Experimental design and maintenance, towards a decision-making approach driven by degradation models, with application to lithium-ion batteries

We discuss an innovative decision-making framework for accelerated degradation tests and predictive maintenance, when information about the state of the system, represented by prior knowledge and experimental data, is encapsulated in a degradation model. We consider dynamic programming and reinforcement learning as the framework for sequential decision making in these areas, also including the degradation model learning when necessary. The application of these methods to the design of life testing experiments and to the maintenance of lithium-ion batteries is proposed.