RESNA Annual Conference - 2019

Using FEA Software To Simulate Lab-Based Failures In Casters

Menachem M. Marcus1, Anand A. Mhatre1, Brady T. Cousins1, Daniel P. Henry1, Samuel E. Zilavy1, David Schmidt1

1University of Pittsburgh


Wheelchairs fail frequently in a short period of time. Part failures with casters, brakes, seats and tires within few weeks to two years of wheelchair use are common. [1–7] Casters are the highest failure prone parts as per research evidence [2,8]. Studies documenting wheelchair incidents and repairs in the United States and Scotland found that nearly one-third of wheelchair part failures are caster failures. [8,9] Caster failures can be catastrophic. For example, caster stem bolt fractures can cause the wheelchair to tip and the user to fall out of the wheelchair, which is the most common wheelchair injury. [10] To improve the quality of wheelchair casters, the University of Pittsburgh’s Department of Rehabilitation Science and Technology (RST) and the International Society of Wheelchair Professionals are developing a caster durability testing standard. [11] They have developed a testing equipment to replicate community failures in a lab-based setting by benchmarking the test exposures to community’s shock and environmental exposures. The testing protocol predicts community failures for most of the caster models. In conjunction to such a physical testing approach, this study aims to develop a simulation-based approach using a finite element analysis (FEA) software to predict caster failures.

The goals of this study are:

1) Evaluate the mechanical properties of the caster including the stem bolt, axle bolt, and the steel caster fork. To perform an accurate fatigue analysis in FEA, the yield strength, fatigue/endurance strength, and ultimate tensile strength are required.

2) Using an FEA analysis and simulation software, develop a model to simulate the impacts and stresses that a caster will experience during testing.  

3) Compare the failures found in the simulation analysis and laboratory-based testing.


For the purpose of this study, the Whirlwind wheelchair casters, popularly known as the Zimbabwe caster was chosen.

Evaluation of caster’s mechanical properties

Due to the computational make-up and requirements of the FEA packages, the material properties of the parts being studied needed to be initially quantified. The yield strength, fatigue/endurance strength, and ultimate tensile strength of the stem bolt, axle bolt, and caster fork of the caster were obtained from information afforded by a Rockwell A hardness test.

FEA analysis

To simulate the caster fork, a FEA structural analysis was performed using ANSYS Mechanical. [12] Some modifications had to be made to the 3-D model to ensure a more accurate outcome. Mainly, the weld was assumed to be fully filled, and as weld failures have not been observed in either field or lab studies, the weld was made overly strong. The analysis included setting up boundary conditions for the axle bolt and stem bolt. Total load on the caster was taken as 30lbm and added in the simulation. Static and transient FEA analyses were conducted and the cycles to fork failure were determined for comparison with lab-based testing failures.


Evaluation of caster’s mechanical properties

Once the hardness values were found, tensile strength values were found from ASTM handbook. The ASTM standard A370 chart was used to translate the hardness value to a Brinell hardness value or straight to tensile strength [13]. For hardness values that do not have the tensile strength shown, the corresponding Brinell hardness value can be used with an equation from Machinery’s Handbook to find tensile strength [14]. This afforded a range of values for the tensile strength, averaging around 30 kpsi. The remaining properties were interpolated from this value.

Figure 1. The caster fork shown as seen in ANSYS Mechanical. The simulated axle bolt is seen as a single element. The coordinate system for the FEA analysis is also shown with the top of the caster aligned in the positive Y direction, and the forward facing forks in the positive X direction.
Figure 1. The caster fork shown with the simulated axle bolt in yellow. The coordinate system for the FEA analysis is also shown.

FEA analysis – FEA boundary conditions

Firstly, for the axle bolt, APDL code was used to simulate the stiffness of the axle bolt and to distribute the load from the center point to the two end nodes in the same way the real axle would. Using the 188 elements to simulate the bolt, will supply an axial force that will prevent the fork ends from moving independent of one another. This setup, along with the applied coordinate system, can be seen in Figure 1.

Figure 2. Cut-away view of the top of the fork. When the load is applied to the fork, separations occur between the fork, stem bolt, and the washer.
Figure 2. When the load is applied to the fork separations occur between the fork (blue), stem bolt (purple), and the washer (orange).
The second boundary condition centered on the top of the fork in two main contacts. The first is the forks contact with the stem bolt and the second is the forks contact with the washer. This washer is between the fork and the wheelchairs main frame. Because the fork is not being pulled from the wheelchair the bolt will not produce any axial forces to the fork. This means that the stem bolt will only stop the fork from moving in the X and Z direction along with eating part of the moment on the fork. This leaves the washer to prevent the fork from moving in the Y direction. This condition can be seen in Figure 2.

