Search

CN-121981365-A - Industrial equipment full life cycle fault prediction and health management method and system

CN121981365ACN 121981365 ACN121981365 ACN 121981365ACN-121981365-A

Abstract

The invention discloses a method and a system for predicting full life cycle faults and managing health of industrial equipment. And (3) building a full-information dimension digital mapping model of the equipment, and completing real-time synchronization of the physical equipment and the digital model. And comprehensively acquiring the running process, state and quality data of the equipment, and generating high-quality characteristic information for health evaluation through storage, abnormal rejection, standardization, noise reduction and characteristic extraction. And (3) building a health assessment model by combining a physical mechanism and a data driving method, and inputting characteristic information and working condition parameters to calculate a health index. And pre-judging the residual service life by adopting a time sequence prediction model, setting a multi-stage early warning threshold value, triggering early warning and pushing early warning information. Matching an optimal operation and maintenance scheme, recording maintenance data, updating a model and a database, constructing a cross-scene fault knowledge system, and carrying out iterative optimization evaluation on a prediction model. The reliability of the equipment is improved, and the operation and maintenance cost is reduced.

Inventors

  • He ningbo
  • LI XUE
  • ZHANG YAN

Assignees

  • 大唐互联科技(武汉)有限公司

Dates

Publication Date
20260505
Application Date
20251211

Claims (10)

  1. 1. A method for predicting and managing full life cycle faults of industrial equipment, comprising: S1, constructing an end-to-end unified network architecture, distributing unique identification addresses for industrial equipment and matched sensors, and completing direct interconnection of the industrial equipment, the sensors and a data processing platform or a cloud platform; S2, based on a unified communication protocol, constructing a digital mapping model covering all information dimensions of the equipment, completing real-time synchronization of the physical equipment and the digital mapping model, and providing digital support for equipment state evaluation; S3, comprehensively acquiring process data, state data and quality data in the running process of industrial equipment, storing the acquired multi-source data by adopting a data storage scheme adapting to the industrial scene, and converting the original data into high-quality characteristic information which can be used for equipment health assessment through abnormal value rejection, data standardization, signal noise reduction and characteristic extraction; S4, a health assessment model is built by combining a physical mechanism and a data driving method, the processed characteristic information and equipment working condition parameters are input into the health assessment model, a quantized equipment health index is obtained through calculation, and the whole health level and the degradation state of a core component of the equipment are reflected through the equipment health index; S5, prejudging the residual service life of the equipment by adopting a time sequence prediction model, setting a multi-level early warning threshold value based on the equipment health index, the fault risk level and the maintenance cost, triggering early warning of the corresponding level when the equipment health index reaches the corresponding early warning threshold value, and pushing early warning information comprising fault positions, influence ranges and processing suggestions to operation and maintenance personnel through multiple terminals; S6, matching an optimal operation and maintenance scheme and an execution main body based on equipment, operation and maintenance and fault related data, supporting fault processing and maintenance operation by means of a technical tool, recording maintenance process data, verifying maintenance effects, and synchronously updating a digital model of the equipment and a related database; S7, constructing a fault knowledge system shared by the cross-scene, storing fault cases and treatment schemes, supporting intelligent retrieval recommendation, periodically fusing newly-added data iteration optimization evaluation, a prediction model and early warning judgment standards, and dynamically adapting to equipment working condition changes.
  2. 2. The method for predicting and managing the whole life cycle failure of industrial equipment according to claim 1, wherein the digitalized mapping model in S2 integrates three types of core information including static attribute, dynamic operation data and history maintenance record of the equipment, establishes a hierarchical association relationship between the equipment and parts and between the parts and sensors, realizes real-time synchronization of physical equipment and the digitalized mapping model, and provides digitalized support for equipment state evaluation.
  3. 3. The method for predicting and managing the complete life cycle failure of industrial equipment according to claim 1, wherein the digitalized mapping model in S2 is constructed by a multi-layer modeling method, and the multi-layer modeling method comprises three-dimensional solid modeling, operation behavior simulation modeling and business rule modeling, so as to respectively implement mapping of physical forms of the equipment, operation dynamic behavior simulation modeling and clear definition of business rules including failure processing and maintenance execution.
  4. 4. The method for predicting and managing full life cycle faults of industrial equipment according to claim 1, wherein the process of converting raw data into high-quality characteristic information usable for equipment health assessment in the step S3 is as follows: firstly, carrying out outlier rejection on original data, and identifying and rejecting abnormal data deviating from a normal distribution range by adopting a statistical criterion; then, the data after eliminating the abnormal value is standardized, the processed data sequence is set as X ′ ＝{x ′ 1 ,x ′ 2 ,...,x ′ m }, wherein X i ′ is the ith data point after eliminating the abnormal value, and m is the total data amount after eliminating the abnormal value, and the data is processed by a standardized formula Conversion to unified dimension data, wherein To eliminate the mean value of the data sequence after outliers, Ensuring the comparability of the data in order to remove the standard deviation of the data sequence after the abnormal value; then, carrying out noise reduction processing on different types of signals, and filtering high-frequency noise and interference components in the data through a signal processing algorithm to obtain a stable and reliable effective signal x i '' ', wherein x i ' '' is the ith data point after noise reduction; Finally, multi-dimensional feature extraction is carried out, and the time domain features comprise peak values X peak ＝max(|x i ' ' ' |) and average values through calculation Variance of Kurtosis of Degree of deviation Acquiring time domain characteristics of data by the internal parameters, wherein X peak is a noise-reduced signal peak value, mu '' 'is a noise-reduced signal mean value, sigma' '' 2 is a noise-reduced signal variance, K is noise-reduced signal kurtosis, and S is noise-reduced signal skewness; Converting a time domain signal into a frequency domain signal X (f) =FFT (X ' ' ' i ) through Fourier transformation, extracting parameters including characteristic frequency amplitude and frequency center of gravity, and obtaining data frequency domain characteristics, wherein X (f) is the frequency domain signal, and f is the frequency value; And integrating the time domain, the frequency domain and the time domain-frequency domain joint characteristics to construct a multidimensional characteristic vector F= [ F 1 ,f 2 ,...,f k ], wherein F 1 to F k are extracted characteristic parameters, k is a characteristic dimension, and high-quality characteristic information which can be used for equipment health assessment is formed.
  5. 5. The method for predicting and managing the full life cycle failure of industrial equipment according to claim 1, wherein the model construction method of the physical mechanism in S4 is as follows: aiming at a gear core component, a gear wear degradation model is constructed based on the coupling of an Archard wear equation and the actual running condition of equipment, and the formula is as follows Wherein W is the actual abrasion loss of the gear, K is the abrasion coefficient of the gear material, F is the gear engagement contact pressure, S is the gear engagement sliding distance, F (ζ) is the working condition correction function, ζ is the load fluctuation coefficient, and the method is used for quantifying the nonlinear influence of load fluctuation on the gear abrasion; aiming at a bearing core component, a bearing fatigue degradation model is built based on Lundberg-Palmgren theory and temperature influence factors, and a residual fatigue life calculation formula is as follows Wherein L 10 is the actual residual fatigue life of the bearing, L 10r is the basic rated life of the bearing, C is the basic rated dynamic load of the bearing, P is the actual running load of the bearing, P is the bearing life index, alpha is the temperature influence coefficient, T is the actual running temperature of the bearing, T 0 is the rated working temperature of the bearing, and the quantitative representation of the influence of the temperature on the service life of the bearing is realized; Constructing a device-level physical mechanism fusion model, weighting and quantifying the degradation states of all core components into health scores, wherein the formula is Wherein PHI is health score output by a physical mechanism model, the value range is 0 to 100, n is the number of core components of the equipment, W i is the weight coefficient of the ith core component and is determined by the influence degree of the component on the operation of the equipment, D i is the actual degradation amount of the ith core component, the abrasion amount W is adopted by a gear, the difference value between the basic rated life and the actual residual fatigue life is adopted by a bearing, D i,max is the maximum allowable degradation amount of the ith core component, and an exponential function is adopted by the bearing And the nonlinear health attenuation law is used for simulating the degradation of the component.
