Training and evaluation of virtual sensors for rooftop units

Jebaraj Vasudevan, Purdue University

Abstract

This thesis focuses on assessing and extending specific virtual sensors for rooftop units with micro-channel condensers, which are a growing part of the market. The rooftop unit virtual sensors provide low-cost measurements of the amount of refrigerant charge, cooling capacity and compressor power and are expected to be embedded within manufactured products in the factory. In addition, a low-cost approach for training the virtual refrigerant charge sensor in an open lab space was proposed and evaluated. The accuracy of virtual rooftop unit sensors were evaluated over a wide range of conditions using measurements obtained in environmental (psychrometric) chambers and were generally within ±10% of the values determined from more direct measurements. The concept of low-cost open lab training for virtual charge sensor along with some guidelines to choose open lab training points was evaluated and found to give similar accuracy as sensors trained using a wide range of operating conditions. The total cost of embedding the three virtual sensors in a rooftop unit at a factory would be in the range of $60 to $120 per unit. This is much less than the cost of directly measuring only two of the three quantities: unit cooling capacity and compressor power. There is no practical direct measurement method for the amount of refrigerant charge to enable cost comparisons with the cost of virtual charge sensing.

Degree

M.S.M.E.

Advisors

Horton, Purdue University.

Subject Area

Mechanical engineering

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

Share

COinS