To simulate the stem bolt, washers, and bearings, a complex APDL code was drawn up using further 188 nodes, as well as compression only elements. The compression only nodes were used to connect the stem to the fork, thus insuring that the two surfaces cannot pull on each other but compression between the two surfaces is allowed wherever the contact occurs. To simulate the stem-bearing, the bearing stiffness was obtained from the manufacturer, and was a complex spring matrix was created to simulated said stiffness. Next, the washer must stop the fork from moving in the Y direction, but because of the separation that occurs, compression only elements are again used to connect the top surface of the fork to a remote node that will be the wheelchair frame. This entire layout can be seen in Figure 3.

The last component required is the mass for resistance in the transient analysis. Using a 30lbm resistance as the baseline, 10 mass elements of 3lbm each were applied to the top surface of the bolt.

Figure 3. The simulated stem bolt, washer, and bearing shown in a cut-away view a seen in ANSYS Mechanical. APDL code was used to create a series of complex spring matrices to represent the stem bolt, washers, and bearings.
Figure 3. The simulated stem bolt, washer, and bearing shown in white, using APDL code to create a series of complex spring matrices.
For the simulation analysis, four distinct layouts were considered: static Y, static X-Y, transient Y, and transient X-Y. The static X-Y instance is when the wheelchair hits the bump, and the static Y instance is when the wheelchair lands after the impact.

Beginning with the simpler of the two options, the static Y instance was approached first. After fixing the model in place using constraints on all the degrees of freedom as explained above, a life-cycle study was performed. As the fork and components are made from ductile material, the Von-Mises stress was used. Additionally, the Goodman criteria was used as the criteria for failure are more concerned with cracking than yielding. A Marin Factor of 0.7 was calculated, and using the 30lbm load, the simulation was run. The fork was seen to last to approximately 90,000 cycles.

The static X-Y application of load is very similar to the analysis of the static Y instance. All conditions were held the same except for the applied force. Assuming that the height of the bump was ½ inch, and calculating the approximate angle the force was applied in to be 40 degrees, the resulting 8.5g seen in the data was from a 6.5g in the Y direction, and a 5.5g in the X direction. This resulted in a larger moment in the fork and yielded a shorter life approximately 39,000 cycles.

Figure 4. Summary of results showing the lab-based results of about 40,000 cycles, and the individual FEA results. The static Y is at about 90,000 cycles, the transient Y is at about 30,000 cycles, and the transient XY is at about 20,000 cycles. The Static XY is seen to come closest to the lab-based results of about 40,000 cycles.
Figure 4. Summary of results showing the lab-based results (Blue) and the individual FEA results. The Static XY (Yellow) is seen to come closest to the lab-based results of about 40,000 cycles.
When changing to the transient analysis some of the boundary conditions had to be changed. Firstly, the washer is no longer fixed in the Y direction because in transient an acceleration is inputted, and a mass is needed to resist that acceleration. This is where the mass elements come into play. In the Y transient the springs simulating the bearings were kept fixed in the X, and Z directions as there is no movement in either of those directions. Since the acceleration command moves the fixed frame, node a (in the center of the axle-bolt) was fixed in all directions and the frame was accelerated in the Y direction with a magnitude of 8.5g. With the mass resisting the motion, we get an approximate life of 30,000 cycles.

When considering the X and Y transient analysis, many more assumptions must be made. This is because the mass resisting the motion in the x direction in unknown.  Because of this, one must assume that the force in the X direction is the same as the force seen in the static analysis and the Y acceleration is only 6.5g. This also allows us to keep the springs fixed in the X direction. The results for this analysis was found to be around 20,000 cycles.

Failure comparison

The results of this study proved to be very promising. With lab-based failures appearing at around 40,000 cycles, the static X-Y simulation comes close to this number. A full table of the results can be seen in Figure 4. Additionally, the location of the failure as seen in the simulation, is one that has been repeatedly seen in the lab. This can be seen in Figure 5. 


