Vibration and current fused variable working condition motor anti-noise intelligent diagnosis method and system

Abstract

The invention discloses a vibration and current fused variable working condition motor anti-noise intelligent diagnosis method and system, belongs to the field of electromechanical fault diagnosis, and realizes instantaneous frequency extraction and fusion of a motor vibration signal and a current signal by adopting a feature fusion method combining adaptive frequency modulation mode decomposition and a maximum absolute value rule. The invention provides an improved convolutional neural network fault diagnosis model based on convolution kernel width and deep optimization, and solves the problems that a traditional deep learning method is not high in motor mechanical and electrical fault diagnosis accuracy and the model performance is degraded. On the basis, an adaptive batch normalization algorithm is used for enhancing the noise immunity of the fault diagnosis model, and the fault diagnosis model has high fault diagnosis accuracy when Gaussian white noise, color noise and random uniformly distributed noise are single or mixed. And an effective solution is provided for anti-noise intelligent diagnosis of mechanical and electrical faults of the induction motor under variable working conditions.

Description

Technical Field

[0001] This invention belongs to the field of electromechanical fault diagnosis, specifically relating to the field of motor mechanical and electrical fault diagnosis, and more specifically to a method and system for intelligent noise reduction diagnosis of variable operating condition motors that integrates vibration and current. Background Technology

[0002] Induction motors are widely used in various industrial fields due to their simple structure and low cost. However, under long-term varying speeds and loads, induction motors are highly susceptible to various mechanical and electrical faults, leading to irreversible mechanical damage and control malfunctions. Therefore, fault diagnosis of induction motors is of great importance.

[0003] Under varying speed and load conditions, the data obtained from testing induction motors exhibits complex distribution patterns, making accurate fault feature extraction difficult. Single vibration signals are susceptible to noise interference in the low-frequency range, affecting fault feature extraction, while single current signals are affected by harmonic components and electromagnetic noise in the high-frequency range. Therefore, feature fusion of vibration and current signals can overcome the shortcomings of using them individually, achieving better motor fault diagnosis results.

[0004] Classical signal processing-based motor fault diagnosis methods require strong expertise in signal analysis and processing, and the algorithms are highly susceptible to varying performance under different operating conditions, making it difficult to develop a universally applicable method. With the rapid development of big data analytics and artificial intelligence, data-driven, deep learning-based intelligent motor fault diagnosis methods have emerged. However, these methods, primarily based on convolutional neural networks, are prone to overfitting due to data distribution patterns and noise. Furthermore, overly complex deep learning models can slow down processing speeds, hindering the widespread application of deep learning methods in motor fault diagnosis. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for intelligent noise diagnosis of motors under varying operating conditions by fusing vibration and current signals, in order to overcome the problem of inaccurate extraction of single signal features of motors under varying operating conditions. This invention designs a method for fusing vibration and current signal features under varying operating conditions, combined with an intelligent fault diagnosis model with strong generalization and high noise resistance, to achieve intelligent and accurate diagnosis of mechanical and electrical faults of motors under varying operating conditions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A smart diagnostic method for noise reduction in variable-condition motors that integrates vibration and current analysis includes the following steps: 1) Construct a motor fault feature dataset Vibration and current signals of the motor under normal, mechanical fault, and electrical fault conditions were collected under variable speed and load conditions. Adaptive frequency modulation decomposition was performed on both types of signals to extract instantaneous frequencies. Then, the maximum absolute value rule was used to fuse the two instantaneous frequencies to obtain a motor fault feature dataset. Finally, the dataset was used... K Cross-validation divides the motor fault feature dataset into training, validation and test sets according to a preset ratio. 2) Construct a fault diagnosis model Using the motor fault feature dataset in step 1), a convolutional neural network fault diagnosis model based on the optimization of convolutional kernel width and depth is constructed. The training set is used for training the fault diagnosis model, and the validation set is used for hyperparameter tuning of the fault diagnosis model. After model training and tuning, the optimal model structure and weight parameters for fault diagnosis are saved. 3) Improve the model's noise resistance Using the optimal fault diagnosis model from step 2), an adaptive batch normalization algorithm is employed to verify the model's noise resistance on noisy test set data. The noise resistance evaluation of the motor fault diagnosis model is completed for Gaussian white noise, colored noise, and random uniformly distributed noise and their mixtures.

