Hardware Reference
In-Depth Information
It also became painfully understood that complex natural phenomena cannot
be modeled so easily, common cause variation being one of the main reasons.
Models can be improved but neglecting variation renders the correlation to
physical findings arbitrary. Hence, the industry is now at a point where it
begins to embrace the concept of stochastic simulation: a single simulation is
no longer adequate, but a number of well-designed simulations are needed to
capture the behavior of the physical process rather than a single realization.
20.2 Science Use Case
Stochastic simulations call for the computation of several realizations of a
process and then processing the results in order to extract conclusions about
the relationship of input and output, and based on different levels of assump-
tions, about process performance and robustness as well as higher-level meta-
modeling. This section is concerned with the particular application of stochas-
tic stamping simulations the automotive industry. Automotive stamping is the
process of deforming flat sheets of material (usually steel or aluminum) into
car body parts using a hydraulic press. The flat sheet is cut in blanks, and
clamped and pressed against dies that are milled to represent the desired
shape. Usually this is a cold process; however, recently there is increased ap-
plication of hot forming.
The process is heavily influenced by the variation of material properties,
blank geometry and position, thickness, rolling angle, forces exerted by the
press, shape and strength of the clamping arrangements, tribological aspects,
to name at least those that may be controlled in practice. Varying such pa-
rameters, in fact encoding the stochastic understanding of these variables,
the stochastic simulation yields valuable information on the behavior of the
production process.
However, a stochastic simulation is also used to model the design process.
Instead of an iterative design, based on successive changes and trial and er-
ror, the designer of the dies might use stochastic simulations to encode her
understanding of the design process variables. In such a case, the number of
variables and possibilities is infinite. The designer may choose to vary every
single geometric detail she thinks might affect performance against the set of
criteria that assist the decision making of when a tool is ready to be released
for milling. Clearly, the number of simulations the automotive engineer needs
to perform has now multiplied. Not only does she need to assess the perfor-
mance of the dies in production, but also perform stochastic simulations to
take decisions on the design of the dies. More than one stochastic simulation is
therefore needed for many processes, while a single engineer designs 20 or more
die sets at the same time, in a department with multiple engineers. Obviously
this simulation load needs to be performed in parallel and one increasingly
