CN-122020266-A - Early fault diagnosis method and device for equipment health management and electronic equipment

CN122020266ACN 122020266 ACN122020266 ACN 122020266ACN-122020266-A

Abstract

The application provides an early fault diagnosis method, device and electronic equipment for equipment health management, which comprise the steps of acquiring equipment operation monitoring data acquired in real time, inputting the equipment operation monitoring data into a pre-trained equipment early fault diagnosis model, wherein the equipment early fault diagnosis model comprises a multi-source feature fusion layer with modulated operation parameters, a multi-head self-adaptive gating attention layer and a multi-scale residual error classification layer, determining fusion feature vectors through the multi-source feature fusion layer, outputting attention weighted feature vectors through the multi-head self-adaptive gating attention layer, weighting the attention weighted feature vectors through the multi-scale residual error classification layer, extracting multi-scale depth features, weighting and polymerizing, outputting fault class probability vectors, and determining early fault diagnosis results of equipment health states according to the fault class probability vectors. The method and the device can improve the early failure prediction accuracy of the equipment.

Inventors

ZHAO CHUANCHAO
Zhu Yunheng
ZHAO CHUANFENG
GUO QIHAO
HAN XIAOLIN
Han Zilei

Assignees

山东能源数智云科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. An early fault diagnosis method for device health management, the method comprising: Acquiring real-time acquired equipment operation monitoring data, wherein the equipment operation monitoring data comprises an original time domain vibration signal corresponding to equipment vibration data and an equipment operation parameter vector; the equipment operation monitoring data is input into a pre-trained equipment early fault diagnosis model, wherein the equipment early fault diagnosis model comprises a multi-source feature fusion layer for operating parameter modulation, a multi-head self-adaptive gating attention layer and a multi-scale residual error classification layer; Performing multiband self-adaptive spectrum reconstruction, joint recursion matrix construction and feature fusion processing on the equipment operation monitoring data through the multisource feature fusion layer to determine fusion feature vectors; dynamically adjusting the activation intensity of each feature dimension in the fusion feature vector according to the equipment operation parameter vector through the multi-head self-adaptive gating attention layer so as to focus on a feature subspace most relevant to faults and output an attention weighted feature vector; And extracting multi-scale depth features from the attention weighted feature vectors through the multi-scale residual error classification layer, weighting and aggregating, outputting fault class probability vectors, and determining early fault diagnosis results of the health state of the equipment according to the fault class probability vectors.
2. The method of claim 1, wherein the step of determining a fused feature vector by the multi-source feature fusion layer performing multi-band adaptive spectral reconstruction, joint recursive matrix construction, and feature fusion processing on the device operational monitoring data comprises: The following steps are executed through the multi-source feature fusion layer: performing multiband self-adaptive spectrum reconstruction on an original time domain vibration signal corresponding to the equipment vibration data to determine a reconstructed spectrum vector; Combining the reconstructed spectrum vector with envelope information of the original time domain vibration signal by adopting a combined recursion matrix construction method to generate a time-frequency combined feature vector capable of simultaneously representing time domain dynamic and frequency domain energy distribution modes; and fusing the frequency spectrum reconstruction vector, the time-frequency joint feature vector and the equipment operation parameter vector to determine a fused feature vector.
3. The method of claim 2, wherein the step of performing multi-band adaptive spectral reconstruction of the original time domain vibration signal corresponding to the device vibration data, and determining a reconstructed spectral vector comprises: performing fast Fourier transform on an original time domain vibration signal corresponding to the equipment vibration data, and taking the amplitude of a single-side frequency spectrum to obtain a corresponding frequency spectrum amplitude vector; according to the spectrum amplitude vector, calculating accumulated energy and gradient of spectrum amplitude, and adaptively determining boundary points dividing frequency bands by locating local maximum points of the accumulated energy gradient so as to separate scattered and weak fault features into independent sub-bands; For each divided sub-band, adaptively calculating an enhancement coefficient according to the relative relation between the average energy of the sub-band and the global average energy, and performing nonlinear enhancement and normalization processing by using a hyperbolic tangent function to obtain the amplitude of a reconstructed spectrum corresponding to each sub-band at each frequency index; And splicing the plurality of reconstructed spectrums according to the original frequency sequence based on the amplitudes of the reconstructed spectrums at the frequency indexes, which are respectively corresponding to all the sub-bands, so as to form a complete and characteristic-enhanced reconstructed spectrum vector.