  6. 6. The method for predicting and managing full life cycle faults of industrial equipment according to claim 1, wherein the data-driven model training process in the step S4 is as follows: set training data set as Wherein F i is a feature vector output by S3, y i is a real value of a corresponding equipment health index, and N is the total sample amount; Constructing a random forest health evaluation model, wherein the model comprises M decision trees, the maximum depth of each decision tree is D max , the minimum sample splitting is s min , and M training subsets are randomly extracted from a data set D by a self-help sampling method In the splitting process of the decision tree nodes, randomly selecting a feature subset from all features of a feature vector F i to perform optimal splitting feature selection, and generating a splitting rule of each decision tree; Dividing the data set D into a training set D tr and a test set D te according to a proportion gamma (1-gamma), adopting K-fold cross-validation optimization model parameters, and dividing the training set D tr into K mutually disjoint subsets K-1 subsets are selected as training data each time, the remaining 1 subsets are used as verification data, and the cross verification is completed in a circulating mode for K times; The objective function of model training is to minimize the mean square error of the predicted value and the true value of the health index, and the formula is: Wherein the method comprises the steps of For the health index predicted value output by the random forest model, N tr is the number of training set samples, and in the model test stage, the prediction accuracy is calculated on a test set D te , and the formula is as follows: Wherein N te is the number of samples in the test set, I (·) is an indication function, the value is 1 when the condition is satisfied, otherwise the value is 0, and ε is a health index prediction error allowance threshold.
  7. 7. The method for predicting full life cycle failure and managing health of industrial equipment according to claim 5 or 6, wherein the calculating process of the equipment health index in S4 is as follows: Firstly, selecting a historical calibration data set of the whole life cycle of equipment, respectively calculating standard deviations std (PHI cal ) and std (DHI cal ) of a physical mechanism model health score sequence PHI cal and a data-driven model health score sequence DHI cal , wherein the smaller the standard deviation is, the more stable the output of a representative model is, the higher the reliability is, and constructing a stability weight factor based on the standard deviation The factor quantifies the stability difference of the two model outputs and provides a basis for the dynamic adjustment of the weight; secondly, calculating a weight adjustment increment delta lambda=lambda-w PHI by taking a preset basic weight as a reference, wherein w PHI is the basic weight of a physical mechanism model, the basic weight of a data driving model is w DHI , the sum of the two basic weights is 1, and then a real-time dynamic weight is obtained, the real-time weight of a health score of the physical mechanism model is w PHI +delta lambda, the real-time weight of a health score of the data driving model is w DHI -delta lambda, the sum of the two weights is always 1, and the normalization of a fusion system is maintained; And finally, substituting the real-time dynamic weight into a fusion formula to obtain an equipment health index EHI= (w PHI +Δλ)·PHI+(w DHI -delta lambda). DHI, automatically improving the weight ratio when the output of the physical mechanism model is more stable, strengthening the interpretability of the fusion result, correspondingly improving the weight ratio when the output of the data driving model is more stable, enhancing the suitability of the fusion result to complex working conditions, and realizing the complementary coordination of the advantages of the two models. Meanwhile, the training output of the data-driven model health score DHI calculation correlation random forest model is provided, the random forest model comprises M decision trees T m , a feature vector F output by S3 is input, the single decision tree output is T m (F), the value range is [0,100], a weighted integration strategy based on the prediction precision of a verification set is adopted, the decision tree weight is alpha m , and the requirements are met The output formula of DHI is: the formula directly outputs a health score with a value range of [0,100 ].
  8. 8. The method for predicting and managing the full life cycle failure of industrial equipment according to claim 1, wherein the construction process of the time sequence prediction model in S5 is as follows: the input layer receives the multidimensional feature vector, and converts the multidimensional feature vector into a high-dimensional embedded vector through the embedded layer, so that feature dimension increase and semantic enhancement are realized; by adopting an additive attention mechanism, obtaining a characteristic attention weight through softmax normalization by calculating the association score of the current hidden state and the hidden state at the historical moment, and highlighting key time sequence information; constructing a multi-layer bidirectional LSTM structure, configuring the number of hidden units, introducing dropout regularization to inhibit overfitting, and deeply mining the time sequence dependency relationship of the characteristics; And the full connection layer performs mapping conversion on the extracted time sequence characteristics, and finally, the output layer outputs the residual service life quantification result of the equipment.
  9. 9. An industrial equipment full life cycle fault prediction and health management system, comprising: The end-to-end interconnection configuration unit is used for constructing an end-to-end unified network architecture, distributing unique identification addresses for industrial equipment and matched sensors, and completing direct interconnection of the industrial equipment, the sensors and a data processing platform or a cloud platform; The digital mapping real-time synchronization unit is used for constructing a digital mapping model covering the full information dimension of the equipment based on a unified communication protocol, completing the real-time synchronization of the physical equipment and the digital mapping model and providing digital support for equipment state evaluation; The multi-source data feature extraction unit is used for comprehensively acquiring process data, state data and quality data in the running process of the industrial equipment, storing the acquired multi-source data by adopting a data storage scheme adapting to the industrial scene, and converting the original data into high-quality feature information which can be used for equipment health assessment through the processing flows of outlier rejection, data standardization, signal noise reduction and feature extraction; The fusion evaluation health calculation unit is used for constructing a health evaluation model by fusing a physical mechanism and a data driving method, inputting the processed characteristic information and equipment working condition parameters into the health evaluation model, and calculating to obtain a quantized equipment health index, wherein the equipment health index reflects the overall health level of the equipment and the degradation state of the core component; The life prediction unit is used for predicting the residual service life of the equipment by adopting a time sequence prediction model, and setting a multi-level early warning threshold value based on the equipment health index, the fault risk level and the maintenance cost; The operation and maintenance scheme model updating unit is used for matching an optimal operation and maintenance scheme with an execution main body based on equipment, operation and maintenance and fault related data, supporting fault processing and maintenance operation by means of a technical tool, recording maintenance process data, verifying maintenance effect and synchronously updating an equipment digital model and a related database; The fault knowledge model iteration unit is used for constructing a fault knowledge system shared by the cross-scene, storing fault cases and treatment schemes, supporting intelligent retrieval recommendation, periodically fusing newly-added data iteration optimization evaluation, a prediction model and early warning judgment standards, and dynamically adapting to equipment working condition changes.
  10. 10. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program is executed by a processor for a full life cycle fault prediction and health management method of an industrial device according to any of claims 1-8.

