The Study On Methods Of Fault Diagnosis For Petrochemical Complex System

Petrochemical production is not only the foundation of the social development but also the pillar industries of economic development, and it is playing a more and more important role in the process of modernization social development. With the increasingly complicated, modern and large-scale of the structure and the scale of the petrochemical process production, probability of production accident also increase gradually. Hence, it is crucial and necessary to implement effective fault diagnosis for petrochemical production process to prevent or avoid the happening of the accidents. In order to carry out fault diagnosis effectively, a hybrid fault diagnosis method including two aspects which are fault monitoring and fault diagnosis is proposed. And unit of taking off isobutane of gas distributary device of a petrochemical company is taken as the example of application in this paper.First of all, the fault diagnosis monitoring model based on Principal Component Analysis(PCA) method is established. Four on-line Working conditions set in the paper are monitored with PCA, and results are consistent with the setting conditions. It shows that using PCA not only can reduce the data dimension and simplify calculation greatly, but also can monitor online effectively and find fault timely.Then, fault diagnosis method based on Back Propagation(BP) neural network is established. It shows that although using BP neural network can determine the fault category which the working condition belongs to, it can not judge out the specific fault situation. Dempster-Shafer(DS) evidence theory is adopted to fuse the recognition results of BP neural network, and the results show that under the premise of determining the fault category which the working condition belongs to, using the DS evidence theory can make the right judge of specific fault situation, but the method of combination of BP neural network and DS evidence theory still exists shortcomings such as long period of time and large amount of calculation.Although Radial Basis Function(RBF) neural network is not widely used as BP neural network, its features such as classification ability, approximation ability and training speed are all better than BP network. Therefore RBF neural network is adopted to establish the fault diagnosis model. The results show that when the demanding of fault threshold is 0.85, using fault diagnosis moedl based on RBF neural network alone can determine the cause of the fault quickly under the condition that diagnostic accuracy is lower.