CN-121977836-A - Permanent magnet synchronous motor bearing fault diagnosis monitoring system and method

CN121977836ACN 121977836 ACN121977836 ACN 121977836ACN-121977836-A

Abstract

The invention relates to the technical field of industrial motors, in particular to a system and a method for diagnosing and monitoring a bearing fault of a permanent magnet synchronous motor, which synchronously collect signals such as vibration, sound, temperature and the like through a multi-source sensor, after preprocessing and double-attention feature fusion, inputting a lightweight depth migration model to realize fault diagnosis and positioning, carrying out real-time reasoning at the edge end, carrying out cloud iteration on an optimization model, and finally providing full-flow monitoring, early warning and maintenance services.

Inventors

MA JIA
DONG LEI
ZHANG JINGWEI
LIU SHUANGJUN
LI YINGNAN
CHANG YUAN
ZHENG HAO
DU ZHENWANG
ZHAO LIN
WANG BAOHUA
WANG HAO
Zheng Huicui

Assignees

唐山三友蓝海科技有限责任公司

Dates

Publication Date: 20260505
Application Date: 20260205

Claims (7)

1. The utility model provides a permanent magnet synchronous motor bearing fault diagnosis monitoring system which characterized in that includes multisource information acquisition module, multisource information preprocessing module, multisource feature fusion module and Light-TRANSFERNET degree of depth migration diagnosis module, wherein: the multisource information acquisition module is used for acquiring signals of vibration, temperature, rotating speed, sound and current aiming at the permanent magnet synchronous motor bearing; The multisource data preprocessing module is used for preprocessing the acquired signals, extracting the characteristics of each signal and carrying out normalization processing on the extracted characteristics; The multi-attention feature fusion module is used for carrying out feature fusion on the normalized features to obtain a fusion feature map; the Light-TRANSFERNET depth migration diagnosis module is used for extracting deep fault classification features and spatial position features by utilizing a MobileNetV backbone network based on a fusion feature map so as to determine fault types and position fault positions, and predicting the residual service life of the bearing by utilizing an LSTM-GRU hybrid model based on trend features in the normalized features.
2. The permanent magnet synchronous motor bearing fault diagnosis monitoring system according to claim 1, further comprising an edge-cloud co-deployment module: the edge end is provided with a multisource information preprocessing module, a multi-attention feature fusion module and a Light-TRANSFERNET depth migration diagnosis module, and a fault early warning program is configured to set an alarm threshold value to realize real-time early warning; uploading diagnosis information real-time data comprising fault types and fault positions to the cloud end through 4G/5G/Ethernet by data transmission; The cloud configuration fault information pushing program pushes diagnosis information to operation and maintenance personnel in real time.
3. The permanent magnet synchronous motor bearing fault diagnosis monitoring system according to claim 1, wherein the multi-source information acquisition module is specifically configured to Vibration, temperature, rotating speed, sound and current sensors are installed by adopting a distributed layout, wherein the vibration sensors are tightly attached to a bearing seat to collect three acceleration signals, the current sensors are fastened and fixed on a stator A/B/C three-phase cable, the temperature sensors are installed on bearing seats of front and rear end covers of a motor by adopting eddy current temperature sensors, the rotating speed sensors are installed at the output shaft ends of the motor, the coaxiality error is smaller than 0.1mm, and the sound sensors are installed at positions which are more than 10cm away from a motor fan; The sampling frequency of the acquisition card for each path of signal is respectively set, a time stamp is added for each path of acquired data in the process of acquiring the signal by the acquisition card, and multi-source data alignment is realized by a linear interpolation algorithm.
4. The permanent magnet synchronous motor bearing fault diagnosis monitoring system according to claim 1, wherein the multisource information preprocessing module is specifically used for Aiming at the vibration signal, drift is eliminated through linear trending, a second-order IIR Notch filter is adopted to inhibit power frequency interference of a power grid, a two-dimensional time-frequency diagram is generated through short-time Fourier transformation, the vibration instantaneous time-frequency fault characteristics are extracted, and the kurtosis of the vibration signal is calculated so as to extract kurtosis characteristics; calculating a current root mean square value and a harmonic distortion rate according to the current signal, and filtering electromagnetic interference noise so as to extract a current smoothing characteristic and a current distortion characteristic; smoothing the temperature signal by adopting a sliding average algorithm to obtain a temperature smoothing characteristic, and calculating a temperature change rate so as to extract the temperature change rate characteristic; Calculating a mel spectrum for the sound signal to extract mel spectrum features; calculating a rotation speed order for the rotation speed signal so as to extract a fault frequency characteristic; And mapping each extracted characteristic to a [0,1] interval through Min-Max normalization, and eliminating dimension difference.
