CN-120974437-B - Equipment health assessment method and device based on multidimensional data fusion and dynamic weight

CN120974437BCN 120974437 BCN120974437 BCN 120974437BCN-120974437-B

Abstract

The embodiment of the invention discloses a device health assessment method and device based on multidimensional data fusion and dynamic weight, which relate to intelligent monitoring and health assessment technology of a device complex system in the field of device health management, and comprise the following steps: the change condition of the health parameter along with the increase of the equipment storage age is described from the angles of regression trend of the health parameter along with time, association relations among different health parameters and probability distribution of the health parameter data, and the difference between the health parameter monitoring data and the baseline data is accurately reflected. The dynamic updating method for the total process weight of the storage is provided, and the weight occupied by different health parameters in the process of calculating the health index is calculated scientifically. Therefore, the problem that the health parameter monitoring data is scarce due to the fact that the equipment cannot be frequently electrified in the actual storage and use processes is effectively solved, and under the condition that the monitoring data is limited, the evaluation accuracy of the health state of the equipment is improved.

Inventors

CHEN JIAYU
LU QINHUA
WANG XUHANG
CHEN MENGXUE
MA YUCHENG
GE HONGJUAN

Assignees

南京航空航天大学

Dates

Publication Date: 20260505
Application Date: 20251020

Claims (6)

1. An equipment health assessment method based on multidimensional data fusion and dynamic weight is characterized by comprising the following steps: s1, establishing an equipment health baseline model aiming at health characteristic parameters of equipment; s2, dynamically adjusting the health index output by the equipment health baseline model based on the weight, and then obtaining the comprehensive health index of the equipment, wherein S2 comprises the steps of adaptively updating the weight of the health index by using CRITIC method and entropy weight method; The self-adaptive update of the weight of the health index by CRITIC method and entropy weight method comprises calculating the initial weight of the health index , , And Respectively representing the weight when the health index is calculated by CRITIC method and entropy weight method based on the ith parameter of the baseline data of the first year; after the parameter data is expanded, iterative calculation is carried out again, and the parameter weight after the kth iteration is as follows: , And Respectively calculating the weight for the ith health characteristic parameter through CRITIC method and entropy weight method in the kth iteration; the process of generating the comprehensive health index comprises the steps of generating the parameter health index of each ith health characteristic parameter Using adaptively updated dynamic weights Weighting and fusing to obtain primary comprehensive health index : , , wherein, The j-th type health index representing the i-th health characteristic parameter, m is the number of types of the health indexes, n represents the number of the health characteristic parameters, and the finally obtained comprehensive health index H of the equipment is equal to H i,MD represents a mahalanobis distance index of the i-th health feature parameter, H i,MAE represents an average absolute error index of the i-th health feature parameter, and H i,OP represents a probability density distribution coincidence percentage index of the i-th health feature parameter; repeating the expansion of the parameter data and the dynamic weight update, and incorporating the new parameter operation data into the weight calculation data set so as to continuously update the parameter data set and adaptively update the weight; S3, determining the health state of equipment according to failure judgment conditions of comprehensive health indexes of the equipment, wherein S3 comprises the steps of screening the equipment with the best health state according to the obtained comprehensive health indexes of the equipment, normalizing the comprehensive health indexes of the equipment with the best health state, determining a failure judgment threshold value of the comprehensive health indexes of the equipment, and defining a failure judgment base line; Determining a health index failure judgment threshold by comparing health states of a plurality of pieces of equipment participating in a test, wherein a first data point of the equipment with the smallest comprehensive health index is taken as a starting point, a numerical value of the equipment with the smallest comprehensive health index at the time of failure starting year is taken as an end point for data normalization, and a failure judgment margin of a unified equipment failure judgment baseline is defined to be 5%, and the health index failure judgment threshold is defined to be 0.95; And then, evaluating the health evaluation capability of the equipment failure judgment baseline through 4 indexes of accuracy, precision, recall and F1 fraction, wherein the accuracy refers to the proportion of the number of correctly classified samples to the total number of samples, the judgment capability of the classifier to the whole sample set is reflected, the precision refers to the proportion of the true positive samples in the samples classified into the positive types, the accuracy of the classifier to the prediction of the positive samples is measured, the recall refers to the proportion of the samples correctly predicted to the positive samples in all the true positive samples, the investigation capability of the classifier to the positive samples is reflected, and the F1 fraction is the harmonic average of the precision and the recall.
2. The method of claim 1, wherein S1 comprises: Collecting key performance monitoring parameters of equipment, and extracting health characteristic parameters representing degradation trend of the equipment, wherein the types of the health characteristic parameters comprise system characteristic voltage, system characteristic current, resistance and system power of the equipment; According to the extracted health characteristic parameters, nominal data of target equipment when leaving a factory and initial storage state data, an equipment health baseline model is established, the equipment health baseline model is used for analyzing health indexes, and the health indexes comprise a Markov distance index, an average absolute error index and a probability density distribution coincidence percentage index.
3. The method of claim 2, wherein a mahalanobis distance indicator is used to characterize the overall degree of deviation between the health characteristic parameter and the standard matrix of the health baseline model, the value of the mahalanobis distance indicator being inversely related to the health status of the equipment; The average absolute error index is used for measuring deviation between the actual value of the health characteristic parameter and the predicted value of the health baseline model, and the value of the average absolute error index is inversely related to the health state of the equipment; The probability density distribution coincidence percentage index is used for evaluating the similarity of the health characteristic parameter and the standard sample distribution shape, and the value of the probability density distribution coincidence percentage index is positively correlated with the health state of the equipment.
4. A method according to claim 2 or 3, wherein the mahalanobis distance indicator of the i-th health characteristic parameter is expressed as: , representing a standard matrix of the healthy baseline model corresponding to the ith parameter, Representing the ith sample data, S representing covariance matrix, and the mean absolute error index of the ith health characteristic parameter being expressed as , wherein, The true value is represented by a value that is true, Representing the predicted value; the probability density distribution coincidence percentage index of the ith health characteristic parameter is expressed as , wherein, Is the probability density value of the health feature parameter over the kth interval, Is the probability density value of the standard sample over the kth interval, Is the interval width of the probability density distribution, and n is the total number of intervals of the probability density distribution.
5. An equipment health assessment device based on multidimensional data fusion and dynamic weights, comprising: The model maintenance module is used for establishing an equipment health baseline model aiming at the health characteristic parameters of the equipment; The analysis module is used for dynamically adjusting the health index output by the equipment health baseline model based on the weight, and then acquiring an equipment comprehensive health index, wherein the weight of the health index is adaptively updated by using a CRITIC method and an entropy weight method; The self-adaptive update of the weight of the health index by CRITIC method and entropy weight method comprises calculating the initial weight of the health index , , And Respectively representing the weight when the health index is calculated by CRITIC method and entropy weight method based on the ith parameter of the baseline data of the first year; after the parameter data is expanded, iterative calculation is carried out again, and the parameter weight after the kth iteration is as follows: , And Respectively calculating the weight for the ith health characteristic parameter through CRITIC method and entropy weight method in the kth iteration; the process of generating the comprehensive health index comprises the steps of generating the parameter health index of each ith health characteristic parameter Using adaptively updated dynamic weights Weighting and fusing to obtain primary comprehensive health index : , , wherein, The j-th type health index representing the i-th health characteristic parameter, m is the number of types of the health indexes, n represents the number of the health characteristic parameters, and the finally obtained comprehensive health index H of the equipment is equal to H i,MD represents a mahalanobis distance index of the i-th health feature parameter, H i,MAE represents an average absolute error index of the i-th health feature parameter, and H i,OP represents a probability density distribution coincidence percentage index of the i-th health feature parameter; repeating the expansion of the parameter data and the dynamic weight update, and incorporating the new parameter operation data into the weight calculation data set so as to continuously update the parameter data set and adaptively update the weight; The monitoring module is used for determining the health state of equipment according to failure judgment conditions of comprehensive health indexes of the equipment, screening the equipment with the best health state according to the obtained comprehensive health indexes of the equipment, determining a failure judgment threshold value of the comprehensive health indexes of the equipment after normalizing the comprehensive health indexes of the equipment with the best health state, and defining a failure judgment base line; Determining a health index failure judgment threshold by comparing health states of a plurality of pieces of equipment participating in a test, wherein a first data point of the equipment with the smallest comprehensive health index is taken as a starting point, a numerical value of the equipment with the smallest comprehensive health index at the time of failure starting year is taken as an end point for data normalization, and a failure judgment margin of a unified equipment failure judgment baseline is defined to be 5%, and the health index failure judgment threshold is defined to be 0.95; And then, evaluating the health evaluation capability of the equipment failure judgment baseline through 4 indexes of accuracy, precision, recall and F1 fraction, wherein the accuracy refers to the proportion of the number of correctly classified samples to the total number of samples, the judgment capability of the classifier to the whole sample set is reflected, the precision refers to the proportion of the true positive samples in the samples classified into the positive types, the accuracy of the classifier to the prediction of the positive samples is measured, the recall refers to the proportion of the samples correctly predicted to the positive samples in all the true positive samples, the investigation capability of the classifier to the positive samples is reflected, and the F1 fraction is the harmonic average of the precision and the recall.
6. The apparatus of claim 5, wherein the model maintenance module is configured to collect key performance monitoring parameters of the equipment, and extract health characteristic parameters representing degradation trends of the equipment, where the health characteristic parameters include system characteristic voltage, system characteristic current, resistance, and system power of the equipment, and establish a health baseline model of the equipment according to the extracted health characteristic parameters and nominal data and initial storage state data of the target equipment when leaving the factory, where the health baseline model is configured to analyze health indexes including a mahalanobis distance index, an average absolute error index, and a probability density distribution coincidence percentage index.

