Bearing fault diagnosis method and model based on neural network

Abstract

The invention relates to the technical field of fault diagnosis, and provides a bearing fault diagnosis method and model based on a neural network, through steps S1-S4, IMF components are extracted through an adaptive noise ensemble empirical mode decomposition module, useful components are screened out according to a minimum envelope entropy criterion, and then time domain features are calculated as initial feature vectors. And expanding the initial feature vector by using a one-dimensional generative adversarial network, and fusing the expanded feature vector and the initial feature vector. Dividing the feature vectors into two parts, namely a training set and a test set, and training by using a CNN-BiLSTM model so as to construct a fault diagnosis model. According to the bearing fault diagnosis method based on the neural network, the diagnosis accuracy is 9%-25% higher than that of other methods, and the operation time is short. According to the bearing fault diagnosis method based on the neural network, fault features are accurately and comprehensively extracted, so that the diagnosis precision and the anti-interference capability of the rolling bearing are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a bearing fault diagnosis method and model based on a neural network. Background Art

[0002] With the rapid development of intelligent mechanical equipment, high speed and heavy load have become important features of modern mechanical systems. As the core component of rotating machinery, the operating state of bearings directly affects the performance and reliability of the entire mechanical system. Vibration signals contain a lot of fault information, so accurately extracting fault features and realizing accurate fault diagnosis in a strong noise environment has become a hot topic in current research. Traditional fault diagnosis methods generally preprocess the collected vibration signals, such as using feature Fourier transform, wavelet transform, empirical mode decomposition, etc. Combined with machine learning algorithms for fault diagnosis, such as Bayesian classification algorithm, support vector machine, logistic regression and random forest, etc. Compared with traditional machine learning algorithms, deep learning algorithms have more powerful ability to approximate complex functions. Such algorithms usually have a multi-layer hidden structure, which can transform data features layer by layer, effectively extract and express information, thereby getting rid of the dependence on signal preprocessing technology in the field of fault diagnosis. Deep learning algorithms can process high-dimensional and nonlinear data, effectively making up for the shortcomings of traditional fault diagnosis models. Among them, Convolutional Neural Networks (CNN), as one of the most typical algorithms in deep learning, has excellent generalization ability in vibration signal feature extraction and has been widely used in the field of bearing fault diagnosis.

[0003] CN118395336A discloses a bearing fault prediction method based on improved VMD-CNN-BiLSTM. The invention comprises the following steps: step 1: obtaining the vibration signal of the bearing parts in the reducer in the manufacturing industry; step 2: optimizing the VMD model by using the improved northern goshawk algorithm UNGO, denoising the vibration signal by using the optimized VMD model, and obtaining several IMFs after decomposition; the improved northern goshawk algorithm UNGO introduces SMP chaos mapping to optimize the initialization population distribution, local optimization in the neighboring search space, and Levy flight strategy to prevent falling into the local optimum; step 3: screening the IMFs of the denoised vibration signal by using the effective weighted standard MBVH, and reconstructing the signal; step 4: dividing the reconstructed signal into a training set and a test set, constructing an improved CNN-BiLSTM prediction model, and completing fault prediction; the improved CNN-BiLSTM prediction model uses the local features extracted by CNN as the input of BiLSTM, and CNN uses the convolution layer of the optimal step length iteratively optimized by the improved northern goshawk algorithm UNGO in step 2 for convolution, and residually connects the output sequence of CNN and the output sequence of BiLSTM.

