FactoryNova: An intelligent Platform for Predictive Maintenance and Energy consumption Optimization in Smart Grid Environment

Abstract

This dissertation addresses the challenges of reactive maintenance and limited exploitation of industrial data in power generation systems. The objective is to design and implement an intelligent framework for predictive maintenance and energy optimization within the M’sila Gas Turbine Production Unit of SONALGAZ. The proposed framework, named FactoryNova, integrates three artificial intelligence modules: an Autoencoder-based anomaly detection system for gas turbine monitoring, a CatBoost-based fault diagnosis and severity classification system for rotating machinery, and a stacked ensemble model for electricity demand forecasting and energy optimization. All modules were integrated into a unified web-based platform providing monitoring, predictive analytics, and decision-support functionalities. Experimental results demonstrated the effectiveness of the proposed approach. The anomaly detection model achieved an accuracy of 94.8% and a ROC-AUC of 0.972. The fault diagnosis module achieved accuracies exceeding 99.9%, while the electricity demand forecasting model obtained a MAPE of 0.2046% and an R² of 0.9994. These results confirm the potential of artificial intelligence to improve equipment reliability, support predictive maintenance, and optimize energy management in industrial environments.