CN-121502172-B - Fault prediction method based on self-adaptive weighting multi-mode joint entropy

CN121502172BCN 121502172 BCN121502172 BCN 121502172BCN-121502172-B

Abstract

The invention belongs to the technical field of intelligent operation and maintenance and fault prediction of industrial equipment, relates to a fault prediction method based on self-adaptive weighting multi-mode joint entropy, and aims to accurately identify equipment states and predict faults. The method comprises the steps of preprocessing historical multi-mode sensing data of industrial equipment, and constructing a condition self-adaptive weight mapping table for representing the corresponding relation between different working condition parameters and optimal basic weights of all modes. Aiming at the multi-mode sensing data acquired in real time, a condition self-adaptive weight mapping table is queried according to real-time working condition parameters, and the dynamic fusion weight of each mode is calculated by combining with real-time signal quality evaluation. Real-time data of each mode is converted into a symbol mode. And adopting a similarity weighted fault-tolerant fusion mechanism to fuse the symbol modes of all modes to generate joint mode probability distribution, calculating a self-adaptive weighted multi-mode joint entropy value, inputting the self-adaptive weighted multi-mode joint entropy value into a pre-trained prediction model, and outputting a state identification or residual service life prediction result of industrial equipment.

Inventors

XIANG JIE
CUI YU
NIU YAN
WANG YILIAN
ZHOU MENGNI
SUN JIE

Assignees

太原理工大学

Dates

Publication Date: 20260512
Application Date: 20251201

Claims (5)

1. A fault prediction method based on self-adaptive weighted multi-mode joint entropy is characterized by comprising the steps of S1, preprocessing historical multi-mode sensing data of industrial equipment and constructing a condition self-adaptive weight mapping table for representing the corresponding relation between different working condition parameters and optimal basic weights of all modes based on the preprocessed historical data; the multi-modal sensing data is multi-channel signal data which are acquired by different types of sensors on industrial equipment and reflect the state of the equipment, wherein S2, the dynamic weight and symbolization are calculated, wherein the multi-modal sensing data acquired in real time are inquired according to real-time working condition parameters, the condition self-adaptive weight mapping table is used for calculating the dynamic fusion weight of each mode by combining with real-time signal quality evaluation, meanwhile, the real-time data of each mode are converted into a symbol mode, S3, fault-tolerant fusion and joint entropy calculation are performed, the symbol modes of each mode are fused by adopting a similarity weighted fault-tolerant fusion mechanism based on the dynamic fusion weight to generate joint mode probability distribution, and the self-adaptive weighted multi-modal joint entropy is used for quantifying the entropy characteristic index of the state complexity of the equipment, and S4, the equipment state prediction is performed by inputting the self-adaptive weighted multi-modal joint entropy into a pre-trained prediction model to output the state identification or the residual service life prediction result of the industrial equipment; The method comprises the steps of S1, constructing a condition self-adaptive weight mapping table, wherein the condition self-adaptive weight mapping table comprises the steps of quantifying discriminative indexes of each mode data, wherein the discriminative indexes are comprehensive indexes which integrate statistical characteristic differences of the mode data in health and fault states and relevance of the statistical characteristic differences and equipment state labels; Step S2, calculating dynamic fusion weights of all modes, wherein the step S comprises the steps of inquiring from the condition self-adaptive weight mapping table according to working condition parameters acquired in real time to obtain basic weights of all modes, estimating real-time instantaneous signal-to-noise ratios of all modes to serve as real-time signal quality evaluation results, and carrying out weighting adjustment and normalization processing by combining the basic weights, the real-time instantaneous signal-to-noise ratios and the discriminant indexes calculated offline to obtain final dynamic fusion weights; The fault-tolerant fusion mechanism in the step S3 comprises the steps of setting a target reference mode, wherein the target reference mode is determined based on a historical symbol mode in a healthy state of equipment, calculating the similarity between a real-time symbol mode of each mode and the target reference mode, multiplying the similarity of each mode and the corresponding dynamic fusion weight, summing the multiplied similarity to obtain the contribution of a joint mode, and reconstructing the joint mode probability distribution based on the contribution sequence.
2. The fault prediction method based on adaptive weighted multi-modal joint entropy according to claim 1, wherein the similarity is calculated by a cosine similarity function.
3. The method for predicting faults based on adaptive weighted multi-modal joint entropy according to claim 1, wherein preprocessing historical multi-modal data in step S1 comprises at least data cleaning, filtering and denoising, normalization and time sequence alignment operations, and wherein time sequence alignment uses a rotating speed signal of equipment as a unified time reference.
4. The fault prediction method based on adaptive weighted multi-mode joint entropy according to claim 1, wherein in step S2, real-time data of each mode is converted into a symbol mode, which means that a symbolization method based on data quantiles is adopted to map a continuous data sequence into a discrete symbol sequence.
5. The method for predicting faults based on adaptive weighted multi-modal joint entropy according to claim 1, wherein the prediction model in the step S4 is a gradient lifting tree model.