Description

Industrial equipment full life cycle fault prediction and health management method and system Technical Field The invention belongs to the technical field of intelligent automobile manufacturing, and particularly relates to a full life cycle fault prediction and health management method and system for industrial equipment. Background In the process of intelligent transformation of industrial production, the stable operation of industrial equipment is directly related to the production efficiency and the safety level, and the full life cycle fault prediction and health management of the equipment are key requirements for industry development. The traditional equipment management mode is dependent on manual inspection and post-maintenance, has obvious limitation, is difficult to master the running state of the equipment in real time, and is easy to miss potential fault hidden dangers due to human negligence, so that the equipment is stopped suddenly, and mass production loss and high maintenance cost are caused. In the existing part of equipment health management schemes, although data acquisition and simple analysis means are introduced, the problems of unsmooth interconnection of equipment and a platform and non-uniform data standards are generally solved, and equipment and sensors of different manufacturers lack uniform identification and communication protocols to form information islands, so that efficient transmission and sharing of data cannot be realized. Meanwhile, the data analysis models of the schemes depend on a physical mechanism or a data driving method, the data analysis models are difficult to adapt to complex and changeable industrial working conditions, and the data analysis models lack of interpretability, so that the accuracy of health assessment results is insufficient, and the real degradation state of the core components of the equipment cannot be effectively reflected. In addition, the existing scheme fails to construct a complete fault prediction, early warning treatment and model iteration closed-loop system, the early warning mechanism is mostly based on fixed threshold values, cannot be dynamically adjusted according to equipment health indexes and maintenance cost, the operation and maintenance scheme is matched and lacks data support, fault knowledge is difficult to share across scenes, similar faults are repeatedly generated, and management efficiency is low. The problems severely restrict the intelligent level of the industrial equipment management, and a full life cycle fault prediction and health management method capable of realizing equipment interconnection, accurate assessment, intelligent early warning and closed loop optimization is urgently needed, so that the running reliability of equipment is improved, the operation and maintenance cost is reduced, and the high-quality development of industrial production is promoted. Disclosure of Invention The invention aims to solve the problems of low efficiency of a traditional industrial equipment management mode, information island existence, single evaluation model and lack of a closed-loop management system in the existing scheme, and provides a full life cycle fault prediction and health management method, which realizes equipment interconnection, accurate evaluation, intelligent early warning and operation and maintenance optimization, improves equipment reliability and reduces operation and maintenance cost. In view of the above-mentioned drawbacks or improvements of the prior art, as a first aspect of the present invention, the present invention provides a method for predicting full life cycle failure and health management of industrial equipment, comprising: S1, constructing an end-to-end unified network architecture, distributing unique identification addresses for industrial equipment and matched sensors, and completing direct interconnection of the industrial equipment, the sensors and a data processing platform or a cloud platform; S2, based on a unified communication protocol, constructing a digital mapping model covering all information dimensions of the equipment, completing real-time synchronization of the physical equipment and the digital mapping model, and providing digital support for equipment state evaluation; S3, comprehensively acquiring process data, state data and quality data in the running process of industrial equipment, storing the acquired multi-source data by adopting a data storage scheme adapting to the industrial scene, and converting the original data into high-quality characteristic information which can be used for equipment health assessment through abnormal value rejection, data standardization, signal noise reduction and characteristic extraction; S4, a health assessment model is built by combining a physical mechanism and a data driving method, the processed characteristic information and equipment working condition parameters are input into the health assessment model, a quantized eq