jaue2019-047: ANN-based Chiller FDD Model for Considering Various Operating Conditions

Abstract

Fault detection and diagnosis (FDD) for a chiller is important to prevent energy waste due to malfunction and to reduce maintenance costs. In recent years, with the development of machine learning, data driven FDD became popular due to the reduced time and expert knowledge required to develop the model. The chiller operates under various operating conditions. However, it is difficult to obtain eligible data under all operating conditions. For practical use, a model trained under limited operating conditions should have acceptable performance under untrained operating conditions. In this study, a chiller FDD model based on artificial neural network for considering various operating conditions was proposed. The proposed model generalizes well under various operating conditions as it can find underlying patterns in the data. Experiments with the data from ASHRAE 1043-RP show that the proposed model has improved generalization performance for operating conditions compared to the SVM-based model.