Data-driven discovery of roughness descriptors for surface characterization and intimate contact modeling of unidirectional composite tapes
Sebastian Rodriguez, Mikhael Tannous, Jad Mounayer, Camilo Cruz, Anais Barasinski, Francisco Chinesta · 2026
Read Data-driven discovery of roughness descriptors for surface characterization and intimate contact modeling of unidirectional composite tapes by Sebastian Rodriguez, Mikhael Tannous, Jad Mounayer, Camilo Cruz, Anais Barasinski, Francisco Chinesta (2026) 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-2603-20418.txt.
Introduction
Rodriguez and co-authors seek better roughness measures for unidirectional composite tapes. Surface roughness drives how fully tapes bond during consolidation. Standard statistical measures describe topology but miss the contact physics. Their strategy uses rank reduction autoencoders with a linear latent space. A truncated SVD on the latent matrix enforces that structure. The learned modes must rebuild the surface accurately after decoding. They must also support tape classification and contact modeling. This links one descriptor set to both process control and physics.
Catalog record
| CREATOR | Sebastian Rodriguez, Mikhael Tannous, Jad Mounayer, Camilo Cruz, Anais Barasinski, Francisco Chinesta |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 53848 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.20418
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Data-driven discovery of roughness descriptors for surface characterization and intimate contact modeling of unidirectional composite tapes Sebastian Rodriguez, Mikhael Tannous, Jad Mounayer, Camilo Cruz, Anais Barasinski, Francisco Chinesta Abstract Unidirectional tapes surface roughness determines the evolution of the degree of intimate contact required for ensuring the thermoplastic molecular diffusion and the associated inter-tapes consolidation during manufacturing of composite structures. However, usual characterization of rough surfaces relies on statistical descriptors that even if they are able to represent the surface topology, they are not necessarily connected with the physics occurring at the interface during inter-tape consolidation. Thus, a key research question could be formulated as follows: Which roughness descriptors simultaneously enable tape classification—crucial for process control—and consolidation modeling via the inference of the evolution of the degree of intimate contact, itself governed by the process parameters? . For providing a valuable response, we propose a novel strategy based on the use of Rank Reduction Autoencoders (RRAEs), autoencoders with a linear latent vector space enforced by applying a truncated Singular Value Decomposition (SVD) to the latent matrix during the encoder-decoder training. In this work, we extract useful roughness descriptors by enforcing the latent SVD modes to (i) accurately represent the roughness after decoding, and (ii) allow the extraction of existing a priori knowledge such as classification or modelling properties. Keywords : Roughness descriptors, Thermoplastic composites tapes, rough surface compaction, Data-driven discovery, Rank Reduction Autoencoder, Machine Learning. 1 Introduction Many composite forming processes for manufacturing structural parts are based on the consolidation of pre-impregnated preforms, e.g., sheets, tapes, …. To obtain mechanical cohesion between thermoplastic pre-impregnated preforms that are brought together during manufacturing, two specific physical conditions at the interface are needed: (a) the development of an effective intimate contact between surfaces that is a requirement for (b) the temperature-driven macro-molecular inter-diffusion (as described by De Gennes reptation theory [ 1 ] ) within the process time window, while avoiding thermal degradation. To reach this goal, both pressure and heat are usually applied during the manufacturing process. Determining the process window for reaching an optimal consolidation without degrading the material, is of critical importance for ensuring the highest productivity while certifying the quality of the resulting structure [ 2 ] . Tapes surface roughness determines the evolution of the degree of intimate contact, which is a sine qua non for the thermoplastic macro-molecular inter-diffusion between tapes. The extent of macro-molecular inter-diffusion dictates the mechanical properties of the polymer-polymer interface and therefore the mechanical integrity of the final composite structure. Typically, the roughness of a composite tape surface is characterized using classical statistical descriptors that even if they are able to represent the surface topology itself, there is not proof that those descriptors are the more suitable ones to describe the physical phenomena taking place during the intimate contact development between tapes. Thus, the following research question appears pertinent: Which roughness descriptors simultaneously enable tape