A preliminary data fusion study to assess the feasibility of Foundation Process-Property Models in Laser Powder Bed Fusion

Oriol Vendrell-Gallart, Nima Negarandeh, Zahra Zanjani Foumani, Mahsa Amiri, Lorenzo Valdevit, Ramin Bostanabad † · 2025

Read A preliminary data fusion study to assess the feasibility of Foundation Process-Property Models in Laser Powder Bed Fusion by Oriol Vendrell-Gallart, Nima Negarandeh, Zahra Zanjani Foumani, Mahsa Amiri, Lorenzo Valdevit, Ramin Bostanabad † (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-16667.txt.

Introduction

Oriol Vendrell-Gallart and co-authors test whether data fusion can support foundation models for metal printing. Their 2025 study deals with laser powder bed fusion, a method that builds metal parts layer by layer. Large foundation models could speed process tuning and new material design. Data scarcity is a major block, though. The team runs controlled experiments on transfer across materials and properties.

Lab data sets were made for two stainless steels. Five process settings were varied, and porosity and hardness were measured. Statistical models called Gaussian processes were fit in many configurations. The tests ask if knowledge of one steel or property helps predict another. Results stress data size, complexity, and noise. The authors call for structured learning that uses domain knowledge.

Catalog record

CREATOROriol Vendrell-Gallart, Nima Negarandeh, Zahra Zanjani Foumani, Mahsa Amiri, Lorenzo Valdevit, Ramin Bostanabad †
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16667

