Thang Xuan Le
Research Subjects
Thang Xuan Le has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Hoa Tran-Ngoc
University of Transport and Communications
7 shared publications - Hung Viet Tran
University of Transport and Communications Faculty of Civil Engineering
3 shared publications - Thanh Bui-Tien
Faculty of Civil and Environmental Engineering, Research Group R2026-UTC-08, University of Transport and Communications, No 3 Cau Giay Street, Lang Ward, Hanoi 10000, Vietnam
3 shared publications - Nguyen Thi Cam Nhung
Faculty of Civil Engineering, University of Transport and Communications, Hanoi 100000, Vietnam
2 shared publications - Quyet Nguyen Huu
University of Transport and Communications
2 shared publications - Thanh Nguyen-Chi
Faculty of Civil Engineering, University of Transport and Communications
2 shared publications
Latest Publications
- 25 May 2026
- Published by Emerald Publishing in Engineering Computations
- Vol. 43 (9) , 1-21
An improved fitness distance balance H5N1 optimization algorithm for seismic design problems
Purpose: This study aims to develop an improved metaheuristic algorithm, FDB-SH5N1, by integrating the Fitness-Distance Balance (FDB) strategy into the SH5N1 framework to optimize seismic design of steel frame structures. The goal is to enhance convergence, stability, and global search ability in high-dimensional, nonlinear, and constrained optimization problems, ensuring structural safety and material efficiency in compliance with LRFD-AISC standards. Design/methodology/approach: The proposed FDB-SH5N1 algorithm integrates the Fitness-Distance Balance (FDB) selection mechanism into the viral-inspired SH5N1 metaheuristic to dynamically balance exploration and exploitation. The algorithm was implemented in MATLAB and validated on two benchmark seismic design problems involving four-story steel frames with 132 and 428 members. Design constraints followed LRFD-AISC and ASCE 7–10 standards. Performance was evaluated based on structural weight, convergence speed, and stability, and compared against several state-of-the-art metaheuristics, including PSO, GWO, QIO, COA, AOA, SH5N1, and AFDB-ARO. Findings: The FDB-SH5N1 algorithm outperformed classical and advanced metaheuristics in optimizing seismic steel frame designs. It achieved up to 96.67% reduction in structural weight compared to PSO and 9.33% over its SH5N1 predecessor, while maintaining compliance with LRFD-AISC constraints. The algorithm demonstrated faster convergence, superior stability, and consistent performance across 30 independent runs. Results confirmed its effectiveness in handling high-dimensional, constrained, and multimodal design problems, especially in preventing premature convergence and improving global search efficiency through the integration of the Fitness-Distance Balance mechanism. Originality/value: This study presents the first integration of the Fitness-Distance Balance (FDB) mechanism into the SH5N1 algorithm, resulting in a novel hybrid approach (FDB-SH5N1) tailored for seismic structural optimization. Unlike existing methods, the proposed algorithm enhances both exploration and exploitation through adaptive selection and mutation strategies inspired by viral dynamics. It offers a significant advancement in solving large-scale, constrained, and multimodal structural design problems. The algorithm’s ability to produce lighter, code-compliant designs with improved convergence and stability highlights its value for engineering applications requiring robust and efficient seismic optimization.
