Decentralized fault diagnosis system using ICA in a complex chemical process

Diagnosing a system fault before it deteriorates the system performance is crucial for the reliability and safety of many industrial systems. This paper develops an online decentralized fault isolation system for a complex chemical process which has several diagnosing agents for different areas of the process. Each diagnostic agent based on multiple adaptive neuro-fuzzy inference system (ANFIS) assigned to different types of fault which use salient features acquired by independent component analysis (ICA). Distributed diagnosis units based on multiple ANFIS reduce the scale and complexity of the system. This structure grants the higher redundancy of one specific fault by confirming from diagnosis-agents of different parts of process. The proposed method is quite simple to be applied to practical application which does not need a priori knowledge about the process dynamics. A simulated nonlinear MIMO distillation column is used as a benchmark problem to evaluate the proposed algorithm. Experimental results demonstrated the effectiveness of this method. This procedure can be relatively easily applicable to a variety of industrial applications in which continuous on-line monitoring, diagnosis and more valuable prognosis, are necessary.