Machine-learning-assisted prediction and optimization of tensile strength in particulate-filled polymer composites

Abstract Particulate-filled polymer composites (PFPCs) are characterized by high specific strength, good formability, and considerable design flexibility, making them key materials for rapid tooling in aerospace equipment manufacturing. In this study, metal-powder/resin composites were investigated, and a tensile-strength dataset comprising 45 independent experimental conditions was established. Three machine-learning regression algorithms, namely linear least absolute shrinkage and selection operator (LASSO) regression, k-nearest neighbors (KNN), and support vector regression (SVR), were employed to develop tensile-strength prediction models. A fixed outer five-fold cross-validation procedure was used to obtain out-of-fold (OOF) predictions, while hyperparameters were selected through inner five-fold cross-validation within each outer training fold; a genetic algorithm (GA) was subsequently used to further optimize the SVR hyperparameters. The OOF RMSE, MAE, and R² values of SVR were 1.67 MPa, 1.32 MPa, and 0.882, respectively, and those of GA-SVR were 1.54 MPa, 1.21 MPa, and 0.901. With increasing particle volume fraction and particle size, the tensile strength initially increased and then decreased, whereas ball-milling time had a relatively limited effect. Single-parameter analysis yielded predicted peak positions of approximately 34.80%, 30.92 μm, and 2.68 h for particle volume fraction, particle size, and ball-milling time, respectively. Continuous searching within the experimental domain identified a high-strength parameter combination of f = 34.63%, D = 30.67 μm, and t = 2.65 h, corresponding to a predicted tensile strength of 61.33 MPa. The measured strengths of three independent validation specimens prepared using a nominal particle size of 31 μm were 61.12, 60.93, and 60.56 MPa, with a mean of 60.87 ± 0.28 MPa, a coefficient of variation of 0.47%, and a relative deviation of 0.75% between the experimental mean and model prediction. Within the current experimental domain, the proposed framework can provide a quantitative reference for PFPC tensile-strength prediction and formulation/process-parameter screening, and offers an efficient strategy for accelerating formulation design and optimization while reducing experimental cost. Its implementation also provided students with practical experience in applying machine-learning algorithms to material-strength prediction and parameter design.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-15
DOI
https://doi.org/10.1038/s41598-026-78358-9
Primary Topic
Machine Learning in Materials Science
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Machine-learning-assisted prediction and optimization of tensile strength in particulate-filled polymer composites

Zhuang Bailiang, Zhang Wen, Zhu Fuxian
5 links
Scientific Reports
Machine Learning in Materials Science
article

Machine-learning-assisted prediction and optimization of tensile strength in particulate-filled polymer composites

Zhuang Bailiang, Zhang Wen, Zhu Fuxian
article en

Abstract

Abstract Particulate-filled polymer composites (PFPCs) are characterized by high specific strength, good formability, and considerable design flexibility, making them key materials for rapid tooling in aerospace equipment manufacturing. In this study, metal-powder/resin composites were investigated, and a tensile-strength dataset comprising 45 independent experimental conditions was established. Three machine-learning regression algorithms, namely linear least absolute shrinkage and selection operator (LASSO) regression, k-nearest neighbors (KNN), and support vector regression (SVR), were employed to develop tensile-strength prediction models. A fixed outer five-fold cross-validation procedure was used to obtain out-of-fold (OOF) predictions, while hyperparameters were selected through inner five-fold cross-validation within each outer training fold; a genetic algorithm (GA) was subsequently used to further optimize the SVR hyperparameters. The OOF RMSE, MAE, and R² values of SVR were 1.67 MPa, 1.32 MPa, and 0.882, respectively, and those of GA-SVR were 1.54 MPa, 1.21 MPa, and 0.901. With increasing particle volume fraction and particle size, the tensile strength initially increased and then decreased, whereas ball-milling time had a relatively limited effect. Single-parameter analysis yielded predicted peak positions of approximately 34.80%, 30.92 μm, and 2.68 h for particle volume fraction, particle size, and ball-milling time, respectively. Continuous searching within the experimental domain identified a high-strength parameter combination of f = 34.63%, D = 30.67 μm, and t = 2.65 h, corresponding to a predicted tensile strength of 61.33 MPa. The measured strengths of three independent validation specimens prepared using a nominal particle size of 31 μm were 61.12, 60.93, and 60.56 MPa, with a mean of 60.87 ± 0.28 MPa, a coefficient of variation of 0.47%, and a relative deviation of 0.75% between the experimental mean and model prediction. Within the current experimental domain, the proposed framework can provide a quantitative reference for PFPC tensile-strength prediction and formulation/process-parameter screening, and offers an efficient strategy for accelerating formulation design and optimization while reducing experimental cost. Its implementation also provided students with practical experience in applying machine-learning algorithms to material-strength prediction and parameter design.

Scientific Reports
Jiangsu University of Technology (CN), China Academy Of Machinery Science & Technology (China) (CN)
Industry, innovation and infrastructure
Machine Learning in Materials Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon