Physics Is As Physics Does

By Kevin Harper and David Smith

Editor’s note: This is the unabridged version of this article appears as promised in DE’s July 2006 Elements of Analysis.

Medical device manufacturers encounter special complexities related tohow their products interact with the human body. Furthermore, theindustry is regulated by exacting standards and requirements. As aresult, these specialized companies spend a lot of time running teststo ensure product integrity, safety, and effectiveness.

While physical testing can never be eliminated, it can be significantlyreduced via the proper use of analysis techniques and tools. Anyreduction in early testing provides a sizable return later in theproduct development cycle. One way to do this is to perform virtualprototyping, which provides unique opportunities for innovation asidefrom its obvious gains in saving money and time on testing.

Simulations can provide a depth of understanding that is not possiblewith testing alone. Not all product behavior can be seen or measuredvia testing, and important information about performance might bemissed otherwise. Virtual prototypes can provide that information. Atany point in the model, an engineer can observe the phenomena from anyangle, and combine information in unique ways to provide knowledgeabout a system and its behavior. That knowledge can be leveraged manytimes over in the product development cycle to further reduce time tomarket and lead to breakthrough innovations.

› ›Reducing early testing can help move manufacturers closer to the idealproduct development cycle curve (pictured) where early knowledge paysoff in reducing the costs of stabilization.

At Ethicon Endo-Surgery we seek to understand the behavior of thephysics of any event related to a product’s function and itsinteraction with biological systems. We try to control the physicswithin the constraints of the most realistic conditions possible. Wecall this “Physics Is as Physics Does” modeling.

These modeling conditions often involve free body dynamics wherecontact is the only constraint. One example of using contact as theonly constraint might be a surgeon’s hand grasping an instrument andsqueezing the handle to actuate the device. Another might be a largeassembly held together by its parts and fasteners and performing likethe actual product without modeling simplifications and assumptionsthat can lead to errors. Many applications also include complexmanufacturing steps that must be modeled in order to assemble a deviceand then actuate it to investigate its function. These factors arecritical to understanding the functions of these devices.

In our simulation work, we apply ABAQUS on distributed memory,high-performance computing clusters to achieve the goal of minimizingassumptions and maximizing knowledge gain. We also use techniques likequasi-static simulation with variable mass scaling and real massmodeling when required to support our analytical needs.

One of our primary products is the Trocar, a device used by surgeonsperforming abdominal surgery to gain access into the abdominal cavity.All other laparoscopic instruments are passed through the Trocar duringprocedures.

Trocar placement begins with penetration of the abdominal wall. SomeTrocars make use of a blade to aid in penetration. It is critical thatthis blade be shielded immediately following penetration to preventdamage to internal organs. The successful operation of the mechanismthat controls blade availability and shielding depends on the precisetiming of the interactions between several floating parts. Insimulating these types of complex interactions, there is littleopportunity to simplify boundary conditions. 

‹ ‹This digital model illustrates the complex internal structure of the Trocar shield mechanism.

In this type of application, mass scaling must be kept to a minimum, asartificially increased inertial effects can drastically skew the timingof how the parts interact in the simulation. Both these criteria,compounded by the relatively long event time associated with the armingand firing of the device as compared to an impact event, make for largerun times.

Another Ethicon product that is widely used in surgical procedures isthe Endocutter. An Endocutter is a device that has a jaw at the distalend that allows the surgeon to grasp, cut, and staple tissues such asthe intestines while leaving the cut ends hemostatic (not bleeding).The jaw contains a cartridge with several rows of staples that providecompression to achieve hemostasis.
To use an Endocutter, the surgeon grasps the pistol grip, inserts thedevice through a Trocar, orients the jaw around the tissue to betransected, and then squeezes the closure trigger to close the jaw andclamp the tissue. After tissue compression is complete, the surgeonsqueezes a second trigger, which staples and cuts the tissue in onemotion. Both these separate mechanisms involve complex assemblies ofparts with critical tolerances and clearances for the proper functionof the device.

Several key factors had to be addressed to assess performance of theEndocutter design under “real-world” operating conditions. First, theeffects of loading on the tissue in the jaw were accounted for via theuse of a nonlinear spring that coupled the internal mechanism of theclosure system to the closure tube that actuates the jaw. It was alsoimportant to consider the simulation of the surgeon’s hand grasping thehandle of the device. These locations were treated as rigid surfaces(yellow in the figure below) and provided only contact as a means of securingthe device.

Using contact in this way provided the proper boundaryconditions for the model so that the entire system would behaveproperly without any artificially induced constraints that could affectthe results. Outside of three rigid surfaces, the rest of thecomponents, including the handle shrouds and all internal mechanisms,were modeled as deformable bodies that were assembled and constrainedonly through contact between the parts. In essence, the model wassimply a digital replication of the actual device without any use ofartificial constraints.

This model provided a large amount of information in just one analysisrun. Every component was studied simultaneously under as close toactual use conditions as possible.
› › External view of complete Endocutter  closure system showing area modeled as a rigid surface (yellow).

This modeling technique allows us to see how the device functions,highlight areas of concern, and easily determine whether the systemwill work as well as required. This reduces the number of singlecomponent models and provides system assembly tolerance insightsbecause deforming bodies can show both static clearances between partsand clearances throughout the actuation cycle under load. This, inturn, reduces the requirement for system stack-up analyses that aredone either by hand or on rigid bodies via CAD.

Another type of modeling issue we frequently face is very fine detailin components while running component and assembly models. Fine detailsrequire very large scale meshing to capture the behavior properly andaccurately. These typically show up in components that are thin or havecomplex features that are integral to the function of the component andcannot be simplified for the analysis. In the Endocutter jaw model, wefaced both issues simultaneously.

The jaw portion of an Endocutter device model was used as part of theclosure mechanism modeling described earlier. It provided a worst-casecondition for jaw load where the surgeon clamps the jaw on a hard tubethat represents another device the surgeon might be using.

In models such as the Endocutter jaw we must make use of thequasi-static modeling process to be able to run the model in areasonable time. Every component in this model is made of steel, andthere are very small elements requiring a very small time step inABAQUS/Explicit. In this case, however, the deformation is quite small(i.e., quasi-static); as the plate pushes the tube forward, it loadsthe jaw against the rigid tube. Very little motion occurs. Primarily,only deflection of components under load takes place after smallamounts of clearance are taken up. Though this model had only contactas the primary constraint outside the prescribed motion of the rigidplate, we can make use of variable mass scaling and increase the massof the model significantly to reduce the run time of the simulation.

All of this makes for very large models, highlighting the need tominimize assumptions and boundary conditions, and the challenge of"real physics” events. Fortunately, with the introduction ofdistributed memory parallel (DMP) processing capabilities in ABAQUS,computational power is available to ABAQUS users like never before. Itis now possible to mitigate quickly rising computational costs at afraction of the expense associated with high-performance symmetricmultiprocessing (SMP) systems.

Our system began as a proof-of-concept built from machines that hadoriginally been deployed throughout the company as individual CADstations for designers and engineers. We connected 10 HP VisualizeWorkstations (64-bit PA-RISC, 400MHz processors) running HP-UX 11i on aprivate 100 MB LAN. The compute nodes had no more than 512MB of RAMalong with a modest internal 9GB hard drive for local scratch space.The head node, used for processing and distributing the job to thecompute nodes and functions to consolidate the fragmented resultsfiles, had 1.5GB of memory and a 74GB hard drive to satisfy theseadditional requirements.

‹ ‹Model of the Endocutter jaw clamping on a rigid tube.

Early performance benchmarks on this system showed anywhere from 6x to8x speed-up factor (depending on the model) on 10 CPUs. Analyses thatonce took a week can now be done in one day. In addition, we were nowable to run cases with very large models that were impossiblepreviously on a single machine. This result was very promising,especially considering the very realizable amounts of disk space andmemory allocated per compute node. We subsequently expanded this HP-UXcluster to 21 CPUs, increasing our performance boost to around 15x, andhave used that system to do a wide array of modeling over the years.

Our current system is a 10-node Red Hat 9 Linux cluster (2.8GHzx86-based processors) with 1.0 gigabit interconnect. As with our firstsystem, these machines were previously deployed as CAD stationsthroughout the company. This new system runs about 2.5 times fasterthan its predecessor. Overall, in the last three years we have beenable to increase the processing power by 40x. This accomplishment is atestament to a promising new model. For those whose budgets do notallow for even the smallest amount of investment in computationalinfrastructure, the recycling and utilization of company businessmachines that have reached their end of life can be very effective.

Along with the enhancements we have made to our computer systems,ABAQUS has continuously improved its code with increased support forparallelization with features like general contact, rigid bodies, andfluid-filled cavities. Even the new licensing structure in ABAQUSembraces the use of compute clusters, so that it becomes acost-effective means of doing these advanced models.

As the trend in computational modeling continues toward more physicallyrealistic simulations at the assembly level, models are getting larger,system requirements more demanding, and run times longer. Computationalrequirements are growing at an exponential rate and show no signs ofslowing. Future advancements in computer and hardware systems, alongwith software systems that run on them, are key to continuing on thepath of “Physics is as Physics Does” modeling and to reaping thebenefits it provides.

Kevin Harper has a BS in mechanical engineering from Michigan StateUniversity and an MS in engineering mechanics/biomechanics from theUniversity of Cincinnati. He leads of the Advanced Analytics Team atEthicon Endo-Surgery. David Smith has a BS in mechanical engineeringfrom Penn State, an MS in biomedical engineering from the University ofMiami, and a Ph.D. in bioengineering specializing in the mechanics ofsoft tissues from the University of Pittsburgh. He heads development ofthe AAT HPC cluster. Send your comments about this article throughe-mail by clicking here. Please reference “As Physics Does, July 2006"in your message.

ABAQUS
Providence, RI

Ethicon Endo-Surgery, Inc.
Cincinnati, OH