Speed Control of Universal Motors in Sewing Machines

With the two controller designs for the VSM Base

Claude (Anthropic), at the direction of Robert Drzyzgula

Development Build v82, November 2026

1 Summary

About this paper. The text, analysis and figures were written by Claude, an AI model made by Anthropic, at the direction of Robert Drzyzgula, who set its scope and questions, supplied the measurements, photographs and hardware, made the design decisions for the VSM Base controllers, and reviewed it. The sources it draws on are listed in the last section; please report errors as issues on the project’s GitHub repository.

A sewing machine’s universal motor is a series-wound machine whose speed depends strongly on load, so the controller matters as much as the motor. A pedal that only lowers the voltage (rheostat, carbon pile, variac, simple triac) leaves speed at the mercy of the load. Only a controller that measures speed and corrects it (back-EMF or tachometer feedback) holds speed when the needle hits a seam.

The carbon pile wins among the open-loop pedals for two reasons. Its resistance varies continuously and steeply with foot pressure, which gives fine control at the bottom of the range. And because its resistance falls as you press, the operator closes the feedback loop with their foot faster than they consciously notice the machine slowing.

This paper covers the physics of the series motor, then each control method: series rheostat, carbon pile, open-loop triac phase control, variac, and feedback controllers including KB Electronics boards and MCU designs. It closes with the two controller designs for the VSM Base: the knee controller built around a repurposed foot-pedal triac board, and a microcontroller phase controller now in design.

A companion paper, Inside the Motor, looks at how these motors are built and how each part maps onto the model used here. It also compares the universal motor with wound-field DC, permanent-magnet and brushless motors and their controls, and shows the waveforms of each control method on the wire. It closes by asking whether a small DC motor could drive a vintage machine from a version of the MCU controller.

A second companion, Inside a Switching Circuit, looks inside the controls: the thyristor and triac, a phase-control pedal circuit part by part, the snubber, the donor board of the knee controller in detail, and the transistor switches (MOSFETs and IGBTs) behind PWM drives and variable-frequency drives.

2 The series motor and why its speed sags under load

In a universal motor the field winding is in series with the armature, so one current II makes both the field flux and the armature current. Torque is flux times current, so below saturation torque rises with the square of the current (Wikipedia: Universal motor). The same series connection is why it runs on AC: field and armature current reverse together, so torque keeps its sign.

A simple steady-state model, with flux proportional to current and the back-EMF proportional to flux times speed:

T=KI2,V=IR+KIωT = K I^2, \qquad V = I R + K I \omega

Here VV is the voltage at the motor terminals, RR the total winding and brush resistance, ω\omega the speed and KK a motor constant. Eliminating II gives the speed–torque curve:

ω(T)=VKT−RK\omega(T) = \frac{V}{\sqrt{K T}} - \frac{R}{K}

Three consequences matter for sewing:

  1. Speed is set by voltage, not commanded. At a fixed torque, speed rises roughly linearly with VV. No pedal sets speed directly; every open-loop pedal sets a voltage, and the load decides the speed.
  2. Speed falls steeply as load rises. The slope is dω/dT=−V/(2KT3/2)d\omega/dT = -V / (2\sqrt{K}\,T^{3/2}), steepest at light load. Series motors slow down as load increases and race when unloaded (Wikipedia; Fay, Motorola AN-443).
  3. Low speed is the fragile region. Writing the fractional sensitivity as (dω/ω)/(dT/T)=−12ω0/ω(d\omega/\omega)/(dT/T) = -\tfrac{1}{2}\,\omega_0/\omega, where ω0=V/KT\omega_0 = V/\sqrt{KT} is the speed with no resistance, shows it grows without bound as ω\omega approaches zero. When most of the voltage is spent on IRI R rather than back-EMF, a small extra load stalls the motor.

A sewing machine’s load is not steady. The needle piercing fabric, the take-up lever, and the hook and feed cams produce a torque pulse every stitch, and a seam crossing adds a step. At full speed the flywheel and armature inertia ride through these pulses. At a few stitches per second the motor sits in the fragile region of point 3, and each pulse produces the familiar surge-and-stall behavior. Each control method below differs mainly in how it handles that region.

3 Series resistance pedals (rheostats)

A wire-wound or stepped-contact rheostat puts a variable resistance RsR_s in series with the motor on the full line voltage VlineV_{line}. It is the worst-regulating of the common methods, because it lowers speed by adding resistance rather than lowering voltage.

With RsR_s in the circuit the speed–torque curve becomes:

ω(T)=VlineKT−R+RsK\omega(T) = \frac{V_{line}}{\sqrt{K T}} - \frac{R + R_s}{K}

The slope dω/dT=−Vline/(2KT3/2)d\omega/dT = -V_{line} / (2\sqrt{K}\,T^{3/2}) does not depend on RsR_s at all. Turning the pedal down shifts the curve toward zero speed but keeps the full-speed slope, so a given torque pulse costs the same absolute rpm at a crawl as at full speed. As a fraction of the remaining speed, the loss is enormous. A voltage-reducing control (variac, phase control) also lowers the slope, because it scales VV.

