This thesis presents an edge-based predictive maintenance system for bearing fault diagnosis using vibration analysis and deep learning. The work focuses on the CWRU bearing dataset, where time-domain vibration signals are converted into cepstrum features and classified with a 1D-CNN into four conditions: Normal, Ball, Inner Race, and Outer Race. The model is then deployed in a Dockerized edge–cloud architecture using FastAPI, a sensor simulator, and an interactive Dash dashboard for real-time monitoring