This research combines Cox proportional hazards models with modern machine learning to accurately predict the remaining useful life of lithium-ion batteries, enabling proactive maintenance and safer energy storage.

DeepCox Technical Support

• Survival Modeling: Developed a hybrid Cox-NN architecture that integrates deep feature extraction with survival likelihood estimation.
• Time-Series Processing: Built LSTM and Transformer-based encoders for multivariate sensor data (voltage, current, temperature).
• Uncertainty Quantification: Implemented Monte Carlo dropout and Bayesian inference for prediction intervals.
• Deployment Tools: Provided a lightweight ONNX runtime for edge devices in industrial settings.