1) Sometimes, as a result of an aggressive strategy of geometric simplification of models, significant aspects that affect the behavior of models are excluded from it. Create a separate local analysis in which only a potentially dangerous area will be localized. Execute the model with minimal or no simplifications at all. It is even permissible to model a separate part of the part, transferring boundary conditions across the borders of the test zone. Analyzing the results of such an analysis, you may discover any new aspects of the operation of your product. For example, mechanical stress concentrators
The potential stress concentrator was removed during the simplification of the mode
2) Structural strength can be understood as a balance in the distribution of stresses. Sometimes the loss of strength characteristics of structures is the loss of this balance. In this case, localization of a potentially dangerous area will not lead you anywhere. It is necessary to analyze the behavior of neighboring components of the structure and their compliance with the declared characteristics. Perhaps the supplier of one of the components of your product has improved their characteristics by using a new material, or for example by changing the characteristics of heat treatment and so on. Identify and eliminate this discrepancy
Unexpectedly high rigidity of the left part passes stress to the right one
3) We use loads obtained by statistical methods for analysis quite often. Sometimes we take loads based on reference data, or values obtained during previous projects. Such approaches are sometimes forced. Despite this, you can try to conduct an audit. Register some data that will help clarify your problem statement. There is a huge variety of options how exactly you can do it. Tensometry, piezo-vibroscopy, speckle interferometry, measurements of temperature or gas dynamic characteristics and much more. After conducting a side test, integrate the results into a mathematical model and check the state of it. Perhaps the working conditions are more aggressive than you expected.
Such simple solutions to the situation will help you eliminate the gap between the model and reality. After identifying the root cause, the solution will become obvious.