Originally Published MDDI December
2004
Product Development
Insight
Virtual Product Development Tools—Innovation and Risk Management: Part
2
Virtual tools such as stochastic simulation remove the uncertainty often associated
with finite element models.
Leslie Rickey
MSC.Software Corp.
 |
| Leslie Rickey |
Virtual product development (VPD) provides engineers with a better understanding
of product performance attributes and eliminates design problems. By using virtual
prototypes to detect problems or performance issues early in the product development
process, problems can be corrected quickly. The end result is innovation, cost-effective
designs, and higher levels of reliability in less time than with traditional
processes validated by physical tests.
The first installment of this article offered an overview of VPD and its uses
for reducing product development risk and cost. This installment focuses on
using VPD to generate analytical data for support of the FDA approval process.
It also examines the role of VPD in identifying uncertainty in factors that
affect design performance and function.
FDA Approval Support
VPD can be used to generate analytical data for subsequent use in support of
the FDA approval process, including premarket approval (PMA) and 510(k) submissions.
Currently, FDA does not require VPD information. However, FDA highly recommends
providing the analysis data from simulation studies as another means of proving
device safety and efficacy. Simulation also helps eliminate the red flags and
stigma created by failed physical tests, which can be very difficult to overcome.
For example, most stent manufacturers run structural and fatigue analyses of
their devices. These analyses include simulating the crimping of a stent onto
a catheter, deployment in a vessel, and fatigue under the pressure cycling caused
by systolic and diastolic pressure cycles. In some cases, the blood vessel is
tested. Assuming the design passes the test, a report is generated and included
as an appendix in the submittal package. Basically, the report shows the type
of virtual testing that was performed and the results of that testing.
Typically, such analysis is run on complex models and is highly nonlinear, including
material and contact. The contact involves crimping the stent onto a catheter,
deploying the catheter into a blood vessel, the catheter and balloon, as well
as pressure cycles. VPD is an additional cost at the beginning of the process;
however, it reduces the number of physical tests required and, in the long term,
it reduces the overall cost of product testing.
Identifying Uncertainty
Material properties, loading conditions, thicknesses, dimensions, and other
factors all vary randomly, affecting the performance and function of a design.
However, most simulation tests are performed on a single model with one or more
load conditions to obtain a single result. In the real world, every value and
feature of a model and its environment have combinations of variability and
uncertainty that must be considered to understand possible outcomes.
Stochastic software is a type of VPD application that helps design engineers
model uncertainty and pinpoint the most influential variables. Equally important,
stochastic simulation identifies which conditions are unimportant and thus can
be ignored. Stochastic simulation software has been used primarily in the finance
and industrial engineering markets for a number of years. It is only now beginning
to catch on within the medical device industry. Stochastic simulation lets designers
consider randomly varied engineering values affecting product performance. The
information it produces enables designers to consider uncertainty. By identifying
the factors or values most affecting performance of a design, as well as identifying
combinations of variables that can lead to failures, this information enables
the development of more-robust designs and enables better risk management of
a design. Stochastic simulation software can scale to run on a desktop or on
supercomputing clusters.
Traditional design methods handle uncertainty in models via safety factors.
But these methods have led to overengineering and excessive costs. The problem
with safety factors is that they provide no measure of the real safety levels
in the structure. However, uncertainty can be taken into account in the same
way it manifests itself in nature—without relying upon safety factors.
The advantage of doing this hinges on a very simple point: models incorporating
uncertainty are extremely realistic. VPD technologies such as stochastic simulation
give engineers the ability to perform realistic tests of designs without relying
upon the limited capabilities of physical prototypes.
Uncertainty is often accommodated in finite element models via safety factors.
Using finite element models turns a randomly varying problem into a deterministic
one. Because safety margins are also computed, these models induce a sense of
security. However, the effects of uncertainty are well known. Something is always
overlooked or unmodeled, and some unfortunate and unanticipated combination
of factors and circumstances often finally leads to catastrophic consequences.
After all, models are only models: they can’t return more than the input
and assumptions on which they are based.
Uncertainty, however, can be taken into account in the very way it manifests
itself in nature. The advantage of doing so is based on a very simple but fundamental
point: models incorporating uncertainty become quite realistic.
Stochastic simulation with a finite element model starts by specifying tolerances
and scatter on all input variables used in the model. This process is known
as model randomization, and it targets, for example, thicknesses, beam cross-sections,
material properties, forces, and imposed displacements. Each tolerance is defined
by certain engineering limits and a particular distribution function.
At this stage, an engineer selects outputs, or observables, such as stresses,
frequencies, displacements, etc. The model is then executed a certain number
of times (50–100 is typical), randomly changing each of the input variables
within assigned tolerances. This process, which is based on Monte Carlo techniques,
is referred to as stochastic simulation. The results are referred to as a meta-model,
and in 2-D or 3-D plots, they appear as a constellation or cloud of points,
which carries enormous amounts of information.
 |
 |
| A stent’s solid mechanical stress on an arterial wall was tested using simulation. The findings showed a compliance mismatch that existed between the stent ends and the arterial wall. Using these findings, a compliance-matching stent was created. |
The more complex a product is, the bigger the chance that some unforeseen combination
of uncertainties will have an unexpected and potentially undesirable effect.
This phenomenon, which can only be observed if the finite element model incorporates
many variables with tolerances, manifests itself in the form of outliers, or
anomalous behavior. In many cases, these outliers can be translated directly
into risk and liability. It is therefore of paramount importance to anticipate
their presence and to understand the circumstances under which they arise. This
is only possible if the model is not biased—the model must be capable of
actually producing these pathologies. Everything in the model must also have
a tolerance, because this reflects reality. The model must be able to produce
not 30 or 40 variables, but thousands. Only in this way can a manufacturer make
credible and founded claims about a model’s robustness.
The primary result of a stochastic simulation is the most likely behavior of
the system. This is crucial to the engineer, because the most likely performance
doesn’t coincide with the nominal one. In addition, with stochastic simulation,
the scatter or dispersion of performance is obtained. This may be readily related
to the manufacturing or assembly quality.
An interesting application of stochastic simulation is possible if one replaces
the real physical tolerances with uniform distributions applied over relatively
wide ranges around their respective nominal values. The result is what is called
a model health check. Of no statistical significance, a model health check is
probably even more important than a proper stochastic simulation. In essence,
the exercise enables quick detection of anomalies or unexpected behavior of
the model and helps focus attention where it is needed. Very often, meta-models
with strange-looking shapes signal a potential problem.
Soft Tissue Modeling and Analysis
A stent is a cylindrical device used in arteries and veins to maintain integrity
of the vessel for acceptable levels of blood flow to specific organs. Their
widespread use in cardiovascular surgical procedures is hindered by 20–30%
failure rates within the first year.
Stent design profoundly influences the postprocedural hemodynamic and solid
mechanical environment of the stented artery by introducing nonphysiologic flow
patterns and elevated vessel strain. This alteration in the mechanical environment
is known to be an important factor in the long-term performance of stented vessels.
Because of their critical function, it is vital that the stent design be thoroughly
validated by methods, such as finite element analysis, that highlight any design
or process problems earlier in the design process.
Clinical evidence has shown an abrupt compliance mismatch existing at the junction
between the stent ends and the host arterial wall, disturbing both the vascular
hemodynamics and the natural wall stress distribution. These alterations caused
by the stent were greatly reduced by smoothing the compliance mismatch between
the stent and the host vessel.
Simulation was used to evaluate the solid mechanical stress on the arterial
wall created by commercially available stents. It was found that stresses were
5–10 times greater than the arterial wall stress under normal physiologic
pressure. A compliance-matching stent was created using these findings and was
manufactured and tested. Preliminary results show the compliance-matching stent
is effective in reducing the unwanted tissue growth associated with the failure
of conventional stents. It is expected that these results will lead to improved
stent designs that will ultimately improve the quality of life for patients
receiving them.1
Changing Product Development Processes
It is challenging to change an organization’s processes and tools, especially
for manufacturers who have extremely exacting requirements and decades of legacy
product development techniques. Managers must demand technology that provides
tangible return-on-investment metrics. The risks and rewards must be considered.
Close attention must be paid to guiding the successful implementation of these
new technologies.
A biomedical device manufacturer’s product development processes are the
most important line item for investment. Manufacturers can generate a great
quantity and quality of repeatable data with which to make product development
and manufacturing decisions. By testing hundreds and even thousands of virtual
designs, design efficacy can be ensured and design flaws eliminated before a
device is submitted to regulatory agencies for approval, and before it is used
by doctors and patients.
Conclusion
The crucial difference between VPD and the other digital enterprise software
tools is that VPD is designed to be quickly and easily implemented. It can provide
return-on-investment metrics in just days or weeks. These software tools are
scalable and built such that anyone from novice designers to PhD analysts can
use them to provide rapid process improvements and measurable cost savings.
Using physical testing for corroboration, VPD processes and tools enable medical
device design engineers to innovate and move products through approval and to
the market as quickly and cost-effectively as possible.
Reference
1. Joel Berry, Finite Element Analysis Used to Design Cardiovascular Stents,
[on-line] (Winston-Salem, NC: Wake Forest University); available from Internet:
www.mscsoftware.com/success.
Copyright ©2004 Medical Device & Diagnostic Industry
