Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating Wide and Deep Neural Networks with Squeeze-and-Excitation Blocks for Multi-Target Property Prediction in Additively Manufactured Fiber Reinforced Composites

Behzad Parvaresh, Rahmat K. Adesunkanmi, Adel Alaeddini

TL;DR

This work tackles the challenge of predicting multiple mechanical and manufacturing properties for CFRC-AM under a large, mixed-parameter design space. It introduces a data-efficient SE-WDNN that fuses a wide and deep architecture with squeeze-and-excitation blocks and uses LHS-guided experimentation to build an eight-target predictive model, achieving a best overall MAPE of $MAPE = 12.33\%$. SHAP analysis reveals that reinforcement strategy (e.g., number of reinforced layers and fiber type) is the dominant factor shaping mechanical performance, guiding design decisions towards fiber-driven optimization. The combination of LHS sampling, SE-WDNN modeling, and interpretability enables efficient, interpretable multi-target predictions that can steer parameter selection and design exploration in CFRC-AM, while highlighting areas for future validation and integration with optimization workflows.

Abstract

Continuous fiber-reinforced composite manufactured by additive manufacturing (CFRC-AM) offers opportunities for printing lightweight materials with high specific strength. However, their performance is sensitive to the interaction of process and material parameters, making exhaustive experimental testing impractical. In this study, we introduce a data-efficient, multi-input, multi-target learning approach that integrates Latin Hypercube Sampling (LHS)-guided experimentation with a squeeze-and-excitation wide and deep neural network (SE-WDNN) to jointly predict multiple mechanical and manufacturing properties of CFRC-AMs based on different manufacturing parameters. We printed and tested 155 specimens selected from a design space of 4,320 combinations using a Markforged Mark Two 3D printer. The processed data formed the input-output set for our proposed model. We compared the results with those from commonly used machine learning models, including feedforward neural networks, Kolmogorov-Arnold networks, XGBoost, CatBoost, and random forests. Our model achieved the lowest overall test error (MAPE = 12.33%) and showed statistically significant improvements over the baseline wide and deep neural network for several target variables (paired t-tests, p <= 0.05). SHapley Additive exPlanations (SHAP) analysis revealed that reinforcement strategy was the major influence on mechanical performance. Overall, this study demonstrates that the integration of LHS and SE-WDNN enables interpretable and sample-efficient multi-target predictions, guiding parameter selection in CFRC-AM with a balance between mechanical behavior and manufacturing metrics.

Integrating Wide and Deep Neural Networks with Squeeze-and-Excitation Blocks for Multi-Target Property Prediction in Additively Manufactured Fiber Reinforced Composites

TL;DR

This work tackles the challenge of predicting multiple mechanical and manufacturing properties for CFRC-AM under a large, mixed-parameter design space. It introduces a data-efficient SE-WDNN that fuses a wide and deep architecture with squeeze-and-excitation blocks and uses LHS-guided experimentation to build an eight-target predictive model, achieving a best overall MAPE of . SHAP analysis reveals that reinforcement strategy (e.g., number of reinforced layers and fiber type) is the dominant factor shaping mechanical performance, guiding design decisions towards fiber-driven optimization. The combination of LHS sampling, SE-WDNN modeling, and interpretability enables efficient, interpretable multi-target predictions that can steer parameter selection and design exploration in CFRC-AM, while highlighting areas for future validation and integration with optimization workflows.

Abstract

Continuous fiber-reinforced composite manufactured by additive manufacturing (CFRC-AM) offers opportunities for printing lightweight materials with high specific strength. However, their performance is sensitive to the interaction of process and material parameters, making exhaustive experimental testing impractical. In this study, we introduce a data-efficient, multi-input, multi-target learning approach that integrates Latin Hypercube Sampling (LHS)-guided experimentation with a squeeze-and-excitation wide and deep neural network (SE-WDNN) to jointly predict multiple mechanical and manufacturing properties of CFRC-AMs based on different manufacturing parameters. We printed and tested 155 specimens selected from a design space of 4,320 combinations using a Markforged Mark Two 3D printer. The processed data formed the input-output set for our proposed model. We compared the results with those from commonly used machine learning models, including feedforward neural networks, Kolmogorov-Arnold networks, XGBoost, CatBoost, and random forests. Our model achieved the lowest overall test error (MAPE = 12.33%) and showed statistically significant improvements over the baseline wide and deep neural network for several target variables (paired t-tests, p <= 0.05). SHapley Additive exPlanations (SHAP) analysis revealed that reinforcement strategy was the major influence on mechanical performance. Overall, this study demonstrates that the integration of LHS and SE-WDNN enables interpretable and sample-efficient multi-target predictions, guiding parameter selection in CFRC-AM with a balance between mechanical behavior and manufacturing metrics.
Paper Structure (35 sections, 18 equations, 14 figures, 8 tables)

This paper contains 35 sections, 18 equations, 14 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Materials and Manufacturing Parameters of CFRC-AM
  4. Plastic and Fiber Types
  5. Fill Density $(\%)$
  6. Infill Pattern
  7. Number of Reinforced layers
  8. Fiber Orientation ($^\circ$)
  9. Concentric Rings
  10. Manufacturing Process
  11. Design of Experiments
  12. LHS formulation.
  13. Implementation details (mixed factors, optimization, diagnostics).
  14. Mechanical Testing
  15. Wide and deep neural network with Squeeze-and-Excitation Prediction Model
  16. ...and 20 more sections

Figures (14)

  • Figure 1: (a) Microscopic images of the thermoplastic matrix filaments used in this work (b) Microscopic images of the continuous-fiber reinforcements
  • Figure 2: (a) Fill density levels for CFRC-AM specimens with hexagonal infill; from 28% to 52% (b) Representative infill patterns for CFRC-AM.
  • Figure 3: (a) Fiber orientation configurations, and (b) concentric ring configurations in CFRC-AM specimens. Each panel shows schematic tool paths on the left and the printed specimens on the right.
  • Figure 4: Markforged Mark Two composite 3D printing setup and test-specimen geometry. Left: the printer used in this study, which combines fused filament fabrication with continuous-fiber reinforcement via a dual-nozzle printhead (insets show the build plate and tool head). Right: CAD model of the ASTM D638 Type I tensile specimen with key dimensions indicated.
  • Figure 5: Distribution of the manufacturing parameters used as input features for CFRC-AM specimens, including categorical parameters (plastic type, fiber type, infill pattern) and numerical parameters (fiber orientation, concentric rings, fill density, number of reinforced layers).
  • ...and 9 more figures