Veco Precision: From manual inspection to an automated measurement system

Assistance with data analysis

Like Veco Precision, do you want to make quick strides with your available data and could use some help doing so? If so, please contact us.

More reliable statements about the quality of our parts

Ard Vlooswijk, Lead Engineer Metrology: “That was still mostly human work – in our keura department, inkjet nozzle plates are checked by eye under a microscope. In addition, we were already working with an automated measurement system, a surface inspection system, but that didn’t work one hundred percent. We were getting a lot of data from the system, but could not draw sufficient conclusions from it. This had to change, so that our machine could start providing more reliable statements about the quality of our parts. To achieve that, we engaged Sjoerd de Vries, Data Scientist and Senior Consultant Manufacturing Master Black Belt at Symbol. After all, automated inspection provides us with many benefits that we were not currently taking advantage of.”

The benefits of automated judging

“To begin with, a measurement system has to make the same statement every time,” says Director of Operations Joris Keizers. “That is not the case with human-eye inspection, because some employees see certain errors and others do not. What then is the truth? An automated measuring system gives the same verdict every time, which makes the quality of our delivered products much more consistent. Another goal was to further automate and therefore make the process cheaper. And our final goal was to make the process easier to scale, something that automated judging makes possible. Through a colleague we came into contact with Sjoerd and thus with Symbol, after which we made the choice to roll out an improvement project with him to achieve these results. And with success.”

Percentage of correct judgment up

A full plan has been laid out to use the measurement system to raise the percentage of correct judgment. “We wanted to raise that percentage in two stages. To achieve this, we put together a team consisting of Sjoerd, a Data Scientist and me,” Vlooswijk says. “Together we made sure the plan went through successfully. Sjoerd was on site with us for six months, three days a week, and that’s how we were able to make quick successes. Sjoerd did data analysis and analyzed images and extracted information from them and reported to us. I then hung human judgment against it.

Not too complex, but complex enough

Keizers adds, “Sjoerd had the gift of not making things too complex, but complex enough. For example, the measurement system takes three hundred pictures per product we make. Sjoerd used clever methods to balance what is statistically relevant and what is just ‘nice to know.’ We are all techies and like to dive into all those images and pixels – into the details, that is. Sjoerd taught us not to look at what is possible, but what is needed.” “For example, some problems Sjoerd saw three times, some problems a hundred times, some problems a thousand times,” Vlooswijk adds. “Sjoerd looked at what distinguishes the most common problems in the measurement system from the other cases, and we implemented those parameters into the system.”

Switching to new software

During the improvement process, the software on which the measurement system runs was also addressed. “Sjoerd was using different image analysis software than our system. His software worked with as many as twenty parameters, where our software could handle “only” three to four parameters. An upgrade of the software also allowed us to deploy those additional parameters. In the first phase of the project, we took a step in improving the percentage of correct judgment, using Sjoerd’s software. This turned out to be technically feasible. The next step was once again to increase this percentage, but also to switch to new software on our measurement system. A big job for which I have to give most credit to our Data Scientist and to which, of course, Sjoerd and I also contributed our bricks. The job is done, and the software is up and running,” Vlooswijk says.

Moving along at the speed of life

The ambition of Keizers and Vlooswijk? Further roll out automated judging to other products. Keizers says: “We started together with Sjoerd with a product that we sell a lot, the inkjet nozzle plate. For this we have (almost) achieved all the desired results. Now we can pick up improvement projects for other similar products ourselves. In short, we are ready for the future. Important, because as Veco Precision we make products with features and holes that are getting smaller and smaller. Inspection is simply no longer possible with the human eye alone. What’s more, customers are placing increasingly stringent demands on the dimensions of our products as well as on how the products should otherwise look. By automating inspection, we can move along with those requirements. And in doing so, our organization is moving with the momentum.”

Do you also have a complex problem at the intersection of process, product, data, software, hardware and/or application?

Sjoerd de Vries likes to think with you, he is a data scientist and helps companies solve complex problems. These problems are so complex because they sit at the intersection of different responsibilities, disciplines & expertise where everyone must understand and interpret each other’s “language.” Sjoerd is the connecting factor in this.

Some projects he has been working on recently:

1. Optimizing multiple Vision systems to fully implement automated detection.
2. Performed online/inline MSAs (measurement system analyses) where the reliability of the analyzers and the measurement itself was determined, then the variation of the measurement and the process was reduced.
3. Achieving reliable reports/data visualizations that include key parameters (CTQs and CTCs). These reports are easily interpreted by the shop floor and lead to a more stable process. Data reduction using descriptor-based modeling has served as the basis for this to get from as much data as possible to right data.
4. Machine builder was at the end of the R&D phase and had to deliver to the customer, however, they did not get beyond 95% of the desired accuracy. Using process mining, TRIZ, data analysis and rheology, the accuracy was brought to the desired level and the machine could be released for industrialization.
5. A customer had a lot of variation in the process due to problems with the raw material, this raw material could not be measured accurately enough by both the customer and supplier. Chemometry was used to reveal that there was contamination in the polyols. A continuous monitoring application was built to detect it and ensure its stability.