USING MACHINE LEARNING TO PREDICT COLLAPSE OF STEEL STRUCTURES UNDER SEISMIC LOADING

2025-01-20 · 1 citations

This study applies machine learning (ML) models to predict the collapse limit state of steel moment resisting frame (SMRF) buildings, considering uncertainties in system parameters and input ground motion characteristics. Structural global collapse is affected by a large number of linear and nonlinear system parameters. One of the main goals of the study is to find the effectiveness of ML methods to predict collapse, as the number of system’s features is reduced. Because of the lack of sufficient experimental data, an ML approach is followed in which three code-compliant SMRF buildings of varying heights (2, 4 and 8 stories), are evaluated up to the collapse limit state, using nonlinear time history analyses. Variability in system parameters and ground motions, as well as potential correlation among some of the parameters, is considered to generate a database of more than 19,000 realizations of collapsed and non-collapsed systems. The ML models are trained and tested with this database, and the efficiency of the models is categorized using different metrics, such as accuracy, F1-score, precision, and recall. Six different ML classification-based techniques are employed to predict collapse, finding that boosting algorithms (e.g., AdaBoost and XGBoost) are the best methods for collapse status classification of the evaluated structural systems. Permutation feature importance is applied to identify the main contributors to collapse. The ML models are then retrained using less features, considering first removal of nonlinear deteriorating parameters, and then removal of the hardening nonlinear parameters. The results show that acceleration amplitude, record-to-record variability, and elastic properties of the system are significant predictors of the collapse limit state, as expected; whereas the importance of nonlinear deteriorating parameters depends on the variability of the data source.

Development of a plane stress projected non-local plasticity model with Tikhonov regularisation

Proceedings of the Institution of Civil Engineers - Engineering and Computational Mechanics · 2025-05-02 · 1 citations

Conventional local plasticity models with a softening slope exhibit a mesh dependency phenomenon, in which outcome parameters (e.g. equivalent plastic strain) depend on the mesh size. This behaviour has prompted the development of non-local plasticity models. The existing non-local models use a radial return algorithm to return the stress to the yield surface. These methods, however, cannot be used for plane stress condition models because the out-of-plane direction is not defined, making the radial return algorithm not uniquely defined. This paper presents a novel non-local model explicitly designed for the two-dimensional (2D) plane stress condition, with the objective of solving models that exhibit strain softening. The proposed model tackles the challenge of an ill-posed softening equation by utilising Tikhonov regularisation within the plane stress projected subspace. A numerical example demonstrates the proposed approach’s effectiveness in providing a stable and efficient solution for strain-softening problems. The results show that the width of the localisation zone is independent of the mesh size and is mainly controlled by the internal length scale parameter. The constant localisation width and the equivalent plastic strain provide reliable results for problems of plasticity with strain softening.

Representative feature contributions when predicting the seismic collapse of steel buildings using an ensemble neural network

Structures · 2025-11-28 · 1 citations

This study implements an ensemble neural network (ENN) to obtain representative and stable feature importance contributions to collapse prediction of steel moment resisting frames (SMRFs). The feature importance assessment includes global sensitivity analyses (GSAs) and feature extraction techniques. To construct the ENN, hundreds of neural network (NN) architectures are generated and an elite set of 50 NNs is initially obtained using a multi-criteria decision analysis (MCDA). A final elite set of 50 NNs is generated after applying a genetic algorithm to the initial elite set, which undergoes several iterations of crossover and mutation. To generate the dataset of SMRF collapse status, thousands of nonlinear time history analyses are carried out on frame systems ranging from 2 to 20 stories. The frames are based on five SMRF baseline systems with variability in input parameters that are randomly selected for each system to incorporate uncertainties in mass, structural damping, nonlinear spring backbone parameters, and ground motion parameters. Then, single NNs and the ENN generate surrogate models that are used to predict the SMRF collapse status. Thereafter, five feature analysis techniques are implemented on the ENN to determine the most important contributors to global collapse of the evaluated SMRFs, including two GSAs (Sobol and delta methods) and three feature importance extraction methods (permutation, integrated gradients, and Shapley adaptive explanation values). The GSA results were obtained for a single-instance (one SMRF) for each NN and, in the most comprehensive approach, as the average of applying this technique to 100 instances (i.e., 100 SMRFs). The study found that GSA results can fluctuate significantly based on the NN architecture and the data instances used in sampling. The use of a multi-instance-based ENN offered a more stable collapse predictor for SMRFs. The results consistently highlight ground motion parameters and elastic system parameters as the most critical contributors to collapse prediction of modern SMRF buildings.

