PHYSICS-INFORMED MACHINE LEARNING FOR MULTI-OBJECTIVE OPTIMIZATION IN ADDITIVE MANUFACTURING: A DATA-EFFICIENT APPROACH

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F61989100%3A27230%2F25%3A10259256" target="_blank" >RIV/61989100:27230/25:10259256 - isvavai.cz</a>

Výsledek na webu

<a href="https://www.webofscience.com/wos/woscc/full-record/WOS:001594510900001" target="_blank" >https://www.webofscience.com/wos/woscc/full-record/WOS:001594510900001</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.17973/MMSJ.2025_10_2025094" target="_blank" >10.17973/MMSJ.2025_10_2025094</a>

Jazyk výsledku

angličtina

Název v původním jazyce

PHYSICS-INFORMED MACHINE LEARNING FOR MULTI-OBJECTIVE OPTIMIZATION IN ADDITIVE MANUFACTURING: A DATA-EFFICIENT APPROACH

Popis výsledku v původním jazyce

Additive manufacturing (AM) quality control relies on empirical approaches due to complex process-property relationships. While machine learning (ML) offers promising solutions, most approaches treat parameters independently without leveraging thermomechanical principles governing the properties of the printed materials. This is essential for understanding the behaviour of fused deposition modelling (FDM) printing. This study investigates whether integrating elementary thermomechanical knowledge into feature engineering improves mechanical property prediction for polylactic acid components under data-constrained conditions. Using 50 experimental samples from controlled printing conditions, three feature engineering strategies were systematically compared: raw process parameters, physics-informed features based on heat transfer and material flow principles, and polynomial interactions across five ML algorithms. Physics-informed features consistently outperformed baseline approaches, with Huber Regressor achieving coefficient of determination equal to 0.817 (51.3% improvement over raw parameters). Feature importance analysis using SHapley Additive exPlanations identified layer height and nozzle temperature as primary predictors, with engineered thermal diffusion and density features contributing significantly to model performance. This study demonstrates the potential of physics-informed feature engineering for improving prediction accuracy in data-constrained AM scenarios, providing methodological insights for thermomechanical integration and actionable guidance for industrial artificial intelligence (AI) implementation.

Název v anglickém jazyce

PHYSICS-INFORMED MACHINE LEARNING FOR MULTI-OBJECTIVE OPTIMIZATION IN ADDITIVE MANUFACTURING: A DATA-EFFICIENT APPROACH

Popis výsledku anglicky

Additive manufacturing (AM) quality control relies on empirical approaches due to complex process-property relationships. While machine learning (ML) offers promising solutions, most approaches treat parameters independently without leveraging thermomechanical principles governing the properties of the printed materials. This is essential for understanding the behaviour of fused deposition modelling (FDM) printing. This study investigates whether integrating elementary thermomechanical knowledge into feature engineering improves mechanical property prediction for polylactic acid components under data-constrained conditions. Using 50 experimental samples from controlled printing conditions, three feature engineering strategies were systematically compared: raw process parameters, physics-informed features based on heat transfer and material flow principles, and polynomial interactions across five ML algorithms. Physics-informed features consistently outperformed baseline approaches, with Huber Regressor achieving coefficient of determination equal to 0.817 (51.3% improvement over raw parameters). Feature importance analysis using SHapley Additive exPlanations identified layer height and nozzle temperature as primary predictors, with engineered thermal diffusion and density features contributing significantly to model performance. This study demonstrates the potential of physics-informed feature engineering for improving prediction accuracy in data-constrained AM scenarios, providing methodological insights for thermomechanical integration and actionable guidance for industrial artificial intelligence (AI) implementation.

Druh

J_imp - Článek v periodiku v databázi Web of Science

CEP obor

—

OECD FORD obor

20300 - Mechanical engineering

Projekt

<a href="/cs/project/EH23_021%2F0010117" target="_blank" >EH23_021/0010117: Inovativní a aditivní technologie pro udržitelný energetický průmysl</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)<br>S - Specificky vyzkum na vysokych skolach

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

MM Science Journal

ISSN

1803-1269

e-ISSN

1805-0476

Svazek periodika

2025

Číslo periodika v rámci svazku

OCT

Stát vydavatele periodika

CZ - Česká republika

Počet stran výsledku

8

Strana od-do

8630-8637

Kód UT WoS článku

001594510900001

EID výsledku v databázi Scopus

—