In an era marked by unprecedented technological advances and a growing global awareness of environmental sustainability, there is a need to develop methodologies that harmonize innovation with ecological responsibility. This work embarks on a journey at the crossroads of cutting-edge research and sustainable design within the MOZART project funded by the European Union's Horizon 2021 research and innovation programme.
The MOZART project is dedicated to pursuing Safe and Sustainable by Design strategies, aiming to reconcile technological progress with environmental consciousness. Central to this endeavor is the challenge of replacing environmentally concerning Hard Chromium (HC) coatings, notorious for their carcinogenic and toxic properties, with safer alternatives. The proposed solution focuses on nickel matrix nano-composite electroplating coatings, which offer high corrosion and wear resistance without the impacts associated with HC coatings. However, the traditional use of boric acid in these processes contradicts sustainability principles outlined in the European Union's REACH regulation.
To find an alternative to boric acid, we introduce the concept of Inverse Molecular Design, leveraging machine learning (ML) generative models such as Conditional Variational Autoencoders (CVAEs) to identify molecular structures with desired functionalities without prior knowledge of those structures. However, the effectiveness of such models heavily relies on the choice of molecular representation. Hence, we explore the importance of selecting a proper representation by comparing Simplified Molecular-Input Line-Entry System (SMILES) and SELF-referencing embedded strings (SELFIES) in training our CVAE models.
In the final phase, the work aimed to explore regions of interest in the latent space of the CVAE, to find alternatives to boric acid. This was done by performing an optimization of already registered molecules in terms of sustainability and similarity to boric acid, and studying how they evolved in terms of location in the latent space. To accomplish this optimization, Genetic Algorithms were implemented.
Our findings not only contribute to the development of sustainable electroplating processes but also show the potential of ML and molecular design in addressing complex environmental challenges. The careful selection of molecular representation ensures the efficacy of ML techniques, thus enhancing the overall success of sustainable materials design initiatives.