This session invites researchers, industry professionals, and stakeholders in structural engineering and construction to explore the integration of artificial intelligence (AI) with traditional engineering methods. AI-driven solutions offer efficiencies and insights that address the limitations of classical approaches, paving the way for advancements in design, materials, and resource management. Attendees will gain an understanding of AI’s transformative potential in construction and structural applications, learning how to harness its capabilities for innovation in materials science and advanced structures. Learner outcomes include insights into AI adoption strategies, enhanced materials control, and interdisciplinary approaches to overcoming technical challenges in modern engineering.
Learning Objectives:
(1) Discuss how to use machine learning and optimization;
(2) Use predictive modeling;
(3) Discuss AI-driven solutions;
(4) Examine applications of Artificial Intelligence.
A Novel Approach for Predicting FRP Debonding Strain in Concrete Using an Optimized Self-learning Model
Presented By: Nima Khodadadi
Affiliation: University of California, Berkeley
Description: Debonding strain is an important parameter for assessing the performance of Fiber-Reinforced Polymer (FRP) externally strengthened concrete structures. It measures the strength and stability of the interface bond, aiding engineers in selecting appropriate FRP materials and construction methods to enhance structural durability and safety. Extensive research has demonstrated that the debonding strain between FRP and concrete is influenced by multiple random variables, exhibiting highly complex nonlinear relationships that cannot be accurately predicted using simple linear regression methods. Therefore, this paper proposes a predictive model using the Support Vector Machine (SVM) method, which is fine-tuned using the Enhanced Cat Swarm Optimization (ECSO) algorithm to predict FRP debonding strain, leveraging its advantages of nonlinear mapping, robustness, and strong generalization capability. The input variables include concrete strength (f’c), shear span ratio (?), tensile steel reinforcement ratio (?s), steel yield strength, steel stirrup reinforcement ratio (?sv), FRP axial stiffness (Eftf), and the ratio of anchorage length to cross-sectional dimension (La/L0) with the output being the debonding strain of FRP. The results indicate that the ECSO-optimized SVM outperforms alternative algorithms in both identification capability and prediction accuracy, proving its effectiveness in assessing the bonding performance of FRP-strengthened concrete structures in infrastructure applications.
Use of Artificial Intelligence to Facilitate Bridge Design for Service Life
Presented By: Atorod Azizinamini
Affiliation: Florida International University
Description: As a result of Strategic Highway Research Program 2 (SHRP2), R19A project a major document was developed for service life design of bridges. Later this document became the source for development of AASHTO Guide for service life design of bridges.
The major deliverable for SHRP2 R19A project was “Design Guide for Bridges for Service Life”, hereafter referred to as Guide.
Guide includes many flowcharts and step by step procedures for design of highway bridges to provide more than 100 years of service life without major interruption.
This presentation will provide an outline of efforts that are underway to utilize data embedded within Guide and other available information for service life design of bridges, in conjunction with AI, to provide an effective tool for bridge engineers. This tool will greatly facilitate the service life design of bridges.
Deep Learning for Earthquake Engineering: Temporal Convolutional Networks for Structural Response Prediction and Ground Motion Reconstruction
Presented By: Kwok Pang
Affiliation: University of California, Berkeley
Description: Temporal Convolutional Networks (TCNs) have shown significant potential in seismic applications, particularly for predicting structural responses and reconstructing ground motion histories. Initial studies highlight their ability to accurately forecast linear structural responses and infer ground motion from response data, demonstrating their practical utility. However, modeling complex nonlinear behaviors, such as damping variations and period elongation, exposes limitations in the TCN framework. To address these challenges, we propose an enhanced TCN that incorporates residual blocks to improve feature representation and combines time-domain and frequency-domain loss functions to capture both temporal and spectral characteristics of structural dynamics. Additionally, the model incorporates an Exponentially Weighted Moving Average feature to emphasize recent data while accounting for historical trends. Validation experiments reveal that the enhanced TCN delivers a better accuracy in predicting nonlinear dynamics. This study underscores the adaptability and scalability of TCNs, establishing a robust foundation for future advancements in seismic response prediction, including the integration of physics-informed neural networks.
Big Data Driven Assessment of Vehicles Crossing a 3-Tower Cable Stayed Bridge
Presented By: Michael Forde
Affiliation: University of Edinburgh
Description: The Queensferry Crossing is the world’s longest 3 tower cable stayed bridge. The towers are the tallest bridge towers in the UK.
With over 2,000 active sensors continuously taking readings – analysis of this “Big Data” presents many challenges. These challenges will be addressed in this paper.
In particular, responses from Weigh-in-Motion sensors will be statistically analysed using Python software.
Transcending Domain Knowledge Limitations through Explainable Artificial Intelligence, Causal Discovery and Inference: A Modern Approach to Predict Fire-induced Spalling of Concrete
Presented By: M. Z. Naser
Affiliation: Clemson
Description: Fire-induced concrete spalling poses a significant threat to the integrity of concrete as it has been shown to trigger collapse in multiple incidents around the world. Despite considerable research efforts, an explicit quantification of the critical factors influencing concrete spalling remains a burning question. This presentation develops a comprehensive multi-method framework to advance the understanding of a complex nonlinear multivariate phenomenon (i.e., fire-induced spalling of concrete). Early in the workflow, quantitative spalling indices are created using statistical transformations and logistic regression modelling to measure spalling propensity empirically. Building on this foundation, a database of over 1000 fire tests is leveraged to train AI-based predictive models that underwent a sensitivity analysis to obtain the best-performing model through various evaluation metrics. Still, the findings obtained so far can primarily be used to predict spalling and not to explain underlying mechanisms. To address this gap, this framework creates a state-of-the-art causal discovery model to visualize the causal relationships among key influencing factors. To further complement this framework, the casual inference was implemented to quantify the critical factors' causal effect on concrete under elevated temperatures, given the contribution of other confounders. Although a similar methodology can be used on all factors, this presentation highlights PP fiber and moisture content due to their codified importance in building codes. Nonetheless, the causal inference results indicate that the inclusion of PP fibers can reduce the probability of spalling by approximately 20–31%, whereas higher moisture levels can increase spalling likelihood by about 15–21%.