CN-121981703-A - Predictive maintenance method and system for electromechanical equipment

CN121981703ACN 121981703 ACN121981703 ACN 121981703ACN-121981703-A

Abstract

The application provides a predictive maintenance method and a predictive maintenance system for electromechanical equipment, which relate to the field of maintenance of the electromechanical equipment, and are characterized in that a comprehensive feature vector is determined based on the three types of features, is input into a trained target neural network model, a predicted fault type is output by the model, and a maintenance scheme is determined based on a model output result. According to the electromechanical equipment predictive maintenance method and system based on multi-mode fusion, the multi-mode data fusion is used for covering the fault characteristics of mechanical, meshing and electrical components of equipment, and the method for extracting the specific characteristics and the neural network prediction are combined, so that the fault prediction accuracy is improved, meanwhile, different operation working conditions of the equipment can be adapted, the maintenance on demand is realized, the maintenance cost is effectively reduced, the service life of the equipment is prolonged, and the operation safety of the equipment is ensured.

Inventors

CHEN SHUANGHUI
JI TIANMING
WANG JIANJUN
ZHANG YI

Assignees

南京土星信息科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260113

Claims (10)

1. The predictive maintenance method for the electromechanical equipment is characterized by being suitable for an equipment monitoring system, wherein the system comprises a sensing module, the sensing module comprises a vibration sensor, a sound sensor, a temperature sensor and a controller, the vibration sensor is arranged at a driving shaft and a supporting point of a ladder way, the sound sensor is close to the ladder way, the temperature sensor is arranged at a motor, and the method is suitable for the controller and comprises the following steps: acquiring a vibration signal, a voiceprint signal and a temperature signal based on the sensing module, and performing time synchronization on the vibration signal, the voiceprint signal and the temperature signal; Decomposing the vibration signal by adopting a wavelet packet, determining energy duty ratios of a plurality of frequency bands based on decomposition results, and determining the energy duty ratios as vibration characteristics; Acquiring a Mel frequency cepstrum coefficient based on the voiceprint signal, acquiring a first-order difference coefficient and a second-order difference coefficient based on the Mel frequency cepstrum coefficient, and taking the first-order difference coefficient and the second-order difference coefficient as sound characteristics; acquiring a hot spot core temperature value based on the temperature signal, and determining the hot spot core temperature value as a temperature characteristic; Determining a comprehensive feature vector based on the vibration feature, the sound feature and the temperature feature, and inputting the comprehensive feature vector into a trained target neural network model, wherein the target neural network model is used for outputting a predicted fault type according to the input comprehensive feature vector; and determining a maintenance scheme based on the output result of the target neural network model.
2. The method according to claim 1, wherein determining a comprehensive feature vector based on the vibration feature, the sound feature, and the temperature feature, and inputting the comprehensive feature vector into a trained target neural network model, wherein before outputting a predicted fault type and a corresponding fault distribution probability according to the input comprehensive feature vector, the method comprises: acquiring a training set and a verification set, wherein the training set and the verification set comprise a plurality of groups of training vibration characteristics, training sound characteristics, training temperature characteristics and fault type labels; inputting each group of the training vibration characteristics, the training sound characteristics, the training temperature characteristics and the fault type labels into an initial neural network model and performing multi-round iterative computation; And after each round of training is completed, evaluating the output result of the initial neural network model by adopting the verification set data, and when the loss difference value of the verification set of the continuous M rounds is smaller than a set threshold value, terminating training and outputting the trained target neural network model.
3. The method of claim 1, wherein the fault type includes a drive system fault, a ladder system fault, and a brake system fault.
4. The method of claim 1, wherein determining a composite feature vector based on the vibration feature, the sound feature, and the temperature feature, and inputting the composite feature vector into a trained target neural network model for outputting a predicted fault type based on the input composite feature vector, comprises: acquiring operation conditions, wherein the operation conditions comprise an uplink condition, a downlink condition, a start-stop condition and load data; determining the correlation weights corresponding to the vibration characteristics, the sound characteristics and the temperature characteristics based on the operation conditions; The integrated feature vector is determined based on the relevance weights.
5. The method of predictive maintenance of an electromechanical device of claim 4, wherein determining the correlation weights for the vibration signature, the sound signature, and the temperature signature based on the operating conditions comprises: respectively presetting a corresponding correlation weight matrix based on the three states of the uplink working condition, the downlink working condition and the start-stop working condition, wherein the correlation weight of the vibration characteristic under the start-stop working condition is higher than that of the uplink working condition and the downlink working condition, and the correlation weight of the sound characteristic under the uplink working condition and the downlink working condition is higher than that of the start-stop working condition; and when the load data is larger than a preset load threshold value, adjusting the correlation weight of the characteristic of the temperature based on the change amplitude of the load data, wherein the higher the load data is, the higher the correlation weight of the temperature characteristic is.