classification—crucial for process control—and consolidation modeling via the inference of the evolution of the degree of intimate contact, itself governed by the process parameters? . Our response to this question follows a novel procedure based on the use of a powerful autoencoder with a linear latent vector space [ 20 , 3 , 4 ] , enforced by applying a truncated Singular Value Decomposition (SVD) at the latent space level during the encoder-decoder training. The resulting SVD modes are enforced to represent any existing a priori knowledge, for instance valuable statistical descriptors, complemented with the ones needed for inference purposes (classification or modeling), and finally for those needed to ensure the profiles representation after decoding. To demonstrate the potential of the proposed technology, we use a database of hundreds of roughness profiles measured on 12 thermoplastic composite tapes produced by different providers. Conventional statistical descriptors were unable to reliably classify the surfaces or predict the evolution of intimate contact during consolidation. Topological Data Analysis (TDA) enabled classification in previous work [ 5 ] , but its success mainly relied on the pronounced differences in macro-roughness between tapes from different suppliers. These large-scale geometrical features generate persistent topological signatures. As illustrated in Fig. 1 , once the macro-scale component is removed, the resulting micro-roughness no longer exhibits clear persistent structures. Consequently, TDA loses its categorization capability when applied to the micro-roughness alone. (a) Macro-roughness profile. (b) Associated micro-roughness after removing the dominant geometrical feature. Figure 1 : Illustration of macro- and micro-roughness measured on a unidirectional composite tape. On the other hand, the data of intimate contact evolution are obtained in-silico by using a 2D simulation of the composite tape surface compaction. A cellular automaton computational method is used to simulate the compaction of the surface roughness profile at constant velocity. Each measured roughness profile is used as initial geometrical domain for a compaction simulation, which produces as output the time-resolved evolution of the degree of intimate contact. The technology proposed in this paper operates with a single latent mode for classification when both macro- and micro-roughness scales are considered, and with four modes when focusing exclusively on the micro-roughness component. The micro-roughness is obtained by removing the macro-scale contribution from the measured surface profiles. As such, the proposed framework constitutes an unprecedented characterization and modeling tool of particular relevance for the community working on composite forming based on pre-impregnated preforms. Remarkably, the same number of latent modes—and only one additional mode when higher accuracy is required—also enables accurate prediction of the evolution of intimate contact, while simultaneously ensuring an almost perfect reconstruction of the roughness profile itself. This demonstrates the strong compactness and physical consistency of the reduced representation, capable of capturing geometry and consolidation kinetics within a unified low-dimensional space. This article is organized into seven sections, including the present introduction. Section 2 presents in detail the numerical method to simulate the composite tape surface compaction, followed by the description of the database of composite tape roughness profiles in Section 3 . Section 4 details the rough surface classification methodology based on Rank Reduction Auto-Encoders (RRAE) and associated results. Section 5 introduces the intimate contact prediction using RRAE along with its associated results, before comparison with standard machine learning architectures in Section 6 . Finally, Section 7 provides conclusions and perspectives. 2 Compaction simulation In former works [ 6 , 7 , 8 ] we proposed simulating the consolidation of as-measured surfaces geometries instead of the sometimes crude reconstructions of them, (for example, based on the use of fractal features or the description of asperities by means of rectangular elements [ 9 , 10 ] ). A numerical strategy to geometrically model an as-measured surface roughness profile considers a Haar-based wavelet representation [ 8 ] , which results in a multi-scale sequence of rectangular patterns from the coarsest scale to the finest one. This geometrical model results in a high-fidelity representation of the as-measured surface. The smoother is the surface, the less levels in the wavelet-based description are required. Given a tape surface geometrical representation, different simulation approaches can be envisaged to emulate the physical phenomena taking place during inter-tape consolidation. For example, in [ 11 ] lubrication flow was considered to compress the rectangular elements (considering a 2D model of a surface profile) from the finest to