Design And Implementaion Of Big Data System For Fault Predicton In Petrochemical Industry

In the field of petrochemical industry,the loss caused by unforeseen facility failures might be enormous.On one hand,the failures might affect the normal operation of enterprises;on the other hand,the failures of some high temperature and high pressure equipments might even cause casualties.Therefore,it is necessary to monitor and control the equipment failures in the petrochemical systems.With the development of big data technologies and applications,it has become a tendency in the field of the petrochemical industry to monitor the status of the equipments,predict failures,earn the time for the fault treatment and prevent maj or accidents.To solve the above problems,the data cleaning methods based on sliding window dectection and Pearson correlation coefficient are proposed firstly.On one hand they help to avoid side effects of noise data;on the other hands they address the high dimension problem of the data analysis and increase the data analysis efficiency.Then,two fault prediction methods based on system health model and GAN and LSTM are brought forward.The system health model based method can deal with the cold start problem when there is little data and GAN and LSTM based approach can reach a better prediction result when the amount of data is large.System health model based fault prediction approach is based on principal component analysis(PCA).The naive PCA based approach dose not take historical data into consideration and indicators of it are potentially inconsistent.Therefore,a system health score algorithm is proposed.This algorithm not only takes historical score into consideration,but also indicates the key metrics which lead fault.The GAN and LSTM based apporach firstly proposes a fault graph,which can visualize the original data.Then,a Generative Adversarial Network(GAN)is applied to deal with the imbalanced dataset.Moreover,the hamming distance between graphes is computed and makes it possible to evaluate the performance of GAN.After that,a Long Short-Term Memory(LSTM)network is applied to transform the fault prediction problem into a multi-classification problem and precisely predict the fault.By integrating the key technologies mentioned above,the big data system for fault predictionin petrochemical industry is designed and implemented.This system could predict the faults of the equipments in petrochemical industry and indicate the key factors which might cause the failures.And this system is verified by test cases and running in factory.