Research On Prognosis And Maintenance Decision Of Equipment Based On Multi Dimension Feature Parameters

Manufacturing is a basis of national development.In recent years,with the rapid development of information technology,automation technology,especially advanced machine tools,sensors,data acquisition devices and other intelligent components,mechanical equipments and production systems are also becoming more and more intelligent,so that the reliability of equipments and production systems is becoming more and more demanding.Prognostics technology can provide health information of mechanical equipments in advance,which is considered as key technology to ensure efficient and reliable operation of mechanical equipments.Therefore,prognositics technology,especially the multifeature fusion based progonostics method can perform performance prediction before failure happens,thus advanced intelligent maintenance strategy can be achieved which has great theory significance and application valure for near-zero down operation.This thesis focuses on the key technology of multi-dimension feature fusion based prognositics and maintenance strategy.Based on the feature extraction results of time domain features,frequency domain features and time-frequency domain features,a feature selection method is proposed based on improved dynamic genetic algorithm.The method obtains candidate features by changing the length of the chromosome and the dynamic search strategy of real-time change to realize the selection of multi-dimension features.Then,feature dimension and feature fusion are conducted using Principal Component Analysis(PCA)method.A performance assessment method based on Support Vector Machine(SVM)is established by using the joint features,and a staged performance evaluation program is designed.Aiming at identifying early rotor unblance,the prognostics of rotor unbalance is analyzed.Experimental system for rotor unbalance identification is established.Vibration and displacement signals of the status with normality,slight unbalance,moderate unbalance and severe unbalance are collected.Then for each status,shaft orbit features are extracted as time domain features and time-frequency domain features are extracted using both wavelet transfomation(WT)and wavelet packet transfomation(WPT).Then multidimension features named as joint features are constructed using all the extracted features.The improved dynamic genetic algorithm-based feature selection scheme and PCA based feature fusion method are implemented to acheve feature dimension.The SVM based prognosis results verify the validity of joint features compared with time-frequency domain features extracted from vibration signal only.The upcoming industrial big data era has raised new requirements and challenges for fault prognositics technology.The value characteristic of industrial big data can be reflected in its comprehensive information.However,it is inevitable that information is filtered and selected artificially during the process of feature extraction and feature selection in traditional prognositics technology.In order to solve the problem of possible effective information loss during the procedure of feature extraction and feature selection,an intelligent prognosis method based on deep learning model is proposed and implemented on a rotating equipment to evaluate its effectiveness.Multi-source heterogeneous data including structured data such as vibration signal and unstructured data such as shaft orbit figures of the aforementioned rotor under the status with normality,slight unbalance,moderate unbalance and severe unbalance are firstly obtained.Then data structurization is conducted and the structurized data are used as input of the deep learning model.Through training of the deep learning model,a nonlinear mapping relationship from data with information without deletion to device performance is established.The experiment results show that the deep learning model based prognosis method has promising accuracy,which provides support for equipment prognosis in industrial big data environment.Based on the prognosis results,an intelligent algorithm-based maintenance decision making strategy is proposed.Firstly,the maintenance strategy of a single equipment is studied.Then research on the maintenance decision method of multimachine system is carried out by introducing the idea of opportunistic maintenance to decision making process and group technology to devision of equipment that need to be maintained,which solves the problem of when the system performs maintenance and what kind of maintenance is required for each maintenance activity.This method reduces the scale of maintenance scheduling problems,improves maintenance efficiency,and thus reduces maintenance costs.Finally,hybrid genetic algorithm is used to schedule the maintenance units.