Remaining Useful Life Prediction of Industrial Consumables Using Wideband Vibration Signals

Abstract

Predictive maintenance (PdM) solutions providing remaining useful life (RUL) of assets in manufacturing plants are increasingly being explored. One of the key challenges in deploying PdM solutions is the non-invasive acquisition of necessary operational data since most legacy assets lack required telemetry. Vibration analysis is one of the promising approaches for monitoring operational aspects of an asset. But this approach, too, has its own challenges arising due to practical constraints such as lack of an ideal mounting area for the sensor. The increased separation between the ideal observation point and the actual observation point leads to signal distortion due to induced mechanical resonances as well as other interfering noise. We present a novel approach for estimating RUL of industrial consumable parts such as cutting tools that overcomes these challenges. The proposed method systematically locates the frequency band of interest containing the cutting tools’ signal among the distorted wideband vibration data by utilizing the cyclic nature of the CNC machine’s operation. The extracted feature values show expected monotonic trends of wear and tear with respect to the usage of cutting tools making accurate RUL estimation for PdM solution possible.