Three further problems follow from the same equation:

4 Carbon-pile pedals: why they feel best

Electrically, a carbon pile is just another series resistance, so it inherits every load-regulation weakness of the rheostat above. Its advantage is entirely in how the resistance is commanded. The pile turns foot force, not foot position, into resistance, and does so continuously over a very wide ratio.

How it works. A stack of graphite discs is squeezed by the pedal through a spring. Resistance lives almost entirely at the disc-to-disc contacts: more force flattens more microscopic asperities into contact, so conducting area grows and resistance falls (UQ Physics Museum). Measurements on voltage-regulator piles fit a power law, R=nC/PmR = nC/P^m, for nn discs under force PP with constants CC and mm (Hayashi, Tanso 1953). Sewing pedals run from several hundred ohms at light pressure to a few tens of ohms at heavy pressure. A brass contact then shorts the pile for full speed (softsolder.com; EEVblog forum).

Four properties explain the feel:

  1. Continuous and logarithmic. A power law means dR/R=−mdP/PdR/R = -m\,dP/P: each equal fractional increase in force gives an equal fractional drop in resistance. Human force perception is also roughly proportional (Weber’s law), so the pedal spreads its resolution evenly across what the foot can distinguish. The bottom of the range, where the motor is most sensitive, gets the same share of resolution as the top. A linear wire-wound rheostat crowds the useful low-speed region into a few millimetres of travel.
  2. Force, not position. The pile and its spring are stiff, so the pedal barely moves. The foot works almost isometrically, and adjusting force needs no limb movement and no travel time. I believe this is a large part of why the pile feels quick, though I found no study measuring it for pedals.
  3. The operator is the feedback loop. When a seam slows the machine, the sewer hears and feels it and presses harder within a few hundred milliseconds. The pile responds instantly, with no mechanical lag between force and resistance. In effect the carbon pile plus the human is a closed-loop controller, and the pile’s fine low-end resolution is what makes that loop stable at a crawl.
  4. No steps, no gap. Resistance changes without breaking the circuit, so there is no contact arcing during speed changes and no jump between taps.

One further effect is plausible but I could not confirm it in the sources: granular carbon contacts are mildly non-ohmic, and their resistance tends to fall as current through them rises. If so, a load that slows the motor (raising current) would slightly lower the pile’s resistance and partly compensate. This is a small IR compensation built into the material. Treat it as a hypothesis to test with a bench measurement, not as established.

The drawbacks are the rheostat’s: heat (the pile can run hot inside its case), weak starting torque at light pressure, and drift as discs wear and the stack loosens. Worn piles give long travel before anything happens, then abrupt starts (softsolder.com).

5 Triac phase-angle control (open loop)

Modern electronic pedals, including the boards examined for this project, are phase-angle controllers. They lower the average voltage by switching, not by burning power in a resistance, so they run cool. Their speed regulation, though, is roughly that of a variac, and their low-speed behavior is worse.

The companion paper Inside a Switching Circuit takes these circuits apart: the thyristor and triac themselves, a minimal pedal circuit part by part, the snubber, and the donor board of §11 in detail. This section keeps to what the method means for the motor.

How it works. A triac sits in series with the motor. Each half-cycle, an RC network charges through the pedal’s resistance until the capacitor reaches the diac’s breakover voltage (about 30 V). The diac then dumps a pulse into the triac gate, and the triac conducts until the motor current falls to zero, a little after the voltage zero because the motor is inductive. Delaying the firing angle α\alpha reduces the share of each half-cycle applied to the motor (Skalka, Freescale AN1662). For a resistive load the RMS output is:

Vrms=Vpk12π(π−α+12sin2α)V_{rms} = V_{pk}\sqrt{\frac{1}{2\pi}\left(\pi - \alpha + \tfrac{1}{2}\sin 2\alpha\right)}

The motor’s inductance stretches the conduction past the voltage zero, so the real curve differs, but the shape holds: a smooth fall from full voltage at α=0\alpha = 0 to nothing at α=π\alpha = \pi.

What it means for sewing:

6 Variac (variable autotransformer)

A variac is the clean reference case: a true, nearly lossless, sinusoidal voltage source you can set anywhere from zero to slightly above line. A brush slides over the exposed turns of a single winding to tap a continuously variable fraction of it (Wikipedia: Autotransformer). Variac was General Radio’s trademark from 1934 to 2002 and is now used generically.

For a universal motor it delivers exactly the model in the physics section with VV reduced: speed–torque slope and stall torque both fall with VV, but there is no added series resistance and no chopping. That makes it the best of the open-loop methods for smoothness and regulation:

As a foot control it fails on practical grounds: it is heavy, its brush is not built for constant motion under the foot, and it is not isolating (one side of the output is tied to the line). A motor-driven variac would work but responds too slowly for sewing.