[0004] CN111679654B discloses a vibration signal diagnosis method and device for a power drive system based on a neural network. The vibration signal diagnosis method for the power drive system based on a neural network includes: Step S1, a plurality of groups of vibration sensors acquire vibration signal data of different components in the power drive system. Among them, the power drive system includes a drive motor, a motor controller, a transmission, a differential, and half shafts. The drive motor includes a motor housing and a motor output shaft. The half shafts include a left half shaft and a right half shaft. The vibration signal at least includes vibration signal sources of the motor housing, the motor output shaft, the left half shaft, and the right half shaft; Step S2, the vibration signal data of different signal sources are respectively input into an empirical mode model for time domain decomposition. The vibration signal of each signal source is decomposed to obtain N intrinsic mode function (IMF) data. Among them, the empirical mode model includes an input vibration signal, outputs multiple groups of intrinsic mode component data and a residue. The empirical mode model is a Hilbert-Huang transform model; Step S3, the N intrinsic mode function (IMF) data are input into a convolution model and convolved with a window function to obtain M first convolution data. Among them, the window function includes a triangular window or a rectangular window, M < N, and both M and N are positive integers; Step S4, the M first convolution data are input into a trained neural network model to output fault occurrence point data and label data. Among them, the neural network model is a machine learning model obtained through training. The training set of the machine learning model includes a first data set and a second data set. The first data set includes multiple groups of input data. The second data includes fault occurrence point data and label data. The label data includes a normal label and a fault code label. The fault occurrence point data has a mapping relationship with the fault code label; In step S1, according to the power transmission direction of the power drive system, a plurality of groups of the vibration sensors are sequentially arranged on the motor controller, the motor housing, the motor output shaft, the transmission input shaft, the transmission housing, the transmission output shaft, the differential input shaft, the differential housing, the left half shaft, and the right half shaft. The plurality of vibration sensors are respectively connected to each component to acquire vibration signal data of different signal sources; The vibration sensors on the motor output shaft, the transmission input shaft, the transmission output shaft, the differential input shaft, the left half shaft, and the right half shaft include eddy current sensors for measuring the vibration signal data of the shaft relative to the bearing; In step S2, when the residue in the empirical mode model is a monotonic signal or a value less than a preset threshold, the vibration signal decomposition of the signal source is stopped; In step S4, the neural network model includes a hidden layer; In step S4, the input data includes the vibration signal data from different components of the power drive system after time domain decomposition by the empirical mode model in step S2 and then after convolution calculation in step S3; In step S3, the convolution model is a one-dimensional convolutional neural network model.

[0005] CN118500732A proposes a rolling bearing fault diagnosis method based on adaptive modal decomposition. The method includes the following steps: (1) Extraction of rolling bearing fault features (1.1) Using an improved sparrow search algorithm to optimize the parameters of the complete set empirical mode decomposition algorithm, the parameters include the noise amplitude weight Nstd and the number of noise additions NR, and using the envelope entropy as the objective function, and then screening the component with the largest envelope peak factor value as the optimal component; (1.2) Using the filter length L, deconvolution period T, and shift number M as parameters, and using the envelope peak factor as the objective function, and then enhancing the fault impact component in the optimal component based on the maximum correlation kurtosis deconvolution algorithm; (1.3) Extracting bearing fault feature information through the envelope spectrum; (2) Dividing the bearing fault interval (2.1) Using the subtraction optimizer optimization algorithm, using the envelope fault feature energy ratio as the objective function to optimize the decomposition parameters of the time-varying filter empirical mode decomposition algorithm, when the envelope fault feature energy ratio reaches the maximum, a set of bandwidth thresholds and B-spline order optimal parameters are obtained. (2.2) The modal components with small Euclidean distance and large kurtosis and mean square error are selected for reconstruction; (2.3) The singular value entropy of the reconstructed signal is finally calculated, and the normal signal, outer ring fault, inner ring fault and rolling element fault are divided into intervals according to the size of the singular value entropy; (3) Separation and extraction of bearing composite faults (3.1) The golden jackal optimization algorithm is used to adaptively select the parameter penalty factor α and decomposition scale K, and the composite index of permutation entropy and mutual information is used as the objective function, and then the vibration signal is subjected to variational mode decomposition and denoising to obtain the reconstructed signal; (3.2) The multi-point optimization adjustment minimum entropy deconvolution is used to perform deconvolution operation on the reconstructed signal, and different periods, filter orders and window functions are set to realize the separation and extraction of rolling bearing composite faults; (3.3) The Teager energy operator is used to enhance the bearing fault characteristics.