Description

Fault prediction method based on self-adaptive weighting multi-mode joint entropy Technical Field The invention relates to the technical field of intelligent operation and maintenance and fault prediction of industrial equipment, in particular to a fault prediction method based on self-adaptive weighting multi-mode joint entropy. Background Industrial equipment is the core foundation of modern industrial production systems, and continuous and stable operation is directly related to production safety, efficiency and economic benefit. Along with the rapid development of industrial Internet of things technology, state monitoring by deploying various sensors on key equipment has become a normal state, so that massive multi-mode data generated by the system provides a data base for predictive maintenance. However, how to extract stable and reliable state features from these heterogeneous and high-dimensional data and realize accurate fault prediction and health management is still a significant challenge in the current industrial intelligent operation and maintenance field. Currently, state monitoring methods based on information entropy are receiving a lot of attention because of their ability to effectively quantify signal complexity. Existing multi-variable entropy methods, such as multi-variable sample entropy and multi-variable permutation entropy, can process multi-channel signals, but have obvious limitations in common. The method usually defaults that the contribution degree of each mode data to the state identification is the same or fixed, and adopts a simple characteristic splicing or static weighting mode for fusion, so that the dynamic changes of different mode signal quality, discriminant and equipment working condition in an actual industrial scene cannot be fully considered. When complex conditions such as fluctuation of working conditions, degradation of sensor performance or partial modal data loss are faced, the characteristic characterization capability of the method is obviously reduced, and the accuracy and the robustness of fault early warning are insufficient. Disclosure of Invention In order to solve the problems in the background art, the invention provides a fault prediction method based on self-adaptive weighted multi-mode joint entropy, which aims to realize efficient and robust fusion of multi-mode data so as to accurately identify equipment states and predict faults. The invention provides a fault prediction method based on self-adaptive weighted multi-mode joint entropy, which comprises the following steps of S1, preprocessing historical multi-mode sensing data of industrial equipment based on data preprocessing and mapping relation construction, and constructing a condition self-adaptive weight mapping table for representing the corresponding relation between different working condition parameters and optimal basis weights of all modes based on the preprocessed historical data. Multimodal sensing data refers to multichannel signal data collected by different types of sensors on an industrial device that reflect the status of the device. S2, calculating dynamic weights and symbolizing, namely inquiring a condition self-adaptive weight mapping table according to real-time working condition parameters aiming at real-time acquired multi-mode sensing data, and calculating dynamic fusion weights of all modes by combining with real-time signal quality evaluation. Simultaneously, real-time data of each mode is converted into a symbol mode. And S3, fault-tolerant fusion and joint entropy calculation, namely fusing symbol modes of all modes by adopting a similarity weighted fault-tolerant fusion mechanism based on dynamic fusion weights to generate joint mode probability distribution, and calculating a self-adaptive weighted multi-mode joint entropy value according to the probability distribution. The adaptively weighted multi-modal joint entropy value is an entropy feature index used to quantify the complexity of the device state. S4, equipment state prediction, namely inputting the self-adaptive weighted multi-mode joint entropy value into a pre-trained prediction model, and outputting a state identification or residual service life prediction result of the industrial equipment. The step S1 is to construct a condition self-adaptive weight mapping table, which comprises the steps of quantifying the discriminant index of each mode data, wherein the discriminant index is a comprehensive index which fuses the statistical characteristic difference of the mode data under the health and fault states and the relevance of the statistical characteristic difference and the equipment state label. Aiming at different working condition parameter intervals, aiming at maximizing the mode distribution difference of equipment in different states, solving the optimal basic weight of each mode in different working conditions, and further constructing a condition self-adaptive weight mapping table. And step