Verification and validation of nonlinear models in civil engineering

The increasing complexity of numerical mechanical models makes it increasingly difficult to guarantee the representativeness or realism of simulation results. The use of numerical simulations as predictive tools is becoming increasingly widespread, as they are a priori less costly, less complex, faster and more flexible than test campaigns. These advantages sometimes make it possible to carry out sensitivity studies that would otherwise be too cumbersome.

In the field of civil engineering, simulations can be used to study problems on scales that are either unattainable experimentally or require unavailable technical and financial resources. In particular, full-scale building tests under earthquake conditions are unthinkable beyond a certain size. In France, for example, the largest seismic table, called Azalée and measuring 6 meters by 6 meters, is currently located at the CEA (Commissariat à l'énergie atomique et aux énergies alternatives) Tamaris facility in Saclay. Only in-situ measurements during actual seismic events provide valuable, but very partial, information on the behavior of these structures.

While such on-site measurements are essential, they are far from sufficient to address the safety issues that arise in the nuclear power generation sector, because observations or measurements characterizing the effective mechanical response of such structures under seismic conditions are rare. In the case of new nuclear power plants, the dimensioning of structures can a priori guarantee quasi-linear behavior under seismic stress. However, the upward reassessment of seismic risk following the Fukushima nuclear accident in 2011 has called into question this quasi-linear behavior in the case of existing structures built prior to this assessment.

As a result, numerical simulations are now routinely incorporated into safety demonstrations. Calculation tools (software and models) must therefore be verified and validated. The main difficulty encountered in these tasks is the absence of a well-defined general formalism. This is probably due to the sheer variety of possible tools, models, situations and applications. Nevertheless, a great deal of work has gone into the creation of guides proposing frameworks, concepts and best practices in this field, but the application of these productions remains a delicate matter.

In order to present a global view of these issues and illustrate them, this article is divided into four parts:

• definitions will be proposed for the two key concepts of verification and validation;

• A history of the various existing guides is presented;

• In the third part, the central role played...