Motion hardware¶

This page gives the machine's geometry, its speed and acceleration limits, the axis conventions, and the stepper drive facts. How the grblHAL planner uses these limits is on the grblHAL driver; how the step stream itself works is on The step engine.

Geometry, speeds and limits¶

The laser PWM carrier (39.98 kHz, 7-bit duty) is on The laser.

Where the limits come from. They are the factory machine's own, decoded from captured factory pulse streams rather than chosen: acceleration about 700 mm/s² on X and 590 on Y under firmware 2.6.0 (2018-era firmware used about 1000), travel moves peaking at 202 mm/s of vector speed (about 8 in/s) at the 28160 Hz travel tick, prints and hunts at a 10 kHz tick, a cut feed of 145 mm/s in the sample print, and a Z cadence of 61 to 115 ms per half-step, about 5.7 mm/s at most.

One correction worth keeping, because the tag names invite the mistake: the HAxr, HAyr and HAar values in a pulse header are not motion acceleration limits. They are head-accelerometer interrupt-generator threshold registers, which is the factory's crash detector (Sensors).

Origin is the back-left corner, and the workspace is all-positive from there. +Y moves the gantry toward the front of the machine. Z counts positive upward, away from the bed.

Z is never driven blind: the lens carriage is referenced against a hall sensor low in its travel, and moves are supervised against it. The sensor and the unit-to-unit variation of its trip point are on Sensors; the lens, its travel, and the focus are below.

The lens and its travel¶

The head carries a 2 in focal length lens (Glowforge's own figure) under a collimated beam, so the focal point moves with the lens, 1:1: a millimeter of lens is a millimeter of focus height. The lens sits in a carriage that travels 0.485 in (12.32 mm) between two mechanical stops, driven by a small stepper through a lead screw. Every head shares the screw, so a half-step is the same height everywhere, about 0.34 mm: ForgeFIRM takes the travel as 36 half-steps ($102 is 2.922 per millimeter), a half-step of leeway at each end of what the bench reference counts (its stops sit 18 half-steps below its hall edge and 20 above, 38 in all).

The hall sensor is the only position reference. It reads home from an edge partway up the travel to the top stop. The rising edge, the first position that reads home going up, found by stepping down out of the zone and back up, is what everything references; it is exact and repeatable, where a count that ends in a stall against a stop is not (the rotor slips whole steps against the stop and re-engages up to three full steps out of phase, so such a count reads low by an even number at random). Going down, the sensor lets go four to six half-steps under the edge; the lens rings a little on every step, which is the jitter in that band. Where along the travel the edge sits differs from head to head, and that is the one per-head number for Z: the commissioning focus card references the lens on the edge, burns a ladder over the head's free travel, and the user's pick on the sheet's thickness gives the focal height when the lens is on the edge (lens_hall_edge_z_mm). The free travel itself is found by the head accelerometer: a free half-step rings, on every second step strongly, and at a stop the ring dies two to four steps before the rotor would slip, so the card steps toward each stop one half-step at a time, calls contact on the first quiet strong step, backs off two, and proves by the count back to the edge that nothing slipped (lens_stop_below_steps, lens_stop_above_steps; 14 below and 20 above on the bench reference machine). Nothing in ForgeFIRM drives the lens onto a stop on a user's machine; when the stops cannot be found on a head, every move keeps a fallback window, ten half-steps below the edge to twelve above, that clears both stops on any head whose edge sits within six half-steps of the bench bench reference's, and the user is told. See Commissioning.

The factory's focus law, from its own Z commands, is a count of full steps up from its zero, where its hunt parks the lens: 4 full steps down from the hall edge. It runs about 2.8 half-steps per millimeter of material and saturates at 30 half-steps, its idea of the usable travel.

In GRBL mode Z is the focal point's height above the tray: Z 0 focuses on the bed, Z 3 on the top of 3 mm material, and +Z is lens up. A home leaves the lens on the hall edge, sets Z to the edge's focal height on the step grid (ten half-steps, Z 3.42 mm, on the bench reference), and parks the focus at lens_park_z_mm, a user setting, 3 mm by default (The grblHAL driver). Z may go below the tray: the tray comes out for tall work.

The lens is never moved without a reference first: in GRBL mode a job's Z moves it within the free travel once a home, or a commissioning card, has referenced it, and the driver refuses Z otherwise.

The lens rises only at the driver's drive current (z_current 0). At the hold current (1) the motor lifts the lens two steps into the service's ramp (630, 164 and 115 ms, then 77 ms per step) and stalls; lowering works at either current (measured on the bench head with single steps at the ramp's timing). Every path that moves the lens sets the drive current first and the hold current after: the homing sweeps, the focus card, and the cloud client's motions.

The machine has no limit or home switches as it ships. What that means in practice, how each mode establishes an origin and how the machine behaves without one, is on Homing (the operator view) and Homing internals (the mechanism).

The Grbl $ settings (steps/mm, rates, accelerations) default to the values above, which are the factory machine's own measured values.

