Enhancing Bearing Lifespan Predictions: Integrating Deep Learning and Higher-Order Spectral Analysis with Acceleration-Based Approaches

This study introduces an analytical tool for predicting the lifespan of bearings, combining machine learning and higher-order spectral analysis. The purpose of this approach is the use of acceleration as the primary parameter, capturing the essence of the bearing's functionality and its diverse states through both horizontal and vertical components. Specifically, this research delves into frequency analysis methods, centering on the Bispectrum of the acceleration signal. This technique offers an insight into the complex spectral relationships among frequencies throughout the bearing's life, by exploring the phase relationships between frequency components and unveiling the nonlinear interactions within the signal. This methodology is particularly valuable for bearing lifespan prediction, as it uncovers intricate patterns that traditional analysis might overlook. The aim is to not only elevate the precision of lifespan predictions but also to deepen the understanding of the failure mechanisms at play, thereby enhancing predictive maintenance strategies and industrial efficiency.