![[winch hyperlapse.gif]] ## The Challenge This winch was part of a larger project where I served as principal engineer to bring pultrusion manufacturing in-house. The process is now transitioning to production under a dedicated manufacturing engineer, but I still design support equipment and solve technical problems as they come up. One of those problems: we needed a custom winch to pull carbon fiber through 130 feet of pultrusion machine. Sounds simple? Not quite. The motor needs to run at 2% of its rated speed (NOT easy), the electronics enclosure needs to survive an environment filled with conductive carbon fiber dust, and the whole system needs to integrate with existing safety circuits on a complicated machine I didn't design. Oh, and it needs to produce enough torque to pull a literal ton of material while operating smoothly enough not to tear delicate fibers. Welcome to the Pultrusion Winch project—10 months of mechanical design, VFD tuning wizardry, and more FEA iterations than I'd like to admit. --- ## What I Built ![[20251209_073009.jpg|500]] A capstan-style winch system with: - **Aluminum extrusion frame** (FEA-validated for motor torque loads and T-nut pullout forces) - **VFD-controlled 3-phase motor** running sensorless vector control at ultra-low speeds - **50-foot remote pendant** with shielded analog control signals surviving next to noisy motor cables - **Vortex cooling system** keeping the VFD alive in a dusty environment without clogging filters - **Dual-channel safety integration** linking emergency stops across two independent machines - **Custom fabricated components:** welded sheet metal brackets and floor mounts, milled/bent/riveted polycarbonate safety enclosure --- ## Concepts ![[Pasted image 20250409090854.png|500]] Originally, I heavily explored the concept of using a capstan drum. Capstans are a beautiful and fascinating example of "primitive" engineering that have stood the test of time. Most often seen in nautical applications, the amount of tension the drum can hold vs the amount of tension you need to pull on the loose end increases exponentially for every wrap around the drum- the diameter of the drum is completely absent from the equation. > [!quote] Fun Fact From Wikipedia > ...(from 5 turns around a capstan with a coefficient of friction of 0.6) means, in theory, that a small child would be capable of holding (but not moving) the weight of two [USS _Nimitz_](https://en.wikipedia.org/wiki/USS_Nimitz "USS Nimitz") supercarriers (97,000 tons each, but for the child it would be only a little more than 1 kg). As there is much less structure to a bundle of loose carbon fiber than there is a rope, I anticipated the need to prevent the bundle from flattening across the drum under tension. Below is a concept of a "guided" capstan drum with turned channels to prevent flattening of the bundle ![[Pasted image 20250422101823.png|250]] ![[Pasted image 20250422102113.png|250]] ![[Pasted image 20250422101921.png|250]] While an interesting design challenge, I opted to go with an off the shelf capstan drum used for pulling fiber optic cabling through underground conduit --- ## The Fun Problems (and How I Solved Them) ### Problem 1: The Motor That Didn't Want to Start ![[20251106_152911.jpg|500]] **The Issue:** First wet pull test, the motor slipped at barely 0.2 tons of load. Not great when you're spec'd for 1 ton. **The Detective Work:** - VFD was using default motor current settings (5.2A) - VFD was under-magnetizing the motor by 21% **The Fix:** Updated motor current parameters, zeroed out slip compensation (it fights with vector mode), re-ran the VFD auto-tune. Motor immediately had way more authority. ### Problem 2: The Motor That Wouldn't Stop Shaking **The Issue:** After the current fix, the motor started *hunting*—oscillating at about 1 Hz like it couldn't make up its mind about speed. Got worse under load. **The Diagnosis:** Too much speed loop gain. I'd made the VFD's control loop too aggressive, so it was overcorrecting and creating a mechanical oscillation. **The Fix:** Dialed back the gain parameter. Smooth as butter. ### Problem 3: Carbon Fiber Dust Wants to Kill Electronics ![[20260105_140138.jpg|500]] **The Issue:** Carbon fiber dust is conductive. Standard cooling fans would suck it into the VFD and create short circuits. The internet in this building goes out every few months because the conductive fibers blow up the router. Not ideal. **The Solution:** Vortex cooler. Uses compressed filtered air to create a cold air stream rotating at almost 1-million RPM *and* positive pressure inside the enclosure. When the motor runs, we're actively pushing air *out* so dust can't infiltrate. When it's idle, everything's sealed up tight with gaskets and IP65 cable glands. Added bonus: no filters to clog or maintain. It just works. ### Problem 4: 50 Feet of Analog Signal in an EMI Nightmare **The Issue:** Operators need to control speed from a pendant 50 feet away, right next to the die. That means running a 0-10V potentiometer signal through a cable that's literally zip-tied next to VFD motor power cables. VFD motor cables are *loud*—electrically speaking. **The Physics:** - Voltage drop? Negligible (the math: 0.003V over 50 feet) - Noise coupling? **Big problem** ![[Pasted image 20250611090107.png|500]] **The Solution:** - EMF optimized component placement in enclosure; proper spacing and minimal parallel wire runs - 12-conductor shielded SOOW cable - Shield connected to VFD ground *at VFD end only* (prevents ground loops) - Pendant end shield isolated - All signal pairs twisted inside the shield Documented backup plans (4-20mA current loop, motor-operated potentiometer) in case the shielding wasn't enough. Turns out it was. Sometimes you get lucky. --- ## The VFD Tuning Saga ![[20260105_142213.jpg|500]] Running a motor at **1.5 Hz** (35 RPM at the output shaft) is not what VFDs are designed for. At that frequency, the motor's internal resistance dominates everything and normal V/Hz control falls apart. **The solution:** Switched to sensorless vector control and spent weeks iteratively tuning motor parameters, voltage boost settings, torque limits, and control loop gains. By December I had a 177-line spreadsheet tracking every test, every setting, every failure mode. On January 5th, after some final tweaks over the holiday break, **it worked**. Full pull, smooth handoff, no oscillation. --- ## The Safety Integration Puzzle ![[20260105_142517.jpg|500]] ![[Pasted image 20260105151057.png|500]] ![[Pasted image 20260105151121.png|500]] The winch needed to play nice with existing safety circuits I didn't design. This involved an exhaustive reverse-engineering of the existing circuit. Final architecture: - **Bidirectional EMO:** Hitting the winch emergency stop shuts down the pullers. Hitting the puller EMO shuts down the winch. - **Guard interlock:** Opening the guard stops the motor but *doesn't* trigger a puller shutdown (prevents nuisance trips when threading fiber). - **Dual-channel monitoring:** Redundant safety relay watches both EMO circuits, requires manual reset after fault. All of this needed to work with the VFD's fault input logic and the puller's existing safety relay. Wiring and debugging multi-system safety circuits is like solving a logic puzzle where mistakes can hurt people—so you triple-check everything. --- ## The Structural Analysis That Saved Us ![[Pasted Image 20250709095316_912.png|500]] ![[Pasted image 20250408085136.png|500]] About halfway through the design, I ran an FEA study on the motor mount. The frame itself could handle the forces but the T-nuts holding the motor mounting plate were showing concerning stress concentrations under full motor torque. **The numbers:** Motor could produce enough torque to literally rip the T-nuts out of the extrusion. That would've been a very expensive and very loud failure. **The fix:** Reinforced mounting plate, increased fastener count, thicker gauge material. Re-ran the FEA, numbers looked good. Dodged that bullet before we ever cut metal. This is why you validate designs *before* you build them. --- ## Fabrication & Assembly **Aluminum Extrusion Frame:** - FEA-validated for motor torque reaction loads - Modular design for assembly and future modifications - Anchor point analysis for floor mounting ![[20250813_122156.jpg|250]] **Welded Sheet Metal Components:** - Motor mounting brackets (reinforced after FEA study) - Floor anchor mounting feet - Cable management brackets - 14-gauge CRS for structural components, 16-gauge for lighter parts ![[20260105_142632.jpg|250]] **Polycarbonate Safety Enclosure:** - Milled, bent, and riveted construction - Integrated guard interlock switch mounting - Access panels for maintenance **Electronics Enclosure:** - IP65-rated cable glands for sealed penetrations - Vortex cooler mounting with compressed air plumbing and pressure equalizing check valves - DIN rail layout for VFD, safety relay, and terminal blocks --- ## What I Learned **Technical Skills:** - VFD sensorless vector control is black magic until you understand motor equivalent circuits - Shielding only works if you ground it correctly (single-point grounding saves lives) - FEA is cheaper than replacing broken equipment - Vortex coolers are underrated for dirty environments **Soft Skills:** - You can't test everything on your desk—sometimes you need to commission it in the field and adapt - Detailed logging pays off when you're troubleshooting (that 177-line VFD table was worth its weight in gold) - Safety integration requires careful planning when modifying existing systems - "It should work in theory" and "it works in production" are very different statements **The Big One:** Ultra-low speed motor control is *hard*. We're talking 2% rated speed (almost too slow to provide feedback) with high torque demand in a sensorless configuration. That's not a textbook problem—that's a "read the VFD manual cover-to-cover and iterate until it works" problem. And it was incredibly satisfying when it finally did. --- ## The Outcome The winch is now transitioning to production operations. The winch reduced the amount of people required for production setup from 3 to 2, and reduced setup time by 45 minutes a run plus the labor of the extra person who is now free to do other tasks. With the pultrusion process being handed off to a dedicated manufacturing engineer, I'm back to focusing on product design work. I started my career turning wrenches so I always enjoy opportunities to use my other non-design skills and get my hands dirty.