Coupled alkali-silica-reaction/seismic simulation of nuclear containment vessels

International Journal of Mechanical Sciences · 2025-04-06 · 5 citations

Performance of a nonlocal plane stress projected algorithm with a truncated weighing function

Engineering Computations · 2025-05-14 · 1 citations

Purpose This study implements and evaluates the effect of incorporating a truncated weighing function on a nonlocal plane stress projected algorithm with Tikhonov regularization for plane stress conditions. Design/methodology/approach Applying the continuum model to materials undergoing softening after yielding often results in mesh-dependent solutions. To address the mesh dependency problem, several integral-type nonlocal plasticity models have been proposed. However, the shortcoming of this approach is that only the points near the strain-softening point significantly impact its stresses and strains. Therefore, a truncated Gaussian distribution weighing function was implemented on a recently developed plane stress projected nonlocal plasticity model. The efficacy of this nonlocal formulation was measured by modifying the truncated Gaussian distribution limits as a function of the number of standard deviations and assessing the effect of this truncation on the computational time and the distribution of equivalent plastic strains. The study also analyzed the impact of the length scale and regularization parameters on the finite element solutions when the truncated weighing function is incorporated into the nonlocal plane stress projected formulation. Findings The output parameters, such as plastic strains, band width and force-deformation curve, showed negligible differences when the Gaussian distribution was truncated to three standard deviations, whereas the computational time decreased by around 5%. However, truncating the weighing function at two standard deviations or less increased the equivalent plastic strain. The output parameters, such as the equivalent plastic strain, are sensitive to length scale modifications and tend to converge as the regularization parameter increases. Originality/value For the first time, the present study uses the truncated Gaussian distribution as a weighing function in a 2D plane stress projected nonlocal plasticity model and studies the effect of truncation of the above function on the numerical accuracy and computational efficiency of the model. The effect of the regularization parameter will also be studied, especially for the length scale factor, since the attenuation of the Gaussian function is controlled by the standard deviation depending on the values of the length scale.

Effect of Boundary Conditions on Buckling Restrained Braced Frame Seismic Performance

2025-11-26

This study evaluates he effect of gusset plate behavior on the seismic performance of Buckling Restrained Braced Frames (BRBFs). The expected failure mode of these frames is yielding of the Buckling Restrained Brace (BRB) core, and although other BRB-related failure mechanisms are commonly reported, the performance and potential failure of BRB-gusset connections are rarely addressed. In this study, finite element (FE) models of BRBFs with diagonally-oriented BRBs braces were first calibrated in the nonlinear interval using the results of experimental BRBFs subjected to quasi-static cyclic loading. The tested specimens consisted of two BRBFs with different unbraced top gusset lengths, each one with two different beam connections (continuous and spliced beams). The calibrated FE models were then used to predict BRBF performance up to failure. A parametric study was performed to address some of the variables required to compute the buckling capacity of gusset plates. The evaluated parameters included the size, thickness, and initial imperfection of the gusset plate, as well as the presence of lug stiffeners. The study shows that cruciform lug stiffeners enhance the buckling capacity of gusset plates by more than 50% because the plates in the out-of-plane (OOP) direction significantly reduce displacements in this direction at the gusset plate free end. Large gusset plate imperfections and thinner plates decrease the buckling capacity and may even modify the critical failure mode, but design practices and imperfection limits imposed by current codes and standards should prevent a significant decrease in BRBF capacity.

Development of a consistent hysteretic model with kinematic hardening and isotropic softening

Structures · 2025-04-04 · 1 citations

A phenomenological plasticity based bilinear hysteretic model is developed to account for strength deterioration in cyclic loading. The kinematic hardening isotropic softening (KHIS) model is based on J2 plasticity theory and considers the material response to harden kinematically, but soften isotropically, according to a backbone curve that includes strength deterioration with a negative tangent slope. Therefore, the yield surface translates when the material hardens, but as the material softens the yield surface center location is anchored in place, while the size of the yield surface shrinks. A set of novel hysteretic rules is proposed to implement the strength deterioration in cyclic reversals to address the inconsistencies in backbone-curve-based phenomenological models. The yield point in a load cycle is always tracked by the backstress, and the maximum strength in any cyclic excursion is associated with the material strength at the point of the reversal. The material response between the yield point and the maximum strength is kinematic hardening, followed by isotropic softening after the peak response. The backstress tensor stays unchanged during softening, fixing the location of the yield surface center. The proposed changes to the hysteretic rules keep the yield point during the reversals consistent with kinematic laws and better represent the material behavior, and can be implemented into 1-, 2-, or 3-dimensional elements. A stress update algorithm is presented for implementing the model in a finite element (FE) code. A series of material point simulations verify the implementation of the model in an open-source FE code. The proposed model can be easily set up in an FE formulation by adding strain softening in the material constitutive relationships without further consideration of specific component failure modes. The model is validated against cyclic experimental tests on steel structural members, showing its capability to accurately simulate strength deterioration.

