Agglomeration Fault Monitoring Of Polyethylene Fluidized Bed Based On Acoustic Signal

Polyethylene is one of the important products in petrochemical industry.The gas-phase polyethylene process is widely used in industrial production due to its performance advantages,and the gas-solid fluidized bed is the core reactor of the polyethylene process.During polymerization,polymer particles may agglomerate together due to various reasons.Agglomeration will continue to affect the reaction state and in severe cases,will let the production cut down.Therefore,the gas-solid fluidized bed monitoring is necessary,which can keep the device running continuoursly by detecting agglomerations timely and processing it early.In the existing monitoring methods,the acoustic method is safe;reliable and easy to implement,and has become a relevant research hotspot.This paper researches the monitoring method of agglomeration fault of polyethylene fluidized bed reactor(FBR)based on acoustic signal.Firstly,a FBR acoustic signal acquisition system was designed and the hardware construction and corresponding software programs were completed.The system uses four acoustic sensors to monitor the particle size of polymers in the reactor.The field monitoring terminal completes the adjustment and collection of the acoustic signal and transmits it wirelessly to the monitoring room.The data is displayed and stored on the monitoring computer in the monitoring room.Based on above data,the acoustic signal feature extraction method was studied.The time domain,frequency domain and entropy features of acoustic signals were compared and analyzed.A weighted principal component analysis method based on ReliefF(R-WPCA)was proposed for feature extraction.R-WPCA method can find the principal components that help to distinguish the agglomeration,and realize the dimension reduction while removing the correlation.In the actual sampling process,there are a large number of unlabeled samples,while the number of fault samples is scarce.Aiming at these problems,a fault diagnosis method based on Active Learning Support Vector Machine(ALSVM)was proposed.In ALSVM,a performance was designed firstly to to select samples that are most helpful in improving the classification accuracy.After being labeled,these samples were used in the support vector machine to obtain a diagnostic model and complete the construction of a fault monitoring system.The monitoring system was tested on a pilot plant.Based on the experimental results,the system was testified that can detect agglomeration faults 12-85 minutes earlier than the traditional method based on pressure and tempreture signals.Furthermore,the alarm results were not affected by normal operation interference.It can be concluded that the monitoring system can operate stably,and the feasibility and reliability of the system have been verified too.