Field Implementation and Performance Evaluation of AI-Based Predictive Maintenance in Smart Industrial Pumps

The paper deals with the field implementation and performance evaluation of artificial intelligence-based predictive maintenance (AI-PdM) systems in smart industrial pump environments. Given the critical role of pumps in industrial operations, ensuring their reliability and minimizing unplanned downtime is of significant importance. Recent advancements in AI and the Internet of Things (IoT) have enabled real-time monitoring, data-driven diagnostics, and anticipatory failure detection in mechanical systems. However, the real-world effectiveness and deployment feasibility of AI-based maintenance remain under-explored in operational contexts. We applied a multi-stage methodology incorporating sensor data acquisition, feature engineering, machine learning model training (including random forest and LSTM algorithms), and in-field validation using a testbed of smart centrifugal pumps in a petrochemical facility. The data collected spanned vibration patterns, motor current signals, pressure variations, and historical failure records over a ۱۲-month period. Model performance was measured using precision, recall, F۱-score, and mean time to failure (MTTF) predictions. We established that the AI-PdM system could predict imminent faults with an accuracy exceeding ۹۲%, significantly reducing unscheduled downtime by ۳۸% and maintenance costs by ۲۷% compared to traditional preventive approaches. Notably, long short-term memory (LSTM) networks outperformed other models in time-series forecasting. The implementation demonstrated the practical viability of AI-PdM under fluctuating operational loads, environmental factors, and legacy infrastructure constraints. This work contributes a comprehensive evaluation of AI-integrated predictive maintenance in industrial contexts, emphasizing data quality, real-time deployment considerations, and human-machine interfacing. It provides a replicable framework for engineers and decision-makers aiming to transition from reactive to predictive asset maintenance paradigms in Industry ۴.۰.

کلیدواژه ها:

Predictive Maintenance ، Artificial Intelligence ، Smart Pumps ، Industrial IoT ، Machine Learning Models