[0006] CN118392491A discloses a fault diagnosis method for AC motor bearings based on MCKD and ICEEMDAN. The method is implemented in the following steps: Step 1, collecting the vibration signal of the motor bearing; Step 2, using the maximum correlation kurtosis deconvolution algorithm to perform the first noise reduction on the collected vibration signal to obtain the first noise reduction vibration signal; Step 3, performing improved adaptive noise complete set empirical mode decomposition on the first noise reduction vibration signal obtained after step 2 to obtain multiple intrinsic mode functions IMF; Step 4, calculating the multi-scale permutation entropy of each IMF component, and setting a threshold, treating the IMF component whose multi-scale permutation entropy is greater than the threshold as an invalid component, removing it, adding the valid components to obtain the final signal for analysis, and performing envelope spectrum analysis to determine the fault type.

[0007] The current AI model for bearing fault diagnosis based on vibration signals has low accuracy, large size, large computing power requirements, and long training and running time. In particular, it has poor noise immunity in strong noise backgrounds. Summary of the invention

[0008] Long-term practice has found that in the context of strong noise, the useful information in the signal is masked by a large amount of noise, making feature extraction and fault identification difficult, with low accuracy, large size, large computing power, and long training and running time. For example, Fourier transform and wavelet transform have limited effects when dealing with complex noise, and it is difficult to effectively separate useful fault features. Traditional feature extraction methods cannot fully capture the subtle features of the fault, resulting in inaccurate diagnosis results. In order to improve the accuracy of diagnosis, traditional models use complex structures and a large number of parameters, resulting in a large model size. Complex models and algorithms usually require more computing resources for feature extraction and fault diagnosis. The hyperparameter adjustment process is cumbersome and time-consuming, and multiple experiments are required to find the optimal parameter combination. In addition, data preprocessing steps, such as denoising and normalization, may also be very time-consuming, especially on large data sets. In particular, there are technical problems such as poor noise resistance under strong noise backgrounds.

[0009] In view of this, the present invention aims to propose a bearing fault diagnosis method based on a neural network, comprising:

[0010] Step S1, obtaining IMF components through an adaptive noise set empirical mode decomposition module, screening them according to the minimum envelope entropy criterion, and calculating the time domain features as the initial feature vector;

[0011] Step S2, select the initial feature vector and expand it in the 1DGAN module to obtain an expanded feature vector;

[0012] Step S3, dividing the expanded feature vector into a training set and a test set, training the CNN-BiLSTM model to obtain a fault prediction model;

[0013] Step S4, obtaining real-time vibration signal data during the operation of the bearing, obtaining a fused feature vector after the real-time vibration signal data passes through steps S1-S2, inputting the fused feature vector into the fault prediction model for prediction, and outputting prediction result data.

[0014] In one embodiment, in step S1, the minimum envelope entropy E is obtained. r for,

[0015]

[0016] Among them, E r is the envelope entropy, f(r) is the IMF component, the envelope signal obtained by Hilbert demodulation, and k IMF components.

[0017] In one embodiment, in step S2, the initial feature vector is selected and expanded in the 1DGAN module at least 5 times.

[0018] In one embodiment, in step S3, the expanded feature vectors are divided into a training set and a test set according to a ratio of 7:3.

[0019] In one embodiment, the CNN-BiLSTM model includes at least an input layer, an output layer, a convolutional layer, and a BiLSTM layer.

[0020] The present invention also discloses a model of the above-mentioned bearing fault diagnosis method based on neural network, and the model includes:

[0021] Adaptive noise ensemble empirical mode decomposition module, used to decompose the bearing vibration signal into IMF components and obtain the initial eigenvector using the minimum envelope entropy criterion;

[0022] A 1DGAN module, comprising at least one generator and one discriminator, for expanding the initial feature vector to obtain an expanded feature vector; wherein the number of the expanded feature vector is greater than the initial feature vector;

[0023] The CNN-BiLSTM model includes an input layer, a one-dimensional convolution layer, a one-dimensional maximum pooling layer, a first random loss layer, a BiLSTM layer, a second random loss layer, a flattening layer, a fully connected layer, a Dropout layer, and an output layer in sequence, and is used to input feature vectors to obtain fault prediction result data.

