Design And Implementation Of Equipment Failure Early Warning System Based On Data Mining

The petrochemical industry has extremely stringent requirements for the generation of safety and stability.China National Petroleum Corporation(CNPC)attaches great importance to equipment management and fault early warning in pipeline projects in Central Asia.In order to discover abnormalities and failures of tracking equipment in a timely manner,it is important to develop a system that can effectively alert equipment failures and abnormalities.This project is an equipment failure warning system for China-Central Asia Pipeline Project of CNPC under this background.The project uses the China-Asian project to accumulate the equipment operation parameters for many years.Based on the existing network architecture,the system realizes the collection,cleaning and dump of the equipment parameters distributed throughout the area without the premise of large network transformation.Through the analysis of the equipment parameters,the system realizes the forecast of the short term operation trend and the early warning of faults and anomalies.This paper mainly introduces the design and implementation of equipment fault early warning system.Users can achieve short and medium term prediction and early warning of equipment running trend through this system.The system is based on the ARIMA algorithm to predict the device parameters.Before the data is predicted,the data needs to be processed through the ETL module,which mainly includes data collection,cleaning,and dumping.Because of the special environment of the system and the wide distribution of log files,the system contains path management functions to assist in the realization of ETL functions.In this system,the ARIMA model is used to predict and analyze the equipment parameters in the middle and short term,and the prediction results are evaluated.After that,the similarity matching between the fault section and the forecast data generated by the fault records is matched,and the fault warning information is finally formed.Users can submit applications for the mid-term(7 days)and short-term(24 hours,48 hours)equipment running trend prediction,the system can detect possible anomalies and failures in the operation and send out alarm in time.This paper introduces the modeling process and evaluation method of ARIMA in detail.After the prediction information is obtained,the similarity matching between the prediction information and the fault information is needed.This paper uses the Pearson correlation algorithm and the exponential smoothing method to achieve the fault information matching function.After testing,forecasting and matching works well.In addition,the system also provides the user with the model customization function.Users can set up a new set of parameter collection and cleaning rules for the corresponding equipment by setting ETL rules.Then they can get the prediction model they want through model training and evaluation.In the pre-commissioning stage of the system,the system is stable in operation,with good prediction effect.The entire system architecture is designed reasonably,the resource consumption is small,and the response is timely,achieving the design goals in the initial stage of system development.