4. The method of claim 2, wherein the step of combining the reconstructed spectral vector with the envelope information of the original time domain vibration signal to generate a time-frequency joint feature vector capable of characterizing both time domain dynamics and frequency domain energy distribution modes using a joint recursive matrix construction method comprises: extracting an envelope signal from the original time domain vibration signal; Dividing the envelope signal and the reconstructed spectrum vector into a plurality of non-overlapping segments respectively, and calculating the root mean square value and the spectrum center of gravity of each segment to obtain an envelope root mean square sequence and a spectrum center of gravity sequence; constructing a joint recursion matrix based on the root mean square sequence and the spectrum barycenter sequence; and reshaping the joint recursion matrix into a one-dimensional vector serving as a time-frequency joint feature vector.
5. The method of claim 1, wherein the training process of the device early-fault diagnosis model is as follows: Acquiring a training set of equipment state monitoring data; training a multisource feature fusion layer, a multi-head self-adaptive gating attention layer and a multiscale residual error classification layer modulated by operating parameters through the equipment state monitoring data training set to obtain an equipment early fault diagnosis model; In the training process, a composite loss function combining the adaptive focus cross entropy loss and the operation parameter perception characteristic consistency loss is adopted to calculate the loss value.
6. The method of claim 5, wherein the step of obtaining a training set of device state monitoring data comprises: the method comprises the steps of acquiring historical equipment operation monitoring data, wherein the historical equipment operation monitoring data comprises continuously acquiring vibration signal samples of equipment in different operation stages and different health states at a fixed sampling frequency, and synchronously acquiring equipment operation parameter vectors corresponding to the vibration signal samples; Labeling health status labels on each section of vibration signal sample and corresponding equipment operation parameter vector in the equipment operation monitoring data; And integrating the marked vibration signal sequence, the corresponding equipment operation parameter vector and the fault type label thereof to construct a structured equipment state monitoring data training set.
7. The method of claim 5, wherein the composite loss function is determined in the following manner: Determining a sample adaptive focus parameter based on the equipment operation parameter vector, and determining adaptive focus cross entropy loss according to the sample adaptive focus parameter and the prediction probability of the sample to the real class; Determining an operational parameter perceived feature consistency loss based on the device operational parameter vector, the bandwidth parameter of the operational parameter similarity, and the indication function based on the real fault class label; and adding the self-adaptive focus cross entropy loss and the operation parameter perception characteristic consistency loss according to weight to obtain a model training composite loss function.
8. An early failure diagnosis apparatus for device health management, the apparatus comprising: the device operation monitoring system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring device operation monitoring data acquired in real time, and the device operation monitoring data comprises an original time domain vibration signal corresponding to the device vibration data and a device operation parameter vector; the model input module is used for inputting the equipment operation monitoring data into a pre-trained equipment early fault diagnosis model, wherein the equipment early fault diagnosis model comprises an operation parameter modulated multi-source feature fusion layer, a multi-head self-adaptive gating attention layer and a multi-scale residual error classification layer; The model prediction module is used for carrying out multiband self-adaptive spectrum reconstruction, joint recursion matrix construction and feature fusion processing on the equipment operation monitoring data through the multisource feature fusion layer to determine fusion feature vectors, dynamically adjusting the activation intensity of each feature dimension in the fusion feature vectors according to the multi-head self-adaptive gating attention layer through the equipment operation parameter vectors so as to focus on the feature subspace most relevant to faults and output attention weighted feature vectors, extracting multi-scale depth features and carrying out weighted aggregation on the attention weighted feature vectors through the multi-scale residual error classification layer, outputting fault class probability vectors, and determining early fault diagnosis results of the health states of the equipment according to the fault class probability vectors.
9. An electronic device comprising a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method of any one of claims 1 to 7.
10. A computer readable storage medium storing computer executable instructions which, when invoked and executed by a processor, cause the processor to implement the method of any one of claims 1 to 7.

