In this paper, we aim at improving the change detection techniques by introducing an adaptive thresholding with a sliding time window. In particular, a real-time optimal sliding time window length is implemented without any preliminary learning step as required in conventional sensorfault detection methods. Based on Shannon's entropy, our method improves the change detection techniques using an adaptive thresholding. The technique can be applied by any change detection technique based on the generalized likelihood ratio (GLR). To validate the robustness of our approach, two commonly used change detection techniques are considered: the cumulative sum (Cusum) and the exponentially weighted Moving average (EWMA) control charts. Experimental validation is experimentally shown considering real data in the context of collaborative mobile robots. In addition, this experiment leads to a fault-tolerant fusion methodology based on the use of an extended Kalman filter (EKF).