What is it about?
Additive manufacturing, commonly known as 3D printing, makes it possible to produce components with complex shapes and tailored properties. However, components produced under different conditions may respond differently when subjected to mechanical stress. Mechanical tests generate several sequences of measurements, such as force and displacement, that describe how a component behaves during bending or compression. Forecasting the future evolution of these signals can help researchers better understand the mechanical performance of additively manufactured materials. This paper introduces a similarity-guided forecasting approach. It first identifies groups of mechanical tests with similar force-response patterns and then trains a separate deep-learning model for each group. Each model forecasts multiple measurements over several future time steps.
Featured Image
Photo by Minku Kang on Unsplash
Why is it important?
A single forecasting model may struggle to represent all the different behaviours found in heterogeneous mechanical testing data. Our approach addresses this problem by allowing specialised predictive models to learn from groups of similar tests. The results show that similarity-guided modelling can provide more accurate forecasts than a model trained on all available test sequences without distinguishing their different behaviours. More accurate forecasting of mechanical responses could support material characterisation, quality assessment and the development of more reliable additively manufactured components. The approach may also be useful in other applications involving heterogeneous groups of multivariate time series.
Perspectives
Our research explores how the similarities among mechanical tests can be used to improve the forecasting of material behaviour. Rather than assuming that all test sequences follow the same pattern, we identify groups of components that respond similarly under mechanical stress. Specialised deep-learning models can then focus on the characteristics of each group. This work connects time-series analysis, artificial intelligence and additive manufacturing, showing how data-driven methods can help extract greater value from complex experimental data. Key takeaways - Mechanical tests of additively manufactured components generate complex, heterogeneous time-series data. - Similar test trajectories are identified by comparing their force-response signals. - A specialised LSTM forecasting model is trained for each group of similar tests. - The models predict several mechanical measurements over multiple future time steps. - Using a relatively small number of groups produced the best forecasting results. - The similarity-guided approach outperformed a model trained on the complete ungrouped dataset.
Prof. Donato Malerba
Universita degli Studi di Bari Aldo Moro
Read the Original
This page is a summary of: Similarity-Guided Forecasting of Multiple Sequences from Mechanical Testing Data of Additively Manufactured Components, July 2026, Springer Science + Business Media,
DOI: 10.1007/978-3-032-32643-0_18.
You can read the full text:
Contributors
The following have contributed to this page
