What are the best practices for validating and verifying CAE results for vibration fatigue and reliability?

Benchmarking competitors' product is the best way to develop and set realistic design target. A competitors' product can be evaluated under operational load/environment and accelerometer/microphone data can be collected for the whole product, sub-system, etc. With Fourier/other appropriate transformation, time-series data can be converted to frequency and competitor product information can be established. From the competitors' benchmarked data, better-than-competitor product target can be set. Some CAE model of competitors' product can be developed at this stage also. Benchmarking also helps coming up better design, cost and manufacturing methods.

CFD is not so common even if there are situations in which vibrations are produced by interaction with the fluids. Mechanism motion, typically handled with multi-body simulation (MBS), it's really important to capture the time history acting on the component subject to fatigue load.

There are so many levels of validation when it comes to CAE . Out of all these, I believe that Model Assurance Criteria is the most effective way to validate your model and improve upon the assumptions that were made. The FEA-Test comparison is one of the methods in MAC. Again this depends on the number of points you choose to run the tests with. We could also make use of MAC Contribution Analysis to improve the performance of the component or Assembly. The significance of MAC comes from the idea of the stiffness of the assembly or system rather than focusing on validation for particular type of loads and boundary conditions.

In my experience, test data from previous experiments/ on-field ongoing processes (by using various sensors and strain gauges) are some of the ways that a virtual system at any level, be it component or subsystem or system level can be checked for validation. Often times, the product development cycle aiming for certain percentage accuracy in life/stress values for a particular part; has closely matching patterns of output contours when compared to the previous design variant of the same part. An extreme change in the output values for minor design variation from one version to next can be a red flag.

In addition to verfiying your output against requirements, specifications and standards; an additional step that needs to be inculcated in every analysis is checking of the mathematical accuracy of the model. This step is related to the FEM itself and little relation to physical reality. It is typically done before external loads are induced in the system and stands true for almost all types of FE analyses. Free-free modal analysis, unit gravity loading, unit enforced displacement check are some of the model verification steps that can help identify bugs early in the process.

To add to Ninad points, a fixed Eigen solution to visually inspect the deformed shape is useful to validate connections between the various component. Even when your free-free model show six rigid body modes and the 7th as elastic, connections might be missed. Looking a forces within your spring of beam elements when used for connections should also give you a hint. When performing a thermo-elastic analysis, applying a 1° delta over a modified model with the same material should yield no internal stresses (apart from the local constraint).