Supervised Machine Learning and Physics based Machine - Learning approach for prediction of peak temperature distribution in Additive Friction Stir Deposition of Aluminium Alloy

Aug 5, 2026

Upload by Anda

Additive friction stir deposition (AFSD) is a solid-state additive manufacturing technique designed to avoid porosity, cracking, and anisotropy issues seen in powder bed fusion and directed energy deposition. Yet the links among process parameters, thermal profiles, and resulting microstructure remain insufficiently understood, limiting property-focused process optimization. This work couples supervised machine learning with physics-informed neural networks to predict AFSD peak temperature distribution from process inputs. Eight regression models and four PINNs are evaluated, with gradient-boosting ensembles achieving the lowest MSE of 165.78. Logistic regression further classifies deposition quality from process factors. The integrated data-driven and physics-guided framework supports thermal management for microstructure tailoring and clarifies AM process–property relationships.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages44

Reading Time1 h 51 min

AccessFree to download

Abstract

1. Introduction

Supervised Machine Learning and Physics based Machine # Learning approach for prediction of peak temperaturedistribution in Additive Friction Stir Deposition of AluminiumAlloy Akshansh Mishra 1 1School of Industrial and Information Engineering, Politecnico Di Milano, Milan, ItalyMail id:akshansh.mishra@mail.polimi.it Abstract: Additive friction stir deposition (AFSD) is a novel solid-state additivemanufacturing technique that circumvents issues of porosity, cracking, and propertiesanisotropy that plague traditional powder bed fusion and directed energy depositionapproaches. However, correlations between process parameters, thermal profiles, andresulting microstructure in AFSD remain poorly understood. This hinders processoptimization for properties. This work employs a cutting-edge framework combiningsupervised machine learning (SML) and physics-informed neural networks (PINNs) topredict peak temperature distribution in AFSD from process parameters. Eight regressionalgorithms were implemented for SML modeling, while four PINNs leveraged governingequations for transport, wave propagation, heat transfer, and quantum mechanics. Acrossmultiple statistical measures, ensemble techniques like gradient boosting proved superior forSML, with lowest MSE of 165.78. The integrated ML approach was also applied to classifydeposition quality from process factors, with logistic regression delivering robust accuracy.By fusing data-driven learning and fundamental physics, this dual methodology providescomprehensive insights into tailoring microstructure through thermal management in AFSD.The work demonstrates the power of bridging statistical and physics-based modeling forelucidating AM process-property relationships. Keywords: Machine Learning; Additive Manufacturing; Additive Friction Stir Deposition;Physics based Neural Networks ## 1. Introduction Additive manufacturing (AM) has revolutionized production across diverse industries byenabling on-demand fabrication of complex geometries directly from digital models [1-5] .While powder-bed fusion and directed energy deposition techniques dominate the metal AMlandscape, they inherently suffer from issues related to the melt-pool mode of materialaddition and subsequent rapid solidification. Porosity, cracking, residual stresses, andanisotropic properties are common [6-9] . In recent years, solid-state metal AM approacheshave emerged seeking to circumvent these challenges by avoiding bulk melting of feedmaterial. Additive Friction Stir Deposition (AFSD) is one such novel technique combiningconcepts of friction stir processing and additive layer manufacturing. First proposed in 2018, it has garnered significant interest for promising superior microstructure and propertiescompared to other metal AM methods [10-15] . One of the primary benefits ofAFSD is its ability to produce fully dense parts with propertiesthat are comparable to those of wrought alloys. This is particularly important in industriessuch as aerospace, automotive, and defense, where high structural integrity and mechanicalperformance are crucial [16-19] . In addition, AFSD allows for localized deposition, enablingthe precise replacement of lost material while maintaining the original properties of thecomponent. This makes it an ideal solution for repairing and refurbishing high-valuecomponents that have been damaged due to wear, corrosion, or impact. Another significantadvantage of AFSD is its capability for solid-state recycling of machining chips and scraps.By converting these materials into usable metal powder feedstock, AFSD enables sustainablein-house recycling, reducing waste and minimizing environmental impact. Moreover, AFSDoffers higher deposition rates compared to other additive manufacturing techniques such aspowder bed fusion and directed energy deposition, allowing for the rapid production of largeparts. The unique deposition mechanism employed by AFSD also produces parts withimproved microstructural propert

FAQs

Why is peak temperature distribution important in AFSD?

Peak temperatures govern microstructure evolution and influence grain size and morphology, directly affecting mechanical properties. They also affect residual stresses and potential distortion.

What modeling approach is used to predict peak temperature distribution in AFSD?

The study uses a combined framework of supervised machine learning regression and physics-informed neural networks, using governing equations for heat transfer and related physical effects.

How is deposition quality classified using the proposed methods?

The integrated ML framework includes a classification task where logistic regression uses process factors to provide robust accuracy for deposition quality.

Aug 5, 2026

Upload by Anda