[0024] In one embodiment, the CNN-BiLSTM model is trained at least 150 times.

[0025] In one embodiment, the learning rate of the CNN-BiLSTM model is 0.01.

[0026] The present invention also discloses an electronic device, at least one processor; and

[0027] a memory communicatively connected to the at least one processor; wherein,

[0028] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned neural network-based bearing fault diagnosis method.

[0029] The present invention further discloses a machine-readable storage medium, on which instructions are stored, and the instructions are used to enable a machine to execute the bearing fault diagnosis method based on neural network as described above in the present invention.

[0030] The present invention discloses a method for bearing fault diagnosis based on a neural network. Through steps S1-S4, an adaptive noise set empirical mode decomposition module is used to extract IMF components, and useful components are screened out according to the minimum envelope entropy criterion, and then the time domain features are calculated as the initial feature vector. The initial feature vector is expanded using a one-dimensional generative adversarial network (1DGAN) to generate a first feature vector. The expanded feature vector is divided into a training set and a test set, and a convolutional neural network-bidirectional long short-term memory network (CNN-BiLSTM) model is used for training to construct a fault prediction model. Real-time vibration signal data is collected during the operation of the bearing, and these data are processed according to steps S1-S2 to obtain a fused feature vector, which is then input into the trained fault prediction model, and the fault prediction result is finally output. The DGAN adversarial network model is used to expand the fault sample, and the CNN-BiLSTM is used to realize efficient diagnosis of bearing faults. The diagnostic accuracy of the bearing fault diagnosis method based on a neural network proposed by the present invention is 9%-25% higher than that of other methods, and the running time is shorter. The bearing fault diagnosis method based on neural network significantly improves the diagnostic accuracy and anti-interference ability of rolling bearings by realizing accurate and comprehensive extraction of fault characteristics.

[0031] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and the description thereof are used to explain the present invention and do not constitute improper limitations on the present invention.

[0033] In the attached picture:

[0034] Figure 1 A schematic diagram of a bearing fault diagnosis method based on a neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0036] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0037] It should be noted that the terms "first", "second", "third", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so as to describe the embodiments of the present invention described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0038] In order to solve the problems in the prior art, in the context of strong noise, traditional methods such as Fourier transform and wavelet transform have limited effects in processing complex noise, and it is difficult to effectively separate useful fault features, resulting in low accuracy; in order to improve diagnostic accuracy, traditional models usually adopt complex structures and a large number of parameters, resulting in a large model size; complex models and algorithms require more computing resources for feature extraction and fault diagnosis, and the training run time is long; the hyperparameter adjustment process is time-consuming and cumbersome, and multiple experiments are required to find the optimal parameter combination; data preprocessing steps (such as denoising, normalization, etc.) are also very time-consuming on large data sets; especially in the context of strong noise, traditional feature extraction methods cannot fully capture the subtle features of the fault, resulting in inaccurate diagnostic results, poor noise resistance and other technical problems. The present invention provides a bearing fault diagnosis method based on a neural network, such as Figure 1 As shown, a schematic diagram of a bearing fault diagnosis method based on a neural network according to an embodiment of the present invention, the bearing fault diagnosis method based on a neural network includes:

[0039] Step S1, obtaining IMF components through an adaptive noise set empirical mode decomposition module, screening them according to the minimum envelope entropy criterion, and calculating the time domain features as the initial feature vector;

[0040] Step S2, select the initial feature vector and expand it in the 1DGAN module to obtain an expanded feature vector;

[0041] Step S3, the expanded feature vector and the initial feature vector are integrated and divided into a training set and a test set, and the CNN-BiLSTM model is trained to obtain a fault prediction model;

[0042] Step S4, obtaining real-time vibration signal data during the operation of the bearing, obtaining a fused feature vector after the real-time vibration signal data passes through steps S1-S2, inputting the fused feature vector into the fault prediction model for prediction, and outputting prediction result data.