Description

Equipment health assessment method and device based on multidimensional data fusion and dynamic weight Technical Field The invention relates to intelligent monitoring and health assessment technology of equipment complex systems in the field of equipment health management, in particular to an equipment health assessment method and device based on multidimensional data fusion and dynamic weights. Background In the long-term storage and use process of a complex system of modern equipment, the performance degradation and reliability guarantee of the complex system directly influence the use efficiency and the whole life cycle cost of the equipment. In order to ensure reliable operation, prolong service life and realize accurate maintenance of the system, fault prediction and health management (Prognostics AND HEALTH MANAGEMENT, PHM) technology has become a core means for improving reliability and safety of the system and reducing operation and maintenance cost. It should be noted that, the health assessment is a core concept in PHM, and the essence is based on deep analysis and decision of state monitoring data, wherein the monitoring is the basis of "data acquisition and state sensing" (such as obtaining parameters of vibration, voltage, resistance, etc.), and the assessment is the key of "state quantification and trend judgment" (such as calculating health index through multidimensional data fusion, judging degradation stage, predicting remaining life). Health state evaluation serves as a core link of PHM technology, and by monitoring multi-source performance parameters in real time, the degradation trend of the system is identified, decision basis is provided for a maintenance-on-demand (CBM) system, and task priority ordering and active guarantee optimization of supporting equipment are carried out. The high-reliability health assessment method has great significance for guaranteeing equipment availability and efficient resource utilization in the face of challenges of complex system multi-level coupling failure. The health assessment method based on machine learning becomes an important research direction in the PHM field through data-driven modeling and self-adaptive feature extraction. Such methods, while capable of quantitatively analyzing equipment health status based on performance parameters, output individual or batch health ratings, still face a number of significant bottlenecks for their practical use. Therefore, in further development, health assessment methods based on multidimensional data fusion are receiving attention. According to the method, multi-source data in the whole life cycle of equipment are integrated, a cross-level and cross-dimension performance degradation analysis framework is constructed, and multi-feature collaborative mining and dynamic fusion are achieved. Compared with a single parameter or model driven evaluation mode, the multidimensional data fusion technology can more comprehensively capture the coupling failure mechanism of the complex system, enhance the integrity of health characterization, and simultaneously promote the characterization capability of the model on the nonlinear degradation mode through multisource feature correlation analysis and self-adaptive weight distribution, thereby providing technical support for accurate health quantification and active maintenance decision of the complex system. Although the multidimensional data fusion method shows theoretical advantages, the existing health assessment method has obvious defects in aspects of feature parameter completeness, feature fusion mechanism, dynamic weight optimization and the like, and restricts the refinement level of the health management of a complex system. The main constraint is that the complex system comprises multiple subsystems such as electronic, mechanical and kinetic energy, and the functional coupling exists among the subsystems, so the complex system belongs to a multi-domain nonlinear coupling system. However, more optimized equipment health assessment schemes often need richer monitoring data, however, the more complex equipment of the system is lack of health parameter monitoring data due to the fact that frequent power-on tests cannot be performed in actual storage and use processes, so that data samples are few, and therefore the batch equipment cannot be tested one by one, and differences between the health parameter monitoring data and baseline data cannot be accurately reflected, and the assessment accuracy of the equipment health state is difficult to improve. Disclosure of Invention The embodiment of the invention provides a device health assessment method and device based on multidimensional data fusion and dynamic weight, which can improve the assessment accuracy of the device health state under the condition of limited monitoring data. In order to achieve the above purpose, the embodiment of the present invention adopts the following technical scheme: in a first aspect,