7-13 October 2022
Jozef Stefan Institute, Ljubljana, Slovenia
Prof. Aleksandar Dimitrov
Faculty of Technology and Metallurgy, St. Cyril and Methodious University, Skopje, North Macedonia
Aim of the mission
My host prof. Dimitrov and I started our research by using experimental data from experimentally produced samples of CNTs at the Laboratory for Nanomaterials at the Faculty of Technology and Metallurgy under different conditions, which was later widened to other carbon nanomaterials samples obtained in the same Lab (graphene).
– First, exploratory data analysis was performed in order to obtain insights and find relevant patterns that are difficult or impossible to be concluded from raw data.
– Then, a comprehensive data pipeline was created in order to prepare the dataset in a suitable from as an input to the machine learning (ML) algorithms.
– Additional feature extraction was performed in order to maximise the quality and quantity of information present in the dataset used as an input to the ML models.
– Existing explainable symbolic ML models were additionally trained on new carbon nanomaterials samples obtained by electrolysis in molten salts, with the aim to improve their accuracy and descriptive power.
– We finally used the best-performing model for deriving rules from it, which was furthermore used to optimise the process of production of low-cost high-quality CNTs as well as graphene at a large scale.
Summary of the Results
By leveraging the datasets gathered from experimentally obtained carbon nanomaterials samoles, it is possible to train a model which can guide and optimise the process of production of CNTs, which is otherwise difficult to control without an expensive equipment. Also, leveraging the descriptive power of symbolic machine learning algorithms enables obtaining high-yield MWCNTs with improved quality and
highly decreased level of defects. Consequently, a decision tree model was produced, which is able to characterise CNTs, while also provide information about the relationship between the quality of the CNT and the input parameters. Encouraged by the outcome, a decision tree model was produced for graphene samples electrolytically obtained.
The obtained experimental results from the employed models were then compared to the theoretical basis in the field, and tested their correlation. Thus, following the descriptive rules from the model that achieved a satisfactorily performance on the evaluation metric (e.g., characterisation accuracy higher than 80%), made it possible to obtain more optimal CNTs, as well as graphene, by electrolysis in molten
salts. Therefore, the best model gave us a valuable novel tool in producing controlled, certain quality high-yield CNT and graphene samples, which can later be used and widely applied.
The obtained results are based on an interdisciplinary field work and research that intertwines fundamental science – physics, chemistry, and information technologies as well as topology into applied nanomaterials and nanotechnologies. Controlling the type and quality of the experimentally obtained carbon nanomaterials, and additionally using the graph theory, enables us to estimate the topological
indices of the produced molecules, which is in accordance to the EUTOPIA description and objectives.
The outputs are very promising and the results will be presented at upcoming conferences and prepared for publishing in relevant scientific journals. Such collaboration and introduction of ML into carbon
nanotechnologies can give a new light and open doors to novel techniques and aspects of existing or future research that could be a base of new project applications by gathering corresponding teams from
people who are members of different WGs of this COST Action.