Early transient fault diagnosis of distillation column based on principle component analysis and adaptive neuro-fuzzy inference system

A novel online algorithm for early fault detection and diagnosis based on statistic method and adaptive neuro-fuzzy inference system (ANFIS) is developed. Principal component analysis (PCA) was used to extract feature vectors of data set of complex chemical plant. The most superior features are fed into ANFIS to identify different abnormal cases. Ability and at the same time the simplicity and rapidity has significantly enhanced. Furthermore the advantage is that no model or structural information about the system is needed. This proposed approach has been implemented on a simulated nonlinear MIMO distillation column.

1 INTRODUCTION

Increasing complexity of modern industrial systems and the high level of process quality, reliability and safety requirements force the automation of diagnostics in order to make it possible to determine the place, reason and time of the fault accurately ^[1]. Different methods of fault diagnosis have been developed and used effectively to detect the complex chemical process faults at an early stage [2]. Early detection of faults can be achieved by model-based fault detection. These methods are based on residual generation by comparison of the estimates of the measured signals with their originals. By model based fault diagnosis method prompt fault detection requires accurate models of processes and leads directly to this problem of system identification. Real processes are usually dynamic, nonlinear and stochastic, and analytical approaches of identification are rarely suitable for them.