6. The method of predictive maintenance of an electromechanical device of claim 5, wherein determining the correlation weights for the vibration signature, the sound signature, and the temperature signature based on the operating conditions comprises: Obtaining kurtosis values corresponding to the vibration signals, the voiceprint signals and the temperature signals; And if any one of the kurtosis values corresponding to the vibration signal, the voiceprint signal and the temperature signal is higher than a corresponding preset value, correcting the correlation weight corresponding to the signal, wherein the higher the corresponding kurtosis value is, the larger the corrected correlation weight lifting amplitude is.
7. The method according to claim 1, wherein obtaining mel-frequency cepstral coefficients based on the voiceprint signal, obtaining first-order differential coefficients and second-order differential coefficients based on the mel-frequency cepstral coefficients, and taking the first-order differential coefficients and the second-order differential coefficients as sound features, comprises: Dividing the sound insulation texture signal into a plurality of signal frames according to a fixed time interval, wherein an overlapping area with a set proportion is reserved between adjacent signal frames; Calculating the power spectrum of each signal frame, and inputting the power spectrum into a preset Mel filter group to obtain a Mel frequency spectrum corresponding to each signal frame; Carrying out logarithmic operation on the Mel frequency spectrum, carrying out discrete cosine transform on the logarithmic operation result, and extracting the first several coefficients in the transformation result to be used as Mel frequency cepstrum coefficients of the voiceprint signal; Arranging the mel frequency cepstrum coefficients of a plurality of continuous signal frames in time sequence, and obtaining the difference value of the corresponding position coefficients of two adjacent signal frames to obtain a first-order difference coefficient of each signal frame; And obtaining the difference value of the first-order difference coefficients of the corresponding positions of the two adjacent signal frames to obtain the second-order difference coefficient of each signal frame.
8. The utility model provides an electromechanical device predictive maintenance system, includes sensing module, sensing module includes vibration sensor, sound sensor and temperature sensor and controller, vibration sensor locates drive shaft and ladder way strong point, sound sensor is close to the ladder way, temperature sensor sets up in the motor, and this system is configured to: acquiring a vibration signal, a voiceprint signal and a temperature signal based on the sensing module, and performing time synchronization on the vibration signal, the voiceprint signal and the temperature signal; Decomposing the vibration signal by adopting a wavelet packet, determining energy duty ratios of a plurality of frequency bands based on decomposition results, and determining the energy duty ratios as vibration characteristics; Acquiring a Mel frequency cepstrum coefficient based on the voiceprint signal, acquiring a first-order difference coefficient and a second-order difference coefficient based on the Mel frequency cepstrum coefficient, and taking the first-order difference coefficient and the second-order difference coefficient as sound characteristics; acquiring a hot spot core temperature value based on the temperature signal, and determining the hot spot core temperature value as a temperature characteristic; Determining a comprehensive feature vector based on the vibration feature, the sound feature and the temperature feature, and inputting the comprehensive feature vector into a trained target neural network model, wherein the target neural network model is used for outputting a predicted fault type according to the input comprehensive feature vector; and determining a maintenance scheme based on the output result of the target neural network model.
9. The predictive maintenance system for an electromechanical device of claim 8, wherein the system is configured to: Determining a comprehensive feature vector based on the vibration feature, the sound feature and the temperature feature, and inputting the comprehensive feature vector into a trained target neural network model, wherein before the target neural network model is used for outputting a predicted fault type and a corresponding fault distribution probability according to the input comprehensive feature vector, the method comprises the following steps: acquiring a training set and a verification set, wherein the training set and the verification set comprise a plurality of groups of training vibration characteristics, training sound characteristics, training temperature characteristics and fault type labels; inputting each group of the training vibration characteristics, the training sound characteristics, the training temperature characteristics and the fault type labels into an initial neural network model and performing multi-round iterative computation; And after each round of training is completed, evaluating the output result of the initial neural network model by adopting the verification set data, and when the loss difference value of the verification set of the continuous M rounds is smaller than a set threshold value, terminating training and outputting the trained target neural network model.
10. An electronic device, comprising: At least one processor; And said memory communicatively coupled to at least one said processor; Wherein the memory stores instructions executable by at least one processor to enable at least one of the processors to perform a method of predictive maintenance of an electromechanical device as set forth in any one of claims 1-7.