[0043] The present invention discloses a method for bearing fault diagnosis based on a neural network. Through steps S1-S4, an adaptive noise set empirical mode decomposition module is used to extract IMF components, and useful components are screened out according to the minimum envelope entropy criterion, and then the time domain features are calculated as initial feature vectors. The initial feature vector is expanded using a one-dimensional generative adversarial network (1DGAN) to generate a feature vector. The fused feature vector is divided into a training set and a test set, and a convolutional neural network-bidirectional long short-term memory network (CNN-BiLSTM) model is used for training to construct a fault prediction model. Real-time vibration signal data is collected during the operation of the bearing, and these data are processed according to steps S1-S2 to obtain a fused feature vector, which is then input into the trained fault prediction model, and the fault prediction result is finally output. The DGAN adversarial network model is used to expand the fault sample, and the CNN-BiLSTM is used to realize efficient diagnosis of bearing faults. The diagnostic accuracy of the bearing fault diagnosis method based on a neural network proposed by the present invention is 9%-25% higher than that of other methods, and the running time is shorter. The bearing fault diagnosis method based on neural network significantly improves the diagnostic accuracy and anti-interference ability of rolling bearings by realizing accurate and comprehensive extraction of fault characteristics.

[0044] The adaptive noise ensemble empirical mode decomposition module adopts the adaptive noise ensemble empirical mode decomposition (ICEEMDAN). ICEEMDAN does not directly add Gaussian white noise during the decomposition process, but selects the kth intrinsic mode component (IMF) component after the white noise is decomposed by empirical mode decomposition (EMD) for decomposition. Compared with the traditional CEEMDAN, it has more accurate decomposition results, better noise control, and wider applicability. Generative adversarial network (GAN) is a deep learning model mainly composed of a generator, a discriminator, a loss function, and an optimizer. The generator network forms an adversarial relationship with the discriminator network to generate new synthetic data instances that can be obtained through real data. One-dimensional Generative Adversarial Network (1DGAN), 1DGAN is specifically aimed at one-dimensional data (such as time series, audio), and uses 1D convolutional layers in the generator and discriminator, enabling it to effectively capture local patterns and sequence information in one-dimensional data.

[0045] The IMF components are obtained by decomposing the adaptive noise set empirical mode decomposition ICEEMDAN, and then screened according to the minimum envelope entropy criterion, and then their time domain features are calculated, and finally the time domain features are used as the feature vector of the sample. The envelope entropy represents the sparse characteristics of the original signal. When there is less noise and more feature information in the IMF, its envelope entropy is smaller. In a preferred embodiment, in step S1, the minimum envelope entropy E is obtained. r for,

[0046]

[0047] Among them, E r is the envelope entropy, f(r) is the IMF component, the envelope signal obtained by Hilbert demodulation, and k IMF components.

[0048] In order to obtain more expanded feature vectors, the generator part is selected from the trained 1DGAN model. The selected initial feature vector is provided to the generator as input. The generator generates an expanded feature vector based on the input. In a preferred embodiment, in step S2, the initial feature vector is selected and expanded in the 1DGAN module at least 5 times. The traditional GAN ​​has problems such as mode collapse and training instability during training. The present invention adopts Wasserstein distance and uses Wasserstein distance as the loss function L of the discriminator. G :

[0049]

[0050] Among them, D(x) represents the discriminator's evaluation of the real data x, and D(G(z)) represents the discriminator's evaluation of the generated data G(z). The goal is to maximize this loss to increase the Wasserstein distance between the real data and the generated data. By optimizing the loss function, the network has stronger stability during the training process and reduces the pattern collapse phenomenon. Even when the generated distribution is very different from the real distribution, the Wasserstein distance can provide effective gradient information, which is helpful for the optimization of the generator.

[0051] In order to ensure that the model has enough data to learn during the training process and to reserve a portion of the data for evaluating the performance of the model, in a preferred embodiment, in step S3, the expanded feature vectors are divided into a training set and a test set in a ratio of 7:3.

