Hydraulic System Abnormality Detection Based On Time Series Similarity And Integrated Classification Analysis

There are many typical characteristics with the systems: high power,small size,fast response,light weight,high precision and high rigidity against load,etc.It is widely used in many fields such as shipbuilding,aerospace,metallurgical industry,construction machinery,agricultural machinery and so on.Nowadays,the hydraulic system is gradually developing in the direction of rapid,high-power,and high-precision.The probability of abnormality in the hydraulic system and its equipment will also increase.If the hydraulic system fails,it will cause significant losses.However,because the hydraulic system is a highly non-linear and complex system,and with the continuous improvement of the hydraulic system's own functions and automatic production level,its complexity is also higher.Therefore,it is difficult to directly check the abnormality of the hydraulic system.The internal observation of the system requires a data-driven method,starting with the historical data obtained by the hydraulic system sensors,and detecting abnormalities in the hydraulic system,so as to detect abnormal changes in the state of the hydraulic system in time and take effective preventive measures to provide corresponding science for inspector analysis.In view of the above-mentioned difficulties and research focus,the work of this paper mainly includes the following contents,Aiming at the complexity and uncertainty of the abnormality of the hydraulic system,a semi-supervised hydraulic system anomaly detection algorithm based on multivariate time series similarity measures is designed.In the selection of feature processing methods,the PAA and PLR methods are used in the hydraulic system.Comparison of the dimensionality reduction effect on the data,the PLR method is selected to represent the dimensionality reduction feature of the data,and the distance measurement method based on the improved DTW is selected for the similarity measurement method.This method can not only meet the distance measurement of unequal time length series,and in the calculation,the point-to-point distance measurement method is converted into the distance calculation between sub-line segment modes to improve the calculation efficiency.The experiment on the hydraulic system detection platform verifies the feasibility and effectiveness of the algorithm.In view of the possible failure of the main components of the hydraulic system,the cooler,the valve,the hydraulic pump,and the accumulator,a supervised method is designed based on the cascade forest integrated classification pressure system anomaly detection algorithm.This method uses the artificial intelligence field Classification method,convert multiple sensor data into an overall feature space,learn the characteristics of each sensor time series data under the four component failure modes,optimize on the single classifier method,and select the integrated classification method based on cascade forest This method selects KNN,SVM,and Naive Bayes as the three base classifiers in composition.The structure is classified layer by layer with the structure of cascade forest,combined with the theory of scale space,selects the appropriate scale to classify the hydraulic system monitoring data Learn to determine whether there is an abnormality,and identify the specific type of abnormality.Experiments on the hydraulic system detection platform verify the feasibility and effectiveness of the algorithm for detecting and identifying the four failure modes of the hydraulic system.