Internal model control and fault detection of time delay systems

Internal Model Control (IMC) as a scheme using parallel process model is extended to achieve an additional ability to generating residuals for the fault detection purposes. After adding a disturbance model to the IMC scheme a residual signal is obtained which can be employed for watching whether a fault in process control has happened. Artificial neural networks (ANNs) are applied to residual signal scanning and evaluation and in this way they serve for predicting its next continuation. This short-time prediction of generated residual signal enhances available sensitivity in distinguishing even small deviations from nominal operation. While a good prediction indicates faultless operation, a decay or break down of the prediction quality signifies a failure. The presented method has been applied on a laboratory-scale heating system where various failures were brought about artificially and detected.