[0007] Further, step 1) involves constructing a motor fault feature dataset, specifically as follows: 1.1) Vibration and current signals are collected under normal, mechanical fault, and electrical fault conditions of the motor. Adaptive frequency modulation mode decomposition is used to extract the instantaneous frequencies of the vibration and current signals. The instantaneous frequency increment is estimated as follows:

[0008] In the formula, j Indicates the number of iterations. , For instantaneous frequency increment, and This is the time series matrix of two orthogonally demodulated signals; 1.2) The feature samples are obtained by fusing two instantaneous frequencies using the maximum absolute value rule. The fusion rule is as follows:

[0009] In the formula, IF For fusion frequency, IF D The instantaneous frequency of the vibration signal. IF A The instantaneous frequency of the current signal; 1.3) All the obtained feature samples are labeled according to different motor states. K-fold cross-validation is used to divide the motor fault feature dataset into slices according to a preset ratio to obtain training set, validation set and test set.

[0010] Furthermore, the preset ratio in step 1.3) is 7:1:2.

[0011] Further, step 2) involves constructing a fault diagnosis model, specifically as follows: 2.1) Set the width of the first convolutional kernel to 45×1; 2.2) Set the kernel width of the subsequent six convolutional layers to 7×1 and the depth to 64; 2.3) Use 1×1 small convolution kernels for cross-short-connection of residual blocks; 2.4) Use the training set to train the fault diagnosis model, and the validation set to tune the hyperparameters of the fault diagnosis model and save the optimal structure and weight parameters of the fault diagnosis model.

[0012] Furthermore, step 3) improves the model's noise resistance specifically as follows: 3.1) Using the obtained optimal fault diagnosis model, the AdaBN algorithm is employed to enhance the model's noise resistance; 3.2) For Gaussian white noise, colored noise, and random uniformly distributed noise and their mixtures, fault diagnosis tests are performed on the noisy test set data to complete the noise immunity evaluation of the motor fault diagnosis model.

[0013] A vibration and current integrated intelligent diagnostic system for variable-condition motor noise reduction includes: The motor fault feature dataset construction module is used to collect vibration and current signals of motors under normal, mechanical, and electrical fault conditions during variable speed and load operation. It performs adaptive frequency modulation decomposition on both types of signals to extract instantaneous frequencies, then uses the maximum absolute value rule to fuse the two instantaneous frequencies to obtain the motor fault feature dataset. Finally, it uses... K Cross-validation divides the motor fault feature dataset into training, validation and test sets according to a preset ratio. Fault diagnosis model construction module: used to build a convolutional neural network fault diagnosis model based on the optimization of convolutional kernel width and depth using motor fault feature dataset. The training set is used for fault diagnosis model training, and the validation set is used for hyperparameter tuning of the fault diagnosis model. After model training and tuning, the optimal fault diagnosis model structure and weight parameters are saved. Model noise immunity enhancement module: Used to verify the noise immunity of the model on noisy test set data by using the best fault diagnosis model and an adaptive batch normalization algorithm. It evaluates the noise immunity of the motor fault diagnosis model for Gaussian white noise, colored noise, random uniformly distributed noise and their mixtures.

