i ∈(D1∪D2∪D3∪D4),r j ∈RS}

[0086] Among them, d i represents the i-th fault characteristic coefficient in the union of four fault characteristic coefficient data sets, r j Represents the jth speed in the speed data set.

[0087] S121, performing outlier processing on the total fault characteristic frequency data set, wherein the outlier processing at least includes outlier screening and outlier deletion, and then calculating the change of the sum of square errors (SSE) of the total fault characteristic frequency data set with the K value, and obtaining the following: Figure 3The elbow diagram shown in FIG. 1 is used to determine the K value, which is the number of fault characteristic frequencies used to construct the characteristic frequency network model. The calculation formula is:

[0088]

[0089] Among them, x_ft i Represents the dataset F fault_total The i-th data point in represents the jth cluster center, C j Represents the cluster in the jth cluster.

[0090] S122. Clustering the total fault feature frequency data set using a k-means++ algorithm to obtain K cluster centers, each of which corresponds to a representative fault feature frequency.

[0091] The clustering method here is similar to the existing solution. The specific operations are as follows:

[0092] Randomly select the initial center point, for each data point, calculate its distance to each center point, and assign the data point to the cluster to which the nearest center point belongs;

[0093] For each cluster, calculate the location of the new center point, which is the mean of all points in the cluster;

[0094] Repeat the above operation until the center point no longer changes or the change is less than the set threshold.

[0095] S131, construct a characteristic frequency network model based on convolutional neural network, such as Figure 4 As shown, the characteristic frequency network model includes a characteristic frequency network and a main network, wherein:

[0096] The characteristic frequency network includes K parallel convolution modules, and the value of each cluster center corresponds to the convolution kernel size and convolution step length of a one-dimensional convolution layer in a convolution module in the characteristic frequency network, that is, the convolution kernel size is equal to the convolution step length. For this reason, the characteristic frequency network has a larger receptive field;

[0097] The main network includes 5 convolution modules in series, and the first convolution module in the main network, i.e. Figure 4 The convolution kernel size of the one-dimensional convolution layer in Main Conv1 in is 5, and the convolution step size is 1. The remaining convolution modules in the main network, i.e. Figure 4 The convolution kernel size of the one-dimensional convolutional layers in Main Conv2~Main Conv5 is 3 and the convolution step size is 1.

[0098] Each of the convolution modules is composed of a serial one-dimensional convolution layer, a batch normalization layer, an activation function layer and a maximum pooling layer, and the specific scheme is as follows.

[0099] The main function of the one-dimensional convolution layer (Conv1d) is to perform a convolution operation on the input signal through a one-dimensional convolution kernel to extract local features; the convolution calculation formula is:

[0100]

[0101] Among them, ksize represents the size of the convolution kernel, m represents the position index in the output feature map, n represents the index within the convolution kernel (or filter), and x_conv [m] It represents the value of the output feature map at position m after convolution, x_input[m+n] represents the value of the input feature map at position m+n, and w[n] represents the weight of the convolution kernel at position n.

[0102] The batch normalization layer (Batch Norm1d) is used to standardize the input signal so that its mean is 0 and its variance is 1, and learn an adjustable scaling and translation parameter; the formula is,

[0103]

[0104] Among them, x_conv represents the output of the one-dimensional convolution layer, that is, the input of the batch normalization layer, x_bn represents the output of the batch normalization layer, μ represents the mean of the input, σ represents the variance of the input, ∈ represents a small value used to prevent the denominator in the formula from being 0, γ and β represent learnable parameters used to scale and translate the normalized input.

[0105] The activation function layer is used to introduce nonlinear characteristics. In this embodiment, the Tanh activation function is used, and the formula is:

[0106] x_than=x_bn*tanh(x_bn)

[0107] in,

[0108] The maximum pooling layer (Max Pool1d) is used to reduce the dimension of the data while retaining important features. The formula is:

[0109] x_mp[i]=max(x_than[i:i+p-1])

[0110] Among them, x_than represents the input signal, p represents the pooling window size, and x_mp[i] represents the i-th element of the output.

[0111] S132. Configure a flattened connection layer and a fully connected layer for the feature frequency network model. The outputs of the feature frequency network and the main network are both used as inputs of the flattened connection layer. The output of the flattened connection layer is used as input of the fully connected layer. The output of the fully connected layer is the output of the feature frequency network model.

