The objective of this work is to design and test a real-time solar hot water (SHW) fault detection system that can reliably detect anticipated and unforeseen faults using only those sensors that are normally required to operate a system. The fault detection system includes a data acquisition module and a hierarchical Adaptive Resonance Theory (ART)-based neural network fault detection module. The data acquisition system logs the collector fin temperature and the water tank heat exchanger outlet temperature. The detection module uses a hierarchical ART neural network that can detect faults and classify them by level of severity. The hierarchical ART neural network is trained with information collected from a verified solar hot water system TRNSYS (Transient Systems Simulation program) model. The TRNSYS model uses weather data for Albuquerque, NM, extracted from the National Solar Radiation Data Base (NSRDB), for the 5-year period 2000–2004. The neural network is trained under different weather conditions. The simulation and experimental results show that the trained fault detection system has the capability to detect expected faults including pump faults, impeller degradation, thermosyphon and potential unexpected ones. Simulated and experimental test results are presented.
Neural Networks Art: Solving Problems with Multiple Solutions and New Teaching Algorithm
