A data-driven method for Fault Detection and Diagnosis in Air Handling Units

David P. Azevedo; Bruno C. Diniz; Leonardo R. Rodrigues

doi:10.20906/sbai.v1i1.2575

David P. Azevedo Instituto Tecnológico de Aeronáutica - ITA, São José dos Campos-SP
Bruno C. Diniz Universidade Federal da Bahia - UFBA, Salvador-BA
Leonardo R. Rodrigues Instituto Tecnológico de Aeronáutica - ITA, São José dos Campos-SP

DOI: https://doi.org/10.20906/sbai.v1i1.2575

Keywords: Fault Detection, Fault Diagnosis, Data-Driven Methods, Artificial Neural Networks, Random Forests, Air Handling Units

Abstract

Air Handling Units (AHUs) are essential refrigeration equipment to provide thermal comfort in buildings and commercial establishments. However, the maintenance strategies adopted for AHUs have low efficiency. In addition to the higher energy consumption, an AHU with a high degradation level has a higher probability of presenting a failure event. Many establishments already have both the automation and the sensors needed to benefit from the data generated by the air handling systems. A fault detection and diagnosis (FDD) method can benefit from these data in order to identify inefficiencies in AHUs, allowing the implementation of an efficient maintenance strategy. This paper presents a data-driven method that uses Artificial Neural Networks with the Random Forest algorithm to identify the operation condition of AHUs, to alert the operator in case of a failure event. The results observed in numerical experiments show an accuracy of 99.3% and a computed F1-score of 0.983. The model was tested with out-of-sample data and the results kept satisfactory.