Prediction of Diesel Generator Oil Pressure and Health Status Using a Hybrid Method
8th Global Power, Energy and Communication Conference, GPECOM 2026, Naples, Italy, 3 - 05 June 2026, pp.226-231, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1109/gpecom70462.2026.11578725
- City: Naples
- Country: Italy
- Page Numbers: pp.226-231
- Keywords: Deep learning, Diesel generator, IoT, Predictive maintenance
- Yıldız Technical University Affiliated: Yes
Abstract
Diesel generators are strategic primary or backup power sources for residential use and critical facilities. Since they are a key component of the energy supply chain, their health status and maintenance requirements are critical issues. This paper presents a predictive maintenance system based on six electrical and electromechanical sensor parameters (frequency, voltage, battery voltage, water temperature, winding temperature, active power) collected at 15-minute intervals from an active field generator. An original hybrid method utilizing eight different deep learning architectures is employed to predict the next-step oil pressure and generate a generator health score. Experimental results show that the Temporal Convolutional Network-Long Short-Term Memory (TCN-LSTM) hybrid architecture achieves the highest accuracy, detecting faults significantly earlier than traditional protection systems. To prevent physical limit violations common in purely data-driven models, a 'Physical Penalty Mechanism' based on technical standards is integrated into the system. The results validate the applicability of predictive maintenance for diesel generators and provide a basis for deep learning architectures compatible with IoT-based remote monitoring systems.