WILLIAM B. FAIRLEY, Analysis and Inference Inc.Swarthmore, PA.
ALAN J. IZENMAN Temple University, Philadelphia, PA.
and A.RHETT WHITLOCK KCI Technologies,Inc.,Manassas,VA.
Investigations of failures of materials or systems aim to determine the causes and the parties responsible. In litigation, causes and responsibilities are always disputed. Most highly disputed, however, is total damage and the cost of remedial action which is often exaggerated by the plaintiffs and understated by the defendants. In order to achieve “a reasonable degree of certainty” in the amount of damage, systematic field and laboratory investigation techniques are needed, coupled with appropriate statistical analyses, modelling and calibration methods. Engineering techniques to determine causes are fairly well established and documented, but statistical methods for enhancing quantity estimates with “reasonable certainty” limits have not been sufficiently welt applied. This paper is a case study of the application of modelling and calibration techniques to determine predicted total damage with confidence limits for a masonry housing project showing spoiling of fired clay, load-bearing hollow blocks. Three types of survey were undertaken: close up, tile-by-tile “scaffold surveys”, and Two other less precise surveys of the entire site, a “photo” survey of the buildings and a “visual” survey obtained by counting spalls from ground level.
Some hollow blocks were obscured in the photo survey but the numbers of spoiled units among them was estimated using a Poisson regression model. This gave a credible goodness of fit. It was then coupled with a calibration technique that related true damage to perceived damage separately for each floor of the buildings. It enabled the photo and the visual survey data to be adjusted to mimic a complete scaffold survey of the entire housing project.
Estimates of the true spall rates for the entire housing project were about 30 spalls/1000 blocks for the parapets only, about 4 spalls/1000 blocks for the walls only and about 6 spalls/1000 blocks for the combined walls and parapets. Confidence limits were derived by fiducial inference for these spall rates. The project showed that statistical methods can provide important and useful contributions to the design and analysis of condition surveys and indicated how costs could be minimized and how reasonable estimates of the extent of damage could be made.