Towards Making Fault Injection on Abstract Models a More Accurate Tool for Predicting RT-Level Effects