Product Reliability Using Simulation

Countless hours of computing time are spent every year in simulation. For example, manufacturers of vehicles need to ensure quality cars and that will last for a given amount of time before breaking down [1]. But in reality, we all know that products from any manufacturer can break down under normal usage and for no apparent reason. On reason for such breakdowns is the failure of simulation techniques–many rooted in the same scientific computing techniques such as matrix inversion and factorization that we are studying in CS 322–to take into account for arbitrary factors of complex systems. Most simulators will produce the same output–or output that varies with a predictable probability–given the same input.

Researchers around the world are trying to create more robust simulations that take into account real life factors in an approach known as Computer Aided Robust Design (AROD) [1]. The idea behind this research is to model factors that can cause unexpected or arbitrary failures in products. One part of this approach is known as the Taguchi method, developed by Genichi Taguchi of Japan [2]. Quality control methods developed by developed for post-manufacture quality control are now used in simulations. The project has a bold approach: Dr. Tanja Clees says that “We are aiming to get as close to the natural manufacturing conditions as possible with our simulations” [1]. For example, variations in the manufacturing process are taken into account and natural variations are considered in the new CAROD simulation system.

How will this simulation software work? This system is in the early stages of its development, but looking at the previous work of the scientists involved may lead to a clue. For example, in [3], iterative methods in sparse linear matrices are applied to the simulation of turbulent flows. In [4], the “classical algebraic multigrid (AGM)” approach is used for industrial and semiconductor processes. In [5], the “partial element equivalent circuits (PEEC)” are used for the simulation of electrical components. Each of these methods are firmly grounded in the basic numerical and scientific computing methods that we are studying in CS 322. It is certain that future simulation techniques and algorithms will also require this background.

Computer simulation [6] is rooted in numerical methods and has already led to many breakthroughs in industry and research. However, there are some pitfalls in such methods, as described in [7, 8]. For example, it is common to have ill-specified problems and simulators that are not validated common, even in the scientific literature. Systems such as the CAROD simulation promise a substantial increase in future product quality at a lower cost to producers and consumers. This shows that investing studying and researching scientific computing can have a long-ranging impact.

[1] http://www.sciencedaily.com/releases/2008/04/080411150948.htm
[2] http://en.wikipedia.org/wiki/Genichi_Taguchi
[3] http://portal.acm.org/citation.cfm?id=1341375&jmp=cit&coll=GUIDE&dl=GUIDE
[4] http://www.scai.fraunhofer.de/fileadmin/download/samg/diss_clees.pdf
[5] http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=4388187
[6] http://en.wikipedia.org/wiki/Computer_simulation
[7] http://en.wikipedia.org/wiki/Computer_simulation#Pitfalls
[8] http://portal.acm.org/citation.cfm?id=304254