7 Feedback controllers: holding speed under load

Every method so far sets a voltage and lets the load set speed. A feedback controller measures speed, compares it with the pedal’s command, and fires earlier when the motor slows. This is the only class that fixes the low-speed weakness at its root: in the fragile region it simply adds voltage until the stitch is through. There are three ways to measure speed.

1. Back-EMF sensed between pulses (half-wave SCR). In a half-wave circuit the motor carries no current on alternate half-cycles. During that gap the spinning armature, in the field’s residual magnetism, generates a small voltage proportional to speed. The firing circuit compares this against the pedal’s reference: a slower motor means a lower back-EMF and an earlier trigger. This trick, used in 1960s–70s drills and some sewing pedals, gives real regulation with a handful of parts. Its limits are half-wave operation (half the power, more torque ripple) and a weak, temperature-dependent residual field. Silicon Chip’s regulated universal-motor controller uses a refined form of the same principle (Silicon Chip, Oct 2002).

2. Armature voltage with IR compensation (full-wave SCR DC drive). A rectifying DC drive powers the motor from a full-wave bridge and regulates the armature voltage. It then adds a correction proportional to current, because the IRI R drop in the windings is the main reason speed falls with load. KB’s KBIC is an example: rated for shunt, PM and AC/DC (universal) motors, with MIN, MAX, ACCEL, IR and current-limit trimpots, and a load regulation of 1% of base speed over a 50:1 range (KB KBIC data sheet D-210). For a series motor the regulation is less exact than for a PM motor, because the back-EMF depends on current as well as speed (E=KIωE = K I \omega), so IR compensation is a first-order correction. In practice it still transforms low-speed behavior.

3. Tachometer (closed-loop phase control). A shaft sensor gives speed directly, and a controller adjusts the triac firing angle to hold it. Washing machines do this with dedicated ICs such as onsemi’s TDA1085C, a phase-angle triac controller with a tachometer input, frequency-to-voltage converter, ramps and current limiting (onsemi TDA1085C). An MCU can do the same with a PI loop; Freescale’s AN1662 is a worked example using a 16-pole tachometer and a triac (Skalka, AN1662). This is the most precise method, but it needs a sensor on the motor or the handwheel.

For the sewer, all three convert the pedal from a voltage command into a speed command. At low speed the machine punches through a seam at the speed you asked for instead of stalling, which is the “full power at slow speed” quality of good electronic machines.

8 KB Electronics controllers

Of KB’s catalog, the KBIC family is the one that fits a sewing-machine universal motor: a full-wave SCR DC drive with IR compensation that is explicitly rated for AC/DC motors. The others are either open-loop or rated only for PM and shunt motors.

Model What it is Motors Speed feedback Fit for the 99
KBIC Full-wave SCR DC drive; MIN, MAX, ACCEL, IR, CL trimpots; auto-inhibit; plug-in horsepower resistor Shunt, PM and AC/DC (universal) Armature voltage + IR compensation; 1% of base speed, 50:1 range Best KB option
KBMM Full-wave SCR DC drive with ACCEL and DECEL ramps Shunt and PM only Armature (1%) or tachometer (1%) Not rated for universal motors
KBWC Triac wall-box fan control; minimum-speed trimpot; RFI filter Shaded pole, PSC, AC/DC None (open-loop phase control) No better than a stock pedal
KBSI-240D Signal isolator: 0–25/120/550 V or 4–20 mA in, up to 10 V out Accessory Not applicable Isolates a pedal or MCU from the drive

Using a KBIC on the 99. The motor runs on rectified DC rather than AC, which a universal motor accepts. Set up this way:

Isolation. KB DC drives are not isolated: their speed-pot terminals are not isolated from the line, which is why KB sells the KBSI-240D to interface non-isolated drives with external signals (KBSI-240D). A pedal wired straight to a KBIC is therefore live. The KBIC also accepts an isolated 0–9 V DC signal in voltage-following mode (KBIC D-210). Feeding that input from an isolated pedal or MCU is the clean way to keep the pedal at a safe voltage.

9 MCU control: what a from-scratch design looks like

An MCU phase controller with a speed sensor can match a KBIC’s regulation while keeping AC drive, an isolated pedal and room for features. The architecture is the one in Freescale’s AN1662, with isolation added (Skalka, AN1662).

Power stage (line side).

Control side (isolated, low voltage). Everything the user touches lives here: pedal, panel switches, speed-limit knob. That removes the live-pedal problem entirely.

