Search

CN-121977639-A - Method, equipment, storage medium and program product for monitoring running state of explosion-proof motor

CN121977639ACN 121977639 ACN121977639 ACN 121977639ACN-121977639-A

Abstract

The invention discloses a method, a system, equipment and a medium for monitoring the running state of an explosion-proof motor, and aims to solve the problem that the prior art cannot effectively strip environmental thermal interference, so that early weak fault signals are submerged. The method comprises the steps of predicting an environmental thermal component in the surface temperature of a motor by utilizing a time convolution network based on historical environmental temperature and motor load data, subtracting the component from the real-time surface temperature to obtain a clean thermodynamic diagram, generating a spatial attention diagram based on the clean thermodynamic diagram to guide and strengthen feature expression of a vibration spectrogram, and finally inputting the fused multi-mode features into a multi-task learning network to achieve accurate classification and positioning of faults. The invention can decouple nonlinear environmental thermal interference and improve the monitoring sensitivity and the diagnosis accuracy of early weak faults.

Inventors

  • CHEN SHANDUO

Assignees

  • 江苏利多电机有限公司

Dates

Publication Date
20260505
Application Date
20260114

Claims (10)

  1. 1. An explosion-proof motor operation state monitoring method, which is characterized by being applied to an explosion-proof motor system, comprising: Synchronously acquiring multi-mode signals output by a motor sensor group arranged on an explosion-proof motor and environment state time sequence data output by an environment sensor arranged in the operation environment of the explosion-proof motor, wherein the multi-mode signals at least comprise a high-resolution surface temperature field diagram acquired by a thermal imaging sensor array on an explosion-proof shell and vibration signals acquired by a vibration sensor, the environment state time sequence data comprise environment parameters of a plurality of sampling moments in a past preset time window, and the environment parameters at least comprise environment temperature, air humidity and air flow rate; Inputting the environmental state time sequence data into a pre-trained space-time heat effect transfer model, and recursively calculating heat conduction hysteresis and nonlinear response of the flameproof housing due to heat capacity and heat resistance characteristics based on the change of the input environmental parameters on the time sequence to obtain a dynamic environment induction heat distribution map; Performing matrix subtraction on the high-resolution surface temperature field map, subtracting the dynamic environment induction thermal distribution map, decoupling environment thermal interference, and calculating to obtain a pure internal source thermal map; performing time-frequency analysis on the vibration signal to obtain a vibration time-frequency diagram; Performing feature fusion on the pure internal source heat map and the vibration time-frequency map to obtain enhanced fusion feature representation, wherein the pure internal source heat map provides space positioning and confidence weighting basis for fault feature frequency components in the vibration time-frequency map; And inputting the enhanced fusion characteristic representation to a preset fault diagnosis model, and determining the specific fault type and position of the explosion-proof motor.
  2. 2. The method of claim 1, wherein the spatiotemporal thermal effect transfer model comprises a cyclic neural network encoder and a convolutional decoder, the step of inputting the environmental state time series data into a pre-trained spatiotemporal thermal effect transfer model, and recursively calculating thermal conduction hysteresis and nonlinear response of the flameproof housing due to thermal capacitance and thermal resistance characteristics based on the change of the input environmental parameters in time series, and obtaining a dynamic environmental induction thermal distribution map, specifically comprising: Inputting the environmental state time sequence data to the cyclic neural network encoder and recursively updating the hidden state of the cyclic neural network encoder, and modeling the thermal conduction hysteresis effect of the flameproof housing to obtain a context feature vector for encoding environmental history information And inputting the context feature vector to the convolution decoder, performing up-sampling and deconvolution operations, and recovering the context feature vector to the same spatial dimension as the high-resolution surface temperature field map to obtain the dynamic environment induction thermal distribution map.
  3. 3. The method according to claim 2, wherein the spatiotemporal thermal effect transfer model is obtained by the offline training step of: Under various changed nonstandard environmental conditions, running a reference explosion-proof motor with a preset health state, and synchronously collecting environmental state time sequence data serving as training input and a high-resolution surface temperature field chart serving as a training label; And performing end-to-end supervised training on the space-time thermal effect transfer model by adopting the training input and the training label and taking a pixel-level mean square error between a dynamic environment induction thermal distribution diagram predicted by a minimized model and a high-resolution surface temperature field diagram acquired in practice as an optimization target.
  4. 4. The method according to claim 1, wherein the feature fusion of the clean internal source heat map and the vibration time-frequency map to obtain an enhanced fusion feature representation specifically comprises: applying a spatial attention module to the pure internal source heat map, and generating a thermal anomaly attention weight in a spatial region corresponding to a preset key component of the explosion-proof motor; Multiplying the thermal anomaly attention weight by the vibration time-frequency plot element by element to amplify energy of fault-signature frequency components spatially correlated to a thermal anomaly region and suppress extraneous frequency components; stacking the weighted vibration time-frequency diagram with the original pure internal source heat diagram along the channel dimension to obtain the enhanced fusion characteristic representation.
  5. 5. The method of claim 4, wherein the step of determining the position of the first electrode is performed, The fault diagnosis model is a multi-task learning network and comprises a shared feature extraction trunk and two parallel output branches; The feature extraction backbone is responsible for extracting depth fusion features from the enhanced fusion feature representation; The two parallel output branches are respectively a fault classification branch and a hot spot positioning branch, and the hot spot positioning branch outputs the central coordinate and the boundary frame of the thermal anomaly region in the pure internal source heat map in a regression mode to serve as the physical position of the fault.
  6. 6. The method of claim 5, wherein the multi-modal signal further comprises: winding temperature data output by a winding temperature sensor arranged in the motor and three-phase power supply current signals acquired by a current transformer; And, prior to inputting the enhanced fusion feature representation into the fault diagnosis model, the method further comprises: and extracting auxiliary features from the winding temperature data and the three-phase power supply current signals, and splicing the auxiliary features with depth features contained in the enhanced fusion feature representation to serve as final input of the fault diagnosis model.
  7. 7. The method according to claim 1, wherein the time-frequency analysis is performed on the vibration signal to obtain a vibration time-frequency diagram, specifically: and performing continuous wavelet transformation on the vibration signal to generate a two-dimensional time-frequency diagram capable of simultaneously characterizing local characteristics of the signal in time and frequency domains as the vibration time-frequency diagram.
  8. 8. An explosion proof electrical machine comprising one or more processors and memory coupled to the one or more processors, the memory for storing computer program code comprising computer instructions that the one or more processors invoke to cause the explosion proof electrical machine to perform the method of any of claims 1-7.
  9. 9. A computer readable storage medium comprising instructions which, when run on an explosion proof machine, cause the explosion proof machine to perform the method of any one of claims 1-7.
  10. 10. A computer program product, characterized in that the computer program product, when run on an explosion-proof machine device, causes the explosion-proof machine device to perform the method of any one of claims 1-7.

Description

Method, equipment, storage medium and program product for monitoring running state of explosion-proof motor Technical Field The application relates to the technical field of fault diagnosis of an explosion-proof motor, in particular to a method, equipment, a storage medium and a program product for monitoring the running state of the explosion-proof motor. Background The explosion-proof motor is a core power device in high-risk explosive environments such as coal mines, petrochemical industry and the like, and the running stability and safety of the explosion-proof motor are directly related to the safety of the whole production system. In order to prevent potential risks, an automatic technology capable of monitoring the health state of the motor in real time and early warning faults is developed, and the method has important value for guaranteeing personnel safety and production continuity. Some related art use multisource sensors for monitoring, particularly temperature sensors. The basic logic is that if early faults (such as bearing abrasion and winding turn-to-turn short circuit) occur in the internal components of the motor, abnormal heating phenomena are often accompanied. Therefore, by arranging the temperature sensor on the surface of the motor, the abnormal temperature rise which is transmitted from inside to outside can be captured. To eliminate the interference of the environmental temperature change on the monitoring result, part of the schemes collect the real-time temperature of the surrounding environment at the same time, and then subtract the environmental temperature value or a compensation quantity related to the environmental temperature value from the temperature reading of the surface of the motor, so as to try to obtain a temperature which only reflects the internal heating of the motor. However, to meet stringent safety standards, the flameproof housing of an explosion-proof motor typically has a thickness and mass far exceeding those of a conventional motor. This results in the housing itself forming a large heat volume, which exhibits a significant hysteresis in the heat transfer, and the temperature changes of the external environment are not instantaneously transferred linearly to the motor surface, but rather undergo a complex damping and delay process. Therefore, this processing mode cannot actually peel off the environmental heat influence correctly, resulting in that the temperature for diagnosis is actually contaminated with uncompensated environmental noise, and it is difficult to accurately distinguish the thermal abnormality signal caused by the internal early weak failure, which really needs attention, which ultimately leads to a reduction in the reliability of the result. Disclosure of Invention The application provides an explosion-proof motor running state monitoring method, equipment, a storage medium and a program product, which are used for improving the reliability of motor running state monitoring. The application provides an explosion-proof motor running state monitoring method, which is applied to an explosion-proof motor system and comprises the steps of synchronously collecting multi-mode signals output by a motor sensor group arranged on the explosion-proof motor and environment state time sequence data output by an environment sensor arranged in the running environment of the explosion-proof motor, wherein the multi-mode signals at least comprise a high-resolution surface temperature field diagram collected by a thermal imaging sensor array on an explosion-proof shell and vibration signals collected by a vibration sensor, and the environment state time sequence data comprise environment parameters of a plurality of sampling moments in a past preset time window, and the environment parameters at least comprise environment temperature, air humidity and air flow rate; the method comprises the steps of inputting environmental state time sequence data into a pre-trained space-time thermal effect transfer model, recursively calculating heat conduction hysteresis and nonlinear response of an explosion-proof shell due to heat capacity and thermal resistance characteristics based on the change of the input environmental parameters on a time sequence to obtain a dynamic environment induction thermal distribution map, performing matrix subtraction on a high-resolution surface temperature field map, subtracting the dynamic environment induction thermal distribution map, decoupling the environment thermal interference, calculating to obtain a pure internal source thermal map, performing time-frequency analysis on vibration signals to obtain a vibration time-frequency map, performing feature fusion on the pure internal source thermal map and the vibration time-frequency map to obtain enhanced fusion feature representation, providing space positioning and confidence weighting basis for fault feature frequency components in the vibration time-frequency map, and inputting