Dataset and Software for an Adaptive Continual-Learning Framework for Tensile Prediction of Multi-Material PETG/PC-ABS Laminates in MEX

This repository provides the experimental validation datasets, model-predicted tensile datasets, and software documentation used for the study “An Adaptive Continual-Learning Framework for Tensile Prediction of Multi-Material PETG/PC-ABS Laminates in MEX”.

The repository supports a physics-informed, parent-material-anchored, phase-aware continual-learning framework for predicting full tensile stress–strain curves of PETG/PC-ABS laminate coupons fabricated by material extrusion (MEX). The framework begins with experimentally measured PETG and PC-ABS parent-material tensile curves, constructs a rule-of-mixtures baseline, and then learns the pointwise residual between the baseline prediction and experimentally measured laminate responses. The model is updated sequentially as new laminate coupon data are introduced, enabling adaptive improvement of full-curve prediction, UTS estimation, and out-of-distribution laminate screening.

The parent PETG, parent PC-ABS, and original COMP experimental datasets used as the input reference dataset for this work are available in the parent Zenodo repository:

DOI: 10.5281/zenodo.18472960

The present repository extends that parent dataset by providing the additional experimental validation data, predicted output data, and software resources used in the adaptive continual-learning study.

The repository is organized into three main components:

1. Experimental Dataset

This folder contains experimentally measured tensile datasets for the laminate validation cases used in the continual-learning workflow. The experimental data are organized by laminate architecture:

- 0.2 mm alternating PETG/PC-ABS laminate- 1.33 mm PC-ABS / 1.33 mm PETG / 1.33 mm PC-ABS laminate- Rank 1 model-suggested laminate architecture- Rank 2 model-suggested laminate architecture- Rank 3 model-suggested laminate architecture

These experimental datasets were used for zero-shot comparison, out-of-distribution validation, coupon-based continual updating, and experimental validation of model-suggested laminate designs.

2. Predicted Dataset

This folder contains the corresponding model-predicted tensile datasets for the same laminate architectures:

- 0.2 mm alternating PETG/PC-ABS laminate- 1.33 mm PC-ABS / 1.33 mm PETG / 1.33 mm PC-ABS laminate- Rank 1 model-suggested laminate architecture- Rank 2 model-suggested laminate architecture- Rank 3 model-suggested laminate architecture

The predicted datasets support comparison between predicted and experimental tensile responses, including full stress–strain curve agreement, UTS prediction, full-curve RMSE, full-curve MAE, and sequential before/after learning behavior.

3. Software and Documentation

This folder contains the primary IPython notebook used to implement the computational workflow, together with pseudocode, README documentation, and technical documentation. The software workflow includes:

- tensile coupon data loading and preprocessing- parent-material anchor construction- rule-of-mixtures baseline prediction- parent-variability and uncertainty handling- mechanical phase detection- adaptive strain-grid construction- phase-aware sample weighting- residual-learning model training- sequential coupon-based model updating- out-of-distribution laminate screening- prediction of full stress–strain curves- UTS, elongation, force, area, and weight calculations- ranking of candidate laminate architectures- experimental comparison using UTS error, RMSE, and MAE

This dataset is intended to support reproducibility of the adaptive continual-learning framework and to provide reusable experimental and predicted tensile-curve data for future research on multi-material MEX laminates, PETG/PC-ABS interfaces, full-curve mechanical prediction, and data-efficient laminate screening.

Associated new dataset DOI:

10.5281/zenodo.19844820

Parent input/reference dataset DOI:

10.5281/zenodo.18472960