Machine Status Tracking Using Vibration via Sparse Sampling and Without Reconstruction

One of the challenges in monitoring machinery vibration is handling the huge amount of data that must be transmitted, stored, and analyzed before transforming it into useful information. Many existing works aim at devising efficient and effective data reduction schemes to minimize this data load. Unfortunately, many of these techniques require the reconstruction of data, even in its reduced form, before proceeding with analysis. This is true even for compressive sampling (CS), the current state-of-the-art method for reducing sampling and data transfer volumes, as the vibration signal must be reconstructed before analysis can be applied. In this work, we capitalize on the nature of manufacturing machines, where vibrations often remain stationary over time and only change when the machine status changes. As such, we propose the use of a multilayer neural network and k-nearest neighbor (kNN) method to analyze the sparsely and randomly sampled data and subsequently identify the machine status without the need for data reconstruction. Experimental results demonstrate that the proposed solution can track a machine's status with an accuracy of 99.88%, using only 15% of the vibration data.