[en] The complexity of modern manufacturing processes in a highly competitive environment forces
the manufacturers to invest massively in automation and monitoring systems. The large data flows
from these new installations are sources of valuable and hidden knowledge that is so far hardly used.
Data mining methods through integrated data analysis tools give a solution to this situation, allowing
easy retrieval of knowledge starting from a data base. This is also a unique opportunity to learn
faster about the process and to detect hidden and complex relationships between parameters involved.
Within this framework we have decided to apply this data analysis method to the straightening process
in shipbuilding. We refer to Caprace et al. (2007) for additional illustrations.
In shipbuilding, the assembly of elements by welding involves temperature gradients within the ma-
terial. These cause deformations which sometimes have to be reduced to obtain an acceptable surface
flatness. The straightening process to eliminate these distortions for esthetical or functional reasons
is labour intensive. Estimating the straightening impact on the production workload is interesting in
the context of production simulation, cost assessment of ship hull, structure optimization, design for
production, etc.
To reach these objectives, the idea was to elaborate, through a data mining approach, a formula
linking the straightening cost to the sections scantlings (plate thickness, dimension and inter-distance
of longitudinal stiffeners, dimension and inter-distance of transversal frames) and to other section
characteristics. This paper describes each stage of the methodology: data description, analysis of
data quality, data exploration and finally choice of discriminatory attributes and the generation of the
data-driven models.
