Transcript
John: Good morning and welcome to the Business of Machining episode number 462. My name is John Saunders.
johngrimsmo: And my name is John Grimsmo.
John: Projects, projects, projects. like it should
johngrimsmo: Never ending stream of projects.
John: But it's all, I would, i would ah there's some fable, I would just, I will die when the projects are over. That's all I do.
johngrimsmo: yeah Yeah, exactly.
John: That's all I do.
johngrimsmo: Do you find yourself starting more projects than you're finishing? Because I do.
John: ah So, yes. but I am much better about being more disciplined these days, including for every reason of just like being, being realistic in the focus and also not spending money, um, or buying stuff.
johngrimsmo: I am too.
John: Like there's no, i think we talked about this. There's no correlation to be buying something and it getting done. Like if it's sitting on my desk or on the side of my office, like, Nope, it could sit there for two years. So, um,
johngrimsmo: It's still a preparatory purchase though.
John: It is a perpetrator purchase and those are fully supported, although they do consume a lot of clutter credits.
johngrimsmo: Yep.
johngrimsmo: Yeah, exactly. um I mean, it's almost like buying the thing is the easy part. And it's kind of fun to do all the research and to get Oh, I need that I need that and I need that and I need that.
johngrimsmo: And then you get all the things and they sit on your desk for a while. But part of you goes at least like that part of the project is ready. And whenever I'm ready to dive into it, I have preparatory purchase. And now I'm ready to go. But you're right, it doesn't like having it sit there doesn't like mean it's going to get done.
John: So you just said something that maybe occurred to me that again, and if anybody is listening and doesn't want to hear me continue to be excited about the basics of LLMs, like probably no longer the podcast for you.
John: Cause like I just remain blown away.
johngrimsmo: Same.
John: um First off, there's a thing i mentioned to you six months ago that I have continued to really find is helpful, although it's not necessarily intuitive or not necessarily what I always remember to do, instinctual, I guess, which is when you do have to step away from your project, leave yourself a really easy kicking off point again.
John: like Because when you're doing it, you and I are the same i think think the same way.
johngrimsmo: Right.
John: like We're like, oh, no, I could fit three more things in before I have to leave or pick up the kids or whatever.
johngrimsmo: Yep.
John: And it's like, nope. actually rewind and just leave yourself the easiest starting off point.
johngrimsmo: Yeah, I forget about that point, which is really good because it's so counterintuitive. Like you you do the easy things because oh I can squeeze in one more easy thing before I go. And then I leave the hard thing left because I don't have time to do the hard thing next.
johngrimsmo: So you're almost saying like cut bait earlier than you could by kicking off those last few things.
John: Yeah.
John: Mm-hmm.
johngrimsmo: But purposely, like there might be a level of pride with that too as you do it because you're like, no, no, I'm going to leave myself a a little treat for when I come back to this project, you know?
John: Yeah.
John: And, you know, it's kind of like the hard part of going to the gym is going to the gym. The hard part is not starting your second set of whatever.
johngrimsmo: Yeah. Yeah.
John: um So, you know, knowing that, okay, all I have to do in the next step is to like plug that wire back into five volts and then turn the pie back on and then SSH into it. Like this is what I've been doing the past few days, a lot of, I forget, I lost my total frame of thought, but yeah.
John: Are you...
John: Well, okay, the last thing I'll say is also the projects I'm doing right now are also very germane to Saunders Machine Works and the team and the automation and the phasing. So they're different than like the general like, oh, I remember my point now. They're different they're different than some the just like things I enjoy or personally or work related but aren't necessarily correlated to like, no, this is like, so it's a lot of the stuff on the which I'll talk about today about the the brother of the automation and metrology. But then the other point, it was like the major point, I guess, was that like,
John: A year ago or three years ago, that project sat at my desk because I didn't necessarily know how to tackle it. And it was going to be a, I knew i was going to be frustrated or have some questions.
johngrimsmo: Yeah.
John: And now you just ask Claude.
johngrimsmo: Yeah.
John: It's like it continues.
johngrimsmo: And it's just done.
John: Well, you're just not alone.
johngrimsmo: And if you're not happy with it,
John: You're not alone. That's the beauty of it.
johngrimsmo: Yeah, no that's what it feels like, right?
John: Yeah. Yeah.
johngrimsmo: And, and You don't need to know how to do it. You just need to know exactly what you want. And if the result is not what you want, you just ask again, ask a better question. And then the problem gets solved.
John: Yeah.
johngrimsmo: Yeah, I continue to add more and more incredible features to GURP. And I'm trying to make sure that they're actually important and not just like cool, because I'm spending a lot of time on it.
John: Yeah.
John: Yeah.
johngrimsmo: So that goes into the whole, like, make sure the projects you're starting are worth starting. um just because you you Just because you have an idea doesn't mean it's a good idea and doesn't mean it's worth your time and doesn't mean it's like going to help.
John: Okay.
John: Mm-hmm.
johngrimsmo: um So I'm really trying to be cautious of that, of making sure that my time is, you know, like I can have fun, but make sure it's like wisely spent. um But yeah, it's crazy.
John: I could see so much ster being relevant though. Are you worried about going overboard? Or...
johngrimsmo: um For sure. I'm worried about putting in, you know, spending dozens of hours on something that nobody's going to use. It doesn't really matter.
John: Got
johngrimsmo: It's just like, it's you know, it's icing on the cake kind of thing. But there there is also that, like, as we've seen with our own projects and with a lot of our friends and and colleagues, they're like, it's very easy to make an 80% project now and not eat not finish it well.
John: Got it. Yeah.
John: Yeah.
johngrimsmo: You know, whereas the companies that are trying to sell software, they need a 95% product that... product that won't piss people off all the time. um and And that it's very easy with LLMs now to get to that 80%, like very easy.
johngrimsmo: But the 80 to 95% of completion of goodness, of of spin polish, you know like it oh it really works. It really does what I want. And it's so intuitive and it's so easy and and it never fails. Like that takes a lot of time. And I like it. I like living in there.
John: Are you being specific about what LLMs you use or even what models within LLMs you're picking?
johngrimsmo: I don't. I don't really care. um
John: Yeah.
johngrimsmo: ah use Cursor for everything, ChatGPT for like brainstorming and and research, and then Cursor for basically executing code.
John: Mm-hmm. even
johngrimsmo: um And that's kind of my workflow.
John: Okay.
johngrimsmo: And I still haven't tried Claude.
John: Oh yeah. I might have to drive up to Toronto and punch you.
johngrimsmo: Is it any different than Cursor? Like, i don't know.
John: Yeah. I don't know what cursor is. I think cursor is just an amalgamation of other models. And I got, I mean, doesn't like, okay.
johngrimsmo: It is, yeah. A lot of the cloud models are like in there, like Opus and whatever else, ah Grok.
John: Okay, yeah.
johngrimsmo: and it just auto like I'm on the auto session where it just picks whatever model like makes sense for my request.
John: Okay.
johngrimsmo: i don't really care. And then in cursor, there's now four sections.
John: Yeah.
johngrimsmo: You can go to the ask section, the plan section, the agent section, and the whatever. so If you don't want it to change your code, you go to the plan session or the ask or whatever.
John: Okay.
John: Okay.
johngrimsmo: And then the planning session is sweet because it can just it outlines this whole like document that's like, okay, we're going step one, we're going to step two, we're going do step three.
John: Okay.
johngrimsmo: And then i can review that. And it's like for brainstorming, okay, I want to add this whole new section to GURP. let's Let's plan it. And I skim through that and be like, yep, yep, no, not step two, not step three. Okay, yeah, go.
johngrimsmo: And then it just goes.
John: Okay.
johngrimsmo: Whereas if you're in the agent mode and you ask a question, it will change your code.
John: Well, to eat my own, uh, whatever words here, I should check out cursor. Cause I really haven't. And I guess to your kind of point, who, who cares?
johngrimsmo: Kinda, yeah.
John: Well, except I just remember being in love with chat using Gemini. And then I asked Claude to do a few things and it, the way it did them was so much more eloquent, so much more helpful that you don't know what you don't know.
John: You don't know what you're missing.
johngrimsmo: Exactly.
John: Um, but again,
johngrimsmo: It's the same when when i went when I went from chatGPT to cursor for code for like doing stuff.
John: Yeah, right.
johngrimsmo: I'm like, oh my gosh, this is a night and day difference.
John: Yeah.
johngrimsmo: So yeah, you don't know.
John: ah The... Go ahead.
johngrimsmo: That's it.
John: When I was ah about to leave on Friday, Alex had a laugh. laugh He's like, yep, so... um Actually, I don't even know. i think he used Claude, but it doesn't matter. He's like, so it wiped all of Lex, I think.
John: Wiped all of something. And he's like, I wish Alex could sort of share exactly what happened specifically, but basically total massive violation overstep.
John: We were fine to Alex's credit. He's done the right way of sandboxing and forking and backing up. So it was like no problem to get back up in like five minutes.
johngrimsmo: Yeah.
John: But when he asked about it, it was just like Claude responded like, Oh yeah, that was a huge overreach on my part.
johngrimsmo: yeah
John: i never should have done that. I'm really sorry. Would you meet like help like fixing it, which is just kind of like, it's kind of like me shooting you and being like, I have a tourniquet. Would you like me to show you how I apply a tourniquet?
John: So I'm like, what? um
johngrimsmo: With no remarks, no nothing.
John: So, yeah, yeah.
johngrimsmo: Like it's, yeah, it's dangerous. Interesting. That's a good, good warning, you know, make your backups.
John: Yeah.
johngrimsmo: Don't give it too much. try It's so easy to trust it nowadays though. Oh my gosh.
John: Yeah, it's scary.
johngrimsmo: Yeah.
John: Agree. Yeah. But it's also like I'm trying to work with some electronic stuff that's outside of my wheelhouse. And I screenshotted a line level shift voltage board, blah, blah, blah.
johngrimsmo: Totally. Totally.
John: And I was just, I was like, I was sure it was the right one. It's like $8 and screenshot it just to make sure it's like, Nope, that's actually the wrong one for this reason. What's up? Like refresh rate or something. It would cause lag issues to be fair.
John: I guess I never even interrogated that to to confirm what it's saying and there was true, but it had me pick a different $8 board.
johngrimsmo: Sure.
John: And like, those are the sort of things that like you or I could have spent, um,
johngrimsmo: Right.
John: We could have stopped a project forever on that.
johngrimsmo: Money and time and yep.
johngrimsmo: Yeah, because those are the things you you don't know what you don't know. And you're probably like me, like I don't need to be an electronics engineer. i don't need to be a software engineer. I don't need to know everything about this stuff. I just want to solve problems.
johngrimsmo: I just want to like do cool stuff.
John: Yeah.
johngrimsmo: I just want to run fast and break things.
John: Yep.
johngrimsmo: And I can do that now. I feel totally, i was thinking last night, I'm like, I feel unrestricted with this kind of access.
John: Yeah.
johngrimsmo: I can do whatever I want, whenever I want. And sure, not to like ah an athletic level of you know software engineering. I don't really care. The fact that I can pull it off is like so cool.
johngrimsmo: And I've always tried to my whole career.
John: Yeah.
johngrimsmo: i mean, I was soldering, my I placed my first DigiKey order in 2002.
John: Mm-hmm.
johngrimsmo: You know, I was like soldering the engine management system for my car when I was 19.
John: Oh, yeah. With all the...
johngrimsmo: Right.
John: Yes.
johngrimsmo: And and I like flash the EEPROM, whatever. And like this was early two thousand And I could get so far with reading on the internet and then I would be just stuck.
johngrimsmo: I'm like, I can't, I don't know how to do any more than this.
