Method for establishing stress/strain curves by means of spline interpolation on the basis of characteristic points and with the use of neural networks

1. Stress/strain curve generator for calculating a stress/strain curve by means of characteristic values and/or corresponding intermediate values, a piecewise approximation being carried out in subintervals to calculate the stress/strain curve, where the characteristic points are calculated by a neural network and principal component inverse transformation on a computer.

2. A method for setting up a neural network comprising:

a) measuring a plurality of stress/strain curves that differ one from the others by at least one parameter,

b) selecting a plurality of characteristic points for each of the stress/strain curves and

c) training a neural network with said at least one parameter as input variable(s) and the corresponding characteristic points as output variables.

3. The method according to claim 2 , wherein the at least one parameter relate to a member selected from the group consisting of formulation, color, component geometry, processing conditions and test conditions.

4. The method according to claim 2 , wherein the selection is based on at least the following values of a stress/strain curve:

a) the yield point,

b) the failure point,

c) one or more minima and/or points of the maximum curvature, residing between the yield point and the failure point on the stress/strain curve.

5. The method according to claim 2 , wherein the selection is based on at least the following values of a stress/strain curve:

a) the yield point,

b) the failure point,

c) one or more minima and/or points of the maximum curvature, residing between the yield point and the failure point an the stress/strain curve and

d) one or more intermediate points, which are generated automatically.

6. The method according to claim 2 , wherein each of the stress/strain curves is established several times without varying the at least one parameter.

7. The method according to claim 2 , wherein the characteristic points are subjected to a transformation and the neural network is trained with the transformed characteristic points as output variables.

8. The method according to claim 7 , wherein the transformation is a factor analysis or principal component transformation.

9. The method according to claim 2 , wherein the stress/strain curves are type-classed and a separate neural network is trained for each type of a stress/strain curve.

10. The method according to claim 9 , wherein a further neural network is trained with the parameter(s) as input variable(s) and the type or the principal components of the type of the stress/strain curve as output variables.

11. The method according to claim 10 , wherein the separate neural networks and the further neural network are connected in parallel and the output of the separate neural network with the type having the highest probability is selected based on the output of the further neural network.

12. A method for establishing a stress/strain curve comprising:

a) setting up a neural network using a method according to claim 2 ,

b) inputting the parameter(s) into the neural network,

c) outputting the characteristic points from the neural network,

d) generating the stress/strain curve from the characteristic points.

13. The method according to claim 12 , wherein the neural network having the type with the highest probability is selected.

14. The method according to claim 2 , wherein the strains are sorted.

15. The method according to claim 14 wherein, the strain is logarithmically scaled in a later subinterval.

16. A neural network prepared by the method according to claim 2 .