What are the key parameters and assumptions that affect the accuracy of your FEA simulation?

The inappropriate use of beam, plate, and shell finite elements can lead to significant inaccuracies in FEA results, especially when the thickness and section properties are not small relative to other dimensions. This occurs because the simplified assumptions underlying these elements do not account for the complex stress states and deformations that arise in thicker or more complex geometries. To mitigate these issues, engineers should consider using solid elements, refine their models appropriately, or validate their simplified models carefully.

Additionally, it’s important to leverage your engineering intuition to simplify your geometry prior to importing it. If you’re running a shear analysis on the characteristic length of a beam, for example, there would be no need to import the complex multidimensional bracket that the beam would attach to, or a hanger loosely adhered to the side of the beam. Only work with the most important mesh bodies to make every residual count!

In linear computational acoustics, sound pressure gradients are primarily influenced by the frequency of the sound wave. For efficiency and accuracy in simulations, it is standard practice to exclude details smaller than the minimum wavelength corresponding to the highest frequency under consideration.

Statistically, much of the work related to geometry and simplification involves preparing thin-walled bodies for shell element meshing. This is particularly true in industries where sheet metal components and structural profiles, like tubes and RHS, dominate designs. In these cases, extracting mid-surfaces from solid bodies and stitching them together for a continuous mesh plays a crucial role in reducing element counts, enabling more efficient non-linear and large-assembly analyses.

Select which parts are critical to the goal of the analysis. For example, bolts can be modeled using continuum elements if bolted joint is the focus of analysis. On the flip side, bolts can be idealized using different connections if bolted joint is not the focus and it should still give you sufficient accuracy in the simulation.

In complex systems like electric machines that have a moving component, it is important to handle the mesh suitably in the regions such as the airgap. Mesh refinement and element order must be appropriately selected to have an optimal trade-off between accuracy and evaluation time. A mesh sensitivity study is generally useful to identify the most suitable mesh settings.

Try to do continuous mesh to replace bonded contacts when modelling inseparable assemblies like weldments, even though bonded contacts appear as an easy way to connect components. Drawbacks of bonded contacts can overweight the benefits. Contacts produce stress concentrators, can dramatically impact non-linear analysis convergence and often loose connectivity or associativity with geometry. In this light, the best alternative is continuous mesh through all connected bodies.

Mesh convergence is tricky and special attention should be given in regions near boundaries or constraints. These constraints can cause singularities which will lead to an increase in stress with every mesh refinement iteration and mesh will never converge.

Nos dias atuais, onde os softwares de FEA s?o muito poderosos para criar modelos geométricos e malhas adequadas, as condi??es de contorno, e carregamentos, que n?o deixam de ser condi??es de contorno também, representam a maior fonte de imprecis?o nas análises de elementos finitos. Enquanto um erro na malha pode apresentar por exemplo, um imprecis?o de 20% no resultado da tens?o, uma defini??o inadequada de condi??o de contorno pode mudar completamente os resultados da análise numérica. Definir condi??es de contorno adequadas depende fortemente do entendimento físico do problema.

Here are two often overlooked examples of how the misrepresentation of boundary conditions can cause inaccuracies: 1. Over-Simplified Support Conditions: Modeling fixed supports (fully restrained, rigid) when, in reality, the ends have some flexibility that should not be neglected. ? 2. Neglecting boundary condition nonlinearities such as frictional contact in structural joints.

Assumptions at the constitutive level may lead to completely wrong outcomes. What kind of assumptions shall we make just for the elastic properties? - Linear or nonlinear? Hyperelastic maybe? - Isotropic or anisotropic? - Elastic or inelastic? Damage, plasticity or both? For instance, in my career I have mostly worked with fiber-reinforced composite materials: 1) Orthotropy. FRP has very different properties in the longitudinal and transverse directions. 2) Damage. Failure criteria or the full damage curve for each stress component. 3) Inelastic deformation. Typically, pressure sensitive plasticity including hardening for in-plane shear. Do we need all these features in our material model? Not always, but we must be aware of them!

With correct material properties and parameters, FEA can bring the results validation close to real world situations. Most of the time, it was observed that the material used is different and material properties considered for the simulation is altogether different, This leads to incorrect results to validate and verify.

