In Sahelian countries, diesel thermal power plants remain the primary means of electricity generation. However, their performance is significantly affected by thermal and hygrometric stresses (high ambient temperatures and extreme relative humidity) as well as by heterogeneous maintenance practices. This study analyzes the operational reliability of a fleet of seven diesel generator units at the Komsilga thermal power plant in Burkina Faso using field data collected between 2022 and 2025 (12,500 operating hours, 248 maintenance interventions, hourly environmental measurements, and historical records of failures and component replacements). The methodological approach combines (i) parametric survival modeling based on the Weibull distribution, extended through an Accelerated Failure Time (AFT) model integrating temperature (T) and humidity (H) as covariates, and (ii) a machine learning approach using a Random Forest model for short-term failure risk prediction (72-hour horizon). The results reveal a strong negative correlation between temperature and MTBF (r = −0.72) and between temperature and reliability (r = −0.75), with an amplifying effect of humidity (r = −0.58 with MTBF). The Weibull-AFT model highlights an exponential decrease in the scale parameter η as temperature and humidity increase, indicating accelerated equipment aging under severe environmental conditions. Under critical operating conditions (T > 40 °C and H > 70 %), MTBF decreases by approximately 55% compared with nominal conditions. The predictive Random Forest model achieves an AUC-ROC of 0.92 and identifies temperature as the most influential variable, followed by humidity and time since the last maintenance operation. A predictive maintenance strategy based on a risk score could reduce unplanned downtime by an estimated 20–30% according to simulation results.