Artificial intelligence (AI) has emerged as a powerful set of tools for engineering innovative materials. However, the AI-aided design of materials textures has not yet been researched in depth. In order to explore the potentials of AI for discovering innovative biointerfaces and engineering materials surfaces, especially for biomedical applications, this study focuses on the control of wettability through design-controlled hierarchical surfaces, whose design is supported and its performance

more »

... dicted thanks to adequately structured and trained artificial neural networks (ANN). The authors explain the creation of a comprehensive library of microtextured surfaces with well-known wettability properties. Such a library is processed and employed for the generation and training of artificial neural networks, which can predict the actual wetting performance of new design biointerfaces. The present research demonstrates that AI can importantly support the engineering of innovative hierarchical or multiscale surfaces when complex-to-model properties and phenomena, such as wettability and wetting, are involved. depth. Questions linked to the AI-aided engineering of materials surfaces and to the optimization of related contact properties and tribological performance, in connection to several mechanical and biomedical engineering challenges, remain unexplored. In fact, materials surface features have a direct influence on properties including friction coefficient [10], wear resistance [11], self-cleaning ability [12, 13] , biocompatible response [14] [15] [16] , ergonomic performance and esthetic aspect [17], among other fundamental characteristics linked to advanced product development in mechanical and biomedical engineering fields. Therefore, they also play determinant roles in materials selection when pursuing innovative functionalities, which can be based on bioinspired design strategies for promoting biological and biomedical applications. The authors hypothesize that the previously introduced holistic approaches to accelerated materials development, relying on the intensive use of AI if adequately researched and developed focusing on materials surfaces, can prove highly transformative towards high performing devices in several industries. The biomedical industry can greatly benefit from innovative hierarchical surfaces and biointerfaces capable of controlling cell-material interactions, improving biodevices compatibility and incorporating innovative sensing and mechanotransduction functionalities through AI-aided bioinspired design strategies. In order to explore and better understand the potentials of AI applied to the discovery of innovative biointerfaces and to the engineering of materials surfaces, especially for biomedical applications, this study focuses on the control of wettability through design-controlled hierarchical surfaces (or microtextured biointerfaces), whose design is supported and its performance predicted thanks to adequately structured and trained artificial neural networks (ANN). Wettability is chosen due to its relevance for functional biomedical (micro-)devices, as further explained. Surface wettability is an interesting property related to surface free energy and to surface topography or geometric micro-/nanostructure [18, 19] . Usually, surface wettability is measured through the water contact angle (CA), which helps to classify surfaces as hydrophobic (CA > 90 • ) or hydrophilic (CA < 90 • ). Values of CA close to 0 • are representative of superhydrophilic surfaces, while values close to 180 • are characteristic of superhydrophobic surfaces. There are two main routes for adjusting the wettability of surfaces: the first focuses on chemical functionalization anchoring appropriate molecules upon flat substrates, the second aims at modifying the shapes or topographies of surfaces. These routes may also be synergically combined. Regarding chemical approaches, the wettability of flat surfaces can be fine-tuned by the formation of a monolayer with appropriate hydrophilic or lipophilic functional groups. For instance, gold surfaces can be modified using thiol [20] or carbene [21] anchors, while hydroxylated surfaces such as silicon oxide, glass, mica, etc. can be modified by conventional siloxane chemistry [22] . Both hydrophilic and hydrophobic biointerfaces are interesting: the former for being usually very adequate for interacting with cells and tissues, hence leading more easily to biocompatible medical devices [23]; the latter for their singular self-cleaning properties and ability to stay dry, which can be applied to the development of easy to clean and sterilize surgical instruments [24] , to cite some examples. Recent research has put forward the potentials of creating hydrophobic and hydrophilic transitions upon the surfaces of biomedical microfluidic systems, capable of controlling fluids upon biointerfaces and hence achieving highly multiplexed systems for a wide set of screening and diagnostic purposes [25, 26] . While significant advances in the monolayer stabilization have been achieved, topology modification results in more robust and highly applicable surfaces. The possibility of controlling cell behavior and fate through modifications of surface topography, in connection with wettability properties, has also been studied in detail [27] . These advances would not have been Nanomaterials 2020, 10, 2287 3 of 19 possible without parallel progress in micro-and nanomanufacturing technologies and combinations thereof, which enable the straightforward, rapid prototyping and even mass-production of biomedical (micro-) devices with three-dimensional design-controlled surface topographies, as previous studies from our team have shown [28] [29] [30] . Although the Cassie-Baxter and Wenzel models can model contact angle under different wetting regimes, it is complex to model the actual performance of a surface a priori. The authors hypothesize that AI can help with predicting the actual behavior of fluids upon biointerfaces. In this study, the authors explain the creation of a comprehensive library of microtextured surfaces with well-known wettability properties. Such a library is processed and employed for the generation and training of artificial neural networks, which can predict the wetting performance of new design biointerfaces. The authors demonstrate that AI can importantly support the engineering of innovative hierarchical or multiscale surfaces when complex-to-model properties and phenomena, such as wettability and wetting, are involved. Materials and Methods Creating a Library of Microtextured Surfaces with Known Wettability Properties Several studies have dealt with the design and manufacture of microtextured surfaces for controlling the wettability and contact angle of materials surfaces. Both subtractive processes (computer numerical control machining, laser ablation, micro-drilling, etc.) and additive methods (laser stereolithography, digital light processing, powder-based laser fusion, lithography-based ceramic manufacture, etc.) have been applied to the creation of such surface topographies in a wide set of materials. Consequently, there is a plethora of scientific publications, including experiences from our team, describing the wettability properties of different synthetic surfaces. In addition, the epidermis of many living organisms from the animal and vegetal realms have shown very interesting wetting performances, which have also been widely reported. For the research purpose, in order to create a comprehensive library of microtextured surfaces with well-known wettability properties, which will subsequently serve as input for generating and training the artificial neural networks capable of predicting surface contact angle, a selection of publications is gathered. The selection includes relevant research works, in which microtextures are designed and manufactured or directly obtained from nature, with enough information about the surface topographies studied so that they can be replicated and with details about water contact angle obtained through systematic testing [30] [31] [32] [33] [34] [35] . After selecting the publications, NX 10 ® (Siemens PLM Software Solutions, Plano, TX, USA) is employed as computer-aided design (CAD) software for modeling the selected microtextured surfaces and completing the CAD library with well-known wettability properties. The CAD models are designed following the descriptions and measurement details included in the consulted references [30] [31] [32] [33] [34] [35] . In most cases, starting from a planar surface of 1 × 1 mm 2 , the combined use of simple solid-based design tools, like extrusions and revolutions of 2D profiles, and Boolean or pattern-based operations leads to the desired CAD models, as shown in Figure 1 (left). Only for 1 specific case of the collection, which mimics the feature of the lotus plant leaves, Brownian-like microtextures are added to the CAD model in order to achieve truly multiscale or hierarchical surfaces, following previous processes published by our team [30] . In addition, in 4 cases of the collection, due to the extremely fine multiscale details of the CAD model, the starting planar surface measures 0.5 × 0.5 mm 2 to avoid final CAD files with extremely large sizes (i.e., more than 1 Gb). This does not affect the study, as training of the ANNs is performed with adimensional parameters (see Section 2.3). The CAD files are stored in .stl (standard tessellation language) files for processing. From 3D CAD Files to Surface Matrices for Further Mathematical Processing The extraction of relevant parameters from the microtextured surfaces is performed with the support of matrix-based operations using MATLAB R2020a (The Mathworks, Inc., Natick, MA, USA). The process for straightforwardly transforming the CAD files into MATLAB surface matrices, which store the information of the surfaces, is schematically presented in Figures 1 and 2.