Data driven models for fault detection - Combining thermal and indoor air quality grey box models

The progressive digitalization is providing more and more measurement data from building operation, in particular from heating, cooling and ventilation (HVAC) systems. This work investigates the potential use of data-driven models to simulate indoor environmental conditions, i.e. temperature and CO2 concentration, for fault detection applications. Herein, a grey-box model, depicting the thermal behaviour of building zones, is coupled with model representing the indoor air quality/ventilation condition in the respective zone allowing the combined use of measurement data from building operation. The models are applied to an office room of a case study building and the model parameters are identified with measurement data for a four-weeks long training period. The identified models are used to predict the timely evolution during a three-day long prediction period. By comparing residual metrics between training phase and prediction phase the model’s capabilities to detect simple exemplary faults are evaluated. Herein, preliminary results with rather simple fault cases, like temperature or CO2 sensor faults or (unintentionally) left-open windows are investigated. Results indicate that indoor temperature anomalies are detected well and that anomalies in CO2-concentration are also detectable with this modelling approach but depend on the available occupancy estimation (or measurement). Further investigations are underway to test possible adaptions to the presented approach to allow for better occupancy estimation and/or account variable ventilation rates.