AI-Driven Anomaly Detection in Component Burn-In & Screening
“In space missions, electronic parts are tested under extreme heat to weed out weak ones. Currently, parts that stay just within safety limits are passed, but many have hidden flaws and fail later; this project builds an AI system to spot these ticking time-bomb components early by analyzing their behavior over time.”
Deconstructed Problem Blueprint
WHAT TO BUILD?
An AI-driven software platform and analytical engine featuring two core modules: a Dynamic Outlier Detection system and a Time-Series Drift Predictor.
WHO IS IT FOR?
ISRO Quality Assurance (QA) inspectors, reliability engineers, and component screening labs testing electronic hardware payloads for space missions.
WHY DOES IT EXIST?
Standard pass/fail testing relies on absolute datasheet boundaries. Components can drift abnormally compared to their manufacturing lot yet still sit inside absolute limits, causing catastrophic in-orbit payload failures.
INPUT → PROCESS → OUTPUT ARCHITECTURE PIPELINE
Time-series parametric test readings including standby current (Iddq), leakage currents, and propagation delays recorded at multi-stage intervals (0h, 24h, 96h, 168h) along with lot metadata.
Lot-level statistical baseline analysis, dynamic thresholding, unsupervised clustering for outlier detection, and time-series regression forecasting to map 0h/24h data to expected 168h values.
An interactive QA dashboard flagging high-risk latent defect components, automated safety-slope breach alerts, predicted 168h drift values, and SHAP/LIME-based explainability reports for inspectors.
CONSTRAINTS
- Zero false negatives tolerance for critical space components
- High reliability and determinism required in predictions
- Explainability must accompany every ML classification for compliance audits
- Must handle missing or irregularly spaced interval readings gracefully
JURY METRICS
- False Negative Rate (minimizing missed defective parts)
- Mean Absolute Error (MAE) between predicted 168h values and actual ground truth
- Model inference speed and interpretability score by QA inspectors
- Dashboard UX intuitiveness and responsiveness
CORE BOTTLENECK
Building an accurate time-series drift predictor using very early stage inputs (0h and 24h) to forecast a long-range state (168h) without overfitting on small lot sample sizes.
Why You Should Choose This PS (Pros)
- Extremely high national strategic value for ISRO and space-grade reliability
- Clear, modular problem scope with distinct requirements (Module A & Module B)
- Strong mathematical and machine learning foundation that allows for brilliant visualization and UI design
Challenges & Risks to Watch (Cons)
- Lack of direct raw space component datasets provided natively in the description requires generating synthetic realistic time-series data
- High penalty for false negatives means models must be carefully tuned
- Domain-specific terminology requires domain research to sound authentic to space engineers
Recommended Tech Stack & Grand Finale Features
Killer Features to Impress the SIH Jury
View Official SIH 2026 Portal Raw Text▼
Background In high-reliability sectors (like space) electronic components undergo rigorous environmental stress screening (ESS), including Burn-In testing (operating components at elevated temperatures, e.g., 125°C for extended periods).
Traditional screening relies on static parametric pass/fail limits. However, 'latent defects'—components that pass the absolute limits but exhibit subtle, anomalous drift over time—often escape into final payloads, leading to catastrophic field failures.
Description Development of a predictive machine learning model that analyzes time-series parametric data (e.g., standby current Iddq, leakage currents, or propagation delays measured at intervals like 0h, 24h, 96h, and 168h to detect anomalous components.
Expected Solution Module A: The outlier detection system Static limits catch obvious failures. Participants need to develop a 'Dynamic' outlier detection system. If a lot has an average leakage current of 10µA, a part showing 45 µA is a massive anomaly, even if the absolute datasheet maximum limit is 50 µA.
Module B: Time-Series Drift Predictor Build a predictive regression model that takes Value_0h and Value_24h as inputs and forecasts Value_168h. If the predicted 168h drift rate exceeds a calculated safety slope, the system flags the component for early rejection.
Evaluation Metrics: