A neuro-fuzzy model for fault detection, prediction and analysis for a petroleum refinery

The paper describes data fusion using a neuro-fuzzy system for fault detection, prediction, and analysis of petroleum refining operations and other process industries. The model described in this paper involves algorithms applied to multi-sensor fusion using historical data to create a trend analysis. The main objective is to detect anomalies in sensor data and to predict future catastrophes. Data mining is applied to find anomalies in data sets. Neuro-fuzzy logic is used to find clusters of inputs using subtractive fuzzy clustering. Fault detection and prognosis are essential in a safety-critical environment such as a refinery. A new set of data is obtained and represented using the fuzzy inference system, with three linguistic values used to define and classify the patterns and failures