Multi-model Condition Monitoring And Fault Prediction Base On Data Fusion

In rescent years, with the development of computer science and artificial intelligence technology, data fusion has been more and more important for research staff. While in industrial control facet, traditional non-parameter model method often cound not react to all kinds of faults in time because of the implicit limitation, the complexity of actual industial system and the multiplicity of faults. On the basis of multi-sensor data fusion technology, I combine data fusion with system status monitoring technology based on model to advance the concept of multi-model fusion. And by taking advantage of the redundance of several models, the reliability of complex system fault prediction can be improved. In this paper, I will introduce the normal knowledge of data fusion technology and several common fusion algorithms, and emphatically depict how to use Bayesian algorithm to achieve multi-model fault prediction. In order to testify the validity of the fusion algorithm, I test two sorts of data, one from the fluid catalytic cracking unit (FCCU) of petrochemistry industry, another from an industrial process fault finding trainer (Feedback 34-250). The result of test has proved using muiti-model to monitor complex system can effectively improve the reliability. In addtion, in this paper several involved methods of non-parameter model monitoring are introduced including Multivariate Statistical Process Control (MSPC) and Neural Network, and detailedly describe how to use Neural Network theory to establish models to monitor system status and predict faults.