Compressed Sensing Based on an Improved K-SVD for Vibration Signal Compression Reconstruction in Wireless Sensor Networks

Compressed sensing (CS) is an effective method to improve the processing of mechanical vibration signals in prognostics health management (PHM). Aiming to address the severe lack of storage and computational resources and the high delay of transmitting massive vibration data in wireless sensor networks (WSNs), a compression reconstruction method is proposed based on an improved k-singular value decomposition algorithm (K-SVD). First, a substantial amount of vibration signals are processed through the K-SVD dictionary training to obtain a highly comprehensive dictionary. Second, the sparse adaptive matching pursuit algorithm (SAMP) is introduced in the sparse coding part to estimate the sparsity of the vibration signal quickly, which provides a significant advantage in terms of reconstruction accuracy. Finally, the alternating direction method of multipliers (ADMMs) is introduced to speed up the optimization process of dictionary training. The experimental outcomes show that the suggested approach displays enhanced sparsity and reconstruction efficacy in comparison to other techniques under identical compression ratio (CR) circumstances.