Classical methods to estimate the Remaining Useful Life propose a distribution of probabilities of failure risk, which consequently makes it possible to give a probability of failure before each moment. However, recent machine learning methods, which use more complex models to better understand the possible causal links between the available data and the target indicator, only propose a regression. In this article, we introduce a transformation of the output value of a regressor based on machine learning by completing it with another one which calculates the estimated error of this model and uses it to create a distribution thanks to a Weibull law. This approach, called double-ML-Weibull, is a much better tool to provide simulation in a stochastic context, instead of using scalar values like "Mean Time To Failure" or "Mean Time Between Failure".