- 25 May 2026
- Published by Emerald Publishing in Engineering Computations
- Vol. 43 (9) , 1-21
An improved fitness distance balance H5N1 optimization algorithm for seismic design problems
Purpose: This study aims to develop an improved metaheuristic algorithm, FDB-SH5N1, by integrating the Fitness-Distance Balance (FDB) strategy into the SH5N1 framework to optimize seismic design of steel frame structures. The goal is to enhance convergence, stability, and global search ability in high-dimensional, nonlinear, and constrained optimization problems, ensuring structural safety and material efficiency in compliance with LRFD-AISC standards. Design/methodology/approach: The proposed FDB-SH5N1 algorithm integrates the Fitness-Distance Balance (FDB) selection mechanism into the viral-inspired SH5N1 metaheuristic to dynamically balance exploration and exploitation. The algorithm was implemented in MATLAB and validated on two benchmark seismic design problems involving four-story steel frames with 132 and 428 members. Design constraints followed LRFD-AISC and ASCE 7–10 standards. Performance was evaluated based on structural weight, convergence speed, and stability, and compared against several state-of-the-art metaheuristics, including PSO, GWO, QIO, COA, AOA, SH5N1, and AFDB-ARO. Findings: The FDB-SH5N1 algorithm outperformed classical and advanced metaheuristics in optimizing seismic steel frame designs. It achieved up to 96.67% reduction in structural weight compared to PSO and 9.33% over its SH5N1 predecessor, while maintaining compliance with LRFD-AISC constraints. The algorithm demonstrated faster convergence, superior stability, and consistent performance across 30 independent runs. Results confirmed its effectiveness in handling high-dimensional, constrained, and multimodal design problems, especially in preventing premature convergence and improving global search efficiency through the integration of the Fitness-Distance Balance mechanism. Originality/value: This study presents the first integration of the Fitness-Distance Balance (FDB) mechanism into the SH5N1 algorithm, resulting in a novel hybrid approach (FDB-SH5N1) tailored for seismic structural optimization. Unlike existing methods, the proposed algorithm enhances both exploration and exploitation through adaptive selection and mutation strategies inspired by viral dynamics. It offers a significant advancement in solving large-scale, constrained, and multimodal structural design problems. The algorithm’s ability to produce lighter, code-compliant designs with improved convergence and stability highlights its value for engineering applications requiring robust and efficient seismic optimization.
- 6 April 2026
- Published by Emerald Publishing in Engineering Computations
- Vol. 43 (9) , 1-23
Deep learning-based time series classification for damage assessment in multi-layer composite structures
Purpose: The purpose of this research is to address existing limitations in structural health monitoring (SHM) by proposing a novel deep learning model specifically designed for accurate time-series data classification. By integrating advanced neural network architectures – including 1-Dimensional Convolutional Neural Networks (1DCNN), Residual Networks (ResNet), Bidirectional Gated Recurrent Units (BiGRU), and Attention Mechanism (AM) – the proposed model effectively captures spatial relationships, highlights critical input sequences, maintains long-term dependencies, and mitigates vanishing gradient issues. The objective is to significantly enhance damage detection accuracy and reliability in multi-layer composite structures, improving upon existing ML methods by achieving superior precision and lower loss values in various structural damage scenarios. Design/methodology/approach: The proposed methodology combines advanced deep learning techniques for classifying structural damage from time-series data. Initially, spatial features are extracted using a 1-Dimensional Convolutional Neural Network (1DCNN). A Bidirectional Gated Recurrent Unit (BiGRU) captures temporal dependencies in both past and future contexts. Residual Networks (ResNet) with double skip connections are employed to address vanishing gradient issues by merging initial inputs with intermediate features. An Attention Mechanism (AM) is integrated to emphasize critical segments within the input sequences. The model is validated experimentally on multi-layer composite structures under various damage scenarios, assessing performance through metrics such as precision and loss. Findings: The study reveals that the proposed 1DCNN-BiGRU-ResNet-AM model significantly outperforms alternative methods, including standalone 1DCNN, GRU, BiGRU, combined 1DCNN-BiGRU, and 1DCNN-BiGRU-ResNet models. Specifically, it achieves superior damage detection accuracy with a precision of 0.902 and a loss value of 0.416. These findings demonstrate that integrating spatial feature extraction, bidirectional temporal analysis, residual connections, and attention mechanisms effectively enhances the identification and classification of structural damage. The results validate the proposed model's ability to overcome limitations related to spatial connectivity, gradient vanishing, and temporal dependencies, proving its practical effectiveness for structural health monitoring tasks involving multi-layer composite structures. Research limitations/implications: This study focuses primarily on controlled laboratory scenarios using multi-layer composite structures, which may limit direct generalization to complex real-world structural health monitoring tasks. Future research should validate the proposed model using diverse field data, considering environmental noise and operational variability. Additionally, while the model addresses vanishing gradients and captures spatial-temporal dependencies effectively, its computational complexity and resource demands might be high, potentially constraining deployment in resource-limited settings. Exploring model optimization, efficiency enhancements, and assessing scalability for practical structural monitoring systems would extend its applicability and robustness. Originality/value: The study's originality lies in developing a novel integrated 1DCNN-BiGRU-ResNet-AM model, uniquely combining advanced deep learning techniques to significantly enhance structural health monitoring accuracy. By leveraging 1DCNN for spatial feature extraction, BiGRU for effective temporal modeling, ResNet's dual skip connections to alleviate vanishing gradients, and Attention Mechanism (AM) for emphasizing critical data segments, the model uniquely addresses existing limitations. The approach also incorporates extensive feature integration, optimized network parameters, and sophisticated data augmentation methods. This comprehensive integration demonstrates notable advancements over traditional methods, providing robust, accurate, and reliable damage prediction in composite structures, substantially contributing to the field of intelligent structural health monitoring.
