By their very nature, empirical models must be treated with care in order to avoid predictions which are not physically possible. One example is the calculation of the Charpy impact toughness of steel welds as a function of composition and processing, where the impact energy should not be negative. However, there is nothing to prevent a user from implementing inputs which lead to nonsensical results. We examine here whether a scheme used in kinetic theory can be generalised to create neural networks which are bounded. It is found that such procedures lead to bias. In the process of doing this work, some interesting trends have been discovered on the role of process parameters in determining the toughness of steel welds.
Materials and Manufacturing Processes 24 (2009) 16-21
 |
|
 |
| Coalesced | Foam | TWIP | FSW | 3D |
| Texture | Properties | Silicon | Synchrotron | Models |
| Neural Networks | Creep | Mechanicallly Alloyed | Theses | Retained Austenite |
| PT Group Home | Materials Algorithms |