A Study of Design of Experiments and Machine Learning Methods to Improve Fault Detection Algorithms

This work presents an industrial application of design of experiments (DOE) and machine learning methods for the development of algorithms applied to fault detection problems in the heating, ventilation, air conditioning and refrigeration (HVAC-R) industry. The framework adopted is an attempt to mitigate the problems which affect the context of machine learning and consists in a sequential approach of DOE and machine learning modelling. The DOE study is performed to ensure the quality of the data that are then used to infer a series of supervised algorithms, both regression and classification. The steps needed for the implementation of the algorithms are described, and the final performance of the models is discussed in terms of pros and cons and results on a test data set.