Techno Press
Tp_Editing System.E (TES.E)
Login Search
You logged in as...

sss
 
CONTENTS
Volume 38, Number 4, October 2026
 

Abstract
The tensile properties of fused deposition modeling (FDM)-printed polymer structures are governed by both internal architecture and printing parameters. This article investigated the influence of infill topology (Grid, Gyroid, and Honeycomb), raster orientation, and infill percentage on tensile properties in FDM-printed PLA samples. Tensile tests were performed in order to determine the ultimate tensile strength (UTS), yield strength, elastic modulus, break elongation, and a quality index. Additionally, the experimental dataset was used to train ensemble machinelearning models for predicting tensile properties directly from printing parameters and calculated geometry. The results indicate that tensile strength and stiffness generally increased with increasing infill percentage (UTS ≈ 60.5 MPa) for 100% infill. Raster orientation had a major influence on anisotropic behavior and infill topology had a strong influence on deformation capacity and failure progression. Out of the models developed, the Histogram-Based Gradient Boosting Regressor (HGBR) had the highest prediction accuracy to predict elongation at break (R² = 0.938) and UTS (R² = 0.866). Nevertheless, the Random Forest (RF) Model had the best prediction for elastic modulus (R² = 0.920). Yield strength was comparatively more difficult to predict (R² = 0.511) as it was sensitive to local deformation mechanism. The presented model demonstrates the potential for AI-enabled techniques to rapid estimation of tensile strength parameters in material-extrusion additive manufacturing and parameter selection.

Key Words
ensemble learning; fused deposition modeling (FDM); infill density; infill topology; polylactic acid; raster angle; SHAP interpretability; tensile behavior

Address
(1) Ayman Nasir, Ahmad Manasrah:
Department of Mechanical Engineering, Al-Zaytoonah University of Jordan, Amman, Jordan;
(2) Yousef Jaradat:
Department of Electrical Engineering, Al-Zaytoonah University of Jordan, Amman, Jordan;
(3) Mohammed Jaradat:
Department of Civil and Infrastructure Engineering, Al-Zaytoonah University of Jordan, Amman, Jordan.

Abstract
Data-driven infrastructure health monitoring depends heavily on the availability and quality of labeled data for machine learning modeling. This paper presents a knowledge-informed, explainable machine learning framework that uses trendability-oriented engineered features together with a multi-objective optimization algorithm, the nondominated sorting genetic algorithm II (NSGA-II), to select domain-invariant features under time-varying environmental and deterioration-related uncertainty. The framework balances computational accuracy against featurespace dimensionality and is validated on a real-world pipeline corrosion dataset acquired using ultrasonic guided wavesunder nonstationary conditions, considering multiple excitation frequencies and propagation modes. Results indicate that the proposed approach improves interpretability and diagnostic performance relative to a single-dimension feature strategy, supporting its feasibility as a decision-aid for early risk detection in energy-related infrastructures. The methodological advances presented in this study are applicable to a wider range of sensing and monitoring problems and offer practical value for infrastructure maintenance planning.

Key Words
AI-enhanced sensing; early warning system; feature fusion; intelligent infrastructure health monitoring; pipeline corrosion detection; predictive maintenance; smart experimental modeling

Address
(1) Kong Fah Tee:
Department of Civil and Environmental Engineering, King Fahd University of Petroleum and Minerals, Dhahran 31261, Saudi Arabia;
(2) Kong Fah Tee:
Interdisciplinary Research Center for Construction and Building Materials, KFUPM, Dhahran 31261, Saudi Arabia;
(3) Milad Zarchi:
Faculty of Mechanical Engineering, Shahid Rajaee Teacher Training University, Tehran, Iran.

Abstract
Mechanical hole-making in carbon fiber reinforced polymer (CFRP) aerospace laminates is essential for structural assembly, yet high experimental costs create severe data scarcity that limits conventional AI models. To overcome this, this study presents an offline predictive framework designed to support digital twin implementations in small-sample environments. A baseline experimental matrix of 27 physical runs was conducted using a standard design of experiments and partitioned into a 70% training split (19 runs) and a 30% testing split (8 runs) to prevent data leakage. A Mega-Trend-Diffusion Virtual Sample Generation (VSG) pipeline was applied exclusively to the 19 training samples, expanding them into 500 high-fidelity virtual data points. The 8 physical test runs were kept completely untouched to ensure unbiased validation. A flat-architecture Broad Learning System (BLS) was trained on the augmented dataset to capture non-linear relationships across surface roughness (Ra), thrust force (Ft), and delamination factor (Fda). To evaluate process variable influences directly within the machine learning domain, Partial Dependence Plots (PDP) and Feature Importance mapping were utilized. Evaluated against the held-out physical test trials, the proposed VSG-BLS framework demonstrated strong predictive accuracy, achieving R2 values of 0.968 for Ra, 0.952 for Ft, and 0.944 for Fda. Compared to an unaugmented BLS baseline, the framework reduced prediction errors by 39.0% for thrust force and 66.2% for surface roughness, proving its capacity to mitigate data scarcity in shop floor applications.

Key Words
aerospace manufacturing; broad learning systems; cart regression; cfrp; drilling-induced damage; surface integrity; virtual sample generation

Address
(1) Shahnaz Mohammed Alkhalil:
Department of Mechanical Engineering, Faculty of Engineering and Technology, Al-Zaytoonah University of Jordan, Queen Alia Airport St 594, Amman 11733 Jordan;
(2) Mai Sharaf Aldeen Bani Younes:
Department of Alternative Energy Technology, Al-Zaytoonah University of Jordan, Amman 11733, Jordan.