Speed sensing, in order of preference:

  1. Handwheel or motor-pulley sensor. A Hall sensor and one or more magnets, or an optical slot. One pulse per handwheel turn is one per stitch, enough for regulation at sewing speeds, and it also gives needle position.
  2. Back-EMF sampling. With phase control there is a dead interval each half-cycle once the current stops. Sampling motor voltage there gives a residual-field back-EMF, as in the analog half-wave designs. It needs no sensor but is noisy and needs calibration per motor.

Loop. One firing decision per half-cycle means 120 updates per second on 60 Hz. A PI loop at that rate gives a closed-loop bandwidth of roughly 10–20 Hz, comfortably faster than a sewer’s foot. A rate limit on the firing angle stops a jammed needle from receiving a sudden full-power pulse, and a stall watchdog (no pulses for some time) cuts the triac, as AN1662 does.

Emulating the carbon pile’s feel. The pile’s virtues (force control, logarithmic response, fine low end) can be reproduced on purpose. Put a force sensor (a small load cell or FSR) under a stiff pedal, take its logarithm, and use that as the speed command. A position pedal (Hall or pot) with an exponential mapping is the cheaper version. Unlike the pile, the command is now speed, so the machine holds it through a seam without the operator compensating.

Features the sensor makes cheap: needle-up/needle-down stop, a slow-start ramp, a hard speed limit for beginners or free-motion work, and a single-stitch button. Several 1970s–80s electronic domestic machines offered similar functions with analog circuits.

10 How the methods compare at the needle

Only the feedback methods hold speed through a seam; among open-loop pedals, the carbon pile has the best command resolution and the variac the best regulation and smoothness.

Method Pedal commands Low-speed resolution Holds speed under load Smoothness at a crawl Heat in controller Pedal at line voltage
Stepped / wire-wound rheostat Resistance (by position) Poor: steps, crowded low end Worst Poor High Yes
Carbon pile Resistance (by force, log-like) Excellent Poor (operator compensates) Good High Yes
Variac Voltage (by position) Good Fair Excellent Negligible Not isolated
Triac, open loop Firing angle (by position) Fair: snap-on hysteresis Fair Poor: pulses, cogging Low Yes, if the pot is in the pedal
Half-wave SCR, back-EMF Speed (approx.) Good Good Fair: half-wave ripple Low Usually yes
KBIC (DC, IR comp) Speed (approx.) Good Good: ~1% of base speed Good Low Yes, unless isolated input
MCU + tachometer Speed Whatever the pedal sensor gives Best Good Low No (isolated by design)

The ratings are qualitative, drawn from the physics above and the cited regulation figures; they are not bench measurements.

Model speed against load torque, with all three set to the same slow speed at nominal load. The resistance pedal and the reduced-voltage controls start at the same speed, but the resistance pedal keeps the full-line slope and stalls at about 1.6× load, while the reduced-voltage curve only halves its speed by 2×. Computed from \omega = V/\sqrt{KT} - R/K with K = 1, R = 0.1 (tools/make_speed_chart.py); illustrative, not measured.

11 The controllers in the VSM Base

The base’s right-hand bay can hold a speed controller (compartment_option = "knee"). Two designs are in progress. The first is built and modeled. The second is in design. This section is kept up to date as they develop. The base’s User’s Guide has the build details for the first.

11.1 The knee controller: a repurposed pedal board

What it is. An inexpensive foot-pedal speed-control board (the XWW KB3151C, 75 × 35 mm) moved into the bay, with a knee lever in place of the pedal. A bar hangs below the table edge and turns a shaft running front to back in the bay. Two gear stages, about 11.75:1, turn a Bourns 91A1A-B28-A25L pot (1 MΩ linear cermet) that replaces the board’s slide pot, VR1. A snap switch (Omron SS-5GL), held open by a cam on the shaft’s sector through the first inch of travel, cuts the motor.

What kind of controller it is. Electrically it is the open-loop triac phase control of §5: a BT136 triac, a DB3 diac and an RC timing network, with the pot setting the firing angle. It therefore behaves like a good stock pedal. It gives no speed regulation; the load sets speed, with the cogging at a crawl and the snap-on hysteresis described there. The 2 MΩ trimmer RV2 sets the minimum speed (below). The knee mechanism improves the command, not the regulation:

The donor board, KB3151C (XWW speed controller, 110–120 V, 1.0 A), traced from photos; KiCad source in hardware/triac-board/. J1 and J2 are its two spade tabs: the board is a two-terminal switch in series with the motor, and the triac’s tab is MT2 (live). Timing: from MT2 through RV1’s pin-1 section, in parallel with RV2 plus RV1’s pin-3 section, to the wiper, then through R2 into C2 and back to MT1. When C2 reaches the DB3’s breakover it fires the BT136 through R1. RV2 lies across RV1’s whole track. At RV1’s pin-3 end is an off zone where the wiper is open and nothing fires; moving toward pin 1 the resistance falls toward zero and speed rises, and RV2, by capping the resistance just past the off zone, sets the slowest speed (as long as it is at least RV1’s 200 kΩ). R3 and C1 are the snubber across the triac. Still to check with a meter: that RV2’s wiper is tied to its pin-3 end, that RV1’s pin 1 and R3 go to the MT2 tab, and which of RV1’s pins is the wiper.