John: Mm-hmm.
johngrimsmo: And I don't want to like deep dive that much and learn it all. And then Raspberry Pis come around and I can like, I can force my way through it the past eight years. And then now it's just like a joke. It's just like so easy. Yeah.
John: I'm so excited to see like what comes next about even like a lot of what I'm doing now is putting together, build materials and projects. And it's like, there's, we're going to look back and think about how it is still somewhat stressful or anxiety inducing to figure out, okay, what's it recommending? Where do I buy it from? Is it the right variant? And so there's some,
John: future that you could see that integrates that better. That probably unfortunately, for better worse, ties in some rev share AdSense type model of how LLM's recommend products, which is really unfortunate if it starts skewing the results.
John: But nevertheless, like a guy like me and you, like, just tell me what board to buy.
johngrimsmo: Interesting.
johngrimsmo: Yeah, I kind of don't care. Just like and help me not make mistakes here.
John: Yeah.
johngrimsmo: and
John: Mm-hmm.
johngrimsmo: And then I don't know if you find as much as I do, because I'm such a tweaker. I love to like like fine tune things. um Once you get kind of the working first version of whatever kind of, whether it's a Raspberry Pi thing or whether it's a featured group or whatever, how much do you massage it and fine tune it after that?
johngrimsmo: Because I do a lot.
John: Yeah, I don't think as much as you.
johngrimsmo: Yeah.
John: I'm more like minimal viable product, get it to do, move on, or maybe hand it off. Exceptions being like where it matters. But no, I think I'm probably much more, i mean, that's who you are for sure as a person, a product company.
johngrimsmo: Sure.
johngrimsmo: Yeah.
John: you will go to those nth degree insanity details. so It's like makes Grimms and Knives awesome, right? Like on so many levels.
johngrimsmo: Yeah, that's true.
John: Okay, well, so to keep it more machining specific to talk a little about what i'm doing, First, i have four projects going on. Number one is a human presence detection. Number two is, um, linear, linear gauge use. Number three is a DIY vision system.
John: Um, which again, totally transparent, ironically is basically going to replicate a lot what you and others, I think like Cyrus have done. I think I shared, I shared that story publicly last week. I don't recall the Keyence visit.
johngrimsmo: No, not.
John: Okay. I'll briefly share on that. And then, um, the raspberry pie, photo archive of products that come the machine, which is pretty simple. Yeah.
johngrimsmo: Nice.
John: The first one is really cool. It is product that I didn't know existed. It was recommended by whatever, some LLM. And it is called a millimeter wave radar detector.
John: In layman's terms, and was I was familiar with like the PIR, time of flight, like some of the other things you'd see on either Fruit or Spark Fund that we just sort of see motion detection or person detection. This again, way misperms, actually uses some wave, millimeter wave, whatever, to detect if the object that it sees in front of it effectively has a pulse.
johngrimsmo: Hmm.
John: So if you move a box in front of it, I will tell you specifically if anyone wants to kind of play around with themselves, it's an $8 chip called the LD2410C.
johngrimsmo: What's it called?
John: two four one zero c
johngrimsmo: because when I was at Hacksmith, he was showing me a product like this, but it was ah it was a brand name. It was like a kit you buy or whatever that could do that.
johngrimsmo: And it was like crazy resolution.
John: Yeah, this is, I bought mine on Amazon, 10 bucks, human presence radar, millimeter wave radar sensor.
johngrimsmo: Cool.
John: So what basically it detects if the object near it has also micro movements that would resemble like the human pulse or fluctuations.
johngrimsmo: Hmm.
John: And I was able to tune the frequency of the c sensor itself to say, I only want to detect if a person is within about one to one and half meters of the object.
John: And then that passes the signal over to a Python script. And on the Python script, I then do, okay, I want that object to, I want the person to be there for five seconds. So basically I want to know if a person walks up to a CNC machine, not a robot moving or not some shadow that's casting.
John: If a human and being walks up to the machine and stays there for three seconds, then I know a human is present.
johngrimsmo: Mm-hmm.
John: And this is doing it now.
johngrimsmo: That's awesome. And then what will you do with that output?
John: You're right.
John: So we we it was a whole back research for just like the idea of like, I want to know if a person is there, which could be things like turning the lights on or presenting information or screens on. And it becomes a little bit more difficult because you have a UR-10 robot in front of the machine that's always moving.
John: So you can't use the traditional motion sensors.
johngrimsmo: But still, what are you going to do with the output? I guess you said a couple things, but...
John: Yeah, there's more I'll share later. But like, yeah, a simple answer could be like the brothers really, those Japanese machines are really aggressive about turning the lights off.
johngrimsmo: Okay, yeah, yeah. Yeah.
John: And so at a simple point, you could just have it update cameras, update lights, turn on iPad, ERP, stuff like that.
johngrimsmo: Yeah. Yeah, exactly. Even like an interesting stat would be hands-on time. Like how much operator time versus runtime, right?
John: oh Oh, yeah, that's a good point. Sure, how much in front of, yeah, right.
johngrimsmo: You know? Because you're like, we'd stand in front of the machine for 5% of the cycle time. Like, yes, please. That's awesome.
John: sure, sure.
johngrimsmo: You know?
John: In the spirit of transparency, but I am not doing this to start tracking people like, or like, you know I mean?
johngrimsmo: Yeah, right?
John: Like, no, that's for sure not the point.
johngrimsmo: yeah why aren't you working hard now and ah those are the kind of questions you got to watch out for from the team because whether you mention it or not you know you can't help people from wonder if that's why you're doing it or whatever so just things you watch out for yeah yeah exactly cool
John: Um, yeah. Yeah. No.
John: Uh, I'm pretty, we're pretty transparent. It's like, that's not what we're, not what we're here to do. Um,
johngrimsmo: So on the topic of that, my latest Sunday update to GURP was, remember I told you a while ago, maybe a year or two ago, I had chat write me a Python script to make our chop grinding code for the speedio, right?
John: so you
John: Yes, love this.
johngrimsmo: And it's super awesome. ah Basically I feed it a 2D profile arc, line, arc, most of them pretty much.
John: Mm-hmm.
johngrimsmo: And then the retract distances from Fusion of that 2D profile are my kind of chop height.
John: Sure.
johngrimsmo: I think I have it programmed as whatever. And then it's it's good, but it was a ah Python like dot PI program that I have to like run in terminal or something like that. And it's, it's messy. I think I made a bin file or whatever to run it automatically, but it's messy. And it's like, you know, be cool is if I just had like a little Python program that opens up that I can run this and update it from. would be cool on top of that is if I could visualize it within the program itself.
John: yeah
johngrimsmo: And then you know would be cool on top of that is why don't I just put this into GURP, like into a custom scripts section. So that escalated quickly. And then basically, I'm sorry, Xander, but my NC Viewer is better than yours.
John: But that's crazy. The fact that that's even remotely doable is nuts.
johngrimsmo: Yeah. Yeah. And it's, it's absolutely fantastic. And then it's it's just me asking constantly more questions. Like, you know, it'd be sweet as if I could enter, um, not just for chop grinding, if I could just open like NC viewer does, if I could open a a full program and see all the tool paths, you don't be sweet on top of that is if I could isolate each tool path and just show the tool path that I want, you know, be sweet as if I click on the, you know, and it's like and on and on and on and on on and yes yes yes yes yes, yes, yes, no problem.
johngrimsmo: And now it works on mobile and it's like, It's like really, really, really clean. And I already used it. I used it yesterday and I saw a mistake that I made in my Post-it program and and I caught it.
John: I guess what I want to wish I understood better is for guys like you and me, is there any situation where we would purchase, subscribe, pay, adopt, embrace new services any anymore, software services?
johngrimsmo: Yeah. For guys like you and me, are like I'm willing to put in the time and it does take time.
John: Oh.
johngrimsmo: And there are situations where, like I said before, you know we can make an 80% solution given the time invested. There might be situations where, you know, you want a 95% solution and you're just willing to just handle it.
John: A
johngrimsmo: Like, like you're about to talk about with the vision system. I'm sort of on the fence money aside that, um, the commercial vision systems are like really thought out, you know, you might want it different because you like to tweak things and I like to tweak things, but some of them are probably just like, you just walk up to it and hit the button and it works.
John: hundred percent.
johngrimsmo: Whereas if it's all written by a Python script and a program and it crashes and it doesn't edge detect properly and all this, like we can, we can mess with it till the end of time, but depending on the situation and the money you want to spend and the result you want, and the operator that's going to use the thing, I think that will always be a case for just turnkey solutions.
John: Yeah.
John: Great.
John: Oh, 100% agree. And especially recognizing that um if you're a you know Fortune 500 manufacturing company, et cetera, you might have in-house R&D that does certain things, but you're also just more in the business of like yeah leverage on the experts to give you turnkey solutions.
johngrimsmo: Yeah, instead of re reinventing every wheel.
John: Exactly.
johngrimsmo: which i'm
John: Right.
johngrimsmo: And that's what I was talking about before, of like being careful which projects I go down, because I could reinvent every wheel, but like of do I need to?
John: Right.
John: Yeah. Well, so the let's come back to the vision system in a second because i for me, that is a call it 50% odds it works type of thing.
John: The linear gauge, i mean, we literally just bought, I just bought an eBay one because if they work on it, getting probably new ones that are modern and match, but an eBay was fine for now.
johngrimsmo: Mm-hmm.
John: And the quadrature encoder, I'm having a soldering cut.
johngrimsmo: This is LVDT basically.
John: No, no. So the problem with LVDTs is that they are effectively a fair, like a piece of steel passing through a variable, their variable voltage tube.
John: So they're infinitely, um,
johngrimsmo: Yeah.
John: refinable or measurable and they're kind of like analog. And so you have to, to the consistency and quality of the measurements relates to how your downstream amplifier or whatever you want to call it, it is is tuned and calibrated and and so forth.
johngrimsmo: OK. Yep,
John: Whereas the um linear gauges that I bought from it to toy you, which effectively look identical. um They actually have a little bit of a bigger housing where they hold the glass scale. So it's just a, it's just a quadrature encoder that you,
John: you You can misread it or screw it up if you exceed the refresh rate of the device you're using to read it. i think you can kind of like lose steps, if you will, which is which is really bad.
johngrimsmo: yep. man
John: um But for the way we will be using it and the hardware we're going to build around it, I don't think that's going to be an issue. And it means us putting the device on it to read it is is just It's not affecting the accuracy.
John: The accuracy that we're getting comes from mid-to-to-you full stop.
johngrimsmo: The device. Oh, interesting.
John: You know what I mean?
johngrimsmo: not Not your amplifier, not your, yeah.
John: Exactly.
johngrimsmo: Interesting.
John: Yeah.
johngrimsmo: Yeah, that's really cool.
John: Maybe a better example is like the record player, which is like the needle and the turntable, but then also you have some preamp or whatever that affects like the quality of how it's interpreting the needle's motions, if you will.
John: um LVDTs are for sure cool, but just not necessary for what we're trying to do.
johngrimsmo: Hmm.
johngrimsmo: Hmm.
John: The Mitsutoyev makes their linear gauges in 1 micron, 0.5 micron, and 0.1 micron, and they are between 6...
johngrimsmo: It's 100 nanometers.
John: Yeah, no joke. I mean, 0.1 micron is, well, 5 microns is 2 tenths.
johngrimsmo: Yeah.
John: Yeah.
johngrimsmo: There you go. Yeah. And it reminds me of the magnetic encoder that I've been using on on my little projects where I have a string of ah cylinder magnets.
John: Oh yeah.
johngrimsmo: And as the sensor goes across, every other magnet reads 2.1 million pulses or raw counts.