Abstract
Internal cavity defects in post-tensioned prestressed pipelines can induce prestress loss and tendon corrosion, while conventional non-destructive methods show limitations in accurately identifying internal grouting defects. This paper presents a piezoelectric impedance-based grouting quality inspection method, with piezoelectric smart aggregates mounted on internal prestressed tendons for high-sensitivity detection. The root mean square deviation (RMSD) is adopted for quantitative evaluation of defect location and degree, and a multi-physics coupling finite element model is established for mechanism verification. The experimental results showed that one day after grouting, the peak frequency of the piezoelectric ceramic transducer (PZT) spectrum curve moved to a higher frequency, and then decreased. After normalization, changes in the peak frequency and peak electrical admittance slowed down. After 7 d, the change in grout material had no effect on the conductance and susceptance signals. The difference between the peak value and peak frequency of electrical admittance 14 d after grouting and that without grouting decreased with increasing grouting defects. The RMSD values of the conductance and susceptance frequency curves of each PZT increased with increasing post-grouting time; the smaller the RMSD value when it reached 14 d after grouting, the greater the degree of structural grouting defects. This study verifies the feasibility of piezoelectric impedance technology for prestressed pipeline grouting detection, clarifies the time-varying signal law during grout hardening, and provides an effective approach for internal defect inspection of closed prestressed ducts.

Key Words
conductance and susceptance; grouting defect; non-destructive testing; piezoelectric impedance technology; prestressed pipeline

Address
(1) Huizhen Shen, Liling Zhou, Shuping Shen, Peileng Zhang, Nana Tong, Feifei Chen:
Zhejiang Institute of Geosciences, Hangzhou 310007, China;
(2) Huizhen Shen, Liling Zhou, Shuping Shen, Peileng Zhang, Nana Tong, Feifei Chen:
Qiantang Institute of Geosciences, Hangzhou 310007, China;
(3) Huizhen Shen, Liling Zhou, Shuping Shen, Peileng Zhang, Nana Tong, Feifei Chen:
Observation and Research Station of Zhejiang Coastal City Geological Safety, Ministry of Natural Resources, Hangzhou 310000, China;
(4) Xinzhe Tang, Zhiming Liu, Yiming Wang, Jingke Liu:
College of Civil Engineering and Architecture, Wenzhou University, Wenzhou 325035, China;
(5) Xinzhe Tang:
Ocean College, Zhejiang University, Zhoushan, 316021, China.

Abstract
The 2017 Pohang earthquake (M 5.4) highlighted the need for rapid assessment of urban transportation vulnerability and evacuation performance following earthquakes. Bridge damage can reduce network connectivity, disrupt traffic flow, and increase evacuation time. Although traffic simulation can evaluate these impacts in detail, its computational cost limits real-time application. This study compares eight machine learning models—Linear Regression, KNN, ANN, SVR, Random Forest, LightGBM, XGBoost, and CatBoost—for predicting evacuation time estimates (ETE) under earthquake-induced bridge damage. Training data were generated from traffic simulations considering various bridge damage scenarios. Linear Regression and KNN showed limited predictive capability, while SVR partially captured nonlinear relationships. In contrast, ANN, Random Forest, and gradient boosting models achieved higher accuracy, with LightGBM, XGBoost, and CatBoost attaining R2 values of approximately 0.92. Among them, LightGBM provided the best balance between accuracy and computational efficiency. Monte Carlo simulations with unseen epicentral locations and earthquake magnitudes confirmed the robustness of gradient boosting models. All machine learning models required less than 0.2 s per scenario, demonstrating substantial computational advantages over conventional traffic simulation. The proposed framework enables rapid and reliable ETE prediction for postearthquake evacuation planning and decision support.

Key Words
bridge damage; disaster response; earthquake; evacuation time estimation; machine learning; traffic simulation

Address
(1) Minjeong Kim, HyungJoo Lee, Jinho Song, Sungsik Yoon:
Department of Artificial Intelligence, Hannam University, 70 Hannam-ro, Daejeon 34430, Republic of Korea;
(2) Segon Kwon:
Korail Research Institute, Korea Railroad Corporation, 240 Jungang‑Ro Daejeon 34618, Republic of Korea.

Abstract
High-rise modular steel structures have been utilized extensively in the current construction industry, possessing the merits of efficiency and mass production. However, these structures present deficiencies in seismic controllability. Therefore, this study takes high-rise fully modular steel structures as the target to conduct relevant research on seismic performance. This study adopts the internal force distribution and deformation rules of layered beams to build up a theoretical model involving multiple connection restraints. An earthquake progressive failure regulation technology based on adjustable braces is adopted for energy dissipation optimization. This technology uses multi-segment yield characteristics as input, and strengthens stiffness focus by introducing core plate deformation design. Plastic controllable joints and yield gradient dual constraints are integrated to calibrate prediction results. Simulation analysis was conducted on high-rise modular steel structures during experiments. In seismic brace performance experiments, the mid-span scheme reached a unidirectional baseline shear force of 11560 kN. Meanwhile, in cross-loading tests, the top-layer displacement angle of the structure was as low as 0.0004 after adopting the topology form of X-shaped cross-synergistic loading. Finally, in ultimate axial force analysis, the bottom axial compression safety of the load-adjustable buckling-restrained brace was reduced to 4405 kN. The results demonstrate the effectiveness of the proposed technology and provide technical support for the seismic design of high-rise modular steel structures.

Key Words
high-rise buildings; load-bearing capacity regulation; modularization; seismic performance; steel structures

Address
College of Civil Engineering, Longdong University, Qingyang, 745000, China.

Techno-Press: Publishers of international journals and conference proceedings.       Copyright © 2026 Techno-Press ALL RIGHTS RESERVED.
P.O. Box 33, Yuseong, Daejeon 34186 Korea.
General Inquiries: info@techno-press.com / Journal Administration: admin@techno-press.com