A hybrid model combining first-principles and data-driven models for on-line condition monitoring

We present an online anomaly detection technique using a hybrid method which combines first-principles (physical) models with data-driven (empirical) models. We use an error propagation scheme for computing the output variance of the proposed hybrid model, which utilizes the input and output measurement errors along with modeling uncertainty. This on-line model output variance estimation technique is used in a statistical test to determine whether observed output measurements are statistically too distant from expected output values (for given inputs) as to declare that an anomaly has occurred. The performance of the proposed error-propagation approach for anomaly detection was successfully tested in a simulated experiment.