[0014] Furthermore, the motor fault feature dataset construction module specifically includes: 1.1) Vibration and current signals are collected under normal, mechanical fault, and electrical fault conditions of the motor. Adaptive frequency modulation mode decomposition is used to extract the instantaneous frequencies of the vibration and current signals. The instantaneous frequency increment is estimated as follows:

[0015] In the formula, j Indicates the number of iterations. , For instantaneous frequency increment, and This is the time series matrix of two orthogonally demodulated signals; 1.2) The feature samples are obtained by fusing two instantaneous frequencies using the maximum absolute value rule. The fusion rule is as follows:

[0016] In the formula, IF For fusion frequency, IF D The instantaneous frequency of the vibration signal. IF A The instantaneous frequency of the current signal; 1.3) All the obtained feature samples are labeled according to different motor states. K-fold cross-validation is used to divide the motor fault feature dataset into slices according to a preset ratio to obtain training set, validation set and test set.

[0017] Furthermore, the preset ratio in step 1.3) is 7:1:2.

[0018] Furthermore, the fault diagnosis model construction module specifically includes: 2.1) Set the width of the first convolutional kernel to 45×1; 2.2) Set the kernel width of the subsequent six convolutional layers to 7×1 and the depth to 64; 2.3) Use 1×1 small convolution kernels for cross-short-connection of residual blocks; 2.4) Use the training set to train the fault diagnosis model, and the validation set to tune the hyperparameters of the fault diagnosis model and save the optimal structure and weight parameters of the fault diagnosis model.

[0019] Furthermore, the model noise resistance improvement module specifically includes: 3.1) Using the obtained optimal fault diagnosis model, the AdaBN algorithm is employed to enhance the model's noise resistance; 3.2) For Gaussian white noise, colored noise, and random uniformly distributed noise and their mixtures, fault diagnosis tests are performed on the noisy test set data to complete the noise immunity evaluation of the motor fault diagnosis model.

[0020] Compared with the prior art, the present invention has the following beneficial technical effects: 1) To address the difficulty and inaccuracy of extracting features from single vibration or current signals under varying operating conditions, a feature fusion method combining ACMD and the maximum absolute value rule is proposed. Compared with traditional single signal feature extraction methods, the feature fusion method can extract complete feature information across the entire frequency band.

[0021] 2) To address the issues of low accuracy and high training costs associated with traditional fault diagnosis models, a convolutional neural network-based intelligent fault diagnosis model with strong generalization and high noise resistance was designed. The model was trained and validated using new samples fused from vibration and current features obtained under varying operating conditions. Noise resistance was verified on noisy test data, evaluating the noise resistance of the motor fault diagnosis method for Gaussian white noise, colored noise, random uniformly distributed noise, and their mixtures. Compared to conventional fault diagnosis methods, this invention demonstrates high accuracy for both mechanical and electrical faults in motors under variable speed and load conditions. Attached Figure Description

[0022] The accompanying drawings are provided to further understand the invention and constitute a part of this invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0023] Figure 1 A smart diagnostic method for noise reduction of motors operating under varying conditions that integrates vibration and current; Figure 2(a) shows the original signals and instantaneous frequencies of eight motor states under no-load conditions; Figure 2(b) shows the original signals and instantaneous frequencies of eight motor states under light load conditions; Figure 2(c) shows the original signals and instantaneous frequencies of eight motor states under medium load conditions; Figure 2(d) shows the original signals and instantaneous frequencies of eight motor states under heavy load conditions; Figure 3 A graph showing the frequency fusion results for eight motor states under four different load conditions; Figure 4 This is a schematic diagram of the convolutional neural network structure constructed in this invention; Figure 5 This is the motor state classification performance of the present invention under varying operating conditions, where (a) is the accuracy of the training set and (b) is the accuracy of the validation set. Figure 6 The graph shows the accuracy results of the fault diagnosis model under seven different noise levels. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be understood that the data used in this way can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in a sequence other than those illustrated or described herein. Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] Example 1 This invention provides an intelligent diagnostic method for noise reduction of motors under varying operating conditions that integrates vibration and current, comprising the following steps: 1) Constructing a motor fault feature dataset: Vibration and current signals of the motor under normal, mechanical, and electrical fault conditions were collected under variable speed and load conditions. Adaptive chirp mode decomposition (ACMD) was performed on both types of signals to extract instantaneous frequencies. The two instantaneous frequencies were then fused using the maximum absolute value rule to obtain the motor fault feature dataset. Finally, the dataset was used... K Cross-validation divides the motor fault feature dataset into slices in a 7:1:2 ratio to obtain training, validation and test sets. Specifically: 1.1) Collect vibration and current signals from motors under normal, mechanical, and electrical fault conditions, and extract the instantaneous frequencies of the vibration and current signals using ACMD. Figures 2(a)-2(d) Its instantaneous frequency increment can be estimated as:

