Sensor fault detection and its efficiency analysis in air handling unit using the combined neural networks

• The basic and auxiliary neural networks were developed to detect AHU faults.
• Dual neural networks detection logic employed the weighting analysis.
• False alarm, missing alarm, and detection time were evaluated for sensor faults.

The supply air temperature control is one of the most important controls in an air handling unit (AHU). Several advanced optimal strategies have been developed and integrated into this controller to ensure better thermal comfort and less energy consumption. However, common faults occurred in the control loop may lead to the optimal control target unachievable. In this paper, the dual neural networks combined strategy is presented to detect the faults of sensors in the supply air temperature control loop of air handling unit. Firstly, the basic and auxiliary neural networks, constructed based on the control relations and the correlation analysis among variables, are developed respectively. In addition, the basic and auxiliary neural networks are combined together through allocating the weighting factors of the two neural networks using the principal component analysis. Finally, the fixed bias, drifting bias, and complete failure of sensors, and coil water valve fault are tested. And the false alarm, missing alarm and detection time of each single neural network and the combined neural networks are analyzed in this paper.