[0112] Likewise Figure 4 As shown, the main function of the flattening and splicing layer (Flatten+Splicing) is to convert multi-dimensional data into a one-dimensional vector (tensor) and connect multiple tensors in a certain dimension.

[0113] The fully connected layer is used to connect each input neuron to the output neuron. Each connection has a weight and a bias. The formula is:

[0114] y=anti'x'+b'

[0115] Among them, x' represents the input vector of the fully connected layer, W' represents the weight matrix, b' represents the bias vector, and y represents the output vector of the fully connected layer.

[0116] S2. Obtain vibration data of the rolling bearing in a normal state and in various fault states, summarize them to form a vibration data set, perform data preprocessing on the vibration data set, and use the vibration data set that has undergone data preprocessing to train the characteristic frequency network model to obtain a fault diagnosis model; this step specifically includes the following process.

[0117] S211, obtaining vibration data of rolling bearings in normal state and various fault states, summarizing to form a vibration data set, dividing the vibration data set according to a preset division ratio to obtain a training set, a verification set and a test set. In this embodiment, the ratio of the training set, the verification set and the test set is 7:2:1. The training set and the verification set are used in the training phase of the characteristic frequency network model, and the test set can be used in the inference phase of the characteristic frequency network model.

[0118] S212, performing data preprocessing on the vibration data set, wherein the data preprocessing at least includes input data length processing, data noise reduction, data enhancement, and data standardization or normalization, to obtain the vibration data set after the data preprocessing.

[0119] The above-mentioned data preprocessing operations are respectively described below.

[0120] a. The input data length processing is to calculate the corresponding relationship between the time for the bearing to rotate one circle and the sampling time according to the sampling frequency, number of sampling points and speed range of the bearing of the data collector. The formula is:

[0121]

[0122] T s ≥k q *T r

[0123] Where N represents the number of sampling points that the collector can obtain in a single sampling; T s Indicates the sampling time of the collector; F s Indicates the sampling frequency of the collector; T r Indicates the time required for the bearing to rotate one circle; RPM indicates the bearing speed (per minute / revolution); k q Indicates multiples.

[0124] In order to extract the complete bearing fault features here, the input data length should be greater than the data length corresponding to one rotation of the bearing.

[0125] According to the actual situation, select the appropriate k q The input data length can be obtained by using a value (at least greater than or equal to 2). For data that does not meet the data length requirement, it can also be met by means of cropping, resampling, etc.

[0126] b. The data denoising is to use Butterworth filter to denoise the data that meets the input length and improve the signal-to-noise ratio.

[0127] The transfer function H(s) of the Butterworth filter can be expressed by the following formula,

[0128]

[0129] Among them, s is a complex frequency domain variable, w c is the cut-off frequency and nr is the order of the filter.

[0130] c. The data enhancement is based on the actual situation, and operations such as inversion and data overlap processing can be selected for data enhancement to expand the data volume of the training set and increase the generalization of subsequent models.

[0131] d. The data standardization or normalization is to standardize or normalize the training set, thereby improving the stability and convergence speed of model training and preventing gradient explosion or disappearance.

[0132] S221. Determine the training set, the validation set and the test set from the vibration data set that has undergone the data preprocessing, process the training set in batches according to a preset total number of training epochs and a batch size per epoch, and input the training set into the characteristic frequency network model batch by batch. In this embodiment, the training process of the characteristic frequency network model is optimized using an Adam optimizer.

[0133] S222. For each batch of the training sets input into the feature frequency network model, forward propagation, loss calculation, back propagation and data recording are performed in sequence, and this step is repeated until the total number of training rounds is reached and the data processing of all the training sets is completed.

[0134] The specific operation of the training process here is as follows:

[0135] Forward propagation: first input the training set into the main network, output the feature tensor extracted by the small convolution kernel, then pass through the global average pooling layer to make the number of channels of the main network output data the same as the number of channels of the feature frequency network model output data, then input the training set into the feature frequency network, output feature frequency tensors of different convolution scales, then concatenate the feature tensor output by the main network and the feature frequency tensor output by the feature frequency network to form a new tensor and flatten it. Apply the Dropout layer to the flattened tensor, and finally input the result of the Dropout layer into the fully connected layer, and after two layers of linear transformation and Tanh activation function, obtain the prediction result of the feature frequency network model.