[0027] In the formula, j Indicates the number of iterations. , For instantaneous frequency increment, and Let be the time series matrix of two orthogonal demodulated signals.

[0028] 1.2) The feature samples are obtained by fusing the two instantaneous frequencies using the maximum absolute value rule, as shown below. Figure 3 Its fusion rules are as follows:

[0029] In the formula, IF For fusion frequency, IF D The instantaneous frequency of the vibration signal. IF A The instantaneous frequency of the current signal.

[0030] 1.3) The obtained motor fault feature dataset is labeled according to different motor states, and K-fold cross-validation is used to divide it into slices in a ratio of 7:1:2 to obtain training set, validation set and test set.

[0031] A single vibration signal is susceptible to noise interference in the low-frequency band, and a single current signal is susceptible to harmonic components and electromagnetic noise in the high-frequency band. Therefore, fusing vibration and current signals can overcome the shortcomings of using them separately.

[0032] 2) Constructing a fault diagnosis model: Using the motor fault feature dataset in step 1), construct a convolutional neural network fault diagnosis model based on the optimization of convolutional kernel width and depth. The training set is used for training the fault diagnosis model, and the validation set is used for hyperparameter tuning of the fault diagnosis model. After model training and tuning, save the optimal fault diagnosis model structure and weight parameters. Specifically: 2.1) Set the width of the first convolutional kernel to 45×1. Increase the receptive field by using a large convolutional kernel to quickly and completely extract short-term features and remove invalid features; 2.2) Set the width of the subsequent six convolutional kernels to 7×1 and the depth to 64. By using small and deep convolutional kernels, the network depth is increased, which enhances the network's ability to express data features and suppresses model overfitting. 2.3) Using 1×1 small convolutional kernels to cross-connect residual blocks enhances the nonlinearity of the network, strengthens feature reuse and reduces the number of parameters, and at the same time, it propagates the gradient back to the previous layers more quickly, thereby alleviating the gradient vanishing problem and helping to train deep networks. 2.4) Use the training set to train the fault diagnosis model, and the validation set to tune the hyperparameters of the fault diagnosis model and save the optimal structure and weight parameters of the fault diagnosis model.

[0033] 3) Improve the noise resistance of the model: Using the best fault diagnosis model in step 2), the Adaptive Batch Normalization (AdaBN) algorithm is used to verify the noise resistance of the model on the noisy test set data. The noise resistance of the motor fault diagnosis method is evaluated for Gaussian white noise, colored noise, random uniformly distributed noise and their mixture. Specifically: 3.1) Using the obtained optimal fault diagnosis model, the AdaBN algorithm is employed to enhance the model's noise resistance; 3.2) For Gaussian white noise, colored noise, and random uniformly distributed noise and their mixtures, fault diagnosis tests are performed on the noisy test set data to complete the noise immunity evaluation of the motor fault diagnosis model.

[0034] Example 2 See appendix Figure 1 The intelligent noise reduction diagnostic method for variable-condition motors that integrates vibration and current analysis of the present invention mainly includes the following steps: 1) Motor fault simulation experiments were conducted using a motor fault simulation test bench. The experiment set up motors in 8 different states, including 1 normal state, 3 mechanical faults of bearings and rotors, and 4 electrical faults of stator and rotor. The three-phase induction motor was set to accelerate from standstill to 1800 r / min and from no-load to heavy-load conditions. Vibration signals and current signals were then collected. 1.1) Collect vibration and current signals from the motor under different operating conditions and states to obtain raw data. Use ACMD to extract the instantaneous frequencies of the vibration and current signals, such as... Figures 2(a)-2(d) Its instantaneous frequency can be estimated as:

[0035] In the formula, j Indicates the number of iterations. , For instantaneous frequency increment, and Let be the time series matrix of two orthogonal demodulated signals.

