Predictive maintenance of Air Compressor Fault using ML

I am instrumenting an air-compressor test rig with thermal, pressure, and acoustic sensors that push measurements as real-time data streams. I now need a complete, reproducible pipeline that turns those streams into actionable fault predictions. Here is what I expect you to handle end-to-end: • Data acquisition – capture, time-stamp, and store each thermal, pressure, and acoustic channel continuously so the same code can be pointed at the live rig or at recorded sessions for offline work. • Feature engineering & model training – build a multi-class classifier that separates normal running from four distinct fault categories (bearing, water-pump, radiator, exhaust-valve). You are free to choose the algorithm and Python stack (scikit-learn, PyTorch, TensorFlow, XGBoost, etc.) provided the training script is fully repeatable. • Performance reporting – deliver class-wise and overall accuracy, precision, recall, F1-score, plus a 5 × 5 confusion matrix generated from held-out data. • Predictive-maintenance validation – produce a concise technical note (diagram, photo, or CAD screenshot) showing exactly where each sensor would be mounted on the physical rig to detect the corresponding faults. Acceptance criteria 1. A Git-based repository containing all source code, requirements.txt (or environment.yml), and a README with step-by-step instructions. 2. A trained model file and a notebook or script that recreates every metric above on demand. 3. The sensor-placement note clearly linking each of the four fault classes to its monitoring points on the compressor. If anything is unclear about the live streaming format or the fault taxonomy, let me know early so we can lock it down before you start coding.