CN-122020394-A - Sensitivity cross-modal sensing-based rotating pair assembly double-domain migration health diagnosis method

CN122020394ACN 122020394 ACN122020394 ACN 122020394ACN-122020394-A

Abstract

The invention discloses a sensitivity cross-modal sensing-based rotating pair assembly body double-domain migration health diagnosis method. Firstly, extracting time domain, frequency domain, variation modal decomposition and other characteristics from vibration signals and voiceprint signals of a revolute pair assembly to construct a cross-modal characteristic set. The accuracy and the robustness of the diagnosis model are further optimized by fusing the acoustic signals with the mechanical state signals by considering the modal drift and the sensitivity index of the voiceprint signals. And then, carrying out deep feature learning by adopting a space-time coupled health characterization network, and combining a sensitivity cross-modal sensing mechanism to jointly optimize the hyper-parameters of the health diagnosis model through an alpha evolution algorithm. And finally, performing cross-modal sensing on the working state data of the rotating pair assembly part to be tested through a health diagnosis model to obtain a fault diagnosis result. The invention can improve the diagnosis accuracy and robustness under the conditions of small samples and variable working conditions, and further enhance the diagnosis capability of the model after the voiceprint signals are fused.

Inventors

XU JINGHUA
JIANG QIANWEN
YE DELIANG
ZHANG SHUYOU

Assignees

浙江大学

Dates

Publication Date: 20260512
Application Date: 20260209

Claims (10)

1. A sensitivity cross-modal sensing-based rotating pair assembly dual-domain migration health diagnosis method is characterized by comprising the following steps: The method comprises the steps of acquiring working state data of a rotating pair assembly part in an original working condition and a variable working condition domain, recording the working state data as first working condition state data and second working condition state data, preprocessing the first working condition state data and the second working condition state data respectively, and acquiring an original domain data set and a variable domain data set; The method comprises the steps of constructing a rotating auxiliary assembly body health diagnosis model, training the rotating auxiliary assembly body health diagnosis model according to a training set and a verification set, and obtaining a trained rotating auxiliary assembly body health diagnosis model, wherein the rotating auxiliary assembly body health diagnosis model comprises a multi-scale space representation extraction sub-module, a time context aggregation sub-module and a classification module which are sequentially connected; and thirdly, collecting working state data of the rotating pair assembly part to be tested, preprocessing the data, inputting the data into a trained rotating pair assembly health diagnosis model, and outputting a health diagnosis result of the rotating pair assembly part to be tested by the model.
2. The rotating pair assembly body double-domain migration health diagnosis method based on sensitivity cross-modal sensing according to claim 1, wherein the multi-scale space characterization extraction submodule comprises a first convolution layer, a first normalization layer, a first activation layer, a second convolution layer, a second normalization layer, a second activation layer and a global maximum pooling layer which are connected in sequence.
3. The method for diagnosing double-domain migration health of a revolute pair assembly based on sensitivity cross-modal sensing according to claim 1, wherein in the second step, the key super parameters of the temporal context aggregation sub-module comprise the number of hidden units of the first layer and the number of hidden units of the second layer of the temporal context aggregation sub-module, a random inactivation rate and a learning rate.
4. The method for diagnosing double-domain migration health of a revolute pair assembly body based on sensitivity cross-modal sensing according to claim 1, wherein in the second step, the alpha evolution algorithm based on sensitivity cross-modal sensing is obtained by combining a sensitivity cross-modal sensing mechanism and improving a fitness function of the alpha evolution algorithm, and the improved fitness function meets the following formula: Wherein, the And Respectively representing the current first fitness and the current second fitness; And Representing a new first fitness and a new second fitness, respectively; And Respectively representing the current key super-parameters and the new key super-parameters; And Respectively representing the health diagnosis error rate of the sample corresponding to the original working condition in the verification set and the health diagnosis error rate of the sample corresponding to the variable working condition; Representing the dynamic adjustment coefficient; representing the number of iterations currently performed; Representing a maximum number of iterations; And Respectively representing the first cost sensitive weight and the second cost sensitive weight after current updating; And Representing first and second initial experience weights, respectively; And Respectively representing normalized first and second cost sensitive weights; And Representing the current cost sensitivity and the new cost sensitivity, respectively.
5. The method for diagnosing double-domain migration health of a revolute pair assembly based on sensitivity cross-modal sensing according to claim 1, wherein in the first step, the preprocessing comprises denoising, data segmentation and feature calculation.
6. The sensitivity cross-modal sensing-based rotating pair assembly two-domain migration health diagnosis method according to claim 5, wherein the features in the feature calculation comprise time domain features, frequency domain features, spectral kurtosis features and variation modal decomposition features.
7. A sensitivity cross-modal sensing-based rotating pair assembly dual-domain migration health diagnostic system, comprising: the data acquisition unit is used for acquiring working state data of the rotating pair assembly body component; the data preprocessing unit is used for preprocessing the working state data; The data set construction unit is used for constructing a training set and a testing set based on the data sets of the original working condition and the variable working condition domain; The system comprises a training parameter optimization unit, a sensitivity cross-modal sensing based alpha evolution algorithm, a time context aggregation sub-module, a test set and a time context aggregation sub-module, wherein the training parameter optimization unit is used for training the rotating auxiliary assembly health diagnosis model based on the training set and the test set; And the diagnosis output unit is used for processing the pretreatment data corresponding to the working state data of the rotating pair assembly part to be detected by utilizing the rotating pair assembly health diagnosis model to obtain the health diagnosis result of the rotating pair assembly part to be detected.
8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of a sensitivity cross-modal awareness based rotating sub-assembly two-domain migration health diagnostic method according to any one of claims 1 to 7.
9. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor implements the steps of a sensitivity cross-modal awareness based rotating sub-assembly two-domain migration health diagnostic method according to any one of claims 1 to 7.
10. A computer program product comprising computer program/instructions which, when executed by a processor, implement the steps of a sensitivity cross-modal awareness based rotating sub-assembly two-domain migration health diagnostic method according to any one of claims 1 to 7.

