Building energy audits are both expensive, on the order of $0.50(US)/sf [1], and there aren’t enough auditors to survey the entire building stock in the U.S. Needed are lower cost automated approaches for rapidly evaluating the energy effectiveness of buildings. A key element of such an approach would be automated measurements of envelope R-values. Proposed is the use of single point-in-time thermal images potentially obtainable from drive-by thermal imaging to infer wall and window R-values. A data mining based approach is proposed, which seeks to calibrate the measured exterior wall temperatures to known and measured R-values for a small subset of residences. In this approach, visual imagery is first used to determine the wall emissivity based on the color of the wall siding in order to yield an estimate of the wall temperature. A Random Forest model is developed using the training set comprised of the residences with known R-value. This model can then be used to estimate R- and C-values of other houses based upon their measured exterior temperatures. The results show that the proposed approach is capable of accurately estimating envelope thermal characteristics over a spectrum of envelope R-values and thermal capacitances present in residences nationally. The resulting error for the houses considered is maximally 12% using as few as nine training houses. The data mining approach has significantly greater accuracy than modeling-based approaches in the literature.

This content is only available via PDF.
You do not currently have access to this content.