Research On Bond Graph Model Based Fault Diagnosis Method For Complex Mechatronics System

Complex mechatronics systems are coupled together by multi-form-structured subsystems on multi-energy field.The condition supervision and fault diagnosis for such system are research focuses for a long time.Model-based fault diagnosis approach has a capacity that fully explores in-depth information reflected from system model,and then performs excellent online diagnosis property without the support of historical operating status data.It becomes one of preferred solutions to maintain a complex mechatronic system continuous operation with high stability.Difficulties that exist on the construction of model-based diagnostic solutions for complex mechatronic system include: structural complexity causes it is difficult to model such system;multi-energy field coupling causes identified model parameters is not precise enough;parameter uncertainty make online residuals do not precisely reflect system's operating status;mode transition of hybrid system brings problems of miss detection and false alarm,etc.Focus on issues related to fault diagnosis method for complex mechatronic system,such as system modeling,system parameter identification,online residual,adaptive threshold of hybrid system,fault detectability and isolability,this study had researched these problems systematically.Main research contents of this paper include:This paper taken bond graph approach as tools for complex mechatronic system modeling,and sequentially model vehicle steering system,analyze structural characteristics and energy transfer law,as well as perform model augmentation and causality assignment;based on conventional bond graph modeling,reflect the mode transition dynamics of hybrid system on Bond Graph model,derive analytical redundancy relation,as a result,acquire model equations.A new type of parameter identification network was designed,and BP network was improved according to parameter identification feature.Takes the results of the Least Squares method as initial value of network training,adopts speed-varying training method to train weight value of networks.Apply modulated modular modeling principle to model vehicle steering system,and then derive model equations,apply proposed identification algorism to identify 6 key parameters,takes results of the least squares method as initial value of network training,perform secondary identification process by improved network.Experimental results show that improved network algorithm is reliable and efficient.System's diagnostic model was buildt to derived ARRs,which are a set of equations containing observed variables,which contains information that are necessary to determine system operating state.It is very effective for fault diagnosis of dynamic system.Clear the objective fault type parameters of this study was abrupt failure instead of gradual failure,or aims at abrupt change of system's internal components parameter occurs in a shorter period,and assume that only single failure exists simultaneously.Experimentally determine diagnostic threshold through system's normal operation,fault diagnosis for continuous system is carry out on automatic vehicle,3 simulated faults are introduced for experiment,analysis of experiment result verify the feasibility of the proposed fault diagnosis system algorithms.Mode-dependent adaptive threshold of hybrid system was proposed according to dynamic model of hybrid system.The cause of false alarm and miss detection during fault diagnosis of hybrid systems was studied to find out drawback of conventional solution and propose new solution.According to estimation of system parameters uncertainties and effect on each part of residual,utilize Linear Fractional Transformation to formulate mode-dependent threshold;establish a fault diagnosis system for hybrid system that includes GARRs,system residual,mode-dependent adaptive threshold algorithm and mode-dependent fault signature matrix.Through simulation circuit building,Bond Graph modeling,simulation experiment performation and experiment result analysis of diagnosis architecture for hybrid system,the feasibility of the proposed diagnostic framework was preliminarily verified.For the purpose of diagnosis experiment for hybrid system,an automatic vehicle was redeveloped on steering system,and then dynamics of hybrid system displayed on it;once system's hybrid Bond Graph model is completed,GARRs are derived from it,mode-dependent thresholds are formulated,MD-FSM is generated for fault detectability and isolabilty analysis.Fault diagnosis experiments of practical system were performed on redeveloped platform,faults were introduced with non-destructive manner,and fault diagnosis performances on different mode were analyzed,diagnostic results of proposed diagnosis architecture for hybrid system are verified on practical system.The experimental results show that diagnostic performance of diagnostic architecture based on diagnosis theory of hybrid systems is coincidence with theoretical analysis and detectability and isolability analysis.This research provides a complete solution and design procedure for fault diagnosis design of complex mechatronics system.