Deep Learning Techniques for Predicting Stress Fields in Composite Materials: A Superior Alternative to Finite Element Analysis

Journal of Composites Science, Journal Year: 2023, Volume and Issue: 7(8), P. 311 - 311

Published: July 28, 2023

Stress evaluation plays a pivotal role in the design of material systems, often accomplished through finite element method (FEM) for intricate structures. However, substantial costs and time requirements associated with multi-scale FEM analyses have prompted growing interest adopting more efficient, machine-learning-driven strategies. This study investigates utilization advanced machine learning techniques predicting local stress fields composite materials, presenting it as superior alternative to traditional approaches. The primary objective this research is develop predictive model field maps components featuring diverse configurations fibers distributed within matrix. To achieve this, we employ Convolutional Neural Network (CNN) specialized U-Net architecture, enabling correlation spatial fiber organization resultant von Mises field. CNN was extensively trained using four distinct data sets, encompassing uniform fibrous structures, non-uniform irregularly shaped comprehensive combination these sets. models demonstrate exceptional proficiency fields, yielding impressive structural similarity index scores (SSIM) 0.977 mean squared errors (MSE) 0.0009 on dedicated test set. harnesses 2D cross-sectional imagery establish surrogate analysis, offering an accurate efficient approach design, irrespective geometric complexity or boundary conditions.

Language: Английский

---

Optimizing bio-hybrid composites for impact resistance using machine learning

Journal of the Brazilian Society of Mechanical Sciences and Engineering, Journal Year: 2025, Volume and Issue: 47(5)

Published: April 4, 2025

Language: Английский

---

Applications of Deep Learning for Composites Materials

Published: Jan. 1, 2024

Composite materials are highly sought-after in various industries for their extraordinary properties. These created by combining two or more different substances, resulting a novel material with improved characteristics. However, composites and laminates display intricate structure patterns, which can be considered as unstructured data. Currently, deep learning is experiencing rapid advancements the field of composite materials, providing prediction enhancement, characterization, structural health monitoring, process optimization, more. It facilitates analysis such complex data patterns effectively automates identification features. This chapter begins high-level overview methods. then explores recent developments use machine depth. To conclude this review, we discuss revolutionary approaches to designing optimizing next generation unprecedented properties, well limitations, challenges, potential growth areas methods context materials.

Language: Английский

---

Time mesh independent framework for learning materials constitutive relationships

Engineering Applications of Artificial Intelligence, Journal Year: 2024, Volume and Issue: 137, P. 109165 - 109165

Published: Sept. 6, 2024

Language: Английский

---

A Deep Learning Framework for Predicting the Heterogeneous Stiffness Map of Brain White Matter Tissue

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: April 8, 2024

Abstract Finding the stiffness map of biological tissues is great importance in evaluating their healthy or pathological conditions. However, due to heterogeneity and anisotropy fibrous tissues, this task presents challenges significant uncertainty when characterized only by single-mode loading experiments. In study, we propose a new theoretical framework landscape specifically focusing on brain white matter tissue. Initially, finite element model tissue was subjected six cases, corresponding stress-strain curves were characterized. By employing multiobjective optimization, material constants an equivalent anisotropic inversely extracted best fit all modes simultaneously. Subsequently, large-scale simulations conducted, incorporating various fiber volume fractions orientations, train convolutional neural network capable predicting properties solely based architecture any given The method applied local imaging data tissue, demonstrating its effectiveness precisely mapping behavior long-term, proposed may find applications traumatic injury, folding studies, neurodegenerative diseases, where accurately capturing crucial for

Language: Английский

---

Prediction of thermo-mechanical properties of PCB conductive layers using convolutional neural networks

Published: April 17, 2023

Deriving the effective material properties of printed circuit board (PCB) layers is a particular challenge due to intricate patterns conductive artwork. In this paper, we develop convolutional neural network model predict orthotropic thermo-mechanical copper PCBs. To ensure necessary level detail for modeling, images copper-resin distributions are extracted from real electronic design files as patches preselected pixel sizes. For each image, corresponding homogenized computed using finite element analyses. After training process on formed dataset completed, developed validated data unseen during training. The obtained results indicate that considered approach suitable prediction patterns, and moreover, promising facilitating present-day time-to-market objectives in industry.

Language: Английский

---

Bending analysis of glass fiber reinforced epoxy composites/copper-clad laminates for multi-layer printed circuit boards

Hybrid Advances, Journal Year: 2023, Volume and Issue: 4, P. 100090 - 100090

Published: Sept. 21, 2023

Printed circuit board (PCB) is the most important part of any electronic device which made copper-clad laminate and glass fiber-reinforced composites. Since ply orientation composites lamina thickness have a significant influence on mechanical properties whole composite it necessary to investigate effects bending PCB so that they can be manufactured meet service requirements. In this work, were investigated for seven different orientations laminas eight combinations (for constant thickness) laminate. A commercially available finite element analysis software (Abaqus) was used simulate three-point test PCBs simulation results validated using experimental results. It found stiffness maximum cross-ply The introduction angle improves von-Mises stress with an insignificant cost (less than 5% reduction). regulated through variation constituent lamina. increased by increasing stiffer or placing towards surface outcome research would provide comprehensive understanding characteristics multi-layered directly utilized manufacturers. However, behavior subject temperature variation, impact loading, vibration, other conditions might affect life should investigated.

Language: Английский

---

Lightweight Bi-LSTM method for the prediction of mechanical properties of concrete

Multimedia Tools and Applications, Journal Year: 2023, Volume and Issue: 83(18), P. 54863 - 54884

Published: Dec. 11, 2023

Language: Английский

---

Deep Learning-Based Prediction of Stress and Strain Maps in Arterial Walls for Improved Cardiovascular Risk Assessment

Applied Sciences, Journal Year: 2023, Volume and Issue: 14(1), P. 379 - 379

Published: Dec. 31, 2023

Conducting computational stress-strain analysis using finite element methods (FEM) is a common approach when dealing with the complex geometries of atherosclerosis, which leading cause global mortality and cardiovascular disease. The considerable expense linked to FEM encourages substitution considerably faster data-driven machine learning (ML) approach. This study investigated potential end-to-end deep tools as more effective substitute for in predicting fields within 2D cross sections arterial walls. We first proposed U-Net-based fully convolutional neural network (CNN) predict von Mises stress strain distribution based on spatial arrangement calcification wall cross-sections. Further, we developed conditional generative adversarial (cGAN) enhance, particularly from perceptual perspective, prediction accuracy field maps walls various quantities configurations. On top U-Net cGAN, also their ensemble approaches improve further. Our dataset, consisting input output images, was generated by implementing boundary conditions extracting maps. trained models can accurately fields, structural similarity index scores (SSIM) 0.854 0.830 mean squared errors 0.017 0.018 strain, respectively, reserved test set. Meanwhile, cGAN combination transfer techniques demonstrate high evidenced SSIM 0.890 0.803 strain. Additionally, 0.008 further support model’s performance designated Overall, this surrogate model analysis, efficiently regardless conditions.

Language: Английский

---

Machine Learning-Enabled Data-Driven Research on Paper-Reinforced Composite Materials

Materials horizons, Journal Year: 2024, Volume and Issue: unknown, P. 327 - 339

Published: Jan. 1, 2024

Language: Английский

---