Having a limited experience working with Abaqus/Nastran and open source programs: 1)Failure is when the program is not working 2)Always perform an elastic analysis before using a more complicated inelastic one. At the bottom line, your program is probably using the modified Newton Raphson method, which is a sequence of elastic analyses. The material cannot fail in an elastic analysis. You can confirm the state of stress simply by reading it from the program. 3)A yielding zone of material does not mean that your foundation/dam etc has failed. You can again confirm the state of stress simply by reading it from the program. 4)If the model is unstable, and you 're using the FE method, than you will see a mechanism appearing and the program will stop (with error messages), because the method cannot deal with it. Be careful because it may be the program's and not the method's fault! 5)Be careful: a foundation may be stable, but it may have settled a meter. To check the failure load of a foundation, you (may) perform multiple analyses, inreasing the foundation load. 6)Ask the program to give you the gradients. If it can do it graphically it is better.