Effective Battery Management Systems (BMS) are crucial for the safe operation, longevity, and optimal performance of batteries in renewable energy and electric vehicle applications. However, accurate real-time estimation of the State of Charge (SOC) and timely fault detection remain challenging due to the nonlinear behaviour of batteries and varying environmental conditions. Conventional methods often suffer from limited accuracy and slow fault diagnosis, which can lead to reduced battery life and system reliability. This paper proposes a hybrid Battery Management System (BMS) approach that combines convolutional neural networks (CNN) with the extended Kalman filter (EKF) to enhance SOC estimation and facilitate early fault detection. The CNN component learns complex non-linear battery dynamics and identifies anomalies from historical data, while the EKF dynamically refines SOC estimates under changing load and temperature conditions. The proposed method is implemented and validated using MATLAB / Simulink simulations, demonstrating enhanced accuracy and faster response compared to traditional approaches. Future work will focus on experimental validation of the proposed framework using real lithium-ion cells and hardware-in-the-loop setups to bridge the simulation-to-reality gap and assess real-time feasibility on embedded BMS platforms.