[0052] In order to better integrate convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM), process sequence data more effectively, and capture local features and long-term dependencies. In a preferred embodiment, the CNN-BiLSTM model includes at least an input layer, an output layer, a convolutional layer, and a BiLSTM layer. It is possible to capture the contextual information in the sequence, so as to better understand the context of the sequence data. The BiLSTM layer consists of two LSTMs in opposite directions, one reading the sequence from front to back and the other reading the sequence from back to front. The fully connected layer is used to map the output of the BiLSTM layer to the output dimension required for the final classification or regression task. One or more fully connected layers are usually used, and activation functions (such as ReLU) are added to the middle layer.

[0053] The present invention also discloses a model of the above-mentioned bearing fault diagnosis method based on neural network, and the model includes:

[0054] Adaptive noise ensemble empirical mode decomposition module, used to decompose the bearing vibration signal into IMF components and obtain the initial eigenvector using the minimum envelope entropy criterion;

[0055] A 1DGAN module, comprising at least one generator and one discriminator, for expanding the initial feature vector to obtain an expanded feature vector; wherein the number of the expanded feature vector is greater than the initial feature vector;

[0056] The CNN-BiLSTM model includes an input layer, a one-dimensional convolution layer, a one-dimensional maximum pooling layer, a first random loss layer, a BiLSTM layer, a second random loss layer, a flattening layer, a fully connected layer, a Dropout layer, and an output layer in sequence, and is used to input feature vectors to obtain fault prediction result data.

[0057] The model of the bearing fault diagnosis method based on neural network disclosed in the present invention, through the collaborative work of the three modules of adaptive noise set empirical mode decomposition module, 1DGAN module and CNN-BiLSTM model, the system can effectively extract features from the original vibration signal and generate a sufficient number of expanded feature vectors, and finally achieve high-precision fault prediction. The adaptive noise set empirical mode decomposition module decomposes the complex nonlinear non-stationary signal into a series of intrinsic mode functions (IMF). The minimum envelope entropy criterion is used to select the initial feature vector. The 1DGAN adversarial network model is used to expand the fault sample, and the CNN-BiLSTM is used to achieve efficient diagnosis of bearing faults. The diagnostic accuracy of the proposed fault diagnosis model is 9%-25% higher than that of other methods, and the running time is short. By realizing accurate and comprehensive extraction of fault features, the diagnostic accuracy and anti-interference ability of rolling bearings are significantly improved.

[0058] In order to ensure that the model fully learns the characteristics and patterns of the data and can obtain better prediction accuracy, in a preferred embodiment, the training number of the CNN-BiLSTM model includes at least 150 times. Input layer: receives feature vectors. One-dimensional convolution layer: extracts local features. One-dimensional maximum pooling layer: reduces the size of features and reduces computational complexity. First random loss layer: prevents overfitting. BiLSTM layer: captures context information in the sequence. Second random loss layer: further prevents overfitting. Flattening layer: flattens multidimensional data into one dimension. Fully connected layer: performs feature mapping. Dropout layer: further prevents overfitting. Output layer: outputs fault prediction result data.

[0059] In order to make the model have better prediction accuracy, in a preferred embodiment, the learning rate of the CNN-BiLSTM model is 0.01. The intrinsic modal components (IMFs) of each experimental signal are found through ICEEMDAN, and the optimal IMF components of each sample are extracted using the principle of minimum envelope entropy, and then the time domain characteristics of the sample are calculated. In order to verify the accuracy of small sample bearing fault prediction, 150 groups of data are selected and the 1DGAN network model is used for data increment. A large amount of synthetic data is generated using a generator trained by the model, and the generated data is reshaped and labeled to form a new synthetic data set. The CNN-BILSTM fault diagnosis model is then used for diagnosis.

[0060] The public bearing data set divides the bearing fault into 10 states based on the fault location and damage depth at a speed of 1797 rad / s as experimental data, as shown in Table 1:

[0061] Table 1 Bearing fault status

[0062]

[0063]

[0064] In order to improve the diagnostic accuracy of the ICEEMDAN-1DGAN-CNN-BiLSTM model proposed in the present invention, as shown in Table 2, the present invention compares and analyzes it with the ICEEMDAN-CNN-BiLSTM and ICEEMDAN-1DGAN-CNN-LSTM models. The former does not use the DGAN network model to increment the fault data, while the latter uses a unidirectional LSTM model instead of BiLSTM for fault diagnosis. The training times of the above three models are all set to 150 times, the learning rate is 0.01, and the regularization parameter is 0.001.

