BatteryIQ - RUL Prediction

BatteryIQ-dashboard

Inspiration

Battery failures are a major issue in electric vehicles and energy storage systems. Unexpected degradation leads to safety risks, high costs, and increased electronic waste. We were inspired by the need for a smarter system that can predict battery life early and support sustainable energy usage aligned with UN SDG 7 and SDG 9.

What it does

BatteryIQ predicts the Remaining Useful Life (RUL) of lithium-ion batteries using deep learning.

Takes battery parameters like voltage, current, temperature, and capacity Extracts advanced Incremental Capacity (IC) features (dQ/dV) Uses a Bidirectional LSTM model Outputs accurate RUL predictions in cycles

It also provides: Degradation trends Prediction graphs Early warning signals (20–30 cycles earlier than traditional methods)

How we built it

We developed a complete 6-layer pipeline:

Data: NASA Li-ion dataset (B0005–B0018)

Preprocessing: IC curve computation RUL labeling (70% EOL threshold) MinMax scaling and sliding window Feature Engineering: 11 features including IC peak metrics

Model: BiLSTM(64) → LSTM(32) → Dense layers Trained on 3 cells, tested on unseen cell

Backend: Flask API (Colab) REST communication Frontend: Interactive UI for predictions

Challenges we ran into

Extracting clean IC curves from noisy battery data Handling nonlinear degradation near end-of-life Avoiding overfitting with limited dataset (only 4 cells) Generalizing model across different battery cells Deploying ML model via Colab and ngrok reliably

Accomplishments that we’re proud of

Achieved high accuracy (~2–5% error) Integrated IC analysis, a research-level feature Built complete end-to-end pipeline (data → UI) Successfully generalized model to unseen battery Early prediction capability (before visible degradation)

What we learned

Deep learning models like LSTM are powerful for time-series degradation Feature engineering (IC curves) is as important as the model itself Real-world data is noisy and requires careful preprocessing Deployment is as challenging as model building Small datasets require smart generalization strategies

What’s next for BatteryIQ - RUL Prediction

Real-time integration with Battery Management Systems (BMS) IoT-based live battery monitoring Explainable AI for better interpretability Support for different battery chemistries