Process Monitoring And Fault Diagnosis For Naphtha Cracking To Ethylene Production

Ethylene is one of the most important industrial chemical.The level of ethylene production has become an important indicator to evaluate the petrochemical industry development of a country.In present petrochemical industry,ethylene is mainly produced by naphtha cracking process.Inside cracking furnace,coils with paralleled arrangement are main field for cracking reaction.Serious safety accidents will be caused by furnace partial faults or operation parameters deviation among different coils..The parameters of the cracking process are adjusted several times for entire production cycle.Each adjustment will produce a new steady condition.The multiple operation condition monitoring and fault diagnosis of the cracking process is one of the research issues.In this work,industrial data characteristics are analyzed.Cross-correlation between variables is found similar among different operating conditions.A process condition identification method is proposed based on wavelet filtering and moving average method,which can effectively identify the active process adjustments and passive fluctuations.In order to timely identify the process fault under multiple operational conditions,a method based on Principal Component Analysis(PCA)is proposed to realize on-line fault identification and diagnosis.The true positive rate of this method is 98%and its false alarm rate is 1.6%.During the naphtha cracking process,coke is continuously generated and attaches to coils inner surface,which gradually reduces the heat transfer performance and increases the coils surface temperature.When the coil surface temperature approaches the material limit,decoking operation must be performed to prevent coil rupture accident.Real-time measurement of coil surface temperature cannot be achieved due to high temperature inside the furnace.In addition,the relative content of key components in cracking products is directly related to the profit of ethylene plant.However,real-time component data cannot be obtained due to the limitation of detection methods.In this work,soft sensor models based on Partial Least Squares(PLS)algorithm are established to achieve real-time prediction of coil surface temperature and relative content of product components.The prediction average relative error of the coil surface temperature and the key components relative content are under 1.1%.