Sensor fault diagnosis and location for small and medium-scale wireless sensor networks

One of the prerequisites for the normal operation of a small and medium-scale wireless sensor network is the ability to diagnose sensor node failures. Considering the characteristics of small scale, low-density node distribution, stable network topology and the fact that each node of the small and medium-scale wireless sensor network can be monitored, a method for detecting and identifying sensor node failure based on Principal Component Analysis (PCA) and wavelet decomposition is proposed in order to improve network stability and reliability. The principal component analysis model is established utilizing historical data and the data collected by sensor nodes are decomposed into principal space and residual sub-space. A faulty sensor can be detected in real time by analyzing the Square Prediction Error (SPE) index and the faulty sensor node identified by extracting high-frequency coefficients of wavelet decomposition. Experimental data indicate that this method of sensor node diagnostic analysis functions in real time operates with low communication overhead and is well adapted for use in small and medium-scale wireless sensor networks.