The introduction of intrinsic compliance in the design of robots allows to reduce the inherent risk for humans working in the vicinity of a robotic cell. Indeed, it permits to decouple the dynamic effects of the links' inertia from those of the rotors' inertia, thus reducing the maximum impact force. However, robot designers are lacking modeling tools to help simulate numerous collision scenarios, analyze the behaviour of a compliant robot and optimize its design. In this article, we introduce a method to reduce the model of a multi-link compliant robot in a simple translationnal mass-spring-mass system. Simulation results show that this reduced model allows to accurately predict the maximal impact force in the case of a collision with a constrained human body part, and thus estimate the severity of such collision. Multiple impact scenarios are conducted on two case-studies, a planar serial elastic robot and the R-Min robot, an underactuated parallel planar robot, designed for collaboration.