John: Yeah. Crazy.
johngrimsmo: over and I'm using eighth inch cylinder magnets. So every quarter inch of movement, I get 2.1 million data points, which I believe equates down to about three nanometers per raw point of data.
John: Yeah. Yeah.
johngrimsmo: And so like I can blow on it and see three nanometers move. And it's not calibrated.
John: Yeah.
johngrimsmo: It's not perfect. It's not accurate, but like it could be.
John: So for the linear gauge, the next step, again, literally, I have to get soldering situation fixed, which I will. And then I bought a a kit from touchdro.com, which seems to have solved some of the potential situations where you could exceed the refresh rate of the data stream, et cetera.
johngrimsmo: is
John: It was also a hundred bucks. Whereas the home project, our project I built that used the quadrature was just directly on a pie. Basically, i think on on a pie, if you were to push the plunger really quick a bunch of times, you will totally overload the system.
johngrimsmo: Yeah.
John: But then it's the nice thing is that we can also just check this against a master. So kind of simple. like I know if it's working or not.
johngrimsmo: Yeah. Very cool.
John: Yep.
johngrimsmo: And the goal is to use that to basically make some sort of measuring gauge where you put the part in and it automatically measures it. And you can have a calibration part where you can verify sanity check all the time, gauge blocker, gauge pin or something like that.
John: who
johngrimsmo: And then with um basically a touchscreen DRO, you can have a zero button, reset button, measure here to here, range, display everything you want, basically, which is so sick.
John: It's kind of converging on what Grant already did with a accurate, but not as accurate dial indicator, digital dial indicator at a USB output that he hooked in, wrote a quad script. I give Grant all the credit for this. It's incredible.
John: And on a web app, he sees the data of the measurements. He can push a button and it uses the FANUC Focus 2 to push tool offset updates to Williman, which I don't even want to do right now.
John: That's like a second new phase to it.
johngrimsmo: Yeah.
John: But what I want to do is have the Willemann to take off a half inch turned part, measure that half inch feature. And if it's off, we can at a minimum, just stop the machine.
johngrimsmo: Yeah. He could text you.
John: Done. And that's, yes.
johngrimsmo: Willimon needs a two tenths offset on tool 15. Be like, okay, I'll be right there.
John: Yeah. Yeah.
johngrimsmo: Pause for now.
John: Yeah.
johngrimsmo: Stay in a warm up cycle until I get there. Like, right?
John: Oh, that's a good point.
johngrimsmo: Right.
John: Yeah. Yeah. But it's stuff like that, that I get too, I get real excited about where it's like, okay, if it measures off, have it put that part in a, uh, inspect bin, have it idle for, or warm up for 10 more minutes, run the next part, measure that one.
John: And you could go into this kind of like triage mode.
johngrimsmo: It is a loop. Yeah. Yeah.
John: Yeah. Yeah.
johngrimsmo: Yeah. Cause I mean, thermal growth is a real thing. If one part's bad and the machine stops and you come back two hours later to run the next one, like it's a different temperature. Your, your two tents is now wildly different. Um, actually it was funny. I was at, uh, an event at the historic house nearby us where Meg was doing a thing. And one of the musicians there was, um,
johngrimsmo: Just an interesting guy. And I get to ask him, he's this retired dude. And I was like, so what'd you do? And he goes, I built um control systems for curling rinks.
johngrimsmo: like ice, you know, curling with the stones that you throw.
John: Oh, yeah.
johngrimsmo: um And he he monitored the ice temperature and like made sure the ice was exactly. And i was like, what do you got Plus or minus one degree? He goes, 0.1 degree, buddy.
John: No kidding.
johngrimsmo: And i was like, I just programmed my first PID or my first um PLC last weekend. He's like, oh, what'd you use? think Random, but funny.
John: Yeah.
John: Well, so this, that reminds me of, did you finish the precision microcast yet?
johngrimsmo: i am I think I'm on the latest or second latest episode. I'm just about there.
John: Okay. That's right. forgot you had a bunch to catch up on.
johngrimsmo: Yeah, yeah.
John: Um, but I'm not really spoiling much. Um, Josh Hacco was talking about building a new facility and my words here, he basically is struggling to get HVC contractors engaged because a lot of the brain power and equipment in the world on for HVC systems that are operating in controlled environments is going to our data centers.
John: So.
johngrimsmo: Oh, interesting.
John: For him to have a metrology room that has a certain parameters that he was looking for is now proving to be quite difficult. And again, i enjoy a lot of what you and I are doing. im sure Josh is more than capable of doing this because it made me think about PIV loops and controlling temperatures range.
John: How do you, how do you run a loop like that? Because there's such a lag as temperature changes.
johngrimsmo: Yeah.
John: But it's just like, no, we just build ourselves.
johngrimsmo: Well, Yeah. I remember reading in the Foundations of Mechanical Accuracy, the book by the Moore Company, written in the, I don't know, 60s or something.
johngrimsmo: They talked about thermally controlled rooms and how they have basically a concrete room within a concrete room underground to like make this perfect.
John: crazy.
johngrimsmo: And they had, this is in the 60s, they had temperature sensors every 12 inches up the wall. And what did they figure out? Air coming in from the floor or something is better than air coming in from the ceiling? Or i forget the details, but...
johngrimsmo: I mean, that's like 80 years ago, 60 years ago.
John: Well there's a whole thing about, oh who is telling this? There's like special ceiling tiles that are made that have holes in them that caused the columns of air to actually flow the correct way.
johngrimsmo: Laminar flow? What?
John: Yeah. when whatever Yeah. Yeah. Crazy.
johngrimsmo: That's awesome.
John: Very.
johngrimsmo: You collimated air. yeah
John: Okay. That's my segue. So, and I'll keep this short. We had a really cool demo from Keyence. There's a chance we buy that system. um But what occurred to me is that a lot of the secret sauce is split half of it. And it may not be half that, but part of it's in the ability to get the image from
johngrimsmo: Yeah.
