Weibull-AFT and Random Forest Hybrid Modeling of Diesel Generators in Burkina Faso

Diesel thermal power plants are an essential link to ensure energy security in the countries of the Sahel, despite the development of renewable energies. In Burkina Faso, these plants contribute greatly to electricity production. However, their operation is influenced by a particularly restrictive environment characterized by ambient temperatures up to 49 °C, humidity variability greater than 70% and high exposure to dust. These conditions accelerate equipment degradation, reduce reliability and increase unexpected failures. This study proposes a contextualized predictive maintenance tool. Indeed, it is based on a Contextualized Global Risk Index (GPRI) combining a climate risk indicator, a probability of failure estimated by Random Forest and a vulnerability indicator based on the history of failures. The tool allows you to classify generators into four levels of criticality (low, moderate, high and critical) and to associate them with appropriate maintenance actions. Simulations show that such an approach could reduce unexpected failures by 20–30% while significantly improving operational reliability. The proposed methodology is therefore based on a hybrid approach combining statistical reliability models and machine learning techniques. Indeed, Weibull-AFT (Accelerated Failure Time) – Random Forest is therefore used to move from a fixed calendar preventive maintenance to a contextualized predictive maintenance, better suited to the environmental contexts of Burkina Faso and the Sahel countries. The results show a strong heterogeneity in the performances of the studied groups. For example, the Diesel Generator Set (G4) and Diesel Generator Set (G6) groups have the best reliability performance, while Diesel Generator Set (G3) appears to be the most vulnerable to degradation.