A cross-method performance study: DOE-based regression, physics-based simulation, and sensor-based machine learning for injection molding quality prediction

Quality monitoring in injection molding can be achieved through various modeling techniques, yet no clear framework exists to guide practitioners in selecting the most appropriate approach for their specific application requirements. Understanding which data sources and modeling strategies are needed, based on the required degree of accuracy, transparency, and cost constraints, remains an important and unresolved research gap in the industry. This study aims to address this challenge by presenting a comparative analysis of three predictive modeling methodologies for evaluating injection molded part quality, each representing a distinct level of complexity and cost: (1) multivariate regression modeling using design of experiments (DOE), the simplest and lowest-cost approach relying solely on machine data, which achieved an average MAPE of 0.95%; (2) physics-based simulation using Moldex3D, which requires commercial software licensing and performed well under normal fill conditions but was substantially penalized by short shot cycles that the simulation could not accurately represent, resulting in an average MAPE of 2.03%; and (3) machine learning (ML), utilizing a nested deep learning architecture based on long short-term memory (LSTM) networks, the most resource-intensive approach requiring in-mold sensors and data acquisition systems, which achieved an average MAPE of 0.27% on the selected comparison cycles, including short shots. These results highlight the complementary capabilities of the three modeling approaches: each approach addresses different needs along the manufacturing lifecycle, and the optimal selection depends on the balance of predictive accuracy, process coverage, and infrastructure cost.

Authors

Institutions

Publication Details

Journal
The International Journal of Advanced Manufacturing Technology
Published
2026-11-21
DOI
https://doi.org/10.1007/s00170-026-19582-5
Primary Topic
Injection Molding Process and Properties
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A cross-method performance study: DOE-based regression, physics-based simulation, and sensor-based machine learning for injection molding quality prediction

Xiangyu Hu, Saeed Davoudabadi Farahani, Yan Tong, Hongpeng Yang et al.
The International Journal of Advanced Manufacturing Technology
Injection Molding Process and Properties
article

A cross-method performance study: DOE-based regression, physics-based simulation, and sensor-based machine learning for injection molding quality prediction

Xiangyu Hu, Saeed Davoudabadi Farahani, Yan Tong, Hongpeng Yang, Sai Aditya Pradeep, Luke Shannon, Dhanashree Shinde, Xiaoyi Liu
article en

Abstract

Quality monitoring in injection molding can be achieved through various modeling techniques, yet no clear framework exists to guide practitioners in selecting the most appropriate approach for their specific application requirements. Understanding which data sources and modeling strategies are needed, based on the required degree of accuracy, transparency, and cost constraints, remains an important and unresolved research gap in the industry. This study aims to address this challenge by presenting a comparative analysis of three predictive modeling methodologies for evaluating injection molded part quality, each representing a distinct level of complexity and cost: (1) multivariate regression modeling using design of experiments (DOE), the simplest and lowest-cost approach relying solely on machine data, which achieved an average MAPE of 0.95%; (2) physics-based simulation using Moldex3D, which requires commercial software licensing and performed well under normal fill conditions but was substantially penalized by short shot cycles that the simulation could not accurately represent, resulting in an average MAPE of 2.03%; and (3) machine learning (ML), utilizing a nested deep learning architecture based on long short-term memory (LSTM) networks, the most resource-intensive approach requiring in-mold sensors and data acquisition systems, which achieved an average MAPE of 0.27% on the selected comparison cycles, including short shots. These results highlight the complementary capabilities of the three modeling approaches: each approach addresses different needs along the manufacturing lifecycle, and the optimal selection depends on the balance of predictive accuracy, process coverage, and infrastructure cost.

The International Journal of Advanced Manufacturing Technology
University of South Carolina (US), Clemson University (US), University of Delaware (US)
Clemson University, Society for Community Research and Action
Industry, innovation and infrastructure
Injection Molding Process and Properties
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