A Data Compression Method for Mechanical Vibration Wireless Sensor Networks Based on Data Segmentation and...

High-frequency sampling in mechanical vibration wireless sensor networks (WSNs) generates large volumes of raw data, imposing substantial demands on transmission bandwidth and node energy consumption. To address this problem, this paper proposes a two-stage compression method based on data segmentation and non-uniform quantization. First, byte-level statistical characteristics of 24-bit raw analog-to-digital converter (ADC) data are analyzed. The results reveal the limitations of conventional transform-domain thresholding and show distinct statistical characteristics among different bit planes. Accordingly, each 24-bit sample is divided into an 8-bit sign-bit channel and a 16-bit content-bit channel. The sign-bit channel exhibits high temporal redundancy and is therefore compressed losslessly using variable run-length encoding (VRLE). An adaptive repetition-ratio criterion is introduced to select the encoding mode and avoid data expansion for short repetition sequences. The content-bit channel is compressed using an eight-interval non-uniform quantization scheme inspired by the A-Law PCM standard. Finer quantization is assigned to small-amplitude components to reduce reconstruction errors. Range coding is subsequently applied to both encoded streams to further remove statistical redundancy. Experiments on periodic rotor vibration signals achieve a compression ratio of 3.22 and a reconstruction signal-to-noise ratio (RSNR) of 37.4 dB. For aperiodic modal testing signals, compression ratios of 3.16 and 2.66 are obtained for force-hammer excitation and frame response signals, with RSNR values of 46.49 and 54.2 dB, respectively. Furthermore, modal identification using the reconstructed data results in a maximum relative error of only 0.03% for the first five natural frequencies. These results demonstrate that the proposed method effectively reduces vibration data volume while preserving critical dynamic characteristics.