Automated Vision and Sensing: Truth or Consequences
June 9, 2011
Simple, reliable vision and sensing systems depend on identifying actual performance requirements up front. Both technical and practical considerations are critical to on-line sensing for factors such as defect detection, gauging and characterization. The earlier the truth about actual performance requirements are identified, the earlier there will be preferred consequences: simplifications and a reliable system.
I was reminded of this recently when I was contacted by the marketing arm of a large corporation about developing an optical inspection system for its extensive carpet product line. The company had selected my firm because of our reputation for solving challenging problems that had resisted internal and external solutions. This interaction provided an example of where lack of experience in practical solutions could easily result in unreliable systems or eliminate solutions entirely.
I was advised that the labor costs for manual inspection would be about $900,000 to $1 Million per year, not including expenses such as capital costs and benefits. Certainly a project well-worth considering. The company had looked at and rejected several commercially available, off-the-shelf systems from other firms.
However, this was the first time in decades that a potential client had detailed the potential savings of a project. I swallowed and swore to not take advantage of this. We could almost certainly provide a system, based on costs and engineering, that would cost considerably less than a more avaricious supplier with this information. It was also a warning to us on the company’s inexperience in this area.
I initially obtained a partial list of requirements by telephone, and requested samples. I always request samples! After some delay and multiple requests for samples, unqualified defects on six different patterned samples arrived.
They did not provide basic technical parameters, such as samples/minute to be inspected, total number of colored patterns of interest and samples with smallest defects to be detected.
Aside from the usual general parameters of the system it was clear that there were different models of products (of the same size) with different color patterns, and with different defects.
In order to further specify, better focus the efforts, and enable simplifications, I asked for the total number of patterns and the total number of defects.
How many patterns should the optical sensing system be required to inspect- five or two hundred?
Are the patterns mixed, or are single models produced on one line in large volume?
Also, what were the frequencies of different defects: did a few defects provide 80% of the rejects?
On a re-inspection of a trial batch basis, what percent of defects were not detected during the first manual inspection?
I offered to visit the company’s plant site, so that possible simplifications and local conditions could be assessed.
Again, it is often the case that although defect detection feasibility can be demonstrated, local conditions can interfere with defect signals, and destroy reliability unless they are taken into account.
After several more delays, I was advised that the company had decided to continue manual inspection with no further interest in development of an automated optical inspection system. With a probable return on investment within three months; with one-million-dollars per year at stake; with no up-front consulting cost; and without talking to me in person, the company was withdrawing. Amazing!
Technical and Practical ConsiderationsWe have discussed technical considerations in several previous blog posts. Techniques that are appropriate to different shapes, colors, textures, materials, etc. have been discussed in previous blog-posts here.
Examples of Local, Practical ConsiderationsIn what follows, we shall provide several examples of where practical considerations were not revealed early by companies, and led to necessary additional work. Earlier development could have been avoided.
In one instance that we designed around, interference from the sunlight through the roof and windows of the factory was neutralized by the use of lasers combined with narrow-band-width optical filters at the photosensors. The environmental lighting condition had made some systems unreliable. This is not an uncommon issue and not an uncommon solution.
However, the use of a narrow parallel laser beam from the curved surface of the inspected product results in a smaller required response time for reliable detection. The response time of many standard industrial photosensors/circuits (about 10 milliseconds) was too long for the actual response needed which was less than 1 millisecond. The electronic response time had made some systems unreliable.]
In another instance, the goal was to gauge the size of bifocal inserts in eyeglass lenses during the process that grinds the outer curvature of the total lens combination. There was a liquid lubricant, with which I was supplied, that was used during this grinding process.
I provided a demonstration of feasibility of this gauging, using the relatively transparent lubricant--that I had requested-to simulate the actual conditions. I then designed and built a system that would automatically detect the presence of lenses. This computer-controlled system would automatically gauge the dimensions of the bifocal insert. The system worked like a charm, taking into account mechanical and spray interferences during the gauging with the moving grinding head.
However, just before I delivered and installed the system, the company sent me a new jar of the lubricant.
This was the liquid after it had been used and reused, so that it was now opaque with small glass chips made it look like white paint, which was similar to polar bear skin looking white despite transparent, hollow hair covering.
The system was unusable with this white paint. Serious redevelopment was required.
In our last example, it was required to measure the diameters of metal tubes moving rapidly after annealing in a furnace. The tubes were glowing between white and red hot. As an initial stage, I prepared a design that distinguished between the sensing radiation and the emitted radiation, and had other adaptive features for speed, diameter statistics, rejection signals, files and displays. I made a presentation to the folks at this plant.
I made certain that I had lunch with an operating engineer at the plant before the presentation. He told me that the only person at the meeting who knew the situation was the plant manager (PM). The PM could be identified because he was the only one in the room who did not wear a tie or formal jacket. The others were the “suits”. In fact, I later found out that there had been a previous system that had failed and that I had not been told about.
After my presentation, I approached the PM and asked him about the problem. He tried to head for the door, but I put myself between him and door. He smiled and told me that the system that they had tried out had melted. Melted!
To calibrate the system, the company had moved the hot tube within 6 inches of the sensing system and it melted. The specifications for our design indicated that the tubes would be five to six feet away from the unit, with no heat issues.