Description

Early fault diagnosis method and device for equipment health management and electronic equipment Technical Field The application relates to the technical field of artificial intelligence, in particular to an early fault diagnosis method and device for equipment health management and electronic equipment. Background With the development of industrial equipment to large-scale, high-speed and continuous, key rotating equipment runs for a long time under complex working conditions, and the health state of the key rotating equipment is directly related to the safety, reliability and running economy of a production system. Before serious faults occur, the equipment often goes through a long early degradation stage, and vibration response characteristic energy generated in the stage is weak and is easily influenced by noise and working condition fluctuation, but if accurate identification can be realized in the stage, a sufficient time window is provided for predictive maintenance and state overhaul, so that sudden shutdown and major accidents are effectively avoided. Therefore, the early fault diagnosis method oriented to the equipment health management has become a research focus and an urgent engineering demand in the field of industrial intelligent operation and maintenance. Disclosure of Invention The application aims to provide an early fault diagnosis method and device for equipment health management and electronic equipment, which can improve the early fault prediction accuracy of the equipment. The early fault diagnosis method for the equipment health management comprises the steps of obtaining equipment operation monitoring data collected in real time, wherein the equipment operation monitoring data comprise original time domain vibration signals corresponding to the equipment vibration data and equipment operation parameter vectors, inputting the equipment operation monitoring data into a pre-trained equipment early fault diagnosis model, the equipment early fault diagnosis model comprises a multi-source feature fusion layer, a multi-head self-adaptive gating attention layer and a multi-scale residual error classification layer which are modulated by operation parameters, carrying out multi-band self-adaptive spectrum reconstruction, joint recursion matrix construction and feature fusion processing on the equipment operation monitoring data to determine fusion feature vectors, dynamically adjusting the activation intensity of each feature dimension in the fusion feature vectors according to the equipment operation parameter vectors through the multi-head self-adaptive gating attention layer to focus on a feature subspace most relevant to faults, outputting attention weighted feature vectors, extracting multi-scale depth features and weighting and aggregating the attention weighted feature vectors through the multi-scale residual error classification layer, outputting fault category probability vectors, and determining an early fault diagnosis result of the equipment health state according to the fault category probability vectors. The method comprises the steps of carrying out multi-band self-adaptive spectrum reconstruction on an original time domain vibration signal corresponding to equipment vibration data through the multi-source feature fusion layer to determine a reconstructed spectrum vector, combining the reconstructed spectrum vector with envelope information of the original time domain vibration signal by adopting a combined recursion matrix construction method to generate a time-frequency combined feature vector capable of simultaneously representing a time domain dynamic energy distribution mode and a frequency domain energy distribution mode, and fusing the spectrum reconstruction vector, the time-frequency combined feature vector and an equipment operation parameter vector to determine a fused feature vector. The method comprises the steps of carrying out multi-band self-adaptive spectrum reconstruction on original time domain vibration signals corresponding to equipment vibration data, determining a reconstructed spectrum vector, carrying out fast Fourier transform on the original time domain vibration signals corresponding to the equipment vibration data, taking the amplitude of a single-side frequency spectrum to obtain a corresponding spectrum amplitude vector, calculating accumulated energy and gradient of the spectrum amplitude according to the spectrum amplitude vector, adaptively determining boundary points of dividing frequency bands by locating local maximum points of the accumulated energy gradient to separate scattered and weak fault features into independent sub-bands, adaptively calculating enhancement coefficients according to the relative relation between average energy of the sub-bands and global average energy for each sub-band, carrying out nonlinear enhancement and normalization processing by using hyperbolic tangent functions to obtain the amplitude of the reconstructe