KPIT developed a hybrid approach to overcome the shortcomings of existing individual methods for SOC and SOH estimation. It combines a battery model and a neural network to predict SOC and then uses the obtained SOC to derive SOH parameters. Deep Learning Toolbox™ and MATLAB^® were used to train a feedforward neural network, which was then extensively validated for robustness. The neural network was then incorporated into Simulink^® and deployed to a PowerPC-based embedded platform using Embedded Coder^® and AUTOSAR Blockset. This workflow has been validated on multiple datasets for LFP and LCO chemistry. It has provided SOC and SOH estimation with improved accuracy well within +- 5% consistently over a different driving cycle range.