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A preliminary data fusion study to assess the feasibility of Foundation Process-Property Models in Laser Powder Bed Fusion Oriol Vendrell-Gallart, Nima Negarandeh, Zahra Zanjani Foumani, Mahsa Amiri, Lorenzo Valdevit, Ramin Bostanabad † Abstract Abstract Foundation models are at the forefront of an increasing number of critical applications. In regards to technologies such as additive manufacturing (AM), these models have the potential to dramatically accelerate process optimization and, in turn, design of next generation materials. A major challenge that impedes the construction of foundation process-property models is data scarcity. To understand the impact of this challenge, and since foundation models rely on data fusion, in this work we conduct controlled experiments where we focus on the transferability of information across different material systems and properties. More specifically, we generate experimental datasets from 17-4 PH and 316L stainless steels (SSs) in Laser Powder Bed Fusion (LPBF) where we measure the effect of five process parameters on porosity and hardness. We then leverage Gaussian processes (GPs) for process-property modeling in various configurations to test if knowledge about one material system or property can be leveraged to build more accurate machine learning models for other material systems or properties. Through extensive cross-validation studies and probing the GPs’ interpretable hyperparameters, we study the intricate relation among data size and dimensionality, complexity of the process-property relations, noise, and characteristics of machine learning models. Our findings highlight the need for structured learning approaches that incorporate domain knowledge in building foundation process-property models rather than relying on uninformed data fusion in data-limited applications. Keywords: Foundation Models; Transfer Learning; Data Fusion; Additive manufacturing; Gaussian process. 1 Introduction The increasing need for lightweight, high-performance materials with complex geometries has driven the advancement of additive manufacturing (AM) technologies [ 1 ] . Among the different AM processes, laser powder bed fusion (LPBF) is a leading technique for producing metallic components that offer outstanding mechanical properties and design versatility [ 2 ] . The properties of LPBF-built parts are heavily influenced by various process parameters such as laser power and speed, scan direction, and deposition layer thickness. Fully understanding the complex relationships between process parameters, microstructural evolution, and mechanical properties in LPBF, with sufficient accuracy for effective optimization, remains a significant challenge [ 3 ] . The vast parameter space inherent in LPBF makes brute-force experimentation impractical. As a result, predictive modeling of process-property relations in LPBF has traditionally relied on domain knowledge and trial-and-error methods [ 4 ] . Therefore, the development of robust, generalizable models capable of accurately predicting material properties from process parameters is crucial for optimizing AM processes and accelerating material design. This motivates the exploration of foundation process-property models to accelerate the optimization and design of materials with desired properties. In recent years, foundation models [ 5 ] —large-scale machine learning (ML) models capable of generalizing across diverse datasets— have shown remarkable success in domains such as natural language processing and computer vision where vast amounts of data are readily available. However, in materials science and manufacturing processes such as LPBF, the availability of large, high-quality datasets is a significant challenge. This raises the issue of whether foundation models can be effectively leveraged for material property prediction despite limited data availability. To address this question, in this work we focus on a controlled environment where we evaluate the transferability of knowledge between two distinct stainless steel (SS) material systems, namely, 17-4 PH SS and 316L SS. Our underlying hypothesis is that these materials contain transferable process-property information, i.e., knowledge from one system can be leveraged to build more accurate process-property ML models for the other material system. However, this assumption remains largely untested in the literature, particularly with respect to LPBF. To test our hypothesis, we model process-property relations via Gaussian processes (GPs) whose training data is based on either one or both steel types. Our studies involve data fusion which we formulate via GPs because they not only are extremely effective in learning from small data, but also have explainable parameters which supports decision making. Our findings have significant implications for materials design and process optimization and they highlight the need for principled transfer learning for building foundation process-property models where domain knowledge, such as physics-informed priors or hierarchical modeling [ 6 ] , is leveraged for data fusion. The rest of the paper is organized as follows. We introduce our material characterization and ML approaches in Section 2 and then assess the feasibility of developing foundation process-property models in Section 3 . We conclude the paper in Section 4 by summarizing our contributions and providing future research directions. 2 Methods We first elaborate on the experimental data generation process in Section 2.1 and then in Sections 2.2 and 2.3 provide some background on GPs which serve as the workhorse of our studies. Finally, our approach is explained in Section 2.4 where we assess the potential advantages of learning two mechanical properties (namely, porosity and hardness) of 17-4 PH SS and 316L SS under various settings. 2.1 Experimental Data Collection The 17-4 PH SS data used in this study is derived from [ 6 ] which consists of 270 process parameter combinations generated using a Sobol sequence. These combinations include variations in laser power ( p ) ranging from 80 to 400 W, laser scan speed ( v ) from 150 to 1500 mm/s, powder layer thickness ( l ) from 20 to 75 \mu m, hatch spacing ( h ) from 70 to 120 \mu m, and two scan rotation angles ( sr ) of 67 or 90 degrees. These parameters were used to fabricate 270 cube-shaped samples ( 2\times 2\times 2 mm), numbered 1 to 270 , using the HT setting outlined in [ 6 ] . Detailed information on the HT setting and the 270 combinations can be found in [ 6 ] . To study the feasibility of information transfer from 17-4 PH SS to 316L SS and vice versa, an additional 270 cube-shaped 316L SS samples are fabricated for this study. These samples are built using the same process conditions and manufacturing approach as 17-4 PH SS, detailed in [ 6 ] . We analyze the microstructural images of the 270 cuboids from each material to extract porosity features. In our studies we consider two cases: extracting the average porosity ( \phi ) using simple thresholding or conducting image analysis to extract a set of features that describe the spatial distribution and geometry of pores (e.g. statistics of pore radius, perimeter, location, etc.). For the latter case, the images are first cropped to eliminate the borders and then binarized via Otsu’s thresholding method [ 7 ] . Morphological dilation [ 8 ] is applied to ensure distinct separation of pores, followed by the watershed algorithm [ 9 ] to segment individual pores, see Figure 5 . Connected component analysis [ 10 ] labels each pore region, and the area of each pore is calculated by counting its pixels. The total porosity ( \phi ) is determined as the ratio of the combined pore area to the total image area, expressed as a percentage. For brevity, in Section 3 we present the results for the former case and suffice to only discussing

Read the complete text