Figure 5. The location of the failure, as seen in the FEA analysis (left) and the lab-based experiments (right). Both cracks seem to propagate from the same location at the top of the caster fork.
Figure 5. The location of the failure, as seen in the FEA analysis (left) and the lab-based experiments (right). Both cracks seem to propagate from the same location.
This FEA analysis is an attempt to replicate caster failures as found in lab-based settings using algorithmic simulations. After confirming the material properties of the caster components and building the simulation parameters in ANSYS, the team was able to simulate fatigue failures in terms of both quantity of cycles to failure, as well as the location of the failure. Where this study does fall short however, is in the many assumptions that were required of the algorithm to replicate these results. Mainly, the applied load was averaged at a constant 8.5g, whereas in the field, the caster will be experiencing a greater range of transient values. The algorithm can be further improved by expanding the number of ‘beams’ in the spring matrices, as their contact points with nodes on the fork display artificially high contact stresses.

The team has plans of further improving this algorithm to include additional caster models, as well as including a wear model for the caster tire. This wear model will allow for significantly more accurate inputs for the loading conditions, and will better simulate the stresses and loads that the caster will be experiencing. Additionally, the team would like to compare simulation failures with community failures when reliable acceleration/strain measurements from the community become available.  


A FEA simulation was performed on a caster model in an attempt to replicate reported field failures as found in a lab-based setting. Based on the results, the model is accurate enough to predict the cycles till failure, as well as the location of the failure.


[1] Saha R. Study of Wheelchair Operations in Rural Areas Covered Under the District Rehabilitation Centre (DRC) Scheme. Indian J Disabil Rehabil. 1990;74–87.

[2] Toro M, Worobey L, Boninger ML, Cooper RA, Pearlman J. Type and Frequency of Reported Wheelchair Repairs and Related Adverse Consequences Among People With Spinal Cord Injury. Arch Phys Med Rehabil. 2016;97(10):1753–60.

[3] Toro ML, Garcia Y, Ojeda AM, Dausey DJ, Pearlman J. Quantitative Exploratory Evaluation of the Frequency, Causes and Consequences of Rehabilitation Wheelchair Breakdowns delivered at a Paediatric Clinic in Mexico. Disabil CBR Incl Dev. 2012;23(3):48–64. Available from:

[4] Mukherjee G, Samanta A. Wheelchair charity: a useless benevolence in community-based rehabilitation. Disabil Rehabil. 2005;27(10):591–6. Available from:

[5] Mhatre A, Martin D, McCambridge M, Reese N, Sullivan M, Schoendorfer D, et al. Developing product quality standards for wheelchairs used in less-resourced environments. African J Disabil. 2017;6(0):15 pages.

[6] Reese N, Rispin K. Assessing Wheelchair Breakdowns In Kenya To Inform Wheelchair Test Standards For Low-Resource Settings. In: RESNA Annual Conference - 2015. 2015.

[7] Rispin K, Riseling K, Wee J. A longitudinal study assessing the maintenance condition of cadres of four types  of wheelchairs provided in low-resource areas. Disabil Rehabil Assist Technol. 2018;13(2):146–56.

[8] Gaal RP, Rebholtz N, Hotchkiss RD, Pfaelzer PF. Wheelchair rider injuries: causes and consequences for wheelchair design and selection. J Rehabil Res Dev. 1997;34(1):58–71.

[9] Mair C. Applied internet-of things technology in the management of wheelchair maintenance at NHS WestMARC: A retrospective. In: European Seating Symposium. 2018.

[10] Kirby RL, Ackroyd-Stolarz SA, Brown MG, Kirkland SA, MacLeod DA. Wheelchair-related accidents caused by tips and falls among noninstitutionalized  users of manually propelled wheelchairs in Nova Scotia. Am J Phys Med Rehabil. 1994;73(5):319–30.

[11] Mhatre, A., Ott J. & Pearlman J., 2017, ‘Development of wheelchair caster testing equipment and preliminary testing of caster models’, African Journal of Disability 6(0), a358.  

[12] ANSYS Mechanical APDL, Version 18.2, 2018, ANSYS

[13] ASTM International. (2017). ASTM A370-17a Standard Test Methods and Definitions for Mechanical Testing of Steel Products. Retrieved from

[14] Oberg, E., Jones, F. D., Horton, H. L., Ryffel, H. H., & McCauley, C. J. (2016). Machinery's handbook: A reference book for the mechanical engineer, designer, manufacturing engineer, draftsman, toolmaker and machinist (30th edition.). New York: Industrial Press.


For this study, a University of Pittsburgh’s Mechanical Engineering Senior Design Group was sponsored by RST to investigate and develop the algorithms for the FEA analysis.