Understanding and detection of process instabilities in wire arc directed energy deposition additive manufacturing using meltpool imaging and machine learning - Research report

Q: What instabilities does the study focus on in WA-DED?

The study targets humping and humping-induced porosity, both treated as common stochastic causes of poor WA-DED part quality.

Q: How is meltpool imaging used in the proposed approach?

In-operando high-speed meltpool imaging captures meltpool dynamics during multi-layer deposition, and physically intuitive morphology signatures are extracted from the data.

Q: How does the process-aware machine learning model perform for detecting instabilities?

The model classifies instability onset with about 85% accuracy (F1-score), outperforming black-box deep learning models with F1-score below 66%.

Aug 5, 2026

Upload by Quinn

This work investigates the wire arc directed energy deposition (WA-DED) additive manufacturing process with two goals: (1) use in-operando high-speed meltpool imaging to reveal and explain the causal dynamics of two common stochastic instabilities—humping and humping-induced porosity—and (2) exploit the imaging data in machine learning for real-time instability detection. Humping and related porosity degrade part quality even after extensive parameter optimization, so understanding and monitoring meltpool phenomena is critical. Meltpool dynamics from multi-layer ER90SG steel deposits are captured, physically intuitive morphology signatures are extracted, and a process-aware model detects instability onset at ~85% accuracy (F1), outperforming black-box deep learning.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Introduction

Objective and scope

Background: directed energy deposition and WA-DED/WAAM process

Materials & Design 258 (2025) 114598 Contents lists available at ScienceDirect Materials & Design journal [homepage:](homepage: www.elsevier.com/locate/matdes)[ www.elsevier.com/locate/matdes](homepage: www.elsevier.com/locate/matdes) | Understanding and detection of process instabilities in wire arc directed energy deposition additive manufacturing using meltpool imaging and machine learning☆
Andr´e Ramalhoa,b,* , Anis Assadb,c, Benjamin Bevansb,d, Fernando Deschamps e, Telmo G. Santos a,f, J.P. Oliveira g,**, Prahalada Rao b
a UNIDEMI, Department of Mechanical and Industrial Engineering, NOVA School of Science and Technology, Universidade NOVA de Lisboa, Caparica 2829-516, Portugal
b Grado Department of Industrial and Systems Engineering, Virginia Tech, Blacksburg, VA, USAc University of Southern Denmark, Department of Technology and Innovation, Sønderborg, Denmark d Sooner Advanced Manufacturing Laboratory, University of Oklahoma, Norman, OK, USA e Pontifícia Universidade Cat´olica do Paran´a, Imaculada Conceiç˜ao 1155, 80215-901 Curitiba, Brazil f Laborat´orio Associado de Sistemas Inteligentes, LASI, 4800-058 Guimar˜aes, Portugal
g CENIMAT/I3N, Department of Materials Science, NOVA School of Science and Technology, Universidade NOVA de Lisboa, Caparica 2829-516, Portugal | | | --- | --- | | A R T I C L E I N F O
Keywords:
Wire arc directed energy deposition Wire arc additive manufacturing (WAAM) Porosity
Humping
Meltpool imaging
Process-aware machine learning | A B S T R A C T | | | This work concerns the wire arc directed energy deposition (WA-DED) additive manufacturing process. The objectives were two-fold: (1) observe and understand, through in-operando high-speed meltpool imaging, the causal dynamics of two common WA-DED process instabilities, namely, humping and humping-induced porosity; and (2) leverage the high-speed meltpool imaging data within machine learning algorithms for real-time detection of process instabilities. Humping and humping-induced porosity are leading stochastic causes of poor WA-DED part quality that occur despite extensive optimization of processing conditions. It is therefore essential to understand, detect and control the causal meltpool phenomena linked to these instabilities. Accordingly, we used a high-speed camera to capture the meltpool dynamics of multi-layer depositions ofER90SG steel parts and meltpool flow behavior related to process instabilities were demarcated and quantified. Next, physically intuitive meltpool morphology signatures were extracted from the imaging data. These signatures were used in a machine learning model trained to autonomously detect process instabilities. This novel processaware machine learning approach classified onset of instabilities with ~85 % accuracy (F1-score), outperforming black-box deep learning models (F1-score <66 %). These results pave the way for a physically intuitive processaware machine learning strategy for monitoring and control of the WA-DED process. | 1. Introduction The objective of this work was two-fold: (1) observe and understand, through in-operando high-speed imaging, the link between wire arc directed energy deposition (WA-DED) meltpool dynamics and two common stochastic process instabilities, namely, humping and humpinginduced porosity that occur despite extensive optimization of processing parameters; and (2) leverage the high-speed meltpool imaging data for in-situ monitoring and detection of process instabilities. The directed energy deposition (DED) family of additive manufacturing (AM) processes use focused energy in the form of electric arc, plasma, laser, or an electron beam to melt and deposit material layer-by-layer [1]. The material can either take the form of powder or wire. The deposition of material in three dimensions is accomplished via ☆ This article is part of a special issue entitled: ‘Additive Manufacturing’ published in Materials & Design. * Corresponding author a

FAQs

What instabilities does the study focus on in WA-DED?

The study targets humping and humping-induced porosity, both treated as common stochastic causes of poor WA-DED part quality.

How is meltpool imaging used in the proposed approach?

In-operando high-speed meltpool imaging captures meltpool dynamics during multi-layer deposition, and physically intuitive morphology signatures are extracted from the data.

How does the process-aware machine learning model perform for detecting instabilities?

The model classifies instability onset with about 85% accuracy (F1-score), outperforming black-box deep learning models with F1-score below 66%.

Aug 5, 2026

Upload by Quinn