A hybrid model-based fault detection strategy for air handling unit sensors

For the air handling unit (AHU) sensor fault detection, the classical fault detection methods based on statistical residual evaluation are difficult to detect small bias fault especially under noisy conditions. On the other hand, a novel technique using fractal correlation dimension (FCD) algorithm can identify a tiny variation of curve fractal characteristic but needs a period of time. To balance their strengths as well as weaknesses, a hybrid model-based fault detection technique is developed by combining these two methods. The simulated data obtained from the TRNSYS simulation platform are used to validate the hybrid fault detection strategy. And a prediction model using support vector regression (SVR) is developed to obtain the fault-free references. Under a noise ranging from −0.3 °C to +0.3 °C, the technique is validated to detect six fixed biases of the supply air temperature sensor under three different load conditions. Given a specified threshold, the hybrid technique can identify the large bias faults such as ±0.5 °C by statistical residuals and can detect small faults of ±0.2 °C by FCD deviations. For the dynamic and nonlinear systems, FCD-based approach is more suitable for fault detection than for fault diagnosis.

► A hybrid fault detection technique combines statistical residual with fractal correlation dimension.
► A support vector regression model is used to predict the fault-free references.
► The technique is tested to detect six fixed biases of temperature sensors.
► The hybrid technique can identify large bias faults and can also detect small faults.