- 6 April 2026
- Published by Emerald Publishing in Engineering Computations
- Vol. 43 (9) , 1-23
Deep learning-based time series classification for damage assessment in multi-layer composite structures
Purpose: The purpose of this research is to address existing limitations in structural health monitoring (SHM) by proposing a novel deep learning model specifically designed for accurate time-series data classification. By integrating advanced neural network architectures – including 1-Dimensional Convolutional Neural Networks (1DCNN), Residual Networks (ResNet), Bidirectional Gated Recurrent Units (BiGRU), and Attention Mechanism (AM) – the proposed model effectively captures spatial relationships, highlights critical input sequences, maintains long-term dependencies, and mitigates vanishing gradient issues. The objective is to significantly enhance damage detection accuracy and reliability in multi-layer composite structures, improving upon existing ML methods by achieving superior precision and lower loss values in various structural damage scenarios. Design/methodology/approach: The proposed methodology combines advanced deep learning techniques for classifying structural damage from time-series data. Initially, spatial features are extracted using a 1-Dimensional Convolutional Neural Network (1DCNN). A Bidirectional Gated Recurrent Unit (BiGRU) captures temporal dependencies in both past and future contexts. Residual Networks (ResNet) with double skip connections are employed to address vanishing gradient issues by merging initial inputs with intermediate features. An Attention Mechanism (AM) is integrated to emphasize critical segments within the input sequences. The model is validated experimentally on multi-layer composite structures under various damage scenarios, assessing performance through metrics such as precision and loss. Findings: The study reveals that the proposed 1DCNN-BiGRU-ResNet-AM model significantly outperforms alternative methods, including standalone 1DCNN, GRU, BiGRU, combined 1DCNN-BiGRU, and 1DCNN-BiGRU-ResNet models. Specifically, it achieves superior damage detection accuracy with a precision of 0.902 and a loss value of 0.416. These findings demonstrate that integrating spatial feature extraction, bidirectional temporal analysis, residual connections, and attention mechanisms effectively enhances the identification and classification of structural damage. The results validate the proposed model's ability to overcome limitations related to spatial connectivity, gradient vanishing, and temporal dependencies, proving its practical effectiveness for structural health monitoring tasks involving multi-layer composite structures. Research limitations/implications: This study focuses primarily on controlled laboratory scenarios using multi-layer composite structures, which may limit direct generalization to complex real-world structural health monitoring tasks. Future research should validate the proposed model using diverse field data, considering environmental noise and operational variability. Additionally, while the model addresses vanishing gradients and captures spatial-temporal dependencies effectively, its computational complexity and resource demands might be high, potentially constraining deployment in resource-limited settings. Exploring model optimization, efficiency enhancements, and assessing scalability for practical structural monitoring systems would extend its applicability and robustness. Originality/value: The study's originality lies in developing a novel integrated 1DCNN-BiGRU-ResNet-AM model, uniquely combining advanced deep learning techniques to significantly enhance structural health monitoring accuracy. By leveraging 1DCNN for spatial feature extraction, BiGRU for effective temporal modeling, ResNet's dual skip connections to alleviate vanishing gradients, and Attention Mechanism (AM) for emphasizing critical data segments, the model uniquely addresses existing limitations. The approach also incorporates extensive feature integration, optimized network parameters, and sophisticated data augmentation methods. This comprehensive integration demonstrates notable advancements over traditional methods, providing robust, accurate, and reliable damage prediction in composite structures, substantially contributing to the field of intelligent structural health monitoring.
