Parts time solution

A google-style software tool that searches for 3D shapes rather than words could soon help reduce the design process by months. Richard Fisher reports.

Like lost socks in a washing machine, CAD-designed parts often disappear into the ether after they are entered into a company library. As staff change, file names are inevitably forgotten.

So while the database may grow, the knowledge is buried. Consequently the part you want is never found and has to be re-designed from scratch.

To help overcome this, engineers at Purdue University, Indiana, have developed a program that searches for the 3D shape of old parts, making the hunt for useful design information a lot simpler.

Prof Karthik Ramani, who heads Purdue’s Research and Education Centre for Information Systems in Engineering (PRECISE), believes that the ability to access such information would present engineers with an invaluable tool.

‘Let’s say there are 1.3 million parts in your inventory. If you are designing a component and you can find something similar that has already been produced, that is half the battle,’ he said. ‘How was the part designed? Have parts like these had problems? What decisions have been made about them? The only way to find out is to use a shape-and-text search.’

Engineers can spend up to six weeks a year looking for old parts information, said Ramani. Purdue’s program could cut this by 80 per cent.

Users can pick from an inventory of designs or sketch a part from memory, and like an internet search engine, receive a cluster of similarly shaped results. They then refine their search to get the closest match. The system also allows the user to specify factors such as how a part was created.

On installation the program self-indexes itself by reading the 3D CAD files and breaking each part into small cubes called voxels. It then uses complex software algorithms to convert the voxels into a retrievable, skeletal graph.

‘We combine the geometry of the part with its topology – for example, how the head is connected to the neck. The part is reduced to a hierarchical representation,’ said Ramani.

However, the key to getting the program to work was making the search multi-levelled, starting with broad criteria, then honing them. The search also uses algorithms that mimic the human brain. The team worked with psychologists to perfect this.

‘The search takes place through multiple representations of the part in a multi-step process,’ explained Ramani. ‘The user says “my part looks more like this” and the computer adjusts the search. This relevance feedback lets the system understand more about what you are looking for and it starts focusing on that,’ he said.

the software, still in prototype form, is not without its limitations. Accuracy was 85 per cent in preliminary experiments, but this figure is based on a small database of hypothetical parts and it is not known whether early versions could cope with the massive diversity of CAD designs across industries.

But while design engineers are unlikely to see universally applicable software in the near future, software company Imaginestics is developing an industry-specific version of the program.

Indeed, president Nainesh Rathod said that industry-specific software for the automotive, aerospace and medical equipment sectors could be available by the end of the year. ‘We’re hoping to develop the software either as a standalone or embedded as an option within one of our existing products. It would integrate very easily with the platforms we already have,’ he said.

meanwhile, Purdue’s software designers hope to put a prototype program demonstrator on the internet towards the end of the summer.

So the CAD designer can expect to get a first glimpse of the Google of design engineering soon. Unfortunately a search program for lost socks is still a long way off.