An easy to set up residual generator based on multilayer perceptron networks and Bayesian optimisation for the application in automated fault detection and diagnosis in building systems

Automatic fault detection and diagnosis (FDD) methods are rarely used in building systems due to their individual design. We present a residual generating FDD approach combining multilayer perceptron networks trained with historical data and Bayesian optimisation for hyperparameter tuning. A comprehensive engineering process has been developed, which is highly automated and applicable by non-machine learning experts. We demonstrate the transferability using datasets from twelve different air handling units and provide an estimation of fault-free behaviour. Applied on a synthetic data set, the approach shows comparably results to a rule-based fault detection, with the advantages of less threshold tuning, detecting unknown faults, and facilitating fault diagnosis based on residuals.