Shah-xai/Engineering-WindTurbine_fault_detection ? reverse-engineered prompt

Build me a wind turbine fault detection project that uses SCADA sensor data to spot early signs of failure.

I want it to load turbine data from CSV, clean it up by removing useless all zero sensors, handle missing values, scale the inputs, and deal with the heavy class imbalance in a smart way. First train an autoencoder only on normal data so it learns what healthy operation looks like. Then use the learned encoder features to train a fault classifier on the balanced data. Please save the trained models, training history, and evaluation results.

I also want a simple way to run the full pipeline from one main script, plus an easy app entry point if possible. Include evaluation with accuracy, precision, recall, F1, AUC, a confusion matrix, and SHAP based feature importance so I can see which sensors matter most. If you need to check current docs for anything, go ahead and look them up online.