I prefer to use experemental data for SIF than FEM results because if we use shell elements FEM couldn't calculate REAL peak stress. To calculate real peak stress we need to use volume FEM elements with very fine mesh at stress concentration zone. FE/Pipe can't do that. There's just special factor that user should enter. Result FEM stress multiplied by this factor to calculate REAL peak stress. Where should we take that factor? It could be obtained only from experiment...
Sometimes FEM analysis give us qualitative incorrect results. For example for 45 degree branch...