[0136] Calculate the loss based on the prediction result of the feature frequency network model and the true label.

[0137] In addition to the above limitations, it should be added that the loss function used in this embodiment is CrossEntropy Loss, which is usually used for classification problems. It combines softmax and negative log-likelihood, so the output no longer needs to go through softmax. The formula is:

[0138]

[0139] Among them, loss represents the calculated loss value and the objective function to be minimized, t i Represents the value of the i-th category in the true label, y i It represents the predicted probability of the i-th category in the model prediction output, and CL represents the total number of categories.

[0140] Back propagation is performed to calculate the gradient, and then the Adam optimizer is used to update the parameters of the characteristic frequency network model.

[0141] Data recording, recording the loss value and other indicators in the previous operation, recording and printing the training progress of the characteristic frequency network model.

[0142] S223. Use the verification set to evaluate the characteristic frequency network model. When the performance indicators in the evaluation results, such as accuracy, meet the preset standards, record the weight file of the current characteristic frequency network model and use the current characteristic frequency network model as a fault diagnosis model.

[0143] S3, obtaining real-time vibration data of the rolling bearing, inputting the real-time vibration data into the fault diagnosis model, and obtaining a fault diagnosis result of the rolling bearing; Figure 5 As shown, this step specifically includes the following process.

[0144] S31. Acquire real-time vibration data of the rolling bearing, and perform data preprocessing on the real-time vibration data. The data preprocessing here is consistent with the data preprocessing mentioned above, and at least includes input data length processing, data noise reduction, data enhancement, and data standardization or normalization.

[0145] S32, loading the fault diagnosis model, inputting the real-time vibration data after the data preprocessing into the fault diagnosis model, and obtaining a fault diagnosis result, wherein the fault diagnosis result at least includes whether there is a fault prompt, the fault type and the fault location.

[0146] Based on the same inventive concept as above, the present application also discloses a rolling bearing fault diagnosis system based on a convolutional neural network, such as Figure 6 As shown, it specifically includes the following modules:

[0147] A characteristic frequency network model construction module is configured to obtain a fault characteristic coefficient data set of a rolling bearing under various fault states, combine the fault characteristic coefficient data set with a pre-constructed speed data set of the rolling bearing to obtain a total fault characteristic frequency data set of the rolling bearing, determine a plurality of representative fault characteristic frequencies from the total fault characteristic frequency data set, and use the plurality of fault characteristic frequencies to construct a characteristic frequency network model based on a convolutional neural network;

[0148] A fault diagnosis model generation module is configured to obtain vibration data of the rolling bearing in a normal state and in a variety of fault states, summarize to form a vibration data set, perform data preprocessing on the vibration data set, and use the vibration data set after the data preprocessing to train the characteristic frequency network model to obtain a fault diagnosis model;

[0149] The rolling bearing fault diagnosis module is configured to obtain real-time vibration data of the rolling bearing, input the real-time vibration data into the fault diagnosis model, and obtain a fault diagnosis result of the rolling bearing.

[0150] In a possible implementation, the characteristic frequency network model construction module specifically includes the following units:

[0151] A speed data set acquisition unit is configured to divide the speed of the rolling bearing into intervals to obtain a plurality of speed intervals, select a specific number of speed values ​​from each speed interval, and aggregate them to form a speed data set RS of the rolling bearing;

[0152] A fault characteristic coefficient data set acquisition unit is configured to retrieve fault characteristic coefficient data sets of rolling bearings under four fault states from a bearing model library, including an outer ring fault characteristic coefficient data set D1, an inner ring fault characteristic coefficient data set D2, a rolling element fault characteristic coefficient data set D3, and a cage fault characteristic coefficient data set D4;

[0153] The fault characteristic frequency total data set acquisition unit is configured to obtain the fault characteristic frequency total data set of the rolling bearing by calculation. The calculation formula is:

[0154] F fault_total ={f ij |f ij =d i × j ,d i ∈(D1∪D2∪D3∪D4),r j ∈RS}

[0155] Among them, d i represents the i-th fault characteristic coefficient in the union of four fault characteristic coefficient data sets, r j Represents the jth speed in the speed data set.