- 11 March 2025
- Published by Emerald Publishing in Engineering Computations
- Vol. 42 (3) , 1024-1096
The H5N1 algorithm: a viral-inspired optimization for solving real-world engineering problems
Purpose: In recent years, the development of metaheuristic algorithms for solving optimization problems within a reasonable timeframe has garnered significant attention from the global scientific community. In this work, a new metaheuristic algorithm inspired by the inflection mechanism of the avian influenza virus H5N1 in poultry and humans, taking into account its mutation mechanism, called H5N1.Design/methodology/approach: This algorithm aims to explore optimal solutions for optimization problems by simulating the adaptive behavior and evolutionary process of the H5N1 virus, thereby enhancing the algorithm’s performance for all types of optimization problems. Additionally, a balanced stochastic probability mechanism derived from the infection probability is presented. Using this mechanism, the H5N1 algorithm can change its phrase, including exploitation and exploration phases. Two versions of H5N1, SH5N1 and MH5N1, are presented to solve single-objective optimization problems (SOPs) and multi-objective optimization problems (MOPs).Findings: The performance of the algorithm is evaluated using a set of benchmark functions, including seven unimodal, six multimodal, ten fixed-dimension multimodal to solve SOPs, ZDT functions and CEC2009 has been used to demonstrate its superiority over other recent algorithms. Finally, six optimization engineering problems have been tested. The results obtained indicate that the proposed algorithm outperformed ten algorithms in SOPs and seven algorithms in MOPs.Originality/value: The experimental findings demonstrate the outstanding convergence of the H5N1 algorithm and its ability to generate solutions of superior quality.
- 11 March 2025
- Published by Emerald Publishing in Engineering Computations
- Vol. 42 (3) , 1024-1096
The H5N1 algorithm: a viral-inspired optimization for solving real-world engineering problems
Purpose: In recent years, the development of metaheuristic algorithms for solving optimization problems within a reasonable timeframe has garnered significant attention from the global scientific community. In this work, a new metaheuristic algorithm inspired by the inflection mechanism of the avian influenza virus H5N1 in poultry and humans, taking into account its mutation mechanism, called H5N1.Design/methodology/approach: This algorithm aims to explore optimal solutions for optimization problems by simulating the adaptive behavior and evolutionary process of the H5N1 virus, thereby enhancing the algorithm’s performance for all types of optimization problems. Additionally, a balanced stochastic probability mechanism derived from the infection probability is presented. Using this mechanism, the H5N1 algorithm can change its phrase, including exploitation and exploration phases. Two versions of H5N1, SH5N1 and MH5N1, are presented to solve single-objective optimization problems (SOPs) and multi-objective optimization problems (MOPs).Findings: The performance of the algorithm is evaluated using a set of benchmark functions, including seven unimodal, six multimodal, ten fixed-dimension multimodal to solve SOPs, ZDT functions and CEC2009 has been used to demonstrate its superiority over other recent algorithms. Finally, six optimization engineering problems have been tested. The results obtained indicate that the proposed algorithm outperformed ten algorithms in SOPs and seven algorithms in MOPs.Originality/value: The experimental findings demonstrate the outstanding convergence of the H5N1 algorithm and its ability to generate solutions of superior quality.
- 8 July 2024
- Published by Elsevier in Construction and Building Materials
- Vol. 439
Enhancing bridge damage assessment: Adaptive cell and deep learning approaches in time-series analysis
- 8 July 2024
- Published by Elsevier in Construction and Building Materials
- Vol. 439
Enhancing bridge damage assessment: Adaptive cell and deep learning approaches in time-series analysis
- Open Access
- 4 April 2024
- Published by Elsevier in KSCE Journal of Civil Engineering
- Vol. 28 (7) , 2920-2933
A Prospective Technique for Damage Detection in Truss Structures Using the Fusion of DNN with AVOA
- 4 April 2024
- Published by Elsevier in KSCE Journal of Civil Engineering
- Vol. 28 (7) , 2920-2933
Open AccessA Prospective Technique for Damage Detection in Truss Structures Using the Fusion of DNN with AVOA
- Open Access
- 9 March 2024
- Published by Elsevier in KSCE Journal of Civil Engineering
- Vol. 28 (6) , 2355-2365
Enhancing Salp Swarm Optimization with Orthogonal Diagonalization Transformation for Damage Detection in Truss Bridge
- 9 March 2024
- Published by Elsevier in KSCE Journal of Civil Engineering
- Vol. 28 (6) , 2355-2365
Open AccessEnhancing Salp Swarm Optimization with Orthogonal Diagonalization Transformation for Damage Detection in Truss Bridge
- 5 December 2021
- Published by Springer Nature
- p. 37-48
Topology Optimization for a Large-Scale Truss Bridge Using a Hybrid Metaheuristic Search Algorithm
- 5 December 2021
- Published by Springer Nature
- p. 37-48
Topology Optimization for a Large-Scale Truss Bridge Using a Hybrid Metaheuristic Search Algorithm