FEA diagnoses machine failures

At our firm, for example, Algor V16 has proven useful in design work and for deciphering why something broke. Rather than just describe a laundry list of features in the latest release (you can read those on the company's Web site), it's more useful to show how the software solves real-world engineering problems.

Verifying the failure mode of a flange-toshaft joint provides a case in point. It was loaded in torsion and bending, and failed sooner than expected. The FEA software is a good selection for this problem because it easily models and analyzes surface-to-surface contact between parts.

A roadmap of the analysis looks like this: I first created a model in Alibre Design, a parametric, feature-based solid modeler. The model easily opens in Fempro, the interface that's used with all of Algor's analysis programs and solvers. Default settings for the hex-dominant, automatic solid mesher put an initial coarse mesh on the model. This shows that the solid model has no flaws and establishes a baseline for further iterations. After meshing, a right click on the menu tree shows material properties available from the software's extensive library.

It only took a couple hours to set up the entire FEA model and generate initial results. Linear static-stress analysis ran to completion on the first try. A report showing calculated details and significant images is just a click away.

After initial analysis, I meshed the model with a substantially finer mesh using the controls built into the automatic solid mesher. A simple slider bar makes it easy to apply smaller or larger elements. Several analysis iterations using increasingly finer meshes established convergence (identified a level of mesh that produces results that have "settled down," so there was little change in results between meshes).

After noting the convergence mesh size, I remeshed the model with default settings and used a mesh-refinement feature. After a couple tries, mesh-refinement points corresponded to the mesh level determined by the convergence studies. This feature is a great time saver when working with large models with many components, as well as models that need a fine mesh in several locations. Final results of the failure analysis were known in less than three weeks after the initial modeling.

The software predicted 83 ksi at the failure points. In 4130 HT steel, that indicated a probable failure mode. The client was soon interested in an accurate identification of the part's failure. As a start, we overlaid images generated by the software from the final analysis on photographs of the broken part. Failure points and propagation of the failure through the shaft as indicated by the FEA software accurately corresponded with the photos.

Knowing stresses were too high made it easier to propose and validate design modifications. Joining a flange to a shaft causes stress risers, so we tried to "bury" the loads inside the flange rather than transfer them at a flangeto-shaft interface. The new design mounts a gear on the end of the shaft. It protrudes through the flange and is welded on the back side. In addition, the flange will be broached to meet the shape of the shaft with a tight (but not interference) fit.

Validating the concept requires a highquality CAD model with close tolerances suitable for FEA. I modeled only a portion of the shaft, the flange, and weld.

A feature in the FEA software allows using "bonded" contact, which gives node-fornode connection over the entire area of a weld. Using the "welded" option limits contacts to the weld edges, which would not have been accurate for this model. Furthermore, the software allowed specifying a tolerance gap of 0.001 in. This prevents bonding the shaft gear to the broached-gear hole in the flange.

Default settings for this operation, on the other hand, would have forced the shaft mesh to match the broached-hole mesh, effectively bonding the parts over their contact area. But that is not what happens. The goal is to let the weld on the back side act as the only connection between the shaft and flange. Access to advanced mesh settings provides flexibility for specifying the mesh exactly as needed.

Because the smallest feature of interest is 0.161-in. long, we instructed the mesher to apply elements about half that size, 0.0805 in. This created a large model of some 500,000 solid-brick elements.

Refining the mesh using the automatic-refinement feature limited the size of the analysis. The feature produces a useful model with less than a half-million degrees of freedom. Manual mesh refinement would have been prohibitive on a problem of this size and complexity. The software's mesh refinement tools turned this into a problem easily solved on a PC with an 850-MHz Athalon processor and 382 Mbytes of RAM running on Windows 98 SE.

After meshing overnight, loads and constraints were applied and the model was verified. The actual analysis took only 90 min and 700 Mbytes of disk space.

Solution results indicated a reduction in maximum stresses from 83 ksi in the failure analysis to 25 ksi. In addition, maximum stress locations were successfully "buried" inside the center fiber of the flange as intended.

I have considerable experience with other widely used FEA programs and first used Algor 16 years ago, but this was my first experience with its V16. It was easy to get up and running with many tutorials and generous help items. In addition, the personnel at the company extend friendly and competent assistance via online help and by telephone.

The software comes from
Algor Inc.
150 Beta Dr.
Pittsburgh, PA 15238
(412) 967-2700
algor.com

— Richard Helms

Richard Helms is president of Industrial Construction,a consulting engineering firm. In addition to analyses, Helms designs machines and tooling, and manages projects.

MAKE CONTACT:
Algor
(412) 967-2700
algor.com
Industrial Construction Co.
(360) 886-1560
[email protected]