Sharpening a knife well is a hand skill: a steady angle, the grits in the right order, and knowing when to stop. We wanted a compact, affordable machine that takes that skill off the user and offers three levels, sharp, sharper and sharpest. The design I developed holds the knife still and carries three abrasive belts of different grit along its edge on one rotating, indexed head, with one drive motor shared by the three belts. The prototype indexed between its stages, travelled the length of the blade and sharpened under supervision. Two parts of the plan did not make it into the time we had: an adjustable knife holder and AI-based tracking of the edge.
WITH
Lukas Fagerholm
WORK
Concept, CAD, printing and machining, electronics, embedded control
OUTCOME
Prototype built · sharpening demonstrated under supervision
Sharpening preview from the workshop. Original recording, 00:00–00:04.
01 / Architecture
One actuator for each motion.
I gave each motion its own actuator so that I could build and test one function at a time. A NEMA 17 stepper turns a T8 lead screw that carries the sharpening head along a linear rail. A small N20 gear motor turns the three-level gantry in 120° steps to select a belt. A second screw-driven slide pushes the belt motor forward to meet the shaft of the selected stage. Only the abrasive drive is shared: one brushed Pittman DC motor runs all three belts, so no high-speed motor has to ride round on the gantry.
Bringing the motions together was where the real work began: on the bench, stiffness, alignment and access mattered as much as the geometry in CAD.
Three abrasive levels · N20 indexing in 120° steps
ENGAGE
Screw-driven motor slide · positive coupling
SHARPEN
One Pittman DC motor · three belts
The built arrangement+
The belts are about 19 × 305 mm, and the feed screw advances 8 mm per turn; both are design values, not measured accuracy. I tried magnets and Hall-effect sensors to confirm which stage was in position. They were not reliable enough, so in the final assembly I still confirmed the position myself before engaging the drive.
Source: thesis, pp. 27–34.
02 / The drive interface
Less rotating mass. More exact alignment.
I kept the drive motor off the gantry so that no high-speed motor and its wiring had to rotate with the abrasive stages. The price is a mechanical handoff repeated at every stage change: the selected shaft and the motor have to line up before the belt can start.
I designed the coupling as a dog-clutch-style positive interface, printed in PETG with chamfered teeth, embedded repelling magnets and O-rings as compliant elements, so that the two halves would find each other instead of meeting tooth on tooth. It engaged with help. It never engaged reliably on its own, and that taught me the lesson the thesis ends on: fewer motors is not the same thing as a simpler machine.
The screw-driven motor slide approaches and withdraws from the coupling. Original presentation, 01:32–01:59.
Why the stopped state matters+
The intended sequence is stop, disengage, index, re-engage, then spin up. The coupling halves should meet at rest, because the energy of an engagement grows with the square of the speed difference. A complete controller would enforce that from confirmed states; my prototype relied on supervision.
Source: thesis, pp. 31–32; original presentation 01:32–01:59.
03 / Iteration in the workshop
The drawing met the load.
In the early test below, a hand drill drives a single belt, so I could see a first sharpening pass before the integrated drive existed. Belt tension then bent the printed gantry levels and tilted the pulley axes. I treated it as three separate problems: for stiffness I cut, drilled and bent metal reinforcement plates; for tracking I designed crowned pulleys; for traction I fitted O-rings to the drive pulleys.
Two ideas were simplified on the way. I first tested a fixed internal cam that tightens only the active belt through a moving idler; it worked on its own, but printed sliding parts and sensitive belt tracking made it too uncertain to carry into the final machine in the time we had. The knife holder went the same way. I built a rigid fixed-angle clamp that closes the load path into the base, but the adjustable holder I had designed, with control of angle and contact, was never finished, so in the filmed demonstrations I held the knife by hand.
An early one-belt sharpening test, driven with a handheld drill. Original recording, 00:29–00:38.
The gantry taking shape on the base, with printed parts and hand tools around the assembly.
Control and manufacture+
I printed the structural parts in PLA and the couplers in PETG, and did the drilling, cutting, filing, tapping and fitting by hand. I wrote the firmware for a NodeMCU-style ESP8266 that accepts serial commands from a browser console I made for the bench, driving one A4988 stepper channel for the feed and three DRV8871 DC channels for indexing, engagement and the belt drive. This was open-loop control, without automatic force or angle adjustment.
Source: thesis, pp. 30–35; first-run footage.
04 / Demonstrated, and still to develop
What it did, and what it still needs.
The machine indexed between its three stages, carried the head along the knife, ran the selected belt and sharpened under supervision. The footage on this page shows that work at the blade; none of it is a render standing in for a test.
Two goals were not completed. The AI-based edge tracking I had planned, so that the machine could follow and judge the edge on its own, stayed a separate image-processing experiment and never reached the controller. A force-to-cut rig to measure sharpness was built mechanically, but its electronics were never calibrated. So the project has no measured before-and-after sharpness. I can say the machine sharpened; I cannot yet say how well.
Supervised sharpening on the physical prototype. Original workshop recording, 00:37–00:42.
The force-to-cut rig: screw feed, guides and clamp. The mechanics were built; the measurement was never calibrated.
Outcome boundaries+
Belt speed, contact force and feed repeatability were never measured. The next version needs four things, in this order: sensing that confirms the stage and the engagement before anything moves, a holder with angle and contact adjustment, a calibrated sharpness measurement, and then the edge tracking. I would also compare the single shared motor against one small motor per belt; three fixed couplings would remove the alignment problem altogether.
Source: thesis, pp. 36–42; workshop footage 00:37–00:42.
Guided scroll · scroll or press Escape to take over
