December 2008 - Issue #4
Defects detected in the blink of a ‘mechanical eye’
Story by Rebecca Thyer
View articles in related topics: Advanced Manufacturing, Advanced Technology, Engineering
By running rabbits through its automotive production line, Ford Australia has been testing the performance of the next generation of inspection systems in a bid to improve quality and, through that, global competitiveness.
However, it is not rabbits of the feral kind that are running loose in the company’s Campbellfield, Victoria, production plant. ‘Rabbit’ is industry jargon for fault.
So, to gauge how well vision-based, non-contact inspection systems work in detecting faults, Ford, which is working with Swinburne University of Technology’s Faculty of Engineering and Industrial Sciences via an Australian Research Council (ARC) funded project, deliberately released some ‘rabbits’ into cars on the production line.
Ford’s final assembly engineering manager Andreà Cavallaro says the rabbits included hard-to-see faults such as missing or unscrewed bolts. “Although results are still being analysed, it showed that these systems can pick up on some, but not all, faults. And that’s the whole purpose of this project – to understand the limitations of the current generation of computer-vision-based inspection systems.”
Essentially, non-contact inspection systems do what they say. Using techniques such as vision, ultrasonics, light-scattering and capacitive sensing, they inspect components on a production line in real time. The use of non-contact inspection systems is growing rapidly in the global automotive, aerospace and agricultural industries, and it is an area that Swinburne’s Non-Contact Inspection Research Group has been involved with since 1991.
Group leader Professor Romesh Nagarajah, whose team includes Tim Barry, Michelle Dunn and Associate Professor Ali Bab-Hadiashar, says that although these systems are applicable in many environments, the variations encountered in manufacturing can limit their use.
The objective of this ARC project is to extend the reliable application of non-contact inspection systems by development and testing for robustness and accuracy in a ‘real-life’ manufacturing environment.
“In other words, the system must be able to accommodate the differences encountered in a manufacturing environment, be that from vibrations, dust or oil, light changes throughout the day, or from natural component variations,” Professor Nagarajah says. “That is why we tend to carry out most of our work on the actual shop floor.”
He says for Australian manufacturers, this type of inspection system must also be flexible enough to cope with small production runs and complex parts.
Despite these demanding requirements – and in a manufacturing environment where Australian companies face determined competition from overseas rivals with lower labour costs – these systems are being looked to with increasing interest. Their use has the potential to reduce labour costs and improve quality – two essential ingredients in remaining internationally competitive.
Mr Cavallaro says that for Ford, improving quality is the ongoing priority. “For us, this project and its outcomes are all about quality: giving the customer what they have paid for, and making sure that the car functions the way it was designed to. Non-contact inspection systems will help us continue to do that and improve on that.
“Ultimately it should improve our competitiveness because any improvement in quality improves the brand,” he says.
Ford already uses various types of vision systems for inspections at its plants in other countries. “At some of the more affluent, high-volume plants Ford has vision systems on the end of robots, but it’s a very expensive set-up,” Mr Cavallaro says.
In Australia the company uses simple vision systems for in-process, but not final, inspections. “Our aim is to use vision systems for final quality inspections to confirm we have built the car to design.”
Professor Nagarajah says that, typically, a vision system comprises cameras, lighting systems and software for image processing, analysis and decision making – that is, deciding whether a component is ‘good’ or ‘bad’.
“In this ARC-funded project, we have used off-the-shelf cameras and lighting systems configured by our research team. All software development was also done by our team, partly using commercial software platforms. We’ve developed a prototype non-contact inspection system that uses vision.”
Using this set-up, Swinburne researchers are exploring the system’s limits by inspecting nine under-body car components, representing varying degrees of complexity.
Professor Nagarajah says the inspection system must inspect, in real time, under-body components to see if they are correct and in the right place. It also must be quick: “If you are going to install these types of systems on a production line, they have to achieve the speeds currently achieved on the line … or faster.”
He says that in most cases automated systems replace humans, which brings positives and negatives. “Human inspectors are very experienced and skillful, but can get tired and therefore miss certain defects. An automated system can avoid this pitfall but, on the other hand, eyes and brains process information very quickly. So, to be viable, an inspection system needs to balance issues of accuracy and speed.”
Professor Nagarajah says inspection systems also need to be flexible enough to cope with different vehicle models and part variations. “This can be further complicated because slight variations in part shape or location do not necessarily signify a defect.”
For example, hoses may not always be the same shape. “They might be turned in or out, but still be OK. Any system has to be able to recognise that. It underlines the difficulty these systems have in assessing what is ‘good’ or ‘bad’.”
By collecting and then analysing numerous images of the nine components under various environmental conditions, Swinburne has developed methods and associated techniques to accommodate these variations. The approach also uses artificial intelligence techniques to help the system make a decision.
For Ford, any system must also be easy to use, Mr Cavallaro says. “If we need a rocket scientist to tweak it or to keep it robust and capable, then it obviously becomes less attractive to implement.”
He says the research is finding good applications for vision systems and it will be just a matter of “how and when” a non-contact inspection system is introduced.
Professor Nagarajah says that once outcomes from the ARC-funded project are implemented the research team will be able to more accurately gauge the capabilities of current systems and associated technologies.
Automation keeps company competitive
For safety equipment manufacturer Autoliv Australia, the use of automated, non-contact inspection systems has allowed the business to remain competitive.
Autoliv produces automotive safety equipment, including seatbelts and airbags, and has 13 technical centres across the globe. With Swinburne’s non-contact inspection group and with Australian Research Council (ARC) support, Autoliv Australia has just completed a project to manufacture complex assemblies with no defects using non-contact inspection systems.
Managing director Seamus Power says the company looked to auto-detection for two reasons – quality and cost. “Quality, for us, is a given. Externally that quality has never changed, but that obviously has a cost if we are rejecting a lot of products internally. So we wanted to build in a non-contact inspection method that became part of our existing production line, thereby keeping labour costs down.”
Mr Power says the company aims to run manufacturing by adhering to the Japanese principle of jidoka. “It means trying to find the error before you make it … instead of inspecting a product later, you inspect in real time. What that means is you may only make one reject or, in many cases, it will stop you from making a reject. You don’t waste time and money making rejects and you don’t have to sort through rejects if you find the problem straight away. It’s cheaper to make things right the first time.”
Autoliv’s work with Swinburne researchers has helped it reduce internal first-time failures by 53 per cent. “The project paid itself back within three months,” Mr Power says. “If we could get that every time, we’d be laughing.”
