Issue link: https://iconnect007.uberflip.com/i/1424540
80 SMT007 MAGAZINE I NOVEMBER 2021 it to a common use, we will create one. What helps is to think about your data, your need to be agnostic to the format. You are right that aer data translation, there is a big difference between data, insights, and actions. We've de- veloped quite the domain expertise in what works, and what doesn't work. What does it mean when this result happens? We put to- gether data on its platform and then add in, algorithmically, our expertise. For example, you know if you get a stream of tested boards going by and you know these are all bed-of-nails contact points. Your audi- ence is probably familiar with that, and one of the things, especially with printed circuit boards bed-of-nails contact points, is they get dirty in the bed of nails and the probes wear out. en your contacts can start to fail, and that creates a measurement failure. It's very common. When people start to see a mea- surement fail consistently, one of the first things they do is to ask if it's the probing. Do I need to replace or clean the probes now? e problem occurs if you wait until it fails; then your line is down. Now imagine if, through analytics, you could see through the data that the measurements were starting to come off what they normally would be. You need to be careful and note the difference between passing and failing, but if you notice the passing cases are slowly getting worse and you combine that with domain ex- pertise, you might see an indication that may- be a probe is getting dirty or will probably start to fail. You could send a predictive alert: "You might want to take a look at this probe at the next shi, change, clean, or fix it at the next opportunity where the line doesn't have to go down." Now the value of giving people that proactive information is huge because it results in zero downtime. It also starts with good, solid measurement data. It's terrible to predict the future with noisy data. Johnson: at's part of what you're driving to- ward, to look at the data and not only figure out what is what, but if it tests pass and fail data. What is product specific data and what is ma- chine maintenance data? You could be looking at data and realize you have a hotspot of fail- ures in one corner of the board. Is that manu- facturing or is that measurement? Cain: Or is it a batch of parts? It could be a lot of things. Is it process related? Is it correlated to a specific set of operators? Yes, and so it's very helpful to not just collect the data on one is- land of information, but correlate it to multi- ple sources, like who was the operator running the tester, which fixture was on the tester, and which tester it was. How does that compare to others testing the same product at near- ly the same time? ose are all very interest- ing points to correlate the data; that gives you some insight into the root cause, so you take action and do better. Johnson: Do you see that analysis as some- thing that your products should be delivering or should this kind of analysis be considered higher level? Does the analysis go to somebody else who is specializing in the management and process control for the line? Where do you see the industry evolving to? Cain: It's at a fun state where there are many options; to be honest, a lot of people are quite rightly doing it themselves because the tools and capabilities are not that hard to obtain. And that's not a bad way to do it, but there are pock- ets where you can purchase a premade solution, We've developed quite the domain expertise in what works, and what doesn't work.