[0036] 1.2) The maximum absolute value rule is used to fuse two instantaneous frequencies to obtain feature samples, such as... Figure 3 Its fusion rules are as follows:

[0037] In the formula, IF For fusion frequency, IF D The instantaneous frequency of the vibration signal. IF A The instantaneous frequency of the current signal.

[0038] 1.3) The obtained motor fault feature dataset is labeled according to different motor states, and then... K Cross-validation is used to divide the dataset into training, validation, and test sets, with the ratio of the three sets set to 7:1:2.

[0039] 2) Constructing a fault diagnosis model, which specifically includes the following steps: 2.1) Construct a convolutional neural network fault diagnosis model using the fault feature dataset from step 1). The designed model structure is as follows: Figure 4 As shown in Table 1, the main parameters are as follows; 2.2) The width of the first convolutional kernel is set to 45×1. The large convolutional kernel increases the receptive field, quickly and completely extracts short-term features, and removes invalid features. The width of the convolutional kernels in the subsequent six layers is set to 7×1 and the depth is set to 64. The small and deep convolutional kernels increase the network depth, enhance the network's ability to express data features, and suppress model overfitting. Small 1×1 convolutional kernels are used to cross-connect residual blocks, which enhances the nonlinearity of the network, strengthens feature reuse, and reduces the number of parameters. At the same time, the gradient is transmitted back to the previous layers more quickly, thereby alleviating the gradient vanishing problem and helping to train deep networks. 2.3) Use the training set to train the fault diagnosis model, and the validation set to tune the hyperparameters of the fault diagnosis model and save the optimal structure and weight parameters of the fault diagnosis model.

[0040] Table 1 Main parameters of the model

[0041] 3) Improving the model's noise resistance specifically includes the following steps: 3.1) Using the obtained optimal fault diagnosis model, the AdaBN algorithm is employed to enhance the model's noise resistance; 3.2) For Gaussian white noise, colored noise, and random uniformly distributed noise and their mixtures, fault diagnosis tests are performed on the noisy test set data to complete the noise immunity evaluation of the motor fault diagnosis model.

[0042] The final model achieved an accuracy of 99.61% on the training set, 97.58% on the validation set, and 97.42% on the test set. To demonstrate the superiority of the method presented in this invention, it is compared with AlexNet and ResNet, and the results are as follows: Figure 5 (a)- Figure 5 (b) As shown in Table 2.

[0043] Table 2 Comparison of results from different diagnostic methods

[0044] With signal-to-noise ratios (SNR) set to -3, 0, and 3 dB, Gaussian white noise (GN), colored noise (CN), and random uniformly distributed noise (RDN) were added to the original signal individually and in combination. Using accuracy as the metric, the trained optimal fault diagnosis model was validated on noisy test data using AdaBN. The results are as follows: Figure 6As shown in Table 3, consistent noise immunity patterns were observed for the three signal-to-noise ratio scenarios: the fault diagnosis accuracy was highest when any of the three noises acted individually, followed by when they were mixed in pairs, and lowest when all three acted simultaneously.