Stepper drives¶

X and Y. The X axis and the two Y motors are driven from the pulse stream (see The step engine). The two Y motors (Y1, Y2) are driven complementary from one Y step and direction pair. The drivers expose:

• Microstepping mode per axis (x_mode, y_mode): 1, 2, 4, 8, 16 or 32. 1 = full steps. The factory runs at ×8; ForgeFIRM's GRBL mode runs at the xy_microsteps setting (8, 16 or 32, default 8) and derives its steps per millimeter, its machine tick and the kernel stop ramp from it (the grblHAL driver). Cloud mode runs at the service's own ×8. The mode is written at the controller's start, at idle: a DRV8825 can re-index its microstep table by up to one full step when MODE changes with the motor energized.
• Current decay mode per axis (x_decay, y_decay): 0 = slow (fast stop, slow response), 1 = mixed (decay pin high impedance), 2 = fast (fast response, slow stop).
• Drive current set through the PIC: x_step_current and y_step_current are 10-bit DAC values (0 to 1023, 0 = minimum). The two axes' DAC scales differ by design; the factory runs X at 135 (33 at hold) and Y at 22 (5 at hold), and ForgeFIRM writes the same.
• Fault lines from the drivers: bit 0 = X, bit 1 = Y1, bit 2 = Y2 (faults); each can be masked (ignored_faults).
• A per-motor lock (motor_lock, bits X, Y1, Y2, Z) that holds an axis still while a program runs.

Z. The Z axis moves the lens carriage in the print head. Its driver enable, current (high or low) and microstep mode (full or half-step) are head controls (z_enable, z_current, z_mode). Z also has a direct single-step control (z_step: 0 = toward the bed, 1 = away from it) that pulses the GPIO outside the pulse stream.

The attribute reference for all of these is on the kernel module.

The 40 V motor rail¶

The stepper drivers run from a 40 V motor rail, switched by the kernel's cnc/enable and cnc/disable.

The DRV8825 drivers on this board wedge on rail glitches. A glitch can leave them unserviceable: SDMA playback and the position counters run normally while the motors produce nothing at all, or stall mid-move. The supply itself is fine; this is a driver failure mode, not a marginal rail. cnc/faults does not flag the state, and whether a given rail power-up wedges them is chance.

The kernel drives their nRESET and nSLEEP pins (reset-gpio gpio3 18, sleep-gpio gpio3 16), but only as a pair inside every enable and disable cycle, together with the rail. The reset pulse therefore rides along on every recovery attempt, and it has never shortened the recovery: a logic reset alone does not clear the state, and only the rail-off duration matters. That fits a latched internal state nRESET does not reach. Recovery is a longer true power-off (ForgeFIRM ladders 5, 15 and 30 seconds) and, at worst, a full machine power cycle.

Three consequences follow, and all three shape how ForgeFIRM is built:

• Position counters, homing anchors and a H:1 status are never proof of motion. The head accelerometer is the motion witness (below, and Sensors).
• The rail stays up while the machine is on. Every power-up is a fresh gamble, so the cheapest policy is not to cycle it. That is why the pulse device is brokered rather than opened per controller, and why there is no idle-rail-off policy (forgectrl).
• The machine proves it can move before the first job of a session (ForgeFIRM internals).

What the motion witness reads¶

The liveness probe commands a small move and reads the head accelerometer across it. On the bench reference, on an identical commanded move (The bench reference):

The noise floor for scale: 1 g reads about 16384 counts. ForgeFIRM gates a controller start at 800 counts, clears any leftover axis mask before its own move so a mask cannot read as a wedge, and settles 300 ms after the run-current step before it samples, because that step jolts the head.

Raw accelerometer reads through sysfs take about 150 ms each, which is enough for a yes-or-no verdict over a multi-second window and useless for a waveform. Reading the part straight over its bus gives about 530 to 800 samples a second, which is what the crash watch and the commissioning lens finder use (see Sensors).

A contact strike, for any future contact sensing. From the retired accelerometer-homing work: creeping toward a rail reads a baseline of about 0.5 to 2 thousand counts, and contact jumps to 29 to 42 thousand within about 4 ms, 20 to 40 times over. But a slow approach is near-silent: belt compliance turns low-speed skipping into sub-threshold grinding. Any contact-sensing scheme has to strike fast.

What the microstep mode changes, and what it does not¶

X and Y run at 8, 16 or 32 microsteps from one setting, and the machine holds its 200 mm/s top speed at all three because the machine tick scales with the mode (the grblHAL driver). Measured on the bench reference, dry and under the laser: position returns exactly at every mode, with no underruns and no clamped events, and the tube current at cruise reads alike across the three.

What does change is vibration, and the finer mode is the quieter one. With every fan, the coolant pump and the TEC commanded off so the machine is silent, running a pattern at 12000 mm/min and reading the head accelerometer over its bus:

The rest floor is about 50 counts with everything off, and about 200 with the coolant pump running: the pump is in the reading. The single-axis legs sit within a few percent of each other at every mode; the gain is on the legs where both axes move, and most of all on a circle.

An arc's tone is the planner's, not the motor's¶

grblHAL traces an arc as chords whose sagitta is $12 (0.002 mm by default), and every chord is a planner block. A 9 inch circle at 12000 mm/min is therefore 531 chords of 1.35 mm and 148 block boundaries a second: an audible tone that no microstep mode changes, because the planner's geometry makes it and the motor's step grid cannot.

Driving $12 finer moves the tone up (148, 209, 296 Hz) and does not lower the vibration. Above about 300 chords a second the feed sags mid-arc and the circle takes longer, with the processor flat at about a third of the core and no underrun: the planner plans to a stop at the end of what it holds, and 100 blocks of 0.3 mm is 30 mm against a 29 mm stopping distance from 200 mm/s. A deeper planner buffer does not move that; the chord rate the protocol loop