The board is a two-terminal series device. It has only two spade tabs and sits in series in one leg of the motor circuit; it has no separate mains input or motor output. In the original pedal, only this control loop enters the box, as a two-wire lead from a standard cordset. The cordset’s junction (a three-pin machine plug or a duplex outlet) joins it to the mains cord, and cordsets are interchangeable, so one pedal fits many machines.

The trimmer RV2 is a 2 MΩ resistor around the slide pot. Its setting decides where in the pedal’s travel the motor snaps on: set high, a quarter of the travel does nothing; set near 430 kΩ, the motor starts as soon as the pedal leaves its off zone; set below the pot’s own 200 kΩ, the pedal stops being monotonic. In the knee controller, where the pot is wired so that RV2 lies across its working section, about 360 kΩ makes the motor start as the off switch closes. Inside a Switching Circuit (§6 there) works the parallel resistance through and simulates the pedal across RV2’s range. Either way, set it by ear.

Wiring the pot, in the board’s own terms:

It is set at full speed: with the bar held at 2”, the wiper is put just at the pin-1 end (0 Ω) and the pinion clamped. At rest it then reads about 445 kΩ. From full speed to the stop the pot turns another 13° into the dead zone past its electrical end, well short of its mechanical stop. The pot’s plastic shaft and the printed pinion are the isolation between the touchable bar and mains, with the pot’s 1,500 VAC rating between element and bushing; a metal-shaft pot must not be substituted.

Power. Three ways in (knee_power), all at mains potential:

Limits. The board is labeled 1.0 A. Its BT136 is a 4 A part but has no heatsink, so a motor over 1 A (one here draws 1.3 A) needs a clip-on heatsink, which is live with the triac’s tab, and vents in the cover. Sewing-machine universal motors here are 110–120 V AC/DC machines, typically 0.5–1.0 A.

What it leaves on the table. Everything the feedback sections describe: it holds no speed through a seam, and at a crawl it pulses. Its virtue is that it is cheap, uses a common part, and gives a vintage machine a knee control in its own base.

Low-end control. With the board’s RC firing network, speed is far from linear in the pot’s resistance: the range from stopped to sewing speed is crowded into a small slice at the high-resistance end. A pot that moves linearly with the knee therefore puts most of the useful control into a fraction of the speed range. Three ways to spread it, none tested yet:

The last can be fitted exactly to a bench measurement of the board’s speed against resistance on a 99 motor, which is the next step. Until then the gearing is linear in resistance.

11.2 The MCU phase controller (in design)

The second design is a feedback controller of the kind described in §9: a microcontroller phase controller with an isolated control side. As it stands:

11.2.1 A first concept circuit

This is a concept, not a design. The circuit below has been drawn but not built, simulated or tested. Its part values are first estimates, not yet checked against the parts’ datasheets. The core board is drawn by GPIO number, not as any particular board. The firmware does not exist. It shows how the pieces of §9 fit together around the chosen board, and is a starting point for bench work, not something to build from. It is drawn by Claude at Bob Drzyzgula’s direction.

The circuit follows the architecture of Freescale’s AN1662: a triac in series with the motor, fired once per half-cycle at a delay measured from the mains zero crossing, with a speed sensor closing the loop. AN1662 runs its microcontroller at line potential, with a resistor from the mains straight to an input pin for zero-cross detection. Here every connection between the mains and the controller passes through an isolating part, so the controller, its pedals and its debug port are all at a safe low voltage. The drawing runs the isolation barrier across the sheet, with the live side above it.

Line side (live). An IEC inlet with earth feeds a 2 A slow fuse and a 150 V MOV. The base’s duplex outlet keeps its two sockets: the light’s is always on, the motor’s is controlled. The motor’s hot lead runs from a motor-enable relay (K1) through the triac (a BTA08-600BW, a snubberless type) to the outlet. The relay lets the firmware cut the motor whatever the triac is doing: it closes last and opens first. Space is left for an RC snubber, to be fitted only if the triac fires falsely or fails to latch (Inside a Switching Circuit, §5, explains both).

Across the barrier. Four parts span it, and an optional fifth:

Control side (low voltage). The RP2350B core board takes 5 V from the module into its VSYS pin through a Schottky diode, so its USB port can stay plugged in; its own regulator supplies 3.3 V to the sensors. A 2N7002 drives the relay coil. The opto-triac’s LED is driven straight from a GPIO set to 8–12 mA drive. The zero-cross output has a pull-up and a small filter capacitor. The hand-wheel index sensor is a DRV5023 Hall switch; one magnet gives one pulse per stitch. An optional encoder beside it, two Hall latches over a ring of alternating magnets, tells the firmware the hand wheel’s angle to 15° (see below). The knee lever, a travel pedal, a pressure pedal (a force-sensing resistor) and a speed-limit knob go to four analog inputs through RC filters. A needle-up/down toggle, a single-stitch button and a heel switch in the pedal use the microcontroller’s internal pull-ups.

