Time Domain Self-Similar Vibration Signal Feature Extraction And Its Application Research On Mechanical Anomaly Detection

In recent years,with the development of science technology,mechanical equip-ments are gradually developing towards automation,large-scale and precision.The stable and efficient operation environment of mechanical equipment is of great signifi-cance for improving efficiency and saving cost.Therefore,the detection of mechanical anomalies is necessary,which enables the anomalies to be discovered in time,so as to take reasonable and effective measurements as soon as possible to avoid property and personal losses.Aiming at the core problem of mechanical anomaly detection,this paper proposes a self-similar vibration signal feature extraction method and researches its feasibility and effectiveness in real application.Generally speaking,the process of mechanical anomaly detection consists of three steps:(1)signal collection,(2)feature extraction and(3)anomaly decision making.Feature extraction is the most important step,which determines the effectiveness of the detection method.This paper proposes an effectiveness self-similar feature extraction method,which fully considers the structural information and time sequence informa-tion of mechanical vibration signal during the process of feature extraction based on Bag-of-Word.Bag-of-Word construction is the first step of self-similar feature extrac-tion.In this paper,K-means clustering method is employed to cluster the training sam-ples and the Bad-of-Word is constructed by clustering center and clustering variance.For a new monitored vibration signal,three strategies(histogram strategy,embedding strategy and correlogram strategy)are used for self-similar feature extraction,so as to extract self-similar feature.By using three strategies to assemble the self-similar features of the monitored vi-bration signal,the structural information can be obtained timely and the mechanical condition is described effectively.However,the self-similar feature extraction can only describe the mechanical condition qualitatively,which is not conducive to the subsequent analysis.For this purpose,the entropy of extracted feature is calculated to describe mechanical condition quantitatively.Statical analysis is used to calculated the anomaly scores to analyse the mechanical condition.Finally,hypothesis test based on 3? is used to judge whether the machine is working in normal condition.In summary,this paper proposes an effective self-similar vibration signal feature ex-traction method.In order to demonstrate the effectiveness of the proposed method in mechanical anomaly detection,we first apply the proposed method to simulation sig-nal.The experiment results show this method has a great detection performance.Fur-thermore,the proposed method is applied to two real engineering scenarios,where the real scenario databases are the vibration database acquired by the laboratory's equip-ment and the open database of CWRU.The experiment results show that the proposed method can detect the abnormal condition timely and accurately,and its detection per-formance is better than some typical feature extraction method.Finally,we conclude the proposed studies,and discuss the future work along with the thesis.