Back to projects
Automotive Health Monitoring System

Automotive Health Monitoring System

Niket Girdhar / April 8, 2026

This is our Capstone Project

Project Overview:

This project presents an interpretable AI-driven predictive diagnostics system for automotive and industrial machinery using multimodal data sources: telemetry signals and acoustic signals.

Telemetry data (temperature, pressure, sensor trends) was used to estimate Remaining Useful Life (RUL), while acoustic signals were used to detect early mechanical faults through anomaly detection. A final fusion model combines both outputs for smarter maintenance decisions. :contentReference[oaicite:0]

Key Highlights:

  • Built Attention-LSTM models for RUL prediction using NASA CMAPSS datasets (FD001–FD004).
  • Implemented Mixture of Experts (MoE) specialist routing for complex fault conditions.
  • Developed a 4-stage acoustic anomaly detection pipeline using Autoencoders on MIMII fan recordings.
  • Used Mel-Spectrograms, Sliding Windows, and Asset-specific Z-normalization for robust acoustic preprocessing.
  • Designed a Linear Meta-Learner Fusion Model combining telemetry and acoustic outputs.
  • Applied SHAP Explainability to validate model reasoning and maintenance logic.
  • Achieved strong benchmark performance:
    • FD001 RMSE: 14.70
    • Acoustic ROC-AUC improved from 0.59 → 0.83
    • Fusion RMSE improved from 41.51 → 39.13 :contentReference[oaicite:1]

Technical Architecture:

1. Telemetry Layer

Used Attention-LSTM networks to estimate RUL from engine sensor sequences.

2. Acoustic Layer

Used Autoencoders to detect abnormal machine sounds using reconstruction error.

3. Fusion Layer

Combined RUL predictions + anomaly score for final maintenance decision.

4. Explainability Layer

Used SHAP values to understand feature impact and ensure trustworthy AI outputs.

Impact:

  • Reduces unexpected machine failures.
  • Enables predictive maintenance scheduling.
  • Improves operational safety.
  • Lowers downtime and maintenance costs.
  • Supports Industry 4.0 smart manufacturing systems. :contentReference[oaicite:2]

This project significantly strengthened my expertise in Machine Learning, Deep Learning, Time-Series Forecasting, Audio Signal Processing, Explainable AI, and Industrial AI Systems.

Project Collaborators:

  • Guide Prof. Ralph Samuel Thangaraj
  • Shreya Gantayat
  • Niket Girdhar (me)
  • Suryakiran R