Development of a Machine Learning Framework for interlayer dwell time prediction in Wire-Arc Additive Manufacturing

June 29, 2026

Wire-Arc Additive Manufacturing (WAAM) enables the production of large-scale metallic components with a high deposition rate, but remains limited by thermal accumulation and geometric variability, which are affected, in part, by the empirical selection of the interlayer dwell time. This dissertation develops an approach to automatically predict, prior to production, the dwell time required to achieve a defined interlayer transition temperature. An experimental database was built comprising 81 straight walls of 316LSi stainless steel, each with 20 layers, varying the torch travel speed, wire feed speed, transition temperature, and deposited bead length. The dwell times were obtained by infrared thermography and used to train machine learning (ML) models suitable for regression problems. Several models were evaluated, and Support Vector Regressors (SVR) were selected, subsequently optimised and integrated into a predictive tool in Python. In the validation trials, the SVRs reproduced the measured dwell times with high accuracy and allowed stable transition temperatures to be achieved. Interpolation and extrapolation conditions were tested, as well as a simple curved geometry. The work demonstrates the feasibility of offline predictions to support thermal and production planning in WAAM, reducing dependence on empirical procedures.

Authors in the community:

• 

  J

  João Pereira Vieira Alexandre

  ist196413

Supervisors of this institution:

• 

  S

  Susana Margarida da Silva Vieira

  ist45346

• 

  R

  Rui Filipe Vieira Sampaio

  ist424864

mechanical-engineering - Mechanical engineering

eng - English

Instituto Superior Técnico