Signal RP2350B pin Use
Triac fire GP4 gate pulse (timer alarm or PIO)
Relay GP5 output
Zero cross GP6 interrupt
Hand-wheel encoder GP2, GP3 quadrature count (PIO)
Hand-wheel index GP7 edge timestamps
Single stitch GP10 input; held, teaches a stop
Heel switch GP11 input
Needle up/down GP12 input
Knee, pedal, pressure, limit GP40–43 (ADC0–3) analog inputs
Motor current (optional) GP44 (ADC4) analog input

ADC5–7 are spare, and GP8/GP9 (UART1) are left for a debug line, though the USB port’s serial link will usually do. Any GPIO can serve for the digital signals, so these are a starting point, to be placed to suit the board chosen. A candidate is WeAct Studio’s RP2350B Core board, about $6, which brings out every GPIO including all eight analog inputs; its USB supply already passes through a diode, so the supply module can feed its 5V pin directly.

Firmware, in outline. As in AN1662:

What a sewist asks of a pedal. Danielle, who has sewn for more than five decades and works with Bob Drzyzgula on repairing and re-homing thrift-store machines, read a draft of this paper and described what she values in a foot control. Her observations, which she agreed to share, set part of this controller’s brief:

The circuit takes these up with a heel switch in the pedal and an encoder at the hand wheel. In the firmware:

The encoder is there because one magnet is not enough. A single index pulse gives the hand wheel’s angle only while it turns, by time since the pulse, so a controller with only the index can’t learn a stop from a hand wheel turned by hand, or check where the needle landed. Two Hall latches placed a quarter-cycle apart over a ring of twelve alternating magnets give 24 counts a turn, 15° each, in either direction, counted by a small PIO program; the index pulse sets and checks the count once a turn. Without the encoder the controller still stops by the index alone, more roughly.

Two practical consequences. Because the control side is isolated, the board’s USB port can stay connected to a computer with the mains on, for flashing (by dragging a UF2 file onto it in BOOTSEL mode) and for the debug serial line; step-through debugging needs a separate SWD probe, such as Raspberry Pi’s Debug Probe. And as with the test rig, a scope’s ground must never be connected to the line side.

The MCU phase controller as a first concept (not built or tested). The isolation barrier runs across the sheet, with the live side above it. Line side, left to right: the IEC inlet and fuse, the surge suppressor, the outlet, the triac stage, and the line sides of the opto-triac (U3), the motor-enable relay (K1), the zero-cross optocoupler (U2), the 5 V supply (PS1) and the optional current transformer (T1). Control side: the triac’s LED drive, the relay driver, the zero-cross input, the RP2350B core board, the operator inputs and heel switch (left), the switches, the hand-wheel index sensor and the optional hand-wheel encoder (right). The KiCad file and its generator are in hardware/mcu-controller/.

The KiCad drawing, a printed copy and the script that generates them are in the repository at hardware/mcu-controller/. The power stage, sensing and control loop will be settled on the bench, and this section will be revised as they are.

11.2.2 Fitting the encoder

Fitting an encoder is an engineering exercise on every machine, and on some it won’t be practical. Where it is, it gives the controller the needle’s angle at all times, and with it taught stops and the heel tap. The electronics are the same everywhere: two Hall latches a quarter-cycle apart over a ring of magnets of alternating pole, plus an index. Each magnet gives two counts, so twelve magnets give 24 counts a turn (15° each). What differs is where the ring can go, and two places look like the best candidates on any machine that can be adapted.

On the arm shaft, inside the arm. Machines with a removable top cover, such as the Singer 306W and many machines of the 1960s, often have room inside the arm casting and places where the arm shaft is open to reach. Set-screw collars on the shaft are common there, and they are left clear because the screws must be reachable. A printed ring that snaps over the shaft or an existing collar, locked by a set screw of its own, carries the magnets in pockets. Its size is limited only by the room in the casting. A 40 mm ring takes about twelve 6 mm magnets, and a larger ring, or smaller magnets, takes more, so resolution is seldom the limit. The sensors sit on a printed bracket fixed to an existing hole in the casting, and their leads run inside the casting down to the bed, out of sight.

On the hand wheel. Where the arm is sealed cast iron, as on the Singer 99K, the shaft can’t be reached without dismantling the machine, so the ring goes on the hand wheel. The sensor lead can’t cross the front of the machine or come near the rim or the bobbin winder’s tire track. On the 99K the one place left is the wheel’s sloped back face, between the tire track and the rim. There the sensor head comes up between the belt and the wheel, or in through the belt from the side, where the belt already keeps hands away.

