Multi-Task Compressive Sensing of Vibration Signal using GMM Clustering for Wireless Transmission

Wireless sensor network has been widely used in the prognostic and health management system. Signal compression is often required during wireless transmission because of the high sampling frequency. In recent years, compressive sensing has great potential to increase the efficiency of wireless transmission. In this paper, we use the multi-task compressive sensing within a multi-task learning setting. Under this modeling, data from all tasks contribute toward inferring a posterior on the parameters. Moreover, a GMM clustering method is used for finding the best classification among signal blocks. Simulation results show that more signal blocks may not necessarily lead to higher reconstruction accuracy. Conversely, reconstruction error may be reduced by concentrating signal blocks with similar distribution together. The performance of proposed method is validated by gearbox data.