5. The permanent magnet synchronous motor bearing fault diagnosis monitoring system according to claim 1, wherein the multi-attention feature fusion module is specifically configured to Taking each extracted characteristic as a channel, and unifying according to the set characteristic diagram size, so as to form a multi-source combined characteristic diagram formed by splicing vibration, current, temperature, sound and rotating speed sub-characteristic diagrams; The method comprises the steps of learning channel attention weights through two full-connection layers aiming at channel-level global features, multiplying the learned channel attention weights with a multi-source combined feature map channel by channel to obtain a channel weighted feature map, and strengthening key fault channel contribution; Performing channel average pooling and maximum pooling on the channel weighted feature images to obtain two single-channel feature images, generating a spatial attention weight by 3×3 convolution processing and Sigmoid activation on the two single-channel feature images, multiplying the spatial attention weight and the channel weighted feature images element by element to obtain a spatial weighted feature image and a focusing fault region; And (3) reducing the dimension of the space weighted feature map through 1X 1 convolution, eliminating feature redundancy, reducing gradient vanishing risk through BN normalization processing, and finally introducing nonlinearity through a ReLU6 activation function so as to output a fusion feature map.
6. The permanent magnet synchronous motor bearing fault diagnosis monitoring system according to claim 4, wherein the Light-TRANSFERNET depth migration diagnosis module is provided with a function for Inputting MobileNetV the fusion feature map into a MobileNetV backbone network, reducing the parameter number through depth separable convolution, and simultaneously embedding an SE attention mechanism to dynamically adjust the weight of each feature channel, so as to extract deep fault classification features and spatial position features; Introducing maximum mean difference MMD loss to reduce the difference between a source domain and a target domain, calculating total loss based on fault classification loss, MMD loss and space position loss, updating backbone network parameters through back propagation based on the total loss, and improving cross-domain generalization capability, wherein the source domain is labs marked fault data, and the target domain is on-site unmarked operation data; Inputting deep fault classification features into two full-connection layers, outputting four-dimensional logits values, converting the four-dimensional logits values into probability distribution through Softmax activation, and selecting fault types corresponding to the maximum probability as diagnosis results, wherein the fault types comprise inner ring faults, outer ring faults, rolling body faults and composite faults; inputting the spatial position characteristics into two fully-connected layers, predicting the fault direction angle and the fault distance, and converting the predicted direction angle and the predicted fault distance into rectangular coordinates so as to position the fault position; And extracting kurtosis characteristics, current smoothing characteristics, temperature change rate characteristics and fault frequency characteristics, and inputting an LSTM-GRU hybrid model to obtain the residual service life of the bearing.
7. A method for diagnosing and monitoring a bearing fault of a permanent magnet synchronous motor, which is characterized by being applied to the permanent magnet synchronous motor bearing fault diagnosis and monitoring system as claimed in any one of claims 1 to 6, and comprising the following specific steps: s1, collecting signals of vibration, temperature, rotating speed, sound and current aiming at a permanent magnet synchronous motor bearing; s2, preprocessing the acquired signals, extracting the characteristics of each signal, and carrying out normalization processing on the extracted characteristics; s3, carrying out feature fusion on the features subjected to normalization processing to obtain a fusion feature map; And S4, extracting deep fault classification features and spatial position features by utilizing MobileNetV backbone networks based on the fusion feature graphs so as to determine fault types and position fault positions, and predicting the residual service life of the bearing by utilizing an LSTM-GRU hybrid model based on trend features in the normalized features.