Description

Sensitivity cross-modal sensing-based rotating pair assembly double-domain migration health diagnosis method Technical Field The invention belongs to the technical field of mechanical health diagnosis, and particularly relates to a sensitivity-cross-modal-sensing-based double-domain migration health diagnosis method for a revolute pair assembly. Background The scale and complexity of the mechanical equipment in the manufacturing industry are increasing, and large, high-speed and high-precision mechanical equipment is increasingly required, such as a large-scale high-speed beam gantry machining center, a heavy AC pendulum five-axis gantry machining center, a five-axis high-speed gantry milling center, a hydrogen compressor, a metal pipe production line and the like, and the traditional intelligent health diagnosis (INTELLIGENT FAULT DIAGNOSIS, IFD) based on a shallow machine learning method is difficult to meet the requirements of equipment health diagnosis. The new generation artificial intelligence technology takes self-induction, self-adaption, self-learning and self-decision as remarkable characteristics, takes knowledge engineering as a core, and develops into IFD to bring new opportunities. Fault prediction and health management (Prognostics AND HEALTH MANAGEMENT, PHM) is a brand new solution for managing health status, which is proposed by comprehensively utilizing the latest research results of modern information technology and artificial intelligence technology, and the IFD is a part of the solution. The PHM can predict the time and the position of the fault in advance, predict the residual service life of the whole system, improve the operation reliability and the safety of the system, reduce the maintenance cost of the system and improve the maintenance accuracy, and realize the state-based maintenance decision capability (Condition-based Maintenance Decision, CMD) of the system. PHM can be divided into eight steps, primary analysis, data acquisition, detection, feature engineering, diagnosis, health assessment, prediction, maintenance and management. The rotating pair assembly body can generate multimode signals of vibration, heat, electromagnetism and the like in the operation process, the signals contain rich equipment state information, the voiceprint signals are used for refining and extracting information of operation characteristic frequency, amplitude, modulation mode and the like, and the intelligent diagnosis of mechanical faults can be realized by accurately monitoring and spectrum analysis of the multimode signals. The sensor of the vibration signal diagnosis method can be not in direct contact with the machine, and the method is flexible, convenient and easy to realize uninterrupted power detection. Representative of the revolute pair assemblies are rolling bearings. In the long-term running process, the rolling bearing is subjected to alternating load, contact fatigue and environmental corrosion continuously, so that the internal rolling body and the raceway interface are easy to undergo progressive performance degradation, and the rolling bearing is in failure modes such as surface abrasion, material aging and fatigue peeling. Bearings in industrial sites are often in complex working conditions of multi-source interference coupling, including strong background vibration, electromagnetic noise and non-stationary excitation, which significantly reduce the signal-to-noise ratio of vibration signals, resulting in early weak fault features being submerged in noise, and traditional feature extraction methods face serious challenges. More complex, the bearing vibration signal under the variable working condition presents obvious non-stationary characteristic, and the statistical distribution of the bearing vibration signal dynamically changes along with parameters such as load, rotating speed and the like. This condition dependence causes a dramatic drop in diagnostic model generalization performance based on static distribution assumptions. In addition, the scarcity of high-quality fault samples in actual engineering further restricts the effectiveness of a data driving method, and particularly under the condition of small samples, the performance of a deep learning model is often difficult to guarantee. Therefore, developing an intelligent diagnosis algorithm with strong noise immunity and cross-working condition adaptability has become a core scientific problem to be solved urgently in the current health diagnosis field. U.S. Pat. No. 5,220,410, (Goodman M A, bishop W, mohr G. System for bearing fault detection: U.S. Pat. No. 5,62,72,p., 2015-12-01.) proposes a bearing fault detection method based on ultrasonic and spectral analysis, wherein bearing high frequency signals are collected by an ultrasonic sensor, demodulated and converted to an expected frequency band, and then a Fast Fourier Transform (FFT) is utilized to generate a frequency spectrum, which is dynamically compared with f