Sensor drift detection by utilizing multi-sensor signals

Sensor accuracy plays a crucial part in the reliability of sensing applications. Calibration is essentially necessary for accuracy of the sensors. In order to reduce the cost of calibration, it is required that drifting sensors can be identified so that only the detected drifting sensors will be calibrated. This paper presents sensor drift detection by using Adaptive network-based fuzzy inference systems (ANFIS). In the experiments, several types of sensors are installed in the same area. The relationship among those sensors are modeled using ANFIS. The models are then used for detecting drifting sensors in the system. Sensors that drifted greater than 10% have successfully and reliably been detected. For drift less than 10%, experimental results show slightly lower reliability in detection.