[0156] The fault characteristic frequency number determination unit is configured to calculate the change of the square sum error of the fault characteristic frequency total data set with the K value, obtain the elbow diagram, and determine the K value from the elbow diagram. The K value is the number of fault characteristic frequencies used to construct the characteristic frequency network model. The calculation formula is:

[0157]

[0158] Among them, x_ft i Represents the dataset F fault_total The i-th data point in represents the jth cluster center, C j represents the cluster in the jth cluster;

[0159] The fault characteristic frequency determination unit is configured to cluster the total fault characteristic frequency data set using a k-means++ algorithm to obtain K cluster centers, each of which corresponds to a representative fault characteristic frequency.

[0160] The characteristic frequency network model preliminary construction unit is configured to construct a characteristic frequency network model based on a convolutional neural network, wherein the characteristic frequency network model includes a characteristic frequency network and a main network, wherein:

[0161] The characteristic frequency network includes K parallel convolution modules, and the value of each cluster center corresponds to the convolution kernel size and convolution step size of a one-dimensional convolution layer in a convolution module in the characteristic frequency network.

[0162] The main network includes 5 serial convolution modules, the convolution kernel size of the one-dimensional convolution layer in the first convolution module in the main network is 5 and the convolution step size is 1, and the convolution kernel size of the one-dimensional convolution layer in the remaining convolution modules in the main network is 3 and the convolution step size is 1.

[0163] Each of the convolution modules is composed of a serial one-dimensional convolution layer, a batch normalization layer, an activation function layer, and a maximum pooling layer;

[0164] The feature frequency network model refinement construction unit is configured to configure a flattened connection layer and a fully connected layer for the feature frequency network model, the outputs of both the feature frequency network and the main network are used as inputs of the flattened connection layer, the output of the flattened connection layer is used as input of the fully connected layer, and the output of the fully connected layer is the output of the feature frequency network model.

[0165] In a possible implementation, the fault diagnosis model generation module specifically includes the following units:

[0166] A vibration data set acquisition and division unit is configured to acquire vibration data of the rolling bearing in a normal state and in various fault states, summarize and form a vibration data set, and divide the vibration data set according to a preset division ratio to obtain a training set, a verification set, and a test set, wherein the training set and the verification set are used in the training phase of the characteristic frequency network model, and the test set can be used in the reasoning phase of the characteristic frequency network model;

[0167] The vibration data set preprocessing unit is configured to perform data preprocessing on the vibration data set, wherein the data preprocessing at least includes input data length processing, data noise reduction, data enhancement, and data standardization or normalization to obtain the vibration data set after the data preprocessing.

[0168] A training set data processing unit is configured to determine the training set, the validation set and the test set from the vibration data set that has undergone the data preprocessing, process the training set in batches according to a preset total number of training rounds and a batch size per round, and input the training set batch by batch into the characteristic frequency network model;

[0169] The characteristic frequency network model training unit is configured to perform forward propagation, loss calculation, back propagation and data recording in order for each batch of the training sets input into the characteristic frequency network model, and repeat the steps until the total number of training rounds is reached and the data processing of all the training sets is completed;

[0170] The fault diagnosis model determination unit is configured to evaluate the characteristic frequency network model using the verification set, and when the performance indicators in the evaluation results meet the preset standards, record the weight file of the current characteristic frequency network model and use the current characteristic frequency network model as the fault diagnosis model.

[0171] In a possible implementation, the rolling bearing fault diagnosis module specifically includes the following units:

[0172] A real-time data preprocessing unit is configured to obtain real-time vibration data of the rolling bearing and perform data preprocessing on the real-time vibration data, wherein the data preprocessing at least includes input data length processing, data noise reduction, data enhancement, and data standardization or normalization;

[0173] The fault diagnosis execution unit is configured to load the fault diagnosis model, input the real-time vibration data after the data preprocessing into the fault diagnosis model, and obtain a fault diagnosis result, which at least includes whether there is a fault prompt, the fault type and the fault location.

[0174] Based on the same inventive concept mentioned above, an embodiment of the present application also discloses an intelligent terminal, including a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the rolling bearing fault diagnosis method based on convolutional neural network as described above.