The index. A third sensor reading a magnet of its own (on another radius, or off to one side of the ring) is the plain answer. Where three sensors won’t fit, a gap in the ring (one magnet left out) can stand in for it, the firmware finding the gap by timing; but then the index is found only once the shaft is turning, so this is a fallback rather than the first choice.

An optical alternative, for a small shaft. Where the shaft can be reached but there is little room around it, a ring of magnets can’t be made large enough for useful resolution. A slotted optical sensor reading a printed slotted ring can: 36 or 60 slots fit where a dozen magnets won’t. Each sensor is a fork with an infrared LED on one side and a detector on the other, and the ring’s teeth pass between them; two forks, or one dual-channel fork, a quarter-slot apart give quadrature, and a separate slot or tooth on another track gives the index. The circuit changes little:

The cost is dirt. The inside of an arm holds oil mist and lint, and an optical path collects both, where a Hall sensor is indifferent to them. So the fitting has to be made for cleaning: the forks open and reachable with the top cover off, the ring’s slots open-ended or wide enough to clear with a brush, and nothing that holds oil. A sewist who keeps a vintage machine well is in the top of it regularly to oil and clean it, and wiping the encoder can become part of that routine. A reflective sensor reading painted or printed bands is also possible but more fragile: it fouls fastest, and printed plastics vary widely in how they reflect infrared (some black filaments look nearly white to it). On a hand wheel, room light adds a further problem, so it is the last resort there.

11.3 Directions under consideration

These are ideas, not designs, recorded so they are not lost.

A pedal-first controller, without feedback. Fitting any sensor to the machine is the largest barrier to the MCU design. A middle course drops machine feedback altogether and puts the effort into the human side: a pedal designed for the foot, supporting all of it and responding to changes in pressure, with response curves that can be switched or tuned. Electrically it is open-loop phase control, like the KB3151C board in the knee controller, whose function it would take over and refine. It would be built in two parts. The mains stays in the base, behind a grounded IEC inlet: the triac stage, its isolation and the isolated supply. The pedal carries only low-voltage sensors, powered from the base, on a cable with a connector that can’t be mistaken for a mains one (a square Molex-style connector, say). One fixed curve could be done in analog; switchable curves, a dead band with no hum below it, soft start and a speed limit are all simple in firmware. So this is best treated as a feedback-free mode of the MCU design, not a separate circuit. Without feedback the load still sets the speed, as with the carbon pile (§4), and the operator compensates; a pressure-responsive pedal makes that compensation natural.

Load compensation without a machine sensor. A tunable compensation mode is mostly firmware, but not entirely. A series motor’s back-EMF depends on its field current (E=KIωE = K I \omega, §2), so between the triac’s pulses, with no current, there is only the weak residual-field voltage of §9: noisy, and it would need the motor voltage carried across the isolation barrier. The more practical route is IR compensation, as in the KBIC (§7, §8): measure the motor current and advance the firing angle as it rises. A small current transformer on the motor lead gives the current with isolation built in. The correction is first-order and must be tuned per motor, either from KK and RR measured on the test rig, or, as with the KBIC’s IR trimpot, turned up in a setup mode until the motor begins to hunt and then backed off. The concept circuit carries the current transformer as an option (T1), so the board can be built with it.

A tabletop controller for accessibility. Many people who sew can’t use a pedal or a knee lever. A tabletop controller on the same electronics would accept a range of assistive interfaces: a slide pot or knob, a grip or pressure sensor, a switch. Assistive switches commonly use 3.5 mm mono jacks, which would let existing switches plug straight in. Switch-only use needs its own firmware modes: start and stop at a preset speed, speed steps up and down, single stitch, needle up and down. Every latched mode must stop on any input, and should probably have a large stop button as well, since the user may not be able to reach for anything else in a hurry. The right set of interfaces is a question for the people who will use them.

11.3.1 A rectified-DC variant

The KBIC (§8) runs a universal motor on DC: it rectifies the line with a controlled bridge and phase-controls the result. The concept circuit can do much the same with one small change in its power path, and no change to its isolation, its firing or its control side. Like the other versions, it is drawn but not built or tested, by Claude at Bob Drzyzgula’s direction.

The change. The triac feeds the motor through a bridge rectifier, with a freewheeling diode across the motor:

Simulated, using Inside the Motor’s representative motor and load: the triac fired at 120° on AC (top) and through a bridge with a freewheeling diode (bottom). On AC the current comes in separate humps with nothing between them. Through the bridge it never reaches zero: it rises at each firing and decays through the diode until the next. The same firing angle gives nearly twice the speed.

What it gains. The simulation gives the size of the effect.

What else changes.

The fully DC alternative, rectifying the mains to a filtered 170 V bus and chopping it with a transistor, is not pursued. It needs a high-voltage switch with an isolated gate drive, and large capacitors that hold a dangerous charge after unplugging. That is far more hazard than this project warrants, for little gain over the bridge.

