Protection systems against natural phenomena in mountains are critical infrastructures that deteriorate over time and necessitate regular maintenance. Debris retention systems, for example, are one type of protection systems that aim to mitigate natural phenomena such as torrential floods and debris flows by storing a specific volume of solid materials and by regulating the passage of the flows. They thus reduce negative consequences and provide protection to downstream exposed assets. The deterioration of a retention system overtime, including its filling by debris, reduces its efficacy in achieving the desired objectives. One key issue in natural risk context is that budgets provided by State or local authorities for the management of protection structures will always be somewhere limited. Consequently, it is essential and expected to develop models that facilitate ensuring the resilience of such critical structures while respecting available human, material and financial resources. This paper proposes a maintenance decision-aiding model that makes it possible to assess and prioritize different maintenance strategies applied to a retention system over its lifetime. The overall framework involves a (1) physical deterioration model, which contributes in building degradation trajectories of the system; (2) a stochastic deterioration surrogate model learnt from the degradation trajectories, developed using stochastic Petri nets and (3) a maintenance model, which is constructed as an additional layer in the stochastic Petri nets degradation model and which contributes in figuring out the most cost-effective maintenance strategy. A numerical analysis is performed using the data from a real retention system located in the Claret torrent in France and subjected to debris flows over a period of 50 years.