In this paper we discuss the components and design of each part of an expert system for fault detection and diagnosis in the refrigeration process of a hydraulic power station. Given a minimum group of measurements, it is possible to extract behavior patterns for the whole system. The knowledge base is obtained by simulating the dynamic behavior of the process. The expert system was designed for a 580MW power plant called Central Pehuenche, located in Chile. The system shows a 96% of detection effectiveness under different operation conditions.