[0045] Table 3 Model accuracy under different signal-to-noise ratios

[0046] Example 3 This invention provides an intelligent noise reduction diagnostic system for variable-condition motors that integrates vibration and current, comprising: The motor fault feature dataset construction module is used to collect vibration and current signals of motors under normal, mechanical, and electrical fault conditions during variable speed and load operation. It performs adaptive frequency modulation decomposition on both types of signals to extract instantaneous frequencies, then uses the maximum absolute value rule to fuse the two instantaneous frequencies to obtain the motor fault feature dataset. Finally, it uses... K Cross-validation divides the motor fault feature dataset into training, validation and test sets according to a preset ratio. Fault diagnosis model construction module: used to build a convolutional neural network fault diagnosis model based on the optimization of convolutional kernel width and depth using motor fault feature dataset. The training set is used for fault diagnosis model training, and the validation set is used for hyperparameter tuning of the fault diagnosis model. After model training and tuning, the optimal fault diagnosis model structure and weight parameters are saved. Model noise immunity enhancement module: Used to verify the noise immunity of the model on noisy test set data by using the best fault diagnosis model and an adaptive batch normalization algorithm. It evaluates the noise immunity of the motor fault diagnosis model for Gaussian white noise, colored noise, random uniformly distributed noise and their mixtures.

[0047] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0048] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0049] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0050] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims of the invention.

Claims

1. A variable-operating-condition motor anti-noise intelligent diagnosis method based on vibration and current fusion, characterized in that: The following steps are involved: 1) Constructing a motor fault feature dataset The vibration and current signals of the motor in normal, mechanical and electrical fault states are collected under variable speed and load conditions, and the two signals are adaptively decomposed into frequency modulation modes to extract the instantaneous frequency. The maximum absolute value rule is then used to fuse the two instantaneous frequencies to obtain the motor fault feature data set. K The motor fault feature data set is sliced ​​and divided according to the preset ratio by cross-validation to obtain the training set, validation set and test set; 2) Build a fault diagnosis model Using the motor fault feature data set in step 1), a convolutional neural network fault diagnosis model based on optimized convolution kernel width and depth is constructed, wherein the training set is used for fault diagnosis model training, and the validation set is used for hyperparameter tuning of the fault diagnosis model. After model training and tuning, the optimal model structure and weight parameters for fault diagnosis are saved; 3) Improve the model’s noise resistance Using the optimal fault diagnosis model in step 2), the adaptive batch normalization algorithm is used to verify the noise resistance of the model on the noisy test set data. The noise resistance evaluation of the motor fault diagnosis model is completed for Gaussian white noise, color noise, random uniform distribution noise and their mixtures.

2. The variable-operating-condition motor anti-noise intelligent diagnosis method with vibration and current fusion according to claim 1 is characterized in that: Step 1) Construct a motor fault feature dataset, specifically: 1.1) Collect the vibration signals and current signals of the motor in normal, mechanical and electrical fault states, and use adaptive frequency modulation mode decomposition to extract the instantaneous frequency of the vibration and current signals. The instantaneous frequency increment is estimated as: In the formula, j represents the number of iterations, , is the instantaneous frequency increment, and is the time series matrix of two orthogonal demodulated signals; 1.2) Use the maximum absolute value rule to fuse the two instantaneous frequencies to obtain feature samples. The fusion rule is: In the formula, IF is the fusion frequency, IF D is the instantaneous frequency of the vibration signal, IF A is the instantaneous frequency of the current signal; 1.3) All feature samples obtained are marked according to different motor states, and K-fold cross validation is used to slice and divide the motor fault feature data set according to a preset ratio to obtain a training set, a validation set, and a test set.

3. The variable-operating-condition motor anti-noise intelligent diagnosis method with vibration and current fusion according to claim 2 is characterized in that: The preset ratio in step 1.3) is 7:1:

2.

4. The variable-operating-condition motor anti-noise intelligent diagnosis method of vibration and current fusion according to claim 1 is characterized in that: Step 2) Build a fault diagnosis model, specifically: 2.1) Set the width of the first convolution kernel to 45×1; 2.2) Set the width of the subsequent six convolutional kernels to 7×1 and the depth to 64; 2.3) Use a small 1×1 convolution kernel to cross-connect the residual blocks; 2.4) The training set is used for fault diagnosis model training, and the validation set is used for hyperparameter tuning of the fault diagnosis model and saving the optimal structure and weight parameters of the fault diagnosis model.