[0175] It should be understood that the "plurality" mentioned in this article refers to two or more. "And / or" describes the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.

[0176] Also based on the same inventive concept mentioned above, an embodiment of the present application also discloses a computer-readable storage medium, in which at least one instruction, at least one program, code set or instruction set is stored, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the rolling bearing fault diagnosis method based on convolutional neural network as described above.

[0177] Those skilled in the art should understand that all or part of the steps to implement the above embodiments may be completed by hardware, or may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium. The above-mentioned storage medium may include, for example: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0178] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A rolling bearing fault diagnosis method based on convolutional neural network, characterized in that: The steps include: Acquire a fault characteristic coefficient data set of a rolling bearing under various fault states, combine the fault characteristic coefficient data set with a pre-constructed speed data set of the rolling bearing to obtain a total fault characteristic frequency data set of the rolling bearing, determine a plurality of representative fault characteristic frequencies from the total fault characteristic frequency data set, and use the plurality of fault characteristic frequencies to construct a characteristic frequency network model based on a convolutional neural network; Acquire vibration data of the rolling bearing in a normal state and in various fault states, summarize and form a vibration data set, perform data preprocessing on the vibration data set, and use the vibration data set after the data preprocessing to train the characteristic frequency network model to obtain a fault diagnosis model; Real-time vibration data of the rolling bearing is acquired, and the real-time vibration data is input into the fault diagnosis model to obtain a fault diagnosis result of the rolling bearing.

2. The rolling bearing fault diagnosis method based on convolutional neural network according to claim 1 is characterized in that: The method of obtaining a fault characteristic coefficient data set of a rolling bearing under various fault conditions, combining the fault characteristic coefficient data set with a pre-constructed speed data set of the rolling bearing to obtain a total fault characteristic frequency data set of the rolling bearing specifically includes the following steps: Divide the speed of the rolling bearing into intervals to obtain a plurality of speed intervals, select a specific number of speed values ​​from each speed interval, and aggregate them to form a speed data set RS of the rolling bearing; Retrieving the fault characteristic coefficient data sets of rolling bearings under four fault states from the bearing model library, including the outer ring fault characteristic coefficient data set D1, the inner ring fault characteristic coefficient data set D2, the rolling element fault characteristic coefficient data set D3, and the cage fault characteristic coefficient data set D4; The total data set of rolling bearing fault characteristic frequency is obtained by calculation, and the calculation formula is: F fault_total ={f ij |f ij =d i ×r j ,d i ∈(D1∪D2∪D3∪D4),r j ∈RS} Among them, d i represents the i-th fault characteristic coefficient in the union of four fault characteristic coefficient data sets, r j Represents the jth speed in the speed data set.

3. The rolling bearing fault diagnosis method based on convolutional neural network according to claim 2 is characterized in that: Determining a plurality of representative fault characteristic frequencies from the total fault characteristic frequency data set specifically comprises the following steps: The change of the sum of square errors of the total data set of the fault characteristic frequency with the K value is calculated to obtain the elbow diagram, and the K value is determined by the elbow diagram. The K value is the number of fault characteristic frequencies used to construct the characteristic frequency network model. The calculation formula is: Among them, x_ft i Represents the dataset F fault_total The i-th data point in represents the jth cluster center, C j represents the cluster in the jth cluster; The total data set of fault characteristic frequencies is clustered using the k-means++ algorithm to obtain K cluster centers, each of which corresponds to a representative fault characteristic frequency.

4. The rolling bearing fault diagnosis method based on convolutional neural network according to claim 3 is characterized in that: The method of using a plurality of the fault characteristic frequencies to construct a characteristic frequency network model based on a convolutional neural network specifically includes the following steps: Construct a characteristic frequency network model based on a convolutional neural network, wherein the characteristic frequency network model includes a characteristic frequency network and a main network, wherein: The characteristic frequency network includes K parallel convolution modules, and the value of each cluster center corresponds to the convolution kernel size and convolution step size of a one-dimensional convolution layer in a convolution module in the characteristic frequency network. The main network includes 5 serial convolution modules, the convolution kernel size of the one-dimensional convolution layer in the first convolution module in the main network is 5 and the convolution step size is 1, and the convolution kernel size of the one-dimensional convolution layer in the remaining convolution modules in the main network is 3 and the convolution step size is 1. Each of the convolution modules is composed of a serial one-dimensional convolution layer, a batch normalization layer, an activation function layer, and a maximum pooling layer; A flattened connection layer and a fully connected layer are configured for the feature frequency network model, the outputs of the feature frequency network and the main network are both used as inputs of the flattened connection layer, the output of the flattened connection layer is used as input of the fully connected layer, and the output of the fully connected layer is the output of the feature frequency network model.