John: period, full stop.
johngrimsmo: The hardware and the software.
John: Just how you get exactly so you cliche to say like that. And then once you have this image, it's almost, it doesn't matter that you have the hardware. You just have this image that I need to now parse. And the latter, I'm quite confident in our ability to experiment with.
John: And in fact, you could just probably use sample images and build the software off of sample images.
johngrimsmo: Yeah. That's what I did.
John: Um,
johngrimsmo: Yep.
John: Bingo. And like, again, LLM, OpenCV, there's existing software and folks have done this. And there is some quote unquote secret sauce. They're interpolating edges to find gradients to get the crispness.
johngrimsmo: Yeah.
John: But, you know, to Keynes' point, you put a gauge pin down on their system and it shows you, I don't know what the stated accuracy specs are, but it's to a tenth they say it can measure this part.
John: um
johngrimsmo: it It depends on zoom zoom value.
John: So I sort of...
johngrimsmo: That's what I'm finding out. Um, and what camera you're using and what your pixel to micron ratio is like when you zoom all the way into one pixel and I have to program mine to like, not, um, dither or whatever, like, like blur the pixels together.
John: Is that right?
John: Yes, yes.
John: Okay.
johngrimsmo: When I zoom in fully, I want to see a square pixel. I want to see like, you know, one pixel and then you measure across that and it's whatever a real number is for me, it was 10 microns per pixel.
John: Mm-hmm.
johngrimsmo: Um, So I can't, you know, i measure less than really 10 microns because that's one pixel.
John: So you can, you can, it
johngrimsmo: Yeah, you can kind of guess and math it and like, don't know.
John: would depend on the shape. Like if ah it was a straight line, you would be able to take a averaging line and get some factor higher.
johngrimsmo: Yep.
johngrimsmo: Right.
John: But that's what I learned, saw too, was in talking to LMS. Again, a major caveat of assuming that there's no hallucination, which is a major risk, but basically sort of told me, okay, if you want,
johngrimsmo: Yep.
John: to handle something like five microns, we can build that system. But going down to like confidently one or two microns is a whole different ballgame. um And
johngrimsmo: Unless you zoom way in and then you're kind of getting free accuracy, but you're losing your field of view.
John: yeah, so to your point, that probably is true.
johngrimsmo: um
John: got it. god Yeah, it makes sense.
johngrimsmo: I mean, to a point you're at some point, you're basically viewing, you know, a two millimeter square and that's like a really tiny,
John: Right.
johngrimsmo: But then you can get half micron per pixel or less.
John: Got it.
johngrimsmo: Well, not like light kind of doesn't split at less than half micron. It's the wavelength of light.
John: Okay.
johngrimsmo: But ah apparently Keyence can zoom in more and it just kind of, it's you're not really splitting light that much smaller, but it they can average it or something.
John: That's crazy.
johngrimsmo: It just gets blurrier.
John: Yeah.
johngrimsmo: I forget.
John: The way the Ellen discussed it, made sense and didn't seem all that difficult. And I think it's something that's commonly done by tools like OpenCD already. So I'm not having to like reinvent the wheel on a lot of it.
johngrimsmo: Yeah.
John: And so the real question, similar to the Mitsutoyo situation, although this is much more wild west was, can I do this system? And so I bought a Basler camera, very specific attributes to that camera.
John: And it was 600 bucks. I bought a,
johngrimsmo: What was it? Was it new?
John: I can't remember. I've been on a few different ones. It was either used or new old new open stock.
johngrimsmo: Okay. And what do you remember what some of those specifics were? Cause I went down this rabbit hole too.
John: It is an ACA2440-20GM. I don't remember. i mean, I can look that up real quick. Actually, we'll take a second.
johngrimsmo: Cause for me, what I found, um, yeah, using, especially for a vision system, you just want black and white, like,
John: Yep.
johngrimsmo: you don't need color and color is actually a detriment because of the Bayer filters on the lenses or on the, um, camera systems.
John: Okay. Okay. Okay.
johngrimsmo: You basically don't need the blue and the red pixels. Um, so you're mostly only using the two green pixels. And now I, when I'm using a color camera, I'm converting the blue and the red pixels to green pixels or something like that. So I'm still using them, but I'm counting it as a green light or whatever.
johngrimsmo: um But for the white light interferometer, the Thor Labs camera, I bought a monochrome camera that technically has kind of twice as many green filters or green pixels as a color camera that has blue and red as well.
John: Oh, sure, it sure. Doubles up on the one color.
johngrimsmo: So yeah, it's interesting.
John: Sure.
johngrimsmo: You can kind of get free accuracy just by either using a monochrome camera or telling the software, all those blue, red pixels, like make them green.
John: Trade off. Right. Makes sense. So camera, I bought a telecentric lens that was also hundreds of dollars.
johngrimsmo: what's What's your... I don't know what the word for it is. This is like... but Mine's 0.275.
John: Specs.
John: Opto engineering TC 12056 half inch F slash 8 156 millimeter C mount. I mean, I need to go play with this stuff.
johngrimsmo: Yeah, yeah. No, because that directly defines your view, your zoom in your field of view, um whatever this number is, whether it's a a two is pretty zoomed in, 0.275 like mine is fairly zoomed out.
John: Correct. Yes.
johngrimsmo: So I can see almost a full inch of viewing area, even though the lens is like five inches in diameter.
John: Okay.
John: Okay.
johngrimsmo: um Yeah.
John: Oh, wow.
johngrimsmo: And that the directly affects your pixel to micron ratio.
John: And then bought it.
John: Yes. Part of the reason I did this too was, and again, this is much more in the, let's just see how we get along.
johngrimsmo: Yeah, it's awesome. fun to see you going through it.
John: um
johngrimsmo: Yeah.
John: A lit columnated LED backlight, which is also very specific to have it be monochromatic.
johngrimsmo: Nice. Yeah.
John: And then I bought a calibration pattern thing as well. But again, the thing I like about this is that
johngrimsmo: night
John: We're not using it in the job shop environment where like if you purchase a system from say Acheans or Zeiss, it needs to just be able to handle any part that's presented to it. I can i can say, hey, here's a good known good part, compare it to the part we're making today.
John: So I'll keep everybody updated to see if we bail on this or if it works.
johngrimsmo: That's awesome, dude. So the the next challenge is more on the hardware side, now that you've bought the components, just like I did, is mounting them and aligning them and having it be like rigid enough and accessible enough, whatever.
John: Okay.
johngrimsmo: And then your height, like your Z focus, basically, is a real thing.
John: Yeah.
johngrimsmo: um
John: Okay.
johngrimsmo: So what I did on Amazon, I bought a Z stage motorized stepper motor.
John: Okay.
johngrimsmo: I haven't used it yet, but for like a couple hundred bucks, it's it's a literal linear rail ball screw Z stage. You just mount that vertically on some base plate and you can program that into your system.
johngrimsmo: And I haven't done it yet, but I want to. There are very simple autofocus algorithms out there. that, especially on a vision system that have edge detect. And it's like, oh, blurry, not blurry, blurry. There's my zero. Now I'm in focus.
johngrimsmo: And you basically hit an autofocus button and it just goes, yep, there it is. And then you hit the calibrate button on your your calibration thing and it goes, yep, okay. And that will get you just better results faster.
John: Yeah, you're way further than me. I'm trying to sort of build a poor man's version of what Keyence brought by, which was left to right camera system.
John: was horizontal, not vertical like the traditional microscope.
johngrimsmo: Oh, okay, yeah.
John: And let's say that this thing had a 12 inch um empty void between the camera and the lens.
johngrimsmo: Interesting. Okay, yeah.
John: There was about a two inch range that was marked in the middle of that where you could put the part anywhere in those two inches. So it was very forgiving.
johngrimsmo: Is that what you're thinking too? A horizontal style? Or whatever that is.
John: Probably. Yeah.
johngrimsmo: side Like the beam is side to side.
John: Yeah. I am.
johngrimsmo: Interesting. I haven't considered that. So your part... You still gotta have the part in focus.
John: No, I like, I'm not fighting.
johngrimsmo: Yeah.
John: and I'm just saying that theirs was very, so long as you were in that relatively forgiving range, it was fine.
johngrimsmo: Yeah,
John: and I think if you moved it out of that range, it didn't so much lose focus as it's just the the measurements were, would have been inaccurate.
johngrimsmo: yeah. I'm sure the key ends has auto focus and just does that. Like you don't even notice.
John: Yeah, that may be the case.
johngrimsmo: Yeah.
John: Um, but the,
johngrimsmo: And then, and then the, the squareness of anything, whether it's that system where the light goes side to side or my system, traditional microscope, we have your edge light on the back on the, on the bottom. and you put your part on top of that and the microscopes above, um, the squareness of your lens affects the telecentricity and the like edge detection.
John: Yeah.
johngrimsmo: Like it needs to be able to be trammed, you know,
John: Our plan right now is to use this on round parts though, unfortunately.
johngrimsmo: Yeah, that'd be better.
John: So it's, it's less of a less of an issue on some, well, yeah.
johngrimsmo: Less of an issue, but. But also as when Cyrus made his first white light interferometer, he did it all laying down. Like where the, you know, the microscope's objectives go left to right, not up and down.
johngrimsmo: And that's just like so much easier because you can just put it on a V block and all of a sudden it's square.
John: Yeah.
John: Yeah.
johngrimsmo: Whereas trying to build something up vertically in space, you have tilt, you have squareness, you have tram, you have all this grap crap that's like not easy to do.
John: Right.
John: But aren't you trying to do it the hard way?
johngrimsmo: I am trying to do it the hard way, which takes so much time.
John: yeah Why? Why?
johngrimsmo: i don't know.
John: Left or right? Okay. I mean, I don't know if I'll need a granite piece, but you buy those like eight by 12 granite blocks for, you know, a hundred bucks, not even.
johngrimsmo: Exactly. So that's funny.
johngrimsmo: ye Yep. Yep.
johngrimsmo: Fun. Good for you, man.
John: Yeah. Yeah. Oh yeah.
johngrimsmo: That's awesome.
johngrimsmo: this is cool
johngrimsmo: Okay, I've got a couple other things. Remember last week I talked about installing the updated laser system on the Speedio, the Bloom lasers.
John: oh yeah
johngrimsmo: um Continued to struggle with that even after the podcast for a little bit. And then eventually I roped in one of my machinists, Jeff Jeff, and i was like, can you just... can I just tell you everything and and let me explain this to you and maybe you'll see something I'm not seeing.
johngrimsmo: Cause it's, I'm running the calibration program and it's not calibrating properly.
John: Yeah. Okay. Yes.
johngrimsmo: And it's still reading off by, I forget how much, but like enough. And I was like, something's wrong. And eventually after him asking enough questions and me, um,
johngrimsmo: answering and thinking about it enough. um I might've told you a few months ago, I added Fjell blade. i bolted it to the top of the bloom laser to give me a hole, a precision perfect hole that I can probe with the touch probe, the spindle probe.
John: yes I do remember this.
johngrimsmo: And so now I have a calibration feature on the machine for the spindle probe, which is so sick.
John: Sure.
johngrimsmo: So I cant i can not only auto calibrate the probe every time for run out, for eccentricity, for whatever.
John: Love that.
johngrimsmo: um I can also verify that it's measuring 0.3109, whatever the hole is. And it turns out I was running that right before the calibration program for the laser, and it just makes the laser calibration not work.
John: Oh my God.
johngrimsmo: And he's like, how would you just block that off for once? All of a sudden, perfect results. We're like, what? And then we turned it on again and did it again, and it didn't revert to the bad way.
John: Oh my god.
johngrimsmo: It just stayed good. And we're like, okay i don't understand this but it's fixed so i'm happy but i'm disappointed about it so i still don't know like really specifically to the line of code what the issue is but it works and that's good to know um
John: It's not overwriting a variable that...
johngrimsmo: i thought it might be but it doesn't seem to be but that's possible maybe i'm just not seeing it and then the other the other one that got me was um
John: Weird. Yeah, sure, sure, sure.
johngrimsmo: I forget exactly what was happening, but when when we do a pallet change with the Auroa, it turns on the air blast for the laser during pallet change, which is, those are separate systems.
John: Which it shouldn't.
johngrimsmo: no those And I'm like, we've always just kind of left and dealt with it.
John: Yeah, right. Okay, yeah. That doesn't make sense.
johngrimsmo: What's funny is Above palette number 64, it doesn't do it. Below palette number 64, it does do it. And I'm like, what? So I started to dig into it. It turns out when the laser guy installed the laser, the bloom laser, he used parameters 1100, 1101, 1102, 1103. When the aroa guy came in a little bit later the AROA, he used parameters And I'm like, oh my gosh.
johngrimsmo: And once I figured this out, and these are in the ladder logic of the program, not just so visible in the code or in whatever, like it's being set by ladder logic stuff. So it took me a while to even find it.
johngrimsmo: And now that I understand it, I can accept and avoid the issues from it. But because it was leaving the shutters open after some pallet changes, which let coolant ingress into the shutters.
John: Ironically, maybe why?
johngrimsmo: that's That's our big problem.
John: Yeah.
johngrimsmo: like
johngrimsmo: And you don't know what you don't know.
John: Yeah.
johngrimsmo: i can't really blame the Aeroa guy. I can't really blame the Bloom guy. It's just they didn't they came six months apart and didn't talk to each other and just happened to overwrite the same things. And we've been living with it for two years.
John: Well, nor will they, nor will they ever talk to each other. Like this is what, if you want to like lose sleep at night and and be a worrisome person, which is not a good thing to do.
johngrimsmo: now
John: It's like the wild west of like who's using what variables and what are they being written to is horrifying.
johngrimsmo: Yeah, I've been very good at documenting how I use them on the machines that matter. But that's not to say that Jeff is going to reference my documents when he goes to program something new.
John: Yeah.
johngrimsmo: you know
John: Yeah. Just let an agent handle it.
johngrimsmo: Exactly, right? The context. It has all the context.
John: Don't. Don't.
johngrimsmo: to To a point.
John: Yeah. Yeah.
johngrimsmo: So yeah, everything everything's all good.
John: Well, our stressor. um This is nothing other than a commiserating story, but our stressor of the week was that um we had that air compressor back up and running that we had to swap the motor out on.
John: And then after it was running for a few days, um both compressors went down at the same time, which stinks.
johngrimsmo: No.
John: And it really does. Talk about crappy timing. The compressor that had not had any issues literally hit a service interval where it just goes into a shutdown maintenance mode, not because anything's wrong.
John: And what are the odds that it did that? The other compressor, we must have somehow... damaged a different part. They call it a transformer. I haven't actually seen it myself. So take that for it's worth. Um, cause I was sort of like, ah, that's weird to have a transformer in a machine and that's weird for it to be broken. And under these circumstances ends up that the new one just came this morning.
John: Um, somewhat frustratingly, they shipped the wrong one. We paid for overnight shipping. and It's like frustrating. You pay for FedEx overnight and the wrong one was sent, but right. One came this morning, swapped it out. it is working, which is great.
John: Um, the other compressor was just simply a quick maintenance kind confirmation that we are doing the maintenance. Um, but, uh, to, to have,
johngrimsmo: So you got that one back up and running right away pretty much, but okay.
John: Correct. But like, don't know, kind of feels like a kick in the pants when you have two compressors specifically for redundancy. And then, and I'm a little bit of a Karen, like when one goes down, then we don't mess with the other one because we were going to move them both.
johngrimsmo: Oh, man.
John: i'm like, nope, that compressor does not get touched. We don't move it. We don't do anything that will create additional risk around its use in the shop and it's still in town.
johngrimsmo: Yep. Oof.
John: Oh well, it's okay.
johngrimsmo: Yeah, because a shop is useless without power or air. You can't do anything, really.
John: Yeah. And we can run most what we need to run on one. um
johngrimsmo: Yeah.
John: So now both are up but until we got that confirmation about two hours ago. I was considering the older one is six years old, which isn't even that old, but i'm considering i look at the math, like what I think I could sell it for and what I could buy a new one for.
John: um and buy one that's a little bit bigger because the we can run on one compressor, but we have to stop running certain machines.
johngrimsmo: and
John: And if the but the new one is big enough, a loan should run the shop. So the second one is, obviously, yeah if the big one breaks, we're still not back.
johngrimsmo: Yeah.
John: But what kind of a cribbing up growth method? And then ultimately we'll replace the other one to get it bigger.
johngrimsmo: Yeah.
johngrimsmo: Yeah. No, I like that, especially something so critical. You're keeping it fresh pretty much, keeping it up to date, you know, sell it for a good deal to a younger shop that just, it's the greatest thing ever for, you know, half price or whatever.
John: Yeah.
johngrimsmo: And then you invest in the newer one for reliability, for service, for all.
John: Yeah.
johngrimsmo: Yeah. It's a good idea.
John: Yeah.
johngrimsmo: Cool. Well, I guess that about does it. I have a lot more notes, but they can save until we get back. So I'm off next week.
John: Okay.
johngrimsmo: little family vacation.
John: Have, yes, love it.
johngrimsmo: And then we'll be back the week after.
John: Good. I will see you August. Well, for the audience, it'll be August 7th. I'll see you on the 5th for the recording. i Have a good time, though.
johngrimsmo: Perfect. right, man.
John: Take care.
johngrimsmo: Take care. Bye.
