Research On Health Management Of Complex System For Latent Failure

Latent failures are important factors that will affect the reliability of complex systems.Latent failures are unobservable and often couple other faults,which will increase the difficulty of identifying,analyzing,and evaluating potential faults.If the potential failure was not detected and discovered before conversion to the actual failure,it would directly affect the operation safety.In addition,complex systems with diversified structures and nonlinear state representation increase the difficulty of considering latent failures.This paper discusses the health management of complex systems for latent failures,including the following research:Firstly,the paper is written by analyzing the characteristics of complex system's failure and latent failure,structuring a health management system for complex systems,and proposing a deep learning method to study latent failure.State monitoring technology is used to keep abreast of the state of the system.And deep learning model is used to process the sensor data,identify the degradation state of the system,mine the potential value of the information,effectively explore potential sources of failure.Therefore,it can lay the foundation for subsequent research.Secondly,considering the data characteristics of potential faults,the combination of principal component analysis and deep belief network model is used to analyze the remaining useful life of the engine.And a deep learning fault prediction process for potential faults is constructed to take active action before potential faults are transformed into faults.Thirdly,the health decision-making for potential faults is discussed.Risk decision considering potential faults is analyzed.The method of importance degree is studied,which provides a means for detecting potential fault sources and offers a theoretical basis for maintenance support.Finally,turbine engine is selected as the research object to prove the effectiveness of the model.The running data of the simulated engine is collected.And the model is trained by the lifecycle data of simulated engine,where sensor multi-dimensional data is used as the input data,and the remaining useful life(in the operation cycle)is output data.Mining data's features to predict the period of latent failure sources.It further analyzes the performance trend of each component in the potential fault source and explores the importance of the impact on the overall output characteristics of the system.Finally,take proactive measures according to the degree of importance to suppress further degradation of component performance,thereby delaying the occurrence of failures.