Sep 27 2005
Researchers at Purdue University who developed the first system capable of searching a company's catalog of three-dimensional parts created with computer-aided design software are now providing a method to evaluate how well such systems work.
Shape-search engines could save time and millions of dollars annually by making it easier for companies to "reuse" previous designs, reducing redundancy and streamlining a company's supply chain. The systems will enable companies to benefit from the lessons learned in creating past parts, said Karthik Ramani, a professor of mechanical engineering and director of the Purdue Research and Education Center for Information Systems in Engineering.
The Purdue mechanical engineers have created a "benchmarking database and process" that enables engineers to evaluate how well their search system is able to retrieve matches, Ramani said.
Findings were detailed in two research papers presented during the 25th Computers and Information in Engineering Conference held by the American Society of Mechanical Engineers in Long Beach, Calif. The papers were written by Ramani, mechanical engineering doctoral students Natraj Iyer and Jayanti Subramaniam, and postdoctoral research associate Jiantao Pu.
The Purdue 3-D shape-search method also has been selected for a special "showcase and reception" from 5-7 p.m. (PDT) on Tuesday (Sept. 27) at the Hyatt Regency hotel. The method, developed by Pu and Ramani, has been commercialized by Imaginestics LLC, a company in the Purdue Research Park. Ramani is chief scientist for the company, which has developed products based on the research.
Computer-aided design, commonly referred to as CAD, was introduced a few decades ago for such daunting projects as creating better ship hulls and airplane wings. The systems are now routinely used to design everything from automotive water pumps to industrial machine parts.
"One of the great disappointments of CAD has been the difficulty of reusing data," Ramani said. "Once CAD information has been created and used, it is often stored and forgotten. As a result, industry loses a lot of money by not being able to reuse earlier parts. The proverbial wheel is reinvented many times."
Parts designers spend about 60 percent of their time searching for the right information, which is one of the most frustrating tasks for engineers, Ramani said.
"The whole power of computers is lost if you are not able to retrieve and 'reuse' what you have created in the past," he said.
The Purdue shape-search system enables people to select an inventoried part that resembles a desired part and retrieve similar items. Users also can sketch the desired part entirely from memory, or they can choose a part that looks similar from the company's catalog and then sketch modifications to that part. The system then assists in finding the desired part.
The Purdue benchmarking system uses an inventory of 1,000 parts and evaluates how well a search system is able to retrieve matches to a part entered into a query. The parts are grouped in 40 categories, such as ringlike parts, T-shaped parts, cylindrical parts and disk-shaped parts.
"If I give a query for a part that's in one of the categories, the top 10 results should ideally be in that category and as close to the queried part as possible," Ramani said. "If the search system found only six matches from the right category and four from some other category, then I know it's not that good."
The Purdue researchers also have created a method for automatically orienting a part as humans would view it - in its "most stable orientation," meaning a position in which the part would not fall over.
"If I placed a part on the top of a table and it fell over, it would be in an unstable position," Ramani said. "If I placed the same part on a table and it remained in place, that would be its most stable position, which is how humans imagine a part and how they draw the part. We came up with a method for very stable pose determination, then we index the part in the database in that pose. Then, for that particular stable position, we project it in various views, such as a side view, a front view and a top view, so that you can really see what the part looks like, which is important for sketch-based queries."
The benchmarking database and parts inventory are available online at http://engineering.purdue.edu/precise. A demonstration software prototype called "Shapelab" also is available online at http://www.purdue.edu/shapelab.
A critical element that makes the most stable pose determination and searchable database possible is using simplified versions of CAD parts. Those simplified versions, called "faceted models" because they are made up of a series of triangular segments, require less computing power than would be needed to process more complex CAD objects. The Purdue researchers created a new method for representing a part, converting it from a flat, two-dimensional drawing into a "two and a half dimension" representation. The 2.5-D method adds individual detail to the drawings and faceted CAD models and represents them in a way that is searchable.
The Purdue methods allow the user to fine-tune the search by changing the sketch.
"The search is a multi-step process, which is very important," Ramani said. "You repeatedly narrow down the characteristics of the part you are looking for to bridge the gap between what's in your head - your idea of what the part looks like - and what's in this huge inventory of parts. This is not a simple, single-step approach that others have tried."
The Purdue researchers used their benchmarking process to test about a dozen shape-search methods, including two from Purdue and the remainder from other universities.
"Our methods consistently performed with 20 to 30 percent higher precision than other methods," Ramani said. "In addition, the Purdue method has the unique capability of being interactive and closer to human perception. They have made search an interactive process rather than a one-shot query."
The Purdue researchers plan to eventually provide their benchmarking system online free of charge.
http://www.purdue.edu/