Description

Predictive maintenance method and system for electromechanical equipment Technical Field The application relates to the field of maintenance of electric equipment, in particular to a predictive maintenance method and system for electromechanical equipment. Background The escalator is used as key traffic equipment in a personnel-intensive place, and the running stability of the escalator directly influences public safety and traffic efficiency. The conventional escalator maintenance mostly adopts a regular maintenance mode, and the mode has two main core problems that firstly, excessive maintenance is carried out, namely, parts which are not failed are frequently disassembled and replaced, so that the maintenance cost is increased and the service life of equipment is shortened, and secondly, maintenance is omitted, namely, the hidden trouble of the failure is not found in time, and shutdown, ladder clamping and even casualties are possibly caused. The existing predictive maintenance technology is mostly dependent on single-mode monitoring, such as monitoring the state of a driving shaft only through a vibration sensor or monitoring the temperature of a motor only through a temperature sensor, but for example, the initial stage of abrasion of a ladder chain only shows fine voiceprint change, single vibration monitoring is difficult to capture, the early stage of overload failure of the motor has the existing temperature rise, the vibration frequency is abnormal, and the single temperature monitoring is easy to misjudge. Disclosure of Invention The embodiment of the application provides a predictive maintenance method and a predictive maintenance system for electromechanical equipment, which are used for solving the problems. In order to achieve the above purpose, the application adopts the following technical scheme: In a first aspect, the present application provides a predictive maintenance method for an electromechanical device, which is suitable for an electric device monitoring system, the system includes a sensing module, the sensing module includes a vibration sensor, a sound sensor, a temperature sensor and a controller, the vibration sensor is disposed at a driving shaft and a supporting point of a ladder path, the sound sensor is close to the ladder path, the temperature sensor is disposed at a motor, the method is suitable for the controller, and includes: Acquiring a vibration signal, a voiceprint signal and a temperature signal based on a sensing module, and performing time synchronization on the vibration signal, the voiceprint signal and the temperature signal; decomposing the vibration signal by adopting a wavelet packet, determining a plurality of frequency band energy duty ratios based on a decomposition result, and determining the plurality of energy duty ratios as vibration characteristics; Acquiring a Mel frequency cepstrum coefficient based on the voiceprint signal, acquiring a first-order difference coefficient and a second-order difference coefficient based on the Mel frequency cepstrum coefficient, and taking the first-order difference coefficient and the second-order difference coefficient as sound characteristics; acquiring a hot spot core temperature value based on the temperature signal, and determining the hot spot core temperature value as a temperature characteristic; Determining a comprehensive feature vector based on vibration features, sound features and temperature features, and inputting the comprehensive feature vector into a trained target neural network model, wherein the target neural network model is used for outputting a predicted fault type according to the input comprehensive feature vector; And determining a maintenance scheme based on the output result of the target neural network model. With reference to the first aspect, in some embodiments, determining a comprehensive feature vector based on the vibration feature, the sound feature, and the temperature feature, and inputting the comprehensive feature vector into a trained target neural network model, before the target neural network model is configured to output a predicted fault type and a corresponding fault distribution probability according to the input comprehensive feature vector, the method includes: acquiring a training set and a verification set, wherein the training set and the verification set comprise a plurality of groups of training vibration characteristics, training sound characteristics, training temperature characteristics and fault type labels; Inputting each group of training vibration characteristics, training sound characteristics, training temperature characteristics and fault type labels into an initial neural network model and performing multi-round iterative computation; after each round of training is completed, the output result of the initial neural network model is evaluated by adopting verification set data, and when the loss difference value of the verification sets of the conti