Scientific advances and innovation

  • To take into account the different sources of error (modeling error, discretization error) in the computation of probability of failure.
  • To develop methods that enable spatial or temporal zooms on critical details for the reliability analysis.
  • To update models from sensor measurements to improve its quality.

Expected technical and economic impact

The expected impact of the MUSCAS project is to develop numerical tools in order to consider coupled phenomena in a stochastic framework on complex structures, such as jacket foundations of wind turbine. To use these tools for reliability analysis and for goal-oriented sensor positioning.

Key project milestones

  • February 2018 - The project begins


Parametric studies on 2-dimensional mechanical problems with random elasticity parameters have shown the influence of the mesh size on the estimation of the probability of failure. The discretization error may not be negligible.

A numerical strategy was developed to adapt the finite element discretization during the estimation of the probability of failure using multi-level kriging. By exploiting a posteriori error estimators, it is possible to optimize the construction of the meta model : computations close to the limit state are done on a fine mesh to guarantee precision whereas computations far for the limit state can be done on a coarse mesh which is sufficient to get a trend and cheap. A second approach, based on a priori error estimators, enables the construction of a multi-level kriging meta model who enables to compute the probability of failure without discretization error. Those two methods have been used on two-dimensionnal mechanical problems with 2 random variables.

In the context of fatigue of offshore structures, the local stress (that can be computed or measured) is polluted bu the measurement error or the discretization error. Considering the local stress is bounded by two values at each time step, a method to build a stress history maximizing the damage and a stress history minimizing the dommage has been developed. From those two histories, two extreme damages (minimum and maximum) are computed using Rainflow counting and S-N curves.

From synthetic data, a stochastic model of the damage evolution based on a cumulative Gamma process has been developed. It enables the computation of probability of failure, conditional probabilities and generation of possible trajectories of damage.


Publications and papers published

Oral communications ans posters
For a non-specialist audience: