Rapid FE-informed surrogate benchmarking for peak-temperature sequence prediction in Ti-6Al-4V laser-directed energy deposition

Purpose This study aims to benchmark next-state reconstruction of one FE-generated scalar peak-temperature sequence for Ti-6Al-4V laser-directed energy deposition (L-DED) as a methodological study, not a deployment-ready process model. Design/methodology/approach A conduction-dominated 10-layer wall was simulated in ANSYS Workbench Mechanical APDL 2025 R1 with temperature-dependent properties, layer activation, a Goldak-type source and convective-radiative losses. The 823-state sequence compared LSTM, XGBoost and DNN using a common target, 20-state context, chronological split, training-only scaling, scoring convention and CPU timing; model-specific inputs are documented. Findings Across the 803-state prediction region, XGBoost achieved RMSE = 12.36 K and MAE = 7.80 K; in the 165-state late validation interval, these rose to 24.58 K and 21.14 K. Cycle- and phase-resolved diagnostics localized the strongest degradation to late-build and rapid heating/near-peak states. XGBoost gave the best accuracy-cost compromise for this data set. Research limitations/implications The high accuracy is strictly an intra-case interpolation result for one FE-generated sequence, not evidence of cross-condition generalization or experimental fidelity. Latent heat and thermo-fluid melt-pool flow are excluded. Practical implications The present model supports only FE-sequence reconstruction and offline algorithm screening. Practical use requires matched experiments, multi-condition training, uncertainty quantification and validation of the complete sensing-control pipeline. Originality/value The transferable protocol combines target-time splitting, training-only preprocessing, explicit context exclusion, documented model-specific features, layer/phase diagnostics and amortized latency reporting, exposing degradation hidden by global R2.

Authors

Institutions

Publication Details

Journal
Rapid Prototyping Journal
Published
2026-11-16
DOI
https://doi.org/10.1108/rpj-05-2026-0266
Primary Topic
Additive Manufacturing Materials and Processes
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Rapid FE-informed surrogate benchmarking for peak-temperature sequence prediction in Ti-6Al-4V laser-directed energy deposition

Amir Guizani, Moncef Hammadi, Mohamed Haddar, Larbi Yousfi et al.
4 links
Rapid Prototyping Journal
Additive Manufacturing Materials and Processes
article

Rapid FE-informed surrogate benchmarking for peak-temperature sequence prediction in Ti-6Al-4V laser-directed energy deposition

Amir Guizani, Moncef Hammadi, Mohamed Haddar, Larbi Yousfi, Slim Bouaziz
article en

Abstract

Purpose This study aims to benchmark next-state reconstruction of one FE-generated scalar peak-temperature sequence for Ti-6Al-4V laser-directed energy deposition (L-DED) as a methodological study, not a deployment-ready process model. Design/methodology/approach A conduction-dominated 10-layer wall was simulated in ANSYS Workbench Mechanical APDL 2025 R1 with temperature-dependent properties, layer activation, a Goldak-type source and convective-radiative losses. The 823-state sequence compared LSTM, XGBoost and DNN using a common target, 20-state context, chronological split, training-only scaling, scoring convention and CPU timing; model-specific inputs are documented. Findings Across the 803-state prediction region, XGBoost achieved RMSE = 12.36 K and MAE = 7.80 K; in the 165-state late validation interval, these rose to 24.58 K and 21.14 K. Cycle- and phase-resolved diagnostics localized the strongest degradation to late-build and rapid heating/near-peak states. XGBoost gave the best accuracy-cost compromise for this data set. Research limitations/implications The high accuracy is strictly an intra-case interpolation result for one FE-generated sequence, not evidence of cross-condition generalization or experimental fidelity. Latent heat and thermo-fluid melt-pool flow are excluded. Practical implications The present model supports only FE-sequence reconstruction and offline algorithm screening. Practical use requires matched experiments, multi-condition training, uncertainty quantification and validation of the complete sensing-control pipeline. Originality/value The transferable protocol combines target-time splitting, training-only preprocessing, explicit context exclusion, documented model-specific features, layer/phase diagnostics and amortized latency reporting, exposing degradation hidden by global R2.

Rapid Prototyping Journal
University of Sfax (TN), Institut Supérieur de Mécanique de Paris (FR)
Additive Manufacturing Materials and Processes
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