Automated Fault Detection, Diagnostics, Impact Evaluation, and Service Decision-Making for Direct Expansion Air Conditioners

Andrew L Hjortland, Purdue University

Abstract

This work describes approaches for automatically detecting, diagnosis, and evaluating the impacts of common faults in unitary rooftop air conditioning equipment. A semi-empirical component-based modeling approach using virtual sensors has been implemented using low-cost microcontrollers and tested on fixed-speed and variable-speed equipment using laboratory psychrometric test chambers. A previously developed virtual refrigerant charge sensor was applied to a fixed-speed rooftop unit with combinations of condenser types and expansion valve types and resulted in average prediction errors less than 10%. In addition, a methodology was developed that can be used to tune the empirical parameters of the model using data collected without psychrometric chambers, greatly reducing the experimental effort and costs required for the model. Virtual sensors previously developed for fixed-speed systems were also implemented for a variable-speed rooftop unit without significant loss of accuracy. Much of this work has been devoted to estimating the performance impacts of faults that grow over time, like heat exchanger fouling or refrigerant charge leakage. To estimate these impacts, semi-empirical models for predicting the normal performance of fixed-speed and variable-speed systems have been developed and evaluated using experimentally collected data. In addition, the virtual sensor approaches for estimating the actual performance of systems using low-cost sensor measurements were evaluated. Together, normal performance models and virtual sensor estimations were used to estimate the overall impacts of several faults on system performance. A methodology for quantifying the performance impacts of simultaneously occurring faults has been developed and tested using a detailed system model and experimental results. While relatively simple, simulated and experimentally collected results showed the fault impact models were accurate within 10% of the actual fault impacts. The fault impact evaluation models could be embedded in an AFDD system and used to determine when performance degradation faults should be serviced from an operating cost perspective. In addition, different service and maintenance strategies are compared in this work using a simulation environment that was developed. A data-driven artificial neural network model of a rooftop unit with faults has been derived for this purpose using a detailed fault impact model for direct expansion cooling equipment. This model was coupled with a building model to simulate operating cost impacts of performance degradations and service over the life of cooling equipment. An optimization problem was formulated with the goal to minimize lifetime energy and service costs and was solved using dynamic programming. Using the optimal solution as a baseline, suboptimal service decision-making strategies were implemented and simulated using the building model. It was found that condition-based maintenance strategies using the outputs of automated fault detection and diagnostics tools can significantly reduce lifetime operating costs over periodic service policies.

Degree

Ph.D.

Advisors

Braun, Purdue University.

Subject Area

Mechanical engineering

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS