Yapay Zeka V8 Motor Çizdi — Transcript
Full transcript
- 0:00AI modeled this V8 engine, guys.
- 0:03In SolidWorks, all night while I was asleep.
- 0:0631 parts.
- 0:07The crankshaft, pistons, camshaft, valves, all of them.
- 0:11And look, it really turns.
- 0:13Then I made ChatGPT the judge and said, "See if you can find a mistake."
- 0:17Did it find one?
- 0:18At the end of the video.
- 0:19Hello, guys.
- 0:21Today I'm going to show you a slightly different video.
- 0:25This time I didn't do the modeling or even explain it; AI did the modeling.
- 0:27Let's play the recording and watch it together.
- 0:30You'll hear every moment I was surprised, laughed, or said, "No way."
- 0:37What did we do?
- 0:38We went online and picked a real V8 that's still on the roads.
- 0:41We found its published dimensions.
- 0:44Then AI modeled the same engine in SolidWorks,
- 0:47part by part, assembled it, and ran it.
- 0:50First, the engine.
- 0:51When I hear V8, one engine comes to mind, guys: Chevrolet's LS3.
- 1:156.2 liters, 8 cylinders, 430 horsepower.
- 1:19The engine in the Corvette and Camaro SS.
- 1:21There's also an "LS swap" culture in America; people put
- 1:24this engine into every car they can get their hands on.
- 1:27The engine comes out of the crate, gets mounted on a stand, and is started up.
- 1:31Honestly guys, I envy the Americans.
- 1:34I wish we could also take cars apart and put them back
- 1:37together in our own garages here in Turkey, and had the
- 1:39money to buy engines like this and swap them in.
- 1:41So why a V8, and why is it so popular?
- 1:53Okay, we won't go that far.
- 1:57But it's also a very nice engine from an engineer's perspective:
- 2:00pushrods, a single camshaft in the middle of the block, and 2 valves per cylinder.
- 2:05Simple, sturdy, and its dimensions are openly published.
- 2:09That's what matters most for our project.
- 2:11There are ready-made CAD files of this engine online;
- 2:14we could have downloaded and opened them.
- 2:16We didn't.
- 2:16If we copied someone else's model, AI wouldn't have modeled anything.
- 2:20We only took the numbers.
- 2:22And here's the nice part: GM's own service document
- 2:25contains this engine's mechanical dimensions.
- 2:28Cylinder bore, crankshaft journals,
- 2:30pin, valve, spring, all with their tolerances.
- 2:34For dimensions with tolerances, we used the exact midpoint of the range.
- 2:37Cylinder bore 103.25 mm, stroke 92, connecting rod length 154.89.
- 2:44Let's calculate the displacement: 8 cylinders
- 2:48times cylinder area times stroke, 6162 cm³.
- 2:526.2 liters.
- 2:53It checks out, guys.
- 2:55But that document doesn't contain everything.
- 2:56The block's outer shape, some journal widths,
- 2:58and the exact cam profile, for example.
- 3:01We estimated those using engineering calculations and
- 3:04clearly labeled them "estimated" in the table.
- 3:06The judge will look at those too.
- 3:08Now for the main question: how will AI use SolidWorks?
- 3:12Normally, AI just talks.
- 3:15You say, "Model a piston," and it tells you how to
- 3:18model one, but you're still the one doing it.
- 3:20You may remember that we solved this for AutoCAD
- 3:24in episode 13; we gave AI a hand, called MCP.
- 3:38Ours isn't that shiny, but it does the job.
- 3:41We built the MCP for SolidWorks ourselves, guys, with Claude.
- 3:45There are ready-made SolidWorks connectors online; we looked at 12 of them.
- 3:50Some are unfinished, and some want permission to
- 3:53run any code they choose on your computer.
- 3:55I don't want that; I want to know what's running on my computer.
- 3:59What can it do?
- 4:01Sketches, extrusions, revolves, holes, patterns, materials,
- 4:04assemblies, mates, motion, mass calculations, and interference checks.
- 4:08What can't it do?
- 4:10It can't delete files, overwrite any files, or access the internet.
- 4:13And this connector is ours, guys.
- 4:16We built it ourselves and use it for our own work.
- 4:19Look, look, it's starting now.
- 4:21It's drawing the lines one by one,
- 4:23adding dimensions, and there, it extruded it.
- 4:26This is the block's initial form.
- 4:28Guys, nobody is touching the mouse right now; I'm just watching.
- 4:32It would take days for a single engineer to
- 4:34model an engine from start to finish.
- 4:36What do you do in real life?
