Abstract

The objective of this work is to design a novel evolving fuzzy prediction interval for modeling a nonlinear system and to implement a novel interval-based algorithm for fault detection. To achieve these goals, three frameworks are studied to provide the basis for the ideas to be proposed. First, the prediction intervals developed in the literature to characterize the uncertainties in modeling systems. Second, the evolving intelligence systems presented in the literature to improve the predictive models through adaptability to temporal changes. Third, the model-based error detection algorithms that detect anomalies in system behavior based on the error of the models. This work proposes a complete recursive design of the evolving fuzzy prediction interval based on interval coverage estimation. This estimation is based on the accumulated number of past time points in which the interval contained the measurements of the modeled system. Additionally, a new interval-based fault detection algorithm is proposed in this work that determines the activation of alarms based on the increment of the proposed interval failure index. This novel index is similar in structure to the interval coverage estimation. The proposed recursively evolving fuzzy prediction interval and interval-based fault detection algorithm were tested on a heat exchanger system whose characteristics were affected by external changes. The experimental results show the effectiveness of the proposed interval model in representing the dynamics of the heat exchanger. Furthermore, the experiment shows that the proposed fault detection algorithm achieves comparable accurate performance to the results of a previous model-based fault detection method. COPY; 2023 Elsevier B.V. All rights reserved.