• Overview :
This thesis is part of an interdisciplinary project between artificial intelligence (AI) and materials for energy conversion for the optimization and acceleration of the production of devices for photovoltaic or solar energy. In this context, the research work will focus more particularly on the foundations of the management and analysis of small data sets.
• Context :
In the energy materials industry (eg materials for photovoltaics or batteries), the actual processes for optimizing devices are so complex that it is often not possible to describe in detail the chemical or physical combinations that have improved their performance. Recently, several studies have shown that the combination of methods of "design of experiments" (DOE) and artificial learning, such as "machine learning" (ML) allows (i) to reduce the time required to optimize a device or system; (ii) to increase the probability of discovering a true optimum.
• Problem :
The method combining DOE and ML requires working with a large number of variables and it is difficult to obtain satisfactory results for reduced sets of data or "small data". It is therefore necessary to develop an additional level of the learning model or consider other AI concepts.
• Goals :
- Exploit performances of machine learning and active learning concepts, respectively on “small data” sets.
- Evaluate new AI concepts such as “deep learning” on the data addressed with the learning methods.
- Study a specific case of experimental optimization by AI in the case of organic photovoltaic cells developed at the ICube laboratory in Strasbourg. External collaborations are considered for experimental data using perovskite materials.