Description

Permanent magnet synchronous motor bearing fault diagnosis monitoring system and method Technical Field The invention relates to the technical field of industrial motor fault diagnosis and intelligent monitoring, in particular to a permanent magnet synchronous motor bearing fault diagnosis monitoring system and method. Background The permanent magnet synchronous motor (PERMANENT MAGNET synchronous motor, PMSM) is a core power system for driving the high-pressure pump to operate by virtue of high efficiency, high power density, quick dynamic response and the like. As a core power device, the health level of a key component, namely a bearing, directly influences the service life of the whole device system and the ion filtering effect in the sea water desalination process, so that a perfect fault detection technology needs to be established so as to realize accurate monitoring of the bearing operation state. According to 2024 industrial electric control fault statistics report, the bearing fault accounts for 40% -55% of the total fault of the permanent magnet synchronous motor, and is the primary cause of unplanned shutdown. The bearing fault has the characteristic of bathtub curve, and is characterized in that only tiny vibration is generated in the early stage, the vibration is difficult to identify in the traditional monitoring, the vibration is aggravated in the middle stage, the current fluctuates, the eccentric risk of the shaft system is increased, and the bearing is blocked or broken in the later stage, so that the motor is damaged in a chained manner. The prior art is largely divided into two types of traditional signal analysis and deep learning intelligent diagnosis, and obvious bottlenecks exist: 1. Traditional signal analysis method Vibration signal analysis, namely, a fast Fourier transform FFT can not process non-stationary signals, wavelet transform relies on empirical basis selection, and the omission ratio exceeds 25%; the current signal analysis is affected by electromagnetic interference, and the accuracy rate is less than 60% when the load changes; temperature signal analysis, namely only monitoring the late fault, and response lag being greater than five minutes; multisource synergetic lack that the vibration-current-temperature correlation model is not established and information complementarity cannot be utilized. 2. Deep learning intelligent diagnosis method The small sample generalization difference is that the fault sample acquisition cost is high, and the generalization error is more than 30% when the sample size is less than 1000; the model parameters such as ResNet and the like are more than 2000 ten thousand, the reasoning time of the edge equipment is more than 200ms, and the real-time performance cannot be met; The positioning accuracy is low, namely, only the fault type is identified, and the specific position cannot be positioned; The accuracy rate of the cross-field adaptation difference permanent magnet synchronous motors of different types is reduced by more than 40 percent during migration. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide a permanent magnet synchronous motor bearing fault diagnosis monitoring system and method, which are used for solving the problems of low diagnosis precision, poor real-time performance, insufficient cross-domain suitability, incapability of accurately positioning fault positions and the like in the prior art and realizing high-efficiency identification, real-time diagnosis, accurate positioning and full life cycle maintenance of early faults of the permanent magnet synchronous motor bearing. The technical scheme adopted for solving the technical problems is as follows: The utility model provides a permanent magnet synchronous motor bearing fault diagnosis monitoring system, includes multisource information acquisition module, multisource information preprocessing module, multisource feature fusion module and Light-TRANSFERNET degree of depth migration diagnostic module, wherein: the multisource information acquisition module is used for acquiring signals of vibration, temperature, rotating speed, sound and current aiming at the permanent magnet synchronous motor bearing; The multisource data preprocessing module is used for preprocessing the acquired signals, extracting the characteristics of each signal and carrying out normalization processing on the extracted characteristics; The multi-attention feature fusion module is used for carrying out feature fusion on the normalized features to obtain a fusion feature map; the Light-TRANSFERNET depth migration diagnosis module is used for extracting deep fault classification features and spatial position features by utilizing a MobileNetV backbone network based on a fusion feature map so as to determine fault types and position fault positions, and predicting the residual service life of the bearing by utilizing an LSTM-GRU hybrid model based on tr