Backbone curves constructed from experimentally derived hysteresis envelopes are often used to evaluate the force-deformation behavior and, thus, seismic residual collapse capacity of structural components under cyclic loading. This paper proposes a novel machine learning–based backbone curve model (ML-BCV) for rapidly predicting these curves for flexure- and shear-critical columns. The model integrates a multioutput least-squares support vector machine to discover the mapping between input and output variables and a grid search optimization algorithm to facilitate the training process. A database including 262 test columns is utilized to train, test, and validate the ML-BCV model by (1) direct comparison with experimental results, (2) a 10-fold cross-validation procedure, and (3) direct comparison with traditional modeling approaches for three columns. The ML-BCV model reduced the root-mean-square error for the four values governing the shape of the backbone curve by 80% (drift ratio at yield shear), 61% (yield shear force), 58% (drift ratio at maximum shear), and 67% (maximum shear force), demonstrating that the ML-BCV is increasingly robust and accurate compared to traditional modeling approaches.

Large concrete structures need to be inspected in order to assess their current physical and functional state, to predict future conditions, to support investment planning and decision making, and to allocate limited maintenance and rehabilitation resources. Current procedures in condition and safety assessment of large concrete structures are performed manually leading to subjective and unreliable results, costly and time-consuming data collection, and safety issues. To address these limitations, automated machine vision-based inspection procedures have increasingly been proposed by the research community. This paper presents current achievements and open challenges in vision-based inspection of large concrete structures. First, the general concept of Building Information Modeling is introduced. Then, vision-based 3D reconstruction and as-built spatial modeling of concrete civil infrastructure are presented. Following that, the focus is set on structural member recognition as well as on concrete damage detection and assessment exemplified for concrete columns. Although some challenges are still under investigation, it can be concluded that vision-based inspection methods have significantly improved over the last 10 years, and now, as-built spatial modeling as well as damage detection and assessment of large concrete structures have the potential to be fully automated.

Drift capacity of reinforced concrete (RC) columns is an important indicator to quantify the seismic vulnerability of RC frame buildings; however, it is challenging to accurately predict this value as the nonlinear behavior can vary greatly by column type. This article proposes a novel, local machine learning (ML) model, called locally weighted least squares support vector machines for regression (LWLS‐SVMR), which integrates LS‐SVMR and locally weighted training criteria to enhance and generalize the prediction of the drift capacity of RC columns, regardless of the type. A database of 160 circular RC columns covering flexure‐, shear‐, and flexure–shear‐critical specimens was developed to train and test the proposed LWLS‐SVMR. The proposed LWLS‐SVMR was validated by comparison with popular existing global and local learning approaches as well as a traditional empirical equation, and the results demonstrated that the proposed LWLS‐SVMR is superior to all other approaches and thus, is a promising artificial intelligence technique for enhancing the prediction of drift capacity, universally across RC flexure‐, shear‐, and flexure–shear‐critical columns. The LWLS‐SVMR exhibits capabilities which may yield it a feasible approach to predict complex, nonlinear behavior in a broad‐spectrum manner.

This is the accepted manuscript. The final version is available from ASCE at http://dx.doi.org/10.1061/(ASCE)ST.1943-541X.000120

This paper presents a methodology for comparing the performance of model-reduction strategies to be used with a diagnostic methodology for leak detection in water distribution networks. The goal is to find reduction strategies that are suitable for error-domain model falsification, a model based data interpretation methodology. Twelve reduction strategies are derived from five strategy categories. Categories differ according to the manner in which nodes are selected for deletion. A node is selected for deletion according to: (1) the diameter of the pipes; (2) the number of pipes linked to a node; (3) the angle of the pipes in the case of two-pipe nodes; (4) the distribution of the water demand; and, (5) a pair-wise combination of some categories.The methodology is illustrated using part of a real network. Performance is evaluated first by judging the equivalency of the reduced network with the initial network (before the application of any reduction procedure) and secondly, by assessing the compatibility with the diagnostic methodology. The results show that for each reduction strategy the equivalency of networks is verified. Computational time can be reduced to less than 20% of the non-reduced network in the best case. Results of diagnostic performance show that the performance decreases when using reduced networks. The reduction strategy with the best diagnostic performance is that based on the angle of two-pipe nodes, with an angle threshold of 165¿. In addition, the sensitivity of the performance of the reduced networks to variation in leak intensity is evaluated. Results show that the reduction strategies where the number of nodes is significantly reduced are the most sensitive.Finally this paper describes a Pareto analysis that is used to select the reduction strategy that is a good compromise between reduction of computational time and performance of the diagnosis. In this context, the extension strategy is the most attractive.

Small data sets are an extremely challenging problem in the machine learning (ML) realm, and in specific, in regression scenarios, as the lack of relevant data can lead to ML models that have large bias. However, there are many applications for which a purely data‐driven procedure would be advantageous, but a large amount of data are not available. This article proposes a novel regression‐based transfer learning (TL) model to address this challenge, where TL is defined as knowledge transfer from a large, relevant data set (source domain data) to a small data set (target domain data). The proposed TL model is termed double‐weighted support vector transfer regression (DW‐SVTR), which couples least squares support vector machines for regression (LS‐SVMR) with two weight functions. The first weight function uses kernel mean matching (KMM) to reweight the source domain data such that the mean values of the source and target domain data in a reproduced kernel Hilbert space (RKHS) are close. In this way, the source domain data points relevant to the target domain points have a larger weight than irrelevant source domain points. The second weight is a function of estimated residuals, which aims to further reduce the negative interference of irrelevant source domain points. The proposed approach is assessed and validated via simulated data and by enhanced shear strength prediction of nonductile columns based on limited availability of nonductile column data. Specifically, the results for the latter show that the proposed DW‐SVTR can reduce the root mean square error (RMSE) by 34% and enhance the coefficient of determination (R²) by 229%. These numerical results demonstrate that the DW‐SVTR significantly reduces the effect of small sample bias and improves prediction performance compared to standard ML methods.