Synthetic Pump Predictive Maintenance Dataset with Physics-Based Degradation for RUL Modeling

This dataset contains synthetic time-series data for pump predictive maintenance in oil and gas operations, designed specifically for machine learning applications such as Remaining Useful Life (RUL) prediction and failure classification.
The dataset includes a comprehensive set of operational, mechanical, electrical, and fluid-related parameters, including pressure, flow rate, temperature, vibration, power consumption, efficiency, water cut, gas-oil ratio (GOR), sand production, and fluid properties. It also incorporates event-based variables such as shutdowns, overload conditions, and pressure drops.
A key feature of this dataset is the integration of physics-informed constraints and domain-specific logic. The data generation process enforces realistic parameter ranges, correlations between variables, and time-dependent degradation behavior. Equipment aging is modeled through non-linear degradation patterns, where early-stage degradation is slow, followed by accelerated wear in later stages.
The dataset also includes maintenance events and failure mechanisms such as bearing failure, seal failure, cavitation, sand erosion, and motor burnout. These are reflected in dynamic changes in system behavior, including increasing vibration, decreasing efficiency, and declining health index.
Target variables are provided for machine learning tasks, including:
Remaining Useful Life (RUL) in daysBinary failure label (failure within 7 days)This dataset is intended for:
Predictive maintenance modelingRUL regression and failure classificationIndustrial AI research and benchmarkingSimulation of real-world pump behavior in artificial lift systemsAlthough synthetic, the dataset is designed to closely mimic real operational conditions in oil and gas production environments, making it suitable for research, prototyping, and model development.
Users should note that while the dataset follows realistic engineering principles, it does not represent any specific field data and should be used primarily for research and development purposes.