The objective of this project is to evaluate and ultimately improve predictive maintenance techniques for the heating, ventilation and air conditioning (HVAC) filter. The filter is a crucial subcomponent to the proper function of the HVAC system. Clogging and other filter malfunctions compromise air quality and can result in costly damages to the HVAC system as a whole. Predictive maintenance (PdM), whereby the filter is replaced on a pre-planned schedule before filter-clogging occurs, is needed in order to avoid such outcomes. The majority of HVAC filter PdM programs have failed in their implantation. Therefore, research is needed to evaluate and explore the use of current and new techniques for predicting HVAC filter failure patterns.
The first step of my research aims to evaluate five predictive techniques that are currently used for HVAC filter PdM: neural network regression, linear regression, bayesian networks regression, boosted regression tree, and decision forest. A case study was used to compare the results of these predictive techniques against the actual failure pattern of a filter installed within the HVAC system at Écoule de Technologie Supérieure. A multi-criteria approach was used to evaluate the predictive techniques according to predictive-accuracy and expert opinion.
A better understanding of why filter clogging occurs is needed to predict filter failure. Therefore, the second part of my research focuses on four parameters pertaining to filterblockage: fan speed, return air temperature, mixed temperature (return and fresh air), and the position of the mixture damper. A case study was developed to examine the extent to which each parameter affects filter performance. Sensors installed within the UTA-104 HVAC unit at ÉTS collected data on these parameters and filter performance over a six-month period. The experimental data gathered was incorporated into the final part of my research: a computational fluid dynamic (CFD) model of an HVAC system.
The third part of my research was devoted to developing a working CFD model using ANSYS FLUENT to simulate airflow within the HVAC system’s filter unit. This CFD model incorporates the best predictive technique as determined by the first part of my research, as well as the experimental data gathered in the second part of my research. If combined with an algorithm that pairs real-time data gathered from an HVAC filter and its environment, this CFD model could theoretically be used to reliably assess filter health and prescribe maintenance procedures.
Therani, M. (Author),
Beauregard, Y. (Supervisor),
Rioux, M. (Co-supervisor) &
Kenné, J.-P. (Co-supervisor),
14 Jun 2023Student thesis: Doctoral thesis › Doctorate in Engineering: Engineering