Reconstruction-Free Vibration Monitoring Using Random Pattern Reuse Sampling

Abstract

Continuous vibration monitoring requires sensor nodes to acquire and transmit high-rate data to the Internet-of-Things (IoT) endpoints, such as cloud platforms or on-premises servers, resulting in substantial communication overhead and computational burden. While compressive sampling (CS) has been widely adopted to alleviate bandwidth and storage constraints, most existing approaches rely on signal reconstruction prior to analysis. This reconstruction step introduces significant processing latency and limits real-time applicability. This work proposes a novel reconstruction-free, decision-centric CS vibration monitoring framework. The proposed method integrates random pattern reuse sampling (RPRS), where an optimal random sampling pattern is identified during an initialization phase in the digital domain, and the same set of sampling indices is reused during runtime. This strategy preserves the randomness required by CS while removing the need to transmit sampling timestamp information, thereby reducing communication overhead. Instead of reconstructing signals, machine state inference is performed directly on compressed measurements, enabling real-time decision-making with minimal computational cost. A hybrid analysis approach integrating the two-sample Kolmogorov–Smirnov (KS) test and sinusoidal curve fitting is employed to detect variations in amplitude distribution and dominant frequency components associated with different operational states. Experimental validation using real manufacturing vibration data demonstrates that the proposed framework achieves a 46.3% reduction in data transmission, approximately two orders of magnitude faster processing, and a classification accuracy of 94.43% $\pm ~0.77$ % without requiring CS reconstruction. These results highlight the effectiveness of the proposed framework as a low-latency, resource-efficient alternative to reconstruction-based CS methods, making it particularly suitable for real-time, resource-constrained IoT vibration monitoring systems.