The rectified-DC variant of the concept (not built or tested). The mains version with the bridge rectifier (BR1) and freewheeling diode (D3) added between the triac and the motor’s own keyed connector (J3). The outlet (J2) keeps only the light, and the current transformer (T1) is in the bridge’s AC lead. The KiCad file and its generator are in hardware/mcu-controller-rectified/.

The KiCad drawing, a printed copy and the generator script are in the repository at hardware/mcu-controller-rectified/. On the motor rig, the experiment is the same motor on AC and on rectified DC. The scope should show the current change from separate humps to a ripple on a steady value.

11.3.2 A DC-motor variant

Inside the Motor, the companion paper, asks whether a small DC motor could drive a vintage machine in place of its universal motor (its §9). A 24 V brushed permanent-magnet motor of the “775” size has about the right power, around 60 W near 7,000 rpm, and fits easily inside a Singer external motor’s envelope. This variant of the concept circuit shows what the controller would become. Like the mains version, it is drawn but not built or tested, by Claude at Bob Drzyzgula’s direction.

The control side is unchanged. The RP2350B core board, the operator inputs, the heel switch, the needle and single-stitch switches, the index sensor and the encoder keep the same parts and the same GPIOs.

The mains side is gone. No triac, opto-triac, zero-cross detector, relay, isolated supply or current transformer is needed. In their place:

Signal RP2350B pin Use
PWM GP4 bridge PWM, 20 kHz (was triac fire)
DIR GP5 direction (was relay)
SLP GP6 driver sleep (was zero cross)
FLT GP13 driver fault
Brake chopper (optional) GP1 output
Motor current GP44 (ADC4) analog input
Motor terminals A, B GP45, GP46 (ADC5, 6) analog inputs
Bus voltage GP47 (ADC7) analog input

What changes in the firmware:

One new hazard: regeneration. A motor slowed by reversing its current, or by a speed loop allowed to drive negative current, returns its energy to the bus. An adapter can’t take energy back. The machine’s inertia holds tens of joules at full speed, enough to drive a 1000 µF bus far past the driver’s 40 V. So the firmware slows the machine only by braking, which spends the energy in the motor’s own resistance. It also watches the bus voltage and, if the bus rises anyway, stops driving and fires the chopper.

The DC-motor variant of the concept (not built or tested). The 24 V power section is above the dashed line, which here separates high current from logic, not live from safe. Left to right: the 24 V input with its fuse, TVS and bulk capacitors; the 5 V buck; the optional brake chopper; the bus-voltage sense; the G2 H-bridge module with its current-limit and current-sense resistors; the 775 motor with its suppression capacitor; and the dividers that read the motor’s terminals. Below the line, the control side is that of the mains version. The KiCad file and its generator are in hardware/mcu-controller-dc/.

The KiCad drawing, a printed copy, the generator script and notes on values and firmware are in the repository at hardware/mcu-controller-dc/. The first bench step is still the one Inside the Motor recommends: run a 775 from a 24 V supply on the motor rig, then put it on a machine on a temporary bracket.

11.4 Other options considered

A bench rig for these measurements is in the repository at standalone/motor-rig/ (its guide is RIG_GUIDE.md). Two bench measurements would settle the open questions for any of these: the 99 motor’s current and speed on a variac at several voltages, giving KK and RR for the model in §2, and its top speed on rectified DC.

12 Sources

  1. Universal motor, Wikipedia.
  2. G. V. Fay, Direction and Speed Control for Series, Universal and Shunt Motors, Motorola AN-443.
  3. Resistance, carbon stack, University of Queensland Physics Museum.
  4. K. Hayashi, Carbon Pile for Automatic Voltage Regulators, Part 2, Tanso 3(2):56–59, 1953.
  5. Old Kenmore sewing machine foot control repair, softsolder.com, March 2009.
  6. Help understanding a sewing machine pedal, EEVblog forum.
  7. I. Skalka, Low-Cost Universal Motor Phase Angle Drive System, Freescale AN1662.
  8. Speed Controller for Universal Motors, Silicon Chip, October 2002.
  9. Autotransformer, Wikipedia.
  10. TDA1085C universal motor speed controller, onsemi data sheet, Rev. 10, July 2007.
  11. KBIC variable speed DC motor control, KB Electronics data sheet D-210, 1998.
  12. KBMM variable speed DC motor control, KB Electronics data sheet D-215.
  13. KBWC solid-state speed controls, KB Electronics / Nidec data sheet.
  14. KBSI-240D signal isolator, Motors & Control product page.

The base’s own parts, used in §11: Bourns 91–99 series panel control datasheet; Omron SS subminiature basic switch, Cat. No. B032-E1-14; Qualtek 723W-X2/03 IEC inlet datasheet; WAGO 221-413. Links are in the User’s Guide’s knee-controller chapter.