Feature Assessment for a Hybrid Model

This paper proposes an assessment of features orientated to improve the accuracy of a hybrid model (HyM) used for detecting faults in a heating, ventilation, and air conditioning (HVAC) system. The HyM combines data collected by sensors embedded in the system with data generated by a physics-based model of the HVAC. The physics-based model includes sensors embedded in the real system and virtual sensors to represent the behaviour of the system when a failure mode (FM) is simulated. This fusion leads to improved maintenance actions to reduce the number of failures and predict the behaviour of the system. HyM can lead to improved fault detection and diagnostics (FDD) processes of critical systems, but multiple fault detection models are sometimes inaccurate. The paper assesses features extracted from synthetic signals. The results of the assessment are used to improve the accuracy of a multiple fault detection model developed in previous research. The assessment of features comprises the following: (1) generation of run-to-failure data using the physics-based model of the HVAC system; the FMs simulated in this paper are dust in the air filter, degradation of the CO2 sensor, degradation of the evaporator fan, and variations in the compression rate of the cooling system; (2) identification of the individual features that strongly distinguish the FM; (3) analysis of how the features selected vary when components degrade.