- 4:38You put together a team.
- 4:39One models the crankshaft, another the piston, another the block.
- 4:42That's what we did too: 10 agents in 2 waves.
- 4:46When everyone works on the same project, they mess up each other's files.
- 4:50Programmers use a method for this called a "worktree."
- 4:53Simply put, the same project has separate desks.
- 4:57Each AI models at its own desk, measures its own part, and checks it.
- 5:01Only a part that passes the checks makes it to the main desk.
- 5:05But there's only 1 instance of SolidWorks.
- 5:07What happens if 10 AIs try to use the mouse at once?
- 5:11While one models the crankshaft, another opens the piston,
- 5:13and a crankshaft line ends up inside the piston.
- 5:16So we put a lock on the connector: whoever has control of SolidWorks
- 5:19keeps it until the job is done, while the others wait their turn.
- 5:23So the brains work in parallel, but there's only one hand.
- 5:25One more thing: after the parts passed their checks overnight, we had it model
- 5:30them all again in SolidWorks from scratch, step by step, to show you.
- 5:34That's what we're about to watch.
- 5:36The block.
- 5:36A 90° V, 2 banks, 4 cylinders each.
- 5:40Look, it just cut all 8 holes at once.
- 5:43Now it's modeling the crankshaft.
- 5:455 main journals, 4 crankpins.
- 5:47The key here is the journal angles: 0, 90, 270, 180°.
- 5:54We call this a cross-plane crankshaft.
- 5:56It's also the reason for that famous V8 rumble.
- 5:59Now it's forming the teeth one by one.
- 6:0258 of them.
- 6:03The engine's brain uses them to tell what angle the crankshaft is at.
- 6:07The piston.
- 6:093 ring grooves: compression at the top, the 2nd ring
- 6:11in the middle, and the oil ring at the bottom.
- 6:13The groove widths match GM's specified clearances exactly.
- 6:17The camshaft.
- 6:1816 cam lobes, 1 for each valve.
- 6:22Look, it's placing each lobe at a different angle.
- 6:24Calculating these by hand would take days; it calculated each lobe's angle from
- 6:29the firing order itself, then measured the geometry it had modeled:
- 6:32the valve opening and closing timing deviated from the target by no more than 0.01°.
- 6:38The lifters and pushrods.
- 6:40They'll carry the motion from the cam up to the rocker arm.
- 6:43Keep this rod in mind; it's going to cause us trouble shortly.
- 6:46The cylinder head.
- 6:47The valve bores are tilted at 15°, just like GM's.
- 6:51The combustion chamber target is 68 cm³; it got 67.97.
- 6:56Now the 2nd wave.
- 6:58Valves, springs, spring retainers, and locks.
- 7:03Rocker arms.
- 7:03The rocker ratio should be 1.7; it measured it and got 1.700000.
- 7:09Honestly, it's more meticulous than I am.
- 7:11Oil pan, front cover, timing sprockets, and chain.
- 7:13The camshaft follows the crankshaft through the chain;
- 7:16when the crankshaft makes 2 revolutions, the camshaft makes 1.
- 7:20So the sprockets are 23 to 46.
- 7:21Flywheel, pulley, valve cover, throttle body, and intake manifold.
- 7:26And the parts are finished, guys.
- 7:28There's something we do here to avoid revisions later.
- 7:32Check each part when it's finished, before assembly.
- 7:35AI measures its own model, compares it with the table, and checks its mass.
- 7:41Look, guys, this isn't just for AI.
- 7:44When you're modeling a project, stop after finishing
- 7:47each part, measure it, then continue.
- 7:50If you save all your checks for the end,
- 7:51you'll get all your revisions at the end.
- 7:53Now we're bringing all the parts together.
- 7:57The block first.
- 7:58The block is fixed; everything will be positioned relative to it.
- 8:00In SolidWorks, we call this a mate, guys.
- 8:03You tell 2 parts, "You stay on this axis, and you sit against this face."
- 8:07Here, mates do what a mechanic's hands do in a real assembly.
- 8:12Look, it took the crankshaft and seated it in its bearings.
- 8:15Now it's fitting the connecting rods onto the crankpins,
- 8:17putting the pistons into the cylinders, and fitting the pins.
- 8:20It also connected the camshaft to the crankshaft with the chain: 23 to 46.
- 8:26The left bank's head is in place.
- 8:28Now the valve assemblies: lifter, pushrod,
- 8:30rocker arm, valve, spring, retainer, and lock.
- 8:338 of them, one by one.
- 8:35It's doing the same thing again on the right bank.
- 8:37It's patient; I would have been bored by now.
- 8:39Oil pan, front cover, flywheel, pulley,
- 8:42valve covers, manifold, and throttle body.
- 8:45It also left the piston rings and connecting rod bolts until the very end.
- 8:49A real mechanic wouldn't do that, of course;
- 8:50you fit the rings before the piston goes into the block.
- 8:53But here everything is already in position, so the order doesn't matter.
- 8:58And the engine is complete, guys.
- 8:59212 parts.
- 9:01Finally, an interference check: do the parts overlap?
- 9:03It checks with the crankshaft at 0, 90, 180, and 270°.
- 9:10There are no collisions among the rotating parts.
- 9:12The remaining minor issues are expected: when the valve is open,
- 9:14the spring appears to overlap the retainer because we modeled the
- 9:16spring as a rigid part, so it doesn't show compression.
- 9:18There's a 0.2 mm³ overlap between the lifter and the block
- 9:22bore; that's due to surface calculation tolerance,
- 9:25and there's actually a 0.02 mm clearance between them.
- 9:29But of course, guys, not everything went this smoothly.
- 9:33I woke up in the morning, and AI had left me a report.
- 9:36A lot of things had come up while preparing the assembly too.
- 9:39Let's see what happened.
- 9:40The agent modeling the valves placed the valve, spring,
- 9:43and guide in the head and found nothing under the spring:
- 9:46there was a 13 mm gap, leaving the spring in midair.
- 9:50The agent modeling the head had forgotten the surface the spring would sit on.
- 9:53They're catching each other's mistakes, guys.
- 9:55The head moved to version 2, with 8 spring seats added.
- 10:00The pulley at the front end of the crankshaft was overlapping the front cover.
- 10:03The snout was too short.
- 10:05The crankshaft also moved to version 2: the snout went from 60 mm to 100 mm.
- 10:09The chain is interesting too.
- 10:10It calculated the length and got 58.6 links.
- 10:13You can't have half a link.
- 10:15It changed the sprockets, but it still didn't work out; it found its
- 10:19own mistake and finally settled on 23 and 46 teeth with 62 links.
- 10:23The first assembly interference check caught something: the flange at
- 10:27the back of the crankshaft was overlapping the block's rear wall.
- 10:31The block moved to version 2, with a pocket cut for the flange.
- 10:34Then it checked again after fitting the flywheel:
- 10:37the flywheel overlapped the block by 2.96 mm.
- 10:39In real life, this engine wouldn't turn; it would lock up.
- 10:42The block moved to version 3, with its rear face recessed by 4 mm.
- 10:47Then, at one point, SolidWorks crashed on its own.
- 10:50There's a special mate type that connects the
- 10:52cam and lifter, called a "cam follower."
- 10:55When AI tried to set it up, SolidWorks abruptly closed.
- 10:59Anyone who uses SolidWorks knows this,
- 11:01guys; it closes at the worst possible moment.
- 11:04What did it do?
- 11:05It disabled that mate type in the connector.
- 11:08Since it had modeled the cam profile itself, it calculates
- 11:11the lifter's rise from that profile at each crank angle
- 11:13and raises the lifter by exactly that amount.
- 11:16So it uses calculations to do what SolidWorks couldn't.
- 11:19Remember I said the pushrod would cause us trouble.
- 11:22Here it is.
- 11:23The sources give this rod's length as 187.6 mm.
- 11:27Starting from the cam, AI lined up the lifter, pushrod,
- 11:31and rocker arm end to end; the rod fell 31 mm short of the rocker arm.
- 11:36Why?
- 11:37Because we don't know the exact interior geometry of the real
- 11:40engine's block or the lifter angles, so we estimated them.
- 11:43That estimate affected the rod length.
- 11:45So what does a real engine builder do?
- 11:47When the cam or head changes, they measure the required
- 11:50pushrod length and buy a rod of that length.
- 11:53AI did the same: it kept all the sourced dimensions
- 11:56and selected the rod by measuring.
- 11:59218.94 mm.
- 12:01That's how it works in the field too, guys; if the drawing's
- 12:04dimension doesn't fit, you measure and cut accordingly.
- 12:07The interference check found something else: the camshaft's bearing
- 12:11journals overlapped the bodies of 2 lifters by 0.5 mm.
- 12:14The journals were narrowed slightly, and now there's nearly 1 mm of clearance.
- 12:19The lobe positions and timing didn't change.
- 12:23Also, on the first attempt, the heads and springs ended up on the wrong side.
- 12:27In SolidWorks, a mate sometimes has 2 solutions
- 12:30and can place a part on the opposite side.
- 12:32AI set itself a rule from then on: after placing each part, it compares its
- 12:37position with the calculation to check whether it's where it should be.
- 12:42If it isn't, it stops.
- 12:44When it closed the valve cover, it found that the
- 12:47pushrods and rocker arms inside didn't fit.
- 12:50The cover was widened internally, and now there's enough room.
- 12:53Also, the rocker arm overlapped the spring retainer by 0.1 mm.
- 12:57About the thickness of a hair.
- 12:58It caught that too and adjusted the arm's edge.
- 13:01So you saw a few revisions in the video, guys.
- 13:03I think that's the best part: the checks caught the mistakes, not us.
- 13:09The engine you're about to see is the final state of this first version.
- 13:12There's another round after the judge.
- 13:14And now for the best moment, guys.
- 13:16All the parts are in place.
- 13:18The crankshaft is about to start turning.
- 13:28Honestly, well done, it turned.
- 13:30Look, the pistons reach the top in the order 1-8-7-2-6-5-4-3.
- 13:37The camshaft turns in the same direction as the crankshaft,
- 13:39but at half the speed, thanks to the chain.
- 13:42The valves open and close.
- 13:44The lifters move up and down on the cam.
- 13:47It has hidden the block and heads, so we're seeing the engine's skeleton.
- 13:51How far do the valves open?
- 13:53The intake valve opens 14.25 mm; GM says 13.97.
- 13:57Exhaust is 13.43; GM says 13.22.
- 14:02About a 2% difference; that's because the ratio rises
- 14:05slightly above 1.7 as the rocker arm rotates.
- 14:07We checked this numerically too, not just visually.
- 14:10The crankshaft stops every 5°, the heights of all 8 pistons are measured,
- 14:13and the results are compared with the crank-and-connecting-rod formula.
- 14:17The largest difference over 2 revolutions: 0.000001 mm.
- 14:23That's 1/1000000 of a millimeter.
- 14:25If you've made it this far, guys, don't forget
- 14:28to subscribe and like the video.
- 14:30And tell us in the comments: which engine should we have it model next?
- 14:34A boxer, a V12, or a diesel?
- 14:36Now we're calling in the judge.
- 14:37I gave ChatGPT all the reports: the dimension table, masses, interference report,
- 14:43piston positions, valve timing, images, and 3D files of all the parts.
- 14:48Also a list of things we deliberately simplified, so it wouldn't
- 14:51get stuck on those and would look for actual mistakes.
- 14:54I said, "Don't be polite; I'm looking for mistakes."
- 14:57The judge opened and measured all the STEP files itself.
- 15:00Cylinder bore, stroke, connecting rod length, valve diameters, they all match.
- 15:05It opened the assembly too: "212 components," it says; it counted them.
- 15:09Its final sentence is: "The V8 geometry of this model is not made up."
- 15:14But it found mistakes too.
- 15:15"The cam profile is mathematically consistent,
- 15:18but it doesn't match the actual factory LS3 cam phasing."
- 15:22It's right, guys.
- 15:23We had estimated the intake centerline angle at 114°.
- 15:27The judge says the published LS3 cam card gives 120°.
- 15:31We checked, and the old LS3 cam card in community listings also says 120.
- 15:36So we had installed the cam about 6° advanced.
- 15:39Second: the main bearing caps.
- 15:41They're steel in the real engine, but we treated
- 15:43them as aluminum within the block.
- 15:45The judge measured their volume from the STEP
- 15:48file and calculated a 3.8 kg difference.
- 15:50That's correct too.
- 15:53Third was our wording mistake: we had written "the LS3
- 15:55piston has valve reliefs" in the simplification list.
- 15:57The judge corrected it: the LS3 has no valve reliefs; it's a flat-top piston.
- 16:01So our model was already correct on that point; the list was wrong.
- 16:05It also questioned the pushrod: a 31 mm extension is too much,
- 16:09it says; the layout needs correcting, not the rod.
- 16:12We had said that too.
- 16:14The judge's point is: the diagnosis is right, but the remedy is wrong;
- 16:18if we reproduce the real layout, the rod goes back to 187.6.
- 16:21The judge says, "There's no 720° row in the piston table."
- 16:25But 720° and 0° are the same position; the engine completes its cycle
- 16:29and returns to the start, so that row already exists at 0°.
- 16:32It also says, "You haven't reported the minimum
- 16:36distance between the valve and piston."
- 16:39It was right about that, so we calculated it.
- 16:40They never touch.
