A comprehensive evaluation of existing and new model-identification approaches for non-destructive concrete strength assessment

•A hybrid conversion model identification system of NDT measurements is proposed.•This system uses the bi-objective approach for mean and variability estimation and quantile regression for local strengths estimation.•Moreover, tabular displays for the selection of the appropriate number of cores have been produced.•Lastly, to ease of use of the proposed methods, a web app was developed. The common practice of concrete strength assessment is to combine non-destructive techniques (NDT) with core test measurements to develop a conversion model that is used to estimate the strengths at NDT test locations. In this work, different model identification approaches to develop this conversion model are investigated, namely regularized regression, quantile regression, Deming regression, bi-objective approach, geometric mean regression, orthogonal regression, and support vector machine (SVM) regression intending to improve the conversion model prediction capability so that accurate estimates of concrete strength can be obtained with a small number of cores. These approaches: in addition to other existing approaches, are first assessed using synthetic datasets and then tested on a real data case study by developing a set of cumulative distribution functions that provides information regarding the risk of a wrong prediction of concrete strength, outside a tolerable level. The results showed that concrete variability was best estimated using Deming regression, orthogonal regression, and bi-objective approach while local strengths were best estimated using quantile and regularized regression. Further, mean concrete strength estimation was observed to be not significantly affected by the choice of a model identification approach. Based on these results, a hybrid conversion model identification system that uses the bi-objective approach for mean and variability estimation and quantile regression for local strengths estimation is proposed. This study also proposes new tabular displays for the selection of the minimum number of cores required for reliable estimation of strength using information related to the NDT measurements and a choice of an accepted risk in the estimation. These tabular displays were produced using extensive analysis of synthetic data to determine the minimum number of cores required to limit the probability of the wrong estimation of strength within an admissible margin of error. Lastly, a web app was developed to easily provide the conversion models an