What is the MAD system?

Technology

MAD uses LSTM autoencoders — a type of neural network designed for time-series data. The model is trained on healthy equipment sensor data and detects anomalies via reconstruction error.

Validation

• 8 public benchmark datasets
• AUC up to 1.000 on individual datasets
• AUC 0.9995 on cross-dataset transfer
• Zero false alarms on transfer test
• 5-24 days detection window before failure

Impact

• 72% downtime reduction in digital twin simulation
• 6x lower maintenance costs
• 86% cost reduction ($1M to $144K per event)
• 40 hours downtime saved per event
• Break-even at 5% detection rate

Related Questions