Gas Turbine Predictive Maintenance Tool

I am expanding an internal application that monitors the health of industrial gas turbines and now needs a robust predictive-maintenance module. The current environment streams high-resolution sensor feeds in JSON and is backed by several years of curated historical performance data. The goal is to turn those two sources into accurate failure-risk forecasts and actionable maintenance recommendations. You will design and implement the full workflow—data ingestion, cleansing, feature engineering, model training, and deployment—inside a Python-based stack (pandas, scikit-learn or similar; feel free to propose alternatives such as PySpark or TensorFlow if they offer clear advantages). Real-time scoring must run fast enough to support control-room dashboards, and the results should be exposed through an API the rest of the system can call. Key deliverables • Data pipeline that pulls JSON sensor streams and historical records into a unified store • Predictive models with documented performance metrics (precision, recall, ROC-AUC) • Automated retraining script tied to new data drops • REST endpoint (or MQTT topic) that returns health indices and recommended maintenance windows • Concise technical documentation so our engineers can maintain and extend the code Acceptance criteria – Model accuracy equal to or better than the current rule-based alarms by at least 20 % – End-to-end latency under 5 s for a single turbine data packet – Code and docs delivered via a Git repository and a short hand-off call Environmental data is not in scope right now, but the architecture should stay flexible enough to add it later. Let me know which libraries or cloud services you prefer so we can align on the tech stack before kickoff.

Project ID: 40517877