[0065] Table 2 Comparison of results in different models of bearing fault diagnosis

[0066]

[0067] The present invention also discloses an electronic device, at least one processor; and

[0068] a memory communicatively connected to the at least one processor; wherein,

[0069] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned neural network-based bearing fault diagnosis method.

[0070] The present invention further discloses a machine-readable storage medium, on which instructions are stored, and the instructions are used to enable a machine to execute the bearing fault diagnosis method based on neural network as described above in the present invention.

[0071] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0072] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0073] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0074] The method and device for providing service information provided by the present invention are described in detail above. The present invention uses specific examples to illustrate the principle and implementation of the present invention. The description of the above embodiments is only used to help understand the method and concept of the present invention. At the same time, for those skilled in the art, according to the concept of the present invention, there will be changes in the specific implementation and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A bearing fault diagnosis method based on neural network, characterized in that: The bearing fault diagnosis method based on neural network includes: Step S1, obtaining IMF components through an adaptive noise set empirical mode decomposition module, screening them according to the minimum envelope entropy criterion, and calculating the time domain features as the initial feature vector; Step S2, select the initial feature vector and expand it in the 1DGAN module to obtain an expanded feature vector; Step S3, fusing the expanded feature vector and the initial feature vector and dividing them into a training set and a test set, training the CNN-BiLSTM model to obtain a fault diagnosis model; Step S4, obtaining real-time vibration signal data during the operation of the bearing, obtaining a fused feature vector after the real-time vibration signal data passes through steps S1-S2, inputting the fused feature vector into the fault diagnosis model for diagnosis, and outputting the diagnosis result data.

2. The bearing fault diagnosis method based on neural network according to claim 1 is characterized in that: In step S1, the minimum envelope entropy E is obtained r for, Among them, E r is the envelope entropy, f(r) is the IMF component, the envelope signal obtained by Hilbert demodulation, and k IMF components.

3. The bearing fault diagnosis method based on neural network according to claim 1 is characterized in that: In step S2, the initial feature vector is selected and expanded in the 1DGAN module at least 5 times.

4. The bearing fault diagnosis method based on neural network according to claim 1 is characterized in that: In step S3, the fused feature vector is divided into a training set and a test set according to a ratio of 7:

3.

5. The method for bearing fault diagnosis based on neural network according to any one of claims 1 to 4, characterized in that: The CNN-BiLSTM model includes at least an input layer, an output layer, a convolutional layer and a BiLSTM layer.

6. A model for the bearing fault diagnosis method based on a neural network as claimed in any one of claims 1 to 5, characterized in that: The model includes, Adaptive noise ensemble empirical mode decomposition module, used to decompose the bearing vibration signal into IMF components and obtain the initial eigenvector using the minimum envelope entropy criterion; A 1DGAN module, comprising at least one generator and one discriminator, for expanding the initial feature vector to obtain an expanded feature vector; wherein the number of the expanded feature vector is greater than the initial feature vector; The CNN-BiLSTM model includes an input layer, a one-dimensional convolution layer, a one-dimensional maximum pooling layer, a first random loss layer, a BiLSTM layer, a second random loss layer, a flattening layer, a fully connected layer, a Dropout layer, and an output layer in sequence, and is used to input feature vectors to obtain fault diagnosis result data.

7. The model according to claim 6, characterized in that The CNN-BiLSTM model is trained at least 150 times.

8. The model according to claim 7, characterized in that The learning rate of the CNN-BiLSTM model is 0.

01.

9. An electronic device, characterized in that: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the neural network-based bearing fault diagnosis method described in any one of claims 1-5.

10. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores instructions for enabling a machine to execute the neural network-based bearing fault diagnosis method according to any one of claims 1 to 5 of the present invention.

Images