5. The rolling bearing fault diagnosis method based on convolutional neural network according to claim 1 is characterized in that: The step of obtaining vibration data of the rolling bearing in a normal state and in various fault states, aggregating the data to form a vibration data set, and performing data preprocessing on the vibration data set specifically includes the following steps: Acquire vibration data of rolling bearings in normal states and various fault states, summarize and form a vibration data set, divide the vibration data set according to a preset division ratio to obtain a training set, a verification set and a test set, wherein the training set and the verification set are used in the training phase of the characteristic frequency network model, and the test set can be used in the reasoning phase of the characteristic frequency network model; The vibration data set is subjected to data preprocessing, wherein the data preprocessing at least includes input data length processing, data noise reduction, data enhancement, and data standardization or normalization, to obtain the vibration data set subjected to the data preprocessing.

6. The rolling bearing fault diagnosis method based on convolutional neural network according to claim 5 is characterized in that: The method of using the vibration data set that has undergone the data preprocessing to train the characteristic frequency network model to obtain a fault diagnosis model specifically includes the following steps: Determining the training set, the validation set, and the test set from the vibration data set that has undergone the data preprocessing, batch processing the training set according to a preset total number of training rounds and a batch size for each round, and inputting the training set batch by batch into the characteristic frequency network model; For each batch of the training sets input into the characteristic frequency network model, forward propagation, loss calculation, back propagation and data recording are performed in sequence, and this step is repeated until the total number of training rounds is reached and the data processing of all the training sets is completed; The characteristic frequency network model is evaluated using the validation set, and when the performance indicators in the evaluation results meet the preset standards, the weight file of the current characteristic frequency network model is recorded and the current characteristic frequency network model is used as a fault diagnosis model.

7. The rolling bearing fault diagnosis method based on convolutional neural network according to claim 1 is characterized in that: The step of acquiring the real-time vibration data of the rolling bearing and inputting the real-time vibration data into the fault diagnosis model to obtain the fault diagnosis result of the rolling bearing specifically comprises the following steps: Acquire real-time vibration data of the rolling bearing, and perform data preprocessing on the real-time vibration data, wherein the data preprocessing at least includes input data length processing, data noise reduction, data enhancement, and data standardization or normalization; The fault diagnosis model is loaded, and the real-time vibration data that has undergone the data preprocessing is input into the fault diagnosis model to obtain a fault diagnosis result, wherein the fault diagnosis result at least includes whether there is a fault prompt, the fault type, and the fault location.

8. A rolling bearing fault diagnosis system based on convolutional neural network, characterized in that: Includes the following modules: A characteristic frequency network model construction module is configured to obtain a fault characteristic coefficient data set of a rolling bearing under various fault states, combine the fault characteristic coefficient data set with a pre-constructed speed data set of the rolling bearing to obtain a total fault characteristic frequency data set of the rolling bearing, determine a plurality of representative fault characteristic frequencies from the total fault characteristic frequency data set, and use the plurality of fault characteristic frequencies to construct a characteristic frequency network model based on a convolutional neural network; A fault diagnosis model generation module is configured to obtain vibration data of the rolling bearing in a normal state and in a variety of fault states, summarize to form a vibration data set, perform data preprocessing on the vibration data set, and use the vibration data set after the data preprocessing to train the characteristic frequency network model to obtain a fault diagnosis model; The rolling bearing fault diagnosis module is configured to obtain real-time vibration data of the rolling bearing, input the real-time vibration data into the fault diagnosis model, and obtain a fault diagnosis result of the rolling bearing.

9. An intelligent terminal, characterized in that: It comprises a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the rolling bearing fault diagnosis method based on convolutional neural network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The readable storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the rolling bearing fault diagnosis method based on a convolutional neural network as described in any one of claims 1 to 7.

Images