Neuromorphic Multiple-Fault Diagnosing System Based on Plant Dynamic Characteristics

In many cases, multiple-fault diagnosis of plant-wide systems based on steady-state data is impossible. To solve this problem, a new diagnosis strategy based on neural networks has been proposed. In the suggested framework, the neural network is used as the diagnoser trained by a hybrid set of steady and dynamic characteristic data of the system. The dynamic characteristic data include overshoot and undershoot values of measured variables and their corresponding occurrence times. To evaluate its performance, the proposed scheme was used in the diagnosis of the concurrent faults of the Tennessee Eastman (TE) process. Various combinations of concurrent faults were considered in this assessment. The results indicate the generality, flexibility, and accuracy of the proposed algorithm such that it is capable of diagnosing various combinations (from single to sextuple) of simultaneous faults, whereas the other diagnosing methods used for the TE process are capable of distinguishing at most three simultaneous faults.