Automated Gantry-Based Disassembly System
Remanufacturing recovers value from worn-out products by taking them apart so their parts can be reused, but that disassembly is still done almost entirely by hand, because every product that comes in is a little different. My bachelor thesis, done together with DTU's national REMANUFACTORY research centre, asked a simple question: could one machine handle the repetitive part of that job and still hand the awkward cases back to a person?
Our answer was a unified tool head. It's a single 3D-printed module, driven by one shaft, that both unscrews fasteners and presses parts out without ever swapping tools. That's trickier than it sounds, since the same assembly has to deliver twisting force for the screws and pushing force for the press at the same time. We managed it with an internal spring that lets the head slide passively from screw mode into press mode as it presses down, with no motor and no reconfiguration. It even swaps its own screwdriver bits at a little parking station, using nothing but the axes the machine already has.
Getting it to work reliably turned out to be mostly a sensing and software problem. A camera finds every screw, corrects for lens distortion and viewing angle with a homography calibration, and reports each screw's position in millimetres along with its size (M4 vs M5). A laser distance sensor watches the head come down and can tell normal surface contact apart from the tiny extra bit of travel that means the bit has actually seated in the screw. A load cell keeps an eye on the press force and stops the instant it hits a set limit. Tying it all together is a state machine that checks the sensors at every step and either carries on, tries again, or flags the part for a human. The machine never just pushes blindly.
We deliberately built the prototype in 3D-printed PLA so a failed part could be redesigned and reprinted the same day, but dimensioned every component so the exact same geometry can be re-cut in 10 mm metal plate for a stronger production version, without moving a single sensor. Code base on GitHub →
- Camera screw localisation & sizing. Homography-calibrated vision reports machine coordinates for every screw and classifies M4 vs M5 from a single image.
- Passive automatic bit changing. The head swaps screwdriver bits using only the existing axes, and reached 100% reliability with the final station design.
- Sensor-driven safety. Force and laser feedback let the machine detect stalls and missed engagements and stop, rather than continuing blindly.
- Honest scope. Validated on purpose-built test objects; the report is clear about what still needs a metal redesign before industrial use.
- Full-system mechatronics. Taking one machine all the way from breaking the problem into functions, through mechanics, electronics and software, to results I could actually verify.
- Sensor selection and calibration. Load cell, laser distance sensor, and camera intrinsics plus pixel-to-mm homography.
- Control logic. Writing a state machine that keeps vision, motion and force feedback working together as one reliable cycle.
- Design for manufacturing. Building a 3D-printed prototype whose geometry carries straight over to machined metal.
- Making disciplined engineering trade-offs and stating limitations honestly instead of overselling results.
- Iterating fast: designing so a failed part could be reprinted and retested the same day.