The Study Of Sensor Condition Monitoring System Based On Online Sequential Extreme Learning Machine

The sensor as the basis of the test and control system collects data for system,the output results of the monitoring have a significant impact on many other aspects of monitoring and decision,such as subsequent data analysis,processing and system operation..And if the sensor is abnormal,the system will lose the ability to monitor,or it may even cause incalculable loss.Therefore,the sensor failure or not,whether the input and output characteristics has been changed,is essential for the entire system.In this paper,the following studies are carried out on the abnormal conditions that may occur in the process of using the sensor:First,building the experimental platform.Based on the water pump system of the water plant,the piezoelectric three axis vibration sensor is selected as the research object,and the representative vibration acceleration signals are collected in the complex environment.And cleaning,sampling,integrated processing this data,then stored it in the database for query and follow-up analysis.Secondly,established the sensor state classification mode.According to the abnormal potential sensor,the sensor state are classified in four cases,as safety state,interference state,impact state and bias state,and describe the reaction state of the sensor;It described the feature extraction method of packet decomposition is used to decompose with three-scale,and the vibration signal is described with the form of energy features in time domain and frequency domain.Then,the classification model of vibration sensor is established by using support vector machine,extreme learning machine and online sequential learning machine algorithm.Aiming at the over fitting phenomenon of support vector machine,the system improves the generalization ability combined with the cross validation method.In the extreme learning machine,comparing the difficult choices of the number of hidden layer neurons and the activation function to find optimal parameters.According to the actual project,establishing online sequential extreme improvement machine learning classification model.The three kinds of classification models are tested and compared.The results show that the results of the online sequential learning machine model are better than the first two.Finally,the on-line monitoring system of vibration sensor is designed and implemented.The virtual instrument Lab VIEW is combined with DAQmx and MATLAB software to realize the real-time data acquisition and storage,It can display the sensor's current state,the warning of abnormal state and the query function of historical data.The results show that the design of the sensor state classification model and vibration sensor on-line monitoring system provides theoretical reference and guidance for the design of sensor for real-time monitoring,maintenance and management,and has far-reaching application value.