Predictive maintenance is based on real-time analysis of sensor data in order to provide a projection of future developments and a prognosis of malfunction. The optimal development of integrated forward-looking maintenance solutions allows hardware manufacturers to offer unified approaches and remote services while securing the budgets of the services provided. The aim is to increase productivity, avoid costly downtimes through better planning of maintenance actions and appropriate management of spare parts and production. Current maintenance models require certain binding assumptions with respect to actual implementation. The objective of the project is to remove methodological locks when maintenance actions are imperfect and of various kinds. The aim is to take into account the history of maintenance actions on a given system. These configurations relate to the effects of memory maintenance. Their dynamic integration into the decision-making process is central to the project. From a scientific point of view, it is a question of finding probabilistic tools developed to enrich existing models. New models promise economic, societal and environmental benefits related to quality control and reduced costs and consequences due to industrial failures and accidents.