This chapter presents an AI-enabled framework for bearing fault diagnosis using vibration signals. The approach achieves an accuracy of 96.84% over several fault classes by employing a Convolutional Neural Network-Attention model with some form of explanation capability. With the aid of explainable artificial intelligence (XAI) techniques, the framework demonstrates that the most important features involved are those with the greatest value of FFT Max and Kurtosis, which correspond to characteristic frequencies and impulsive behavior associated with fault type discrimination. This study illustrates the importance of adding explainability to predictive maintenance systems with an industrial focus that impacts how fault diagnosis systems are designed, enhancing strategic reliability.