5. The variable-operating-condition motor anti-noise intelligent diagnosis method with vibration and current fusion according to claim 1 is characterized in that: In step 3), the model's anti-noise performance is improved, specifically: 3.1) Using the obtained optimal fault diagnosis model, the AdaBN algorithm is used to enhance the model's noise resistance; 3.2) For Gaussian white noise, color noise, random uniformly distributed noise and their mixtures, fault diagnosis tests are performed on the noisy test set data to complete the noise resistance evaluation of the motor fault diagnosis model.

6. The variable-operating-condition motor anti-noise intelligent diagnostic system integrating vibration and current is characterized by: include: Motor fault feature data set construction module: It is used to collect vibration and current signals of normal, mechanical and electrical fault states of the motor under variable speed and load conditions, and extract the instantaneous frequency by adaptive frequency modulation mode decomposition of the two signals, and then use the maximum absolute value rule to fuse the two instantaneous frequencies to obtain the motor fault feature data set. K The motor fault feature data set is sliced ​​and divided according to the preset ratio by cross-validation to obtain the training set, validation set and test set; Fault diagnosis model construction module: used to construct a convolutional neural network fault diagnosis model based on convolution kernel width and depth optimization using the motor fault feature data set. The training set is used for fault diagnosis model training, and the validation set is used for hyperparameter tuning of the fault diagnosis model. After model training and tuning, the optimal model structure and weight parameters for fault diagnosis are saved. Model noise resistance performance improvement module: It is used to use the optimal fault diagnosis model and the adaptive batch normalization algorithm to verify the model noise resistance of the noisy test set data, and complete the noise resistance evaluation of the motor fault diagnosis model for Gaussian white noise, color noise, random uniformly distributed noise and their mixtures.

7. The variable-operating-condition motor anti-noise intelligent diagnosis system with vibration and current fusion according to claim 6 is characterized in that: The motor fault feature data set construction module specifically includes: 1.1) Collect the vibration signals and current signals of the motor in normal, mechanical and electrical fault states, and use adaptive frequency modulation mode decomposition to extract the instantaneous frequency of the vibration and current signals. The instantaneous frequency increment is estimated as: In the formula, j represents the number of iterations, , is the instantaneous frequency increment, and is the time series matrix of two orthogonal demodulated signals; 1.2) Use the maximum absolute value rule to fuse the two instantaneous frequencies to obtain feature samples. The fusion rule is: In the formula, IF is the fusion frequency, IF D is the instantaneous frequency of the vibration signal, IF A is the instantaneous frequency of the current signal; 1.3) All feature samples obtained are marked according to different motor states, and K-fold cross validation is used to slice and divide the motor fault feature data set according to a preset ratio to obtain a training set, a validation set, and a test set.

8. The variable-operating-condition motor anti-noise intelligent diagnosis system with vibration and current fusion according to claim 7 is characterized in that: The preset ratio in step 1.3) is 7:1:

2.

9. The variable-operating-condition motor anti-noise intelligent diagnosis system with vibration and current fusion according to claim 6 is characterized in that: The fault diagnosis model construction module specifically includes: 2.1) Set the width of the first convolution kernel to 45×1; 2.2) Set the width of the subsequent six convolutional kernels to 7×1 and the depth to 64; 2.3) Use a small 1×1 convolution kernel to cross-connect the residual blocks; 2.4) The training set is used for fault diagnosis model training, and the validation set is used for hyperparameter tuning of the fault diagnosis model and saving the optimal structure and weight parameters of the fault diagnosis model.

10. The variable-operating-condition motor anti-noise intelligent diagnosis system with vibration and current fusion according to claim 6 is characterized in that: The model anti-noise performance improvement module is specifically as follows: 3.1) Using the obtained optimal fault diagnosis model, the AdaBN algorithm is used to enhance the model's noise resistance; 3.2) For Gaussian white noise, color noise, random uniformly distributed noise and their mixtures, fault diagnosis tests are performed on the noisy test set data to complete the noise resistance evaluation of the motor fault diagnosis model.

Images