Page 1 of 7 · Infotact Solutions Internship · IoT Edge AI

Manufacturing & Automotive — Contextual Predictive Maintenance

AI4I 2020 Dataset · 10,000 IoT machine cycles · Real-time edge failure intelligence

Machines Monitored
10,000
Total cycles in dataset
Total Failures
339
Across all failure modes
Failure Rate
3.39%
Overall machine failures
Uptime Rate
96.61%
Non-failure cycles
Failure vs Non-Failure Distribution
Failure Count by Machine Type (H / L / M)
Failure Mode Breakdown — TWF · HDF · PWF · OSF · RNF
Failure Rate % by Machine Type
Page 2 of 7

Sensor Monitoring

Live telemetry across 200 machine cycles · Failure events marked with red triangles

Air Temperature (K)
Process Temperature (K)
Rotational Speed (RPM)
Torque (Nm)
Tool Wear (min) — Failure events highlighted ▲
Page 3 of 7

Feature Engineering Analysis

Engineered features: Temperature Difference · Power Estimate (RPM×Torque) · Tool Wear Rate

Temp Difference Distribution (K)
Power Estimate Distribution
Tool Wear Rate Distribution
Correlation Heatmap — All Features × Machine Failure
Page 4 of 7

Model Performance Dashboard

Logistic Regression · Random Forest · XGBoost — 8,000 train / 2,000 test · Stratified split

Accuracy Comparison (%)
Precision · Recall · F1 Score
Cross-Validation Mean F1 (5-Fold)
Model Performance Summary Table
ModelAccuracyPrecisionRecallF1 ScoreCV F1 (5-fold)Verdict
Logistic Regression85.95%17.82%86.76%29.57%27.61%Best Recall
Random Forest ⭐98.60%90.00%66.18%76.27%78.01%Best Overall
XGBoost98.35%77.78%72.06%74.81%77.91%Balanced
Page 5 of 7

Explainable AI — SHAP Analysis

Feature importance via Random Forest · Mean comparison: Failure vs No-Failure cycles

SHAP Summary — Feature Importances
Feature Impact: % Deviation from No-Failure Baseline
Page 6 of 7

Failure Prediction System

Adjust sensor sliders to simulate machine conditions · Rule-based RF threshold model

Input Sensor Parameters
Live Failure Probability Gauge
0%
Failure Probability
Engineered Features (Live)
ΔTemp
10.0K
Power
61.5K
Wear Rate
3.93
Page 7 of 7

Insights & Recommendations

Actionable IoT edge maintenance strategies · Manufacturing & Automotive · AI4I 2020 analysis

Top 8 Predictive Features (Random Forest)
Maintenance Recommendations
🔴
High Torque Alert CRITICAL
Torque > 65 Nm → 4× higher failure rate. Alert immediately when sustained torque exceeds 60 Nm. Primary driver of OSF failures (98 events).
🔴
Low RPM + High Torque CRITICAL
RPM < 1300 + torque > 55 Nm triggers Power Failure (PWF). Maintain speed above 1350 under load. 95 PWF events recorded.
🟡
Tool Replacement Schedule HIGH
Mean wear at failure = 143.8 min vs 106.7 min healthy. Replace tools proactively at 120–130 min to prevent TWF (46 events).
🟡
Temperature Differential Alert MEDIUM
ΔTemp < 8.6 K causes HDF. 115 events = 34% of all failures. Trigger cooling check when ΔT drops below 8.6 K.
🟢
Deploy Random Forest RECOMMENDED
98.6% accuracy · 90% precision · 78% CV F1. Run inference every 10 cycles. ~76% of failures preventable.
Key Business Insights
Root Cause
HDF (34%) is most common. Caused when process–air ΔTemp < 8.5 K — indicating cooling degradation. Continuously monitor ΔT.
Top Feature
Power Estimate (RPM × Torque) explains 18.43% of variance. Combined with Rotational Speed (22.27%), these two features drive most predictions.
ROI Impact
L-type machines lead with 235 failures (3.92% rate). RF's 76.3% F1 enables early detection of 76% of failures — dramatically reducing downtime costs.