Power to change everyday life

released 08.18.09

Modeling and simulation are an integral part of modern research and development. Tom Lange, director of corporate research and development modeling and simulation at NCSA's newest Private Sector Program (PSP) partner, Procter & Gamble, presented at the PSP partners meeting in May. He shares with Access' Barbara Jewett some key points from that presentation as well as other thoughts on how high-performance computing impacts not just his work with one of the world's largest corporations, but also U.S. innovation and competitiveness.

Q: You were part of a Council on Competitiveness working group that examined high-performance computing (HPC) and U.S. industry. Why is HPC important to industry in general, and to Procter & Gamble in particular?

A: The Council on Competitiveness is a nonpartisan, nongovernmental group of corporate CEOs, university presidents, and labor leaders working to ensure U.S. prosperity. Why is HPC important to the council and to P&G? It really comes back to this innovation cycle: design, build, fly, crash, fix. People usually understand this for an automobile or an airplane, but it is really hard for most people to wrap their minds around this for disposable diapers or laundry detergent. Consumer products manufacturers have the same motivation to do modeling and simulation as most durable goods manufacturers.

If you are stuck in a physical "only" learning cycle, development costs too much and takes too long. It takes time to build prototypes, which are a one-time use, and you need a huge testing infrastructure. Sometimes that limits innovation because you can't bet it all on the next idea; you have to be a little more disciplined than that for your shareholders.

Q: What could P&G and other companies do if you had unlimited advanced computing resources in the modeling and simulation area?

A: What if we could predict, from ab initio quantum methods, chemical erosion kinetics? Why would P&G care about that? Hairspray in metal cans is expensive. I'd like to put it in a less expensive bottle, but I can't put it there for chemical corrosion reasons.

What if we could do with less wrapping, less cardboard, less plastic, and less metal in our packaging? I'll bet up to 20 percent of the packaging in our grocery stores is insurance. The truth is, it's there because we haven't done the engineering to take it out.

What about car crashes? I'm glad safety experts do the crash tests with the dummy. But what if we could predict hematoma and stop predicting just the G-forces to a dummy's head. Can we predict hematoma? That requires more computing. You don't need just models of the head, you also have to have models of the brain, and a fluid structure interaction where you account for that.

We could solve the biggest, the most complex problems with more computing power. And maybe someone could finally create a plastic fork where that bottom tine doesn't break off!

Q: Why aren't we doing some of those things now? We have the compute power to tackle many of these problems.

A: Lack of application software. That's the issue from my chair. Software for parallel processing is our problem. For example, we don't have the software to simultaneously do spatial and temporal decomposition so I can affordably track molecules and machinery at the same time. There is also an affordability issue related to business models.

Education is also an opportunity. We don't have engineering and science graduates that are computationally aware at the bachelor's and master's levels. The Ph.D. level is fine. But at the lower levels they are being taught, I regret to say, very similarly to the way I was taught a long time ago—and I took my first college chemistry test with a slide rule.

Why is this happening? Because the universities don't have the software either. You know, there are a lot of "why's." Like why are colleges and universities still emphasizing teaching serial programming in computer science classes when parallel is the way the world is going?

Q: So what are you doing with modeling and simulation at P&G?

A: When consumers get products home, the products need to perform as expected or as advertised. Performance in my world often means leveraging fundamental science and engineering contradictions. I need Charmin to be soft but strong. Diapers that breathe yet contain. Moisturizing lotions that stay put when applied but are easy to squeeze. Who wants toothpaste that falls off the brush before reaching the mouth?

With HPC, modeling and simulation lets us more quickly create these products, transforming everyday life. I can replace slow and expensive learning cycles with faster and cheaper virtual realism. We don't yet do everything virtually, but I'm trying.

I have computing machines that are about a decade behind leadership-class machines. I don't apologize for this. P&G is not going to be on the leading edge like government labs or NSF machines like Blue Waters. But through collaborations with government labs and with centers like NCSA, we can access leadership-class compute power for problems like very large molecular dynamics simulations (oil-water surfactant mixtures), for large computational fluid dynamics problems with flow-through porous media, or finite element analysis for very complex machines with hundreds of thousands of moving parts.

Q: How do you justify the return on investment in HPC within an industrial setting?

A: Every year our department does a business impact study that is reviewed with our finance department. This study is shown to our senior leadership. It includes things like capital avoidance and innovation cost savings. It includes the business impact of new products or businesses that modeling helped make possible. We might not have been the only contributor but we were a necessary contributor.

Q: Can you give a specific example where P&G benefitted from HPC in R&D?

A: We used HPC, primarily finite element analysis, when we transitioned Folgers coffee from metal cans to plastic containers a few years ago. Consumers think it's great we made a container with handles and flat panels, giving it a cool look. The truth is the flat panels on the front were put there to solve a stress problem.

After we seal the can at the factory, the coffee continues to give off gas, which builds up and exerts pressure against the container. A metal can is able to handle that pressure, but not an affordable plastic one.

Why not leave it in a metal can? Oil from the coffee in the presence of oxygen makes it go stale in metal. So a one-way check valve in the plastic container that swings open to relieve the internal pressure but doesn't let oxygen back in the container solved the problem. However, the check valve gives you another problem during shipping.

When the truck goes up over the mountains, the pressure drops, gas builds in the container and is released by the valve. Going down, the outside pressure rises again, greatly exceeding the now lower pressure within the canister. Because the air can't get back in to equalize the inside and outside pressure, the canister implodes, basically crushing itself.

Stores don't want to put crushed containers on their shelves, and consumers certainly don't want to buy them. Using modeling and simulation, we helped engineer a solution: flat panels that fit on the side of the canister. When the canister starts to implode, these panels shrink as well and evenly distribute the stress. The bottom and top remain round, and the side panels flatten out a bit and look slightly square. Implosion problem solved.

Simulations showed there was still enough stress generated that an area around the canister lid would crack. As the crack spread, the whole lid collapsed and came off. You could potentially wind up with a disaster; imagine an entire truckload of coffee canisters with no tops. The traditional approach to solving this problem would be to assemble a team to discuss and devise designs, make numerous prototypes, fill them with coffee, load them in a truck, and ship them over the mountains to see what happens. That takes four to six months before you see your results. By using HPC to model various canister/lid designs and try them out in simulated environments, very shortly we had a canister and lid that performs as it should.

Q: But your modeling and simulation department is about more than just new product development and package designs, isn't it?

A: We use a variety of computational science and engineering codes for all sorts of problems.

We do computational chemistry to better understand our formulations. This was very important during Hurricane Katrina when our surfactant supply was threatened. We used models to reformulate very frequently and maintained our supply to our customers through that entire period. And maintained the cleaning performance of a pivotal brand.

More information:
http://www.pg.com/science/index.shtml
http://www.compete.org/