The golden rule in regards to applying the material properties is to validate them using experimental material analysis testing methods. This can be done using multiple methods depending on the material that is in question In addition, if assumptions regarding material properties has to be made, then these assumptions must be properly reported as part of the methodology and/or results in order to put these simulation results within their proper context.

??The material #constitutive model is very important for #FEA solution #accuracy and #convergence ??The #selection of material model (linear,hyper,visco,elastic-plastic,etc) should be a close match to the observed behavior of the material, with model assumptions ??The material model must be valid over the anticipated range of loading. No discontinuities jumps ??Adjacent topological components must be similar in stiffness to aid model convergence ??It is important to understand that all FEA models are #approximations, by virtue of defined polynomial element shape functions ??#Hourglassing may occur with “soft” materials; this is dealt with by adding artificial stiffness via hourglassing protection #FEA #engineering #computation ??

I would add to [future Dr.] Hijazi's comment, knowing your governing equation and applicable non-dimensional numbers can also help in making assumptioms. For example, you can calculate non-dimensional numbers such as Reynolds or Mach that will allow you to assume laminar, incompressible fluid flow (or not!). Of course, you still need to validate your model with the applicable experimental data.

1. Complex models that require a large number of elements for accurate predictions can consume significant amounts of memory, pushing the limits of computer resources; the engineer can switch from a default direct solver to an iterative solver such as preconditioned CG or GMRES, which requires less memory to solve the FEA model effectively. 2. Large models with many degrees of freedom can result in long computation times. Enable multi-threading/parallel processing to ensure that the solver utilizes all available CPU cores. If available in hardware, enable GPU acceleration to significantly speed up FEA computations, especially for large models.

My experience of 18+ years in simulation segment tells me that, 90% time default settings of solvers works very well, if one follows the standard Best Practices for meshing/discretization. Most of the time, the users don't use common sense, physics , load transfer path, boundary conditions, loadings, etc. and only blaming to simulation software. This thought process needs to be corrected in first place. Simulation software deployed or launched only after thousands of load cases and examples.

We cannot just blindly 'trust' the FEA software's default settings! While it may not be necessary to know how to code, we do need to understand the correct way of computational thinking (i.e. the algorithm). We must know how the solver behaves and solves our problem (e.g., implicit vs. explicit solvers). Not to mention, there are many settings that introduce 'numerical manipulations' into the model. Also, be sure to read the user guide and other references!

Having conducted many random vibration/shock analysis in my 42 year career I found modal testing to be invaluable in validating FE models! Complex MIMO testing or simple roving impact testing for linear dynamic systems should be considered when FE accuracy is demanded.

In FEA, solver settings crucially impact accuracy and performance: -Convergence Tolerance: Determines the acceptable error level; lower tolerance means higher accuracy but more computation. -Solver Type: Direct solvers offer precise results, while iterative solvers (e.g., Conjugate Gradient) are better for large systems. -Time Step Size: In transient analyses, smaller steps enhance accuracy but increase computational time. -Preconditioning: Improves efficiency of iterative solvers by transforming the problem for easier solving. -Maximum Iterations: Sets a limit for iterative processes, balancing between convergence and computation time. Adjusting these settings ensures optimal accuracy and efficiency in simulations.

The choice of element type can affect the accuracy of the FEA simulation. Different element types have different levels of accuracy and computational requirements.

I think what I'm missing here is: STEP 0: Pen & Paper Draw a rigid body diagram with loads and constraints, and think of the objectives. Do not touch your computer till you have this step-0 very clear and understood.

Remember that to have a good model it must: 1) Be based on experimental/observed data from the system to be modeled 2) Reproduce known experimental findings (validation) 3) Be able to make meaningful predictions about your system Happy modeling!

I agree with Dr. Torres-Rodriguez. In fact I think that this is an excellent list of things to consider when doing FEA modeling.

Assumptions would be an important part of your FEA simulation too. Your model is only as good as the assumptions it's built on. For example, if you are performing a linear static analysis; ensure that the underlying assumptions of linear material behaviour, small displacements, and Quasi-static loading applies to the problem. If you use linear static analysis for a problem with sinusoidal forces or large deformations, it might not give very accurate or reliable results.