On Multivariate Time Series Classification Based Concurrent Faults Diagnosis Of Hydraulic Systems

Hydraulic systems are a type of system with outstanding nonlinear characteristics,the system have the characteristics of multi-functional components,multiple operation modes,multi-time-space coupling,and multi-parameter monitoring signals.The automation level and system scale of such objects are increasing day by day,and the functional requirements are more specialized and diversified,the complexity of operation and monitoring is much higher than that of traditional hydraulic systems,therefore,the possibility of failure during operation is greatly increased,which poses risks to the safety and reliability of the hydraulic system and higher-level systems.After analyzing a large number of hydraulic system failure cases,it is found that due to the complex characteristics of the hydraulic system’s own structure and mechanism,the internal structure of the system cannot be directly observed,the failure mechanism is difficult to obtain completely,and there are many types of failures,especially in scale effects and the above characteristics In the case of compound action,the failure mode formed by multiple anomalies is more subtle but can appear at any time and location.For this kind of spatiotemporal correlation coupled concurrent fault diagnosis,it has always been the focus of the academic and industrial concerns.This article uses this as an entry point to carry out related research on the fault diagnosis of hydraulic systems based on multivariate time series classification.The main contents are as follows:The failure mechanism of hydraulic systems failure and the structural characteristics of the hydraulic systems are studied,the main failures and their failure characteristics are analyzed,a data-driven fault diagnosis method is introduced for the problems in the concurrent failure of hydraulic systems,which relationship realizes the characterization of the fault.Using the fault state monitoring data extracted from the sensor network of the hydraulic systems fault simulation platform for analysis and verification,the feasibility of the mapping method is expressed,and on this basis,a technical route for concurrent fault diagnosis of the hydraulic system is formed.Aiming at the characteristics of multivariate time series of fault status monitoring data,multivariate time series classification is selected as the method of hydraulic system concurrent fault diagnosis,and a multivariate time series classification method based on1NN-SVM is proposed.Use machine learning algorithm to extract related features in multivariate time series data,introduce nearest-neighbor 1NN algorithm to extract time-series features in data,and convert related features into feature tags;After converting the feature labels,the multi-class OVO-SVM is used to complete the entire multi-class time series classification process.Finally,the proposed method is introduced into the single fault experiment of hydraulic systems fault components to verify the effectiveness of the proposed hydraulic system fault diagnosis algorithm based on multivariate time series classification.Aiming at the influence of noise on the sensor when acquiring fault status monitoring data during the hydraulic system work,a new multivariate time series feature extraction structure is proposed to improve the accuracy of feature extraction by combining domain-independent classification algorithm and domain-related classification algorithm.An adaptive multivariate time-series feature extraction module based on Attention is proposed,Long Short-Term Memory(LSTM)is used as the decoder and encoder in the Attention structure to extract features from the original data.The 1NN-SVM method is used to complete the multivariate time series feature classification in the feature space,eliminating the impact of noise on the multivariate time series classification results,further improving the fault diagnosis accuracy and verifying the improved method through the single fault experiment of the hydraulic system fault component.Aiming at the single failure and concurrent failure of the actual hydraulic system,the related experiments and results analysis of the two methods are carried out.The experiment is divided into two parts.The first part is the fault diagnosis and identification of all single faults of each fault element;the second part is the diagnosis and all possible faults of two fault elements in the concurrent fault condition Fault identification of the occurrence of concurrent faults.Experimental results verify the feasibility of the algorithm proposed in this paper in fault diagnosis.At the end of the paper,the comparison results of the fault diagnosis method of hydraulic system based on multivariate time series classification and common methods are given,and the advantages of the proposed algorithm are explained.