Effect and Implications of Feature Selection on Machine Learning Prediction of Seismic Drift Demands and Collapse of Steel Moment Resisting Frames

Journal of Earthquake Engineering · 2025-09-16 · 2 citations

Machine learning (ML) techniques were generated for different information levels to identify the minimum set of system parameters required for predicting collapse and maximum interstory drift (SDRmax) of steel moment resisting frame (SMRF) buildings. Five baseline modern SMRFs were evaluated under seismic loading with varying system and ground motion (GM) parameters to generate a database. Classification and regression-based ML models were tested at three system information levels to predict collapse and SDRmax, respectively. The ML predictions were mainly controlled by GM parameters and were relatively insensitive to system parameters defining nonlinear behavior, such as spring backbone curve features. Machine learning methods indicate that nonlinear parameters have a minor effect on the prediction collapse and maximum interstory drift ratios (SDRmax) of modern SRMFS.The ML-based study predicts SDRmax and SMRF collapse considering multiple buildings.The ML-based study accounts for uncertainties and variable correlations in modeling multiple multistory SMRF buildings.XGBoost resulted the best ML model to predict SDRmax and collapse for the evaluated SMRFs. Machine learning methods indicate that nonlinear parameters have a minor effect on the prediction collapse and maximum interstory drift ratios (SDRmax) of modern SRMFS. The ML-based study predicts SDRmax and SMRF collapse considering multiple buildings. The ML-based study accounts for uncertainties and variable correlations in modeling multiple multistory SMRF buildings. XGBoost resulted the best ML model to predict SDRmax and collapse for the evaluated SMRFs.

Using Incremental Learning With Rehearsal to Enhance Global Collapse Prediction Machine Learning Models Across Diverse Steel Building Datasets

The Structural Design of Tall and Special Buildings · 2025-12-02 · 1 citations

ABSTRACT Generalization of machine learning (ML) surrogate models across distinct databases is underexplored, despite being crucial as retraining the entire model every time new data become available is inefficient. This study proposes an incremental learning methodology to improve ML models' prediction of seismic collapse of steel moment‐resisting frames (SMRFs) across distinct datasets. Three boosting algorithms, XGBoost, LightGBM, and CatBoost, were trained on a source dataset to generate surrogate ML models that can predict the SMRF's seismic response. Thereafter, the ML models were used to predict the response on a new (target) dataset of SMRFs that differ in geometric dimensions and design approaches. Initially, boosting models trained on one dataset performed poorly on another dataset, even if the datasets displayed similar characteristics and consistent feature importance rankings. Incorporation of incremental learning improved the prediction on the target dataset, but introduced catastrophic forgetting that reduced the effectiveness of the ML model on the source dataset, a problem mitigated with a rehearsal strategy. Incremental learning with rehearsal yields results comparable to those obtained by fully retraining with both source and target datasets, resulting in an effective method for ML transferability, without having to retrain entire databases and without reducing the effectiveness of ML models on the source database.

Truncated Weighted Singular Value Decomposition Solution of Tikhonov Regularized Softening Shear Band Problems in Finite Elements

International Journal for Numerical Methods in Engineering · 2025-10-08

ABSTRACT A Truncated Weighted Singular Value Decomposition (TWSVD) based approach is proposed for the Tikhonov regularized solution of the ill‐posed softening type nonlocal plasticity model. Tikhonov regularization provides a stable, smooth, and mesh‐independent solution of integral‐type nonlocal plasticity, but is computationally expensive for large‐scale problems, particularly in three‐dimensional models. The proposed solution technique offers an efficient method for the solution of the large‐scale Euler equation resulting from the Tikhonov regularization, which is significant for practical applications of softening problems. The TWSVD approach modifies the truncated SVD method to apply to the numerical integration method commonly used in finite element (FE) models. The mathematical formulation of the proposed approach is presented with a linear isotropic plasticity model, and a stress‐update algorithm is developed for the FE application of the approach. The solution algorithm is implemented in a three‐dimensional model with isoparametric brick elements. The solution approach is validated with a benchmark shear‐band softening problem. Results show that the TWSVD method can significantly reduce the computational time compared to conventional Tikhonov‐regularized solutions, without sacrificing accuracy, making it a practical and efficient solution tool for large‐scale systems in nonlocal plasticity models with softening behavior.