This model repository contains a Physics-Guided Bayesian Neural Network (PG-BNN) built for fault detection and operational anomaly prediction in wind turbine sub-systems (gearbox, generator, rotor, and bearings).

Key Features

Bayesian Variational Inference: Quantifies epistemic uncertainty via Monte Carlo sampling.
Physics Regularization: Enforces thermodynamic and aerodynamic physical bounds (energy balance, torque-speed dynamic constraints) during training to prevent unphysical predictions.
SCADA Feature Support: Designed to ingest 40+ SCADA channels along with engineered rolling statistics and exponential moving averages.

Quick Usage

python

from inference import PGBNNPredictor
import torch

predictor = PGBNNPredictor.from_pretrained("AerovigilAI/wind-turbine-pg-bnn")
sample_scada = torch.randn(1, 40) # 40 SCADA sensor features

mean_pred, uncertainty = predictor.predict(sample_scada, num_samples=30)
print(f"Fault Probability: {mean_pred.item():.4f}, Uncertainty (Std Dev): {uncertainty.item():.4f}")
---

rajaram-90/AerovigilAI

Physics-Guided Bayesian Neural Network (PG-BNN) for Wind Turbine Predictive Maintenance

Key Features

Quick Usage

rajaram-90/
AerovigilAI