4 research repos · 20+ SOTA models · 15+ benchmark datasets · 20+ tutorial notebooks
Research Projects
Four independent frameworks covering the core domains of industrial AI
Industrial Predictive Maintenance
AI pipelines for Remaining Useful Life (RUL) prediction and fault detection on NASA CMAPSS, CWRU, IMS, and Paderborn bearing datasets. Includes LSTM, Transformer, TCN, and LSTM Autoencoder with a unified fit / predict / evaluate API and ONNX edge-deployment support.
Industrial Time-Series AI
SOTA benchmark suite for multivariate industrial sensor stream analysis — forecasting and anomaly detection. Implements PatchTST (ICLR 2023), DLinear (AAAI 2023), TCN, Transformer, and LSTM Autoencoder on ETT, SWaT, PSM, and SMAP datasets.
AI Power Electronics Diagnostics
Deep learning pipelines for fault detection in inverters and motor drives from voltage, current, and harmonic signals. Covers FFT, STFT spectrograms, wavelet analysis, and 5 model architectures (1D CNN, Spectrogram CNN, Transformer, BiLSTM, Autoencoder).
Smart Manufacturing AI
AI pipelines for vision-based defect detection (MVTec AD, NEU Steel), robot joint anomaly detection, RL-based production scheduling with PPO, and discrete-event digital twin simulation with MQTT/OPC-UA sync.
Benchmark Datasets
Real industrial datasets used across the four frameworks
| Dataset | Domain | Task | Used In |
|---|---|---|---|
| NASA CMAPSS | Turbofan engine | RUL prediction | Predictive Maintenance |
| CWRU Bearing | Rolling bearing | Fault classification | Predictive Maintenance |
| IMS Bearing | Rolling bearing | RUL / anomaly | Predictive Maintenance |
| Paderborn Bearing | Rolling bearing | Fault classification | Predictive Maintenance |
| ETT (h1/h2/m1/m2) | Power grid | Forecasting | Time-Series AI |
| PSM / SMAP / MSL | Server / NASA telemetry | Anomaly detection | Time-Series AI |
| SWaT / WADI | Water treatment / distribution | Both | Time-Series AI |
| Kaggle Motor Temp | PMSM motor drive | Temp / anomaly | Power Electronics |
| MVTec AD | Industrial surfaces | Defect detection | Smart Manufacturing |
| NEU Surface Defect | Hot-rolled steel | Defect classification | Smart Manufacturing |
Key Results
Reproducible benchmark results — run any benchmark with a single command
| Framework | Dataset | Model | Metric | Result |
|---|---|---|---|---|
| Predictive Maintenance | CMAPSS FD001 | Transformer | RMSE ↓ | 12.89 |
| Predictive Maintenance | CMAPSS FD001 | Transformer | NASA Score ↓ | 198.7 |
| Time-Series AI | SWaT (synthetic) | LSTM Autoencoder | ROC-AUC ↑ | 0.9999 |
| Time-Series AI | SWaT (synthetic) | LSTM Autoencoder | F1-PA ↑ | 1.000 |
| Power Electronics | Inverter (9 classes) | 1D CNN | Accuracy ↑ | ~99% |
| Smart Manufacturing | MVTec AD — bottle | ViT-B/16 | AUROC ↑ | 0.982 |
About the Lab
The IEEE IES Industrial AI Lab develops open-source, research-grade AI frameworks for the IEEE Industrial Electronics Society community. Our goal is to provide reproducible baselines and benchmarks that bridge the gap between academic research and industrial deployment.
Each framework is designed as a mini research platform — not just code, but reproducible experiments, standardized evaluation protocols, and tutorial notebooks that make the work accessible to both engineers and researchers.