- 16:42Its score: 6.5.
- 16:446.5 out of 10, guys.
- 16:45Basic geometry 9, kinematics 9, dimensional consistency 9; but valvetrain similarity
- 16:51to the real LS3 is 4, materials and mass 5, and dynamic validation 3.
- 16:56We got 6.5, but we didn't stop.
- 16:58We corrected all the judge's findings one by one.
- 17:01We tilted the valves toward the pushrod side like in the real LS,
- 17:06shortened the rocker arm, and the pushrod returned to its actual length: 187.6.
- 17:13We set the cam to 120°, made the bearing caps steel and the
- 17:16liners cast iron, and made the intake valve stem hollow.
- 17:19Meanwhile, AI caught 3 more mistakes on its own.
- 17:23First: as the new rocker arm rotated, its ratio
- 17:26increased, and the valve opened 4% too far.
- 17:29It recalculated the arms so they would be perpendicular at half lift.
- 17:32Second: the new liners extended too far down,
- 17:35and the crankshaft was hitting them.
- 17:37Third: the seat underneath the rocker arm was touching the angled valve stem.
- 17:42The result: the intake valve opens 14.007 mm; GM says 13.97.
- 17:49The difference is 3‰.
- 17:50There isn't a single collision among the rotating
- 17:52parts at 8 different crank angles.
- 17:54And we sent it back to the same judge.
- 17:56This time it really measured it: it found a rocker
- 18:01ratio of 1.6999996 from the STEP file.
- 18:05It measured the valve-to-piston clearance itself from
- 18:09the solid model: 2.8, the same as our report.
- 18:12But it caught something again.
- 18:13The real LS3 cam has different base circles for
- 18:16the intake and exhaust lobes: 19 and 19.3 mm.
- 18:20Both of ours are 15.9.
- 18:22We had estimated that.
- 18:24It's also stated in GM's own description; the judge is right.
- 18:27The second point is more subtle: "duration at 0.050 inch" on the cam
- 18:31card is measured at the lifter, but we measured it at the valve.
- 18:35So the cam duration matches on paper,
- 18:36but it's too short when measured at the lifter.
- 18:38The judge found this by calculating it from the formula alone, without seeing
- 18:42the code; we checked it with our own code and got exactly the same number.
- 18:46Also, the real LS3 rocker arm doesn't have
- 18:48a roller at the valve tip, but ours does.
- 18:51Its score: 7.5.
- 18:53From 6.5 to 7.5.
- 18:55The valvetrain went from 4 to 6.5, and materials from 5 to 7.
- 19:00Its final statement is: "I accept it as an LS3-based kinematic CAD
- 19:04model that has improved significantly over its previous version.
- 19:08However, I do not yet approve it as a geometric replica
- 19:11of an OEM LS3 or as a validated digital twin."
- 19:14The remaining findings are for the next version, guys: the cam base circle,
- 19:18duration measurement, and a rocker arm without a roller tip.
- 19:21They've all been noted.
- 19:23We're putting the engine files on GitHub, guys: the dimension
- 19:26table, each part's description, and the STEP files.
- 19:29We haven't put GM's document there, of course; the table lists
- 19:32where each dimension came from and which ones are estimates.
- 19:36The link is in the description.
- 19:37You can download the STEP files and open them in your own
- 19:41program: SolidWorks, Fusion, FreeCAD, whichever you use.
- 19:44If you try it, let us know in the comments.
- 19:45I'm not sharing the SolidWorks connector itself, the MCP, publicly, guys.
- 19:51We fed ISO standards into it; I was told that making it public in
- 19:54this form posed a copyright risk and that I shouldn't share it.
- 19:58That's also why I couldn't share the one from the previous AutoCAD video.
- 20:01But you can write your own MCP: as you saw in this video,
- 20:05you tell AI what you want, and it writes the code.
- 20:09Scan your own textbooks and notes and give them to it as sources on
- 20:13your own computer; you'll have a tool that works for your needs.
- 20:17One more thing: AI modeled this engine, but we provided the checklist,
- 20:20the dimension table, and the question, "Where did we make estimates?"
- 20:24Keep the engineering in your hands; AI is a tool.
- 20:27If you want more engineering videos like this, tell us in the comments.
- 20:32Which engine should be next?
- 20:33A boxer, a V12, or a diesel?
- 20:35Tell us in the comments.
- 20:36See you.
About this transcript
This page contains the full transcript of Yapay Zeka V8 Motor Çizdi by Baretli Kafa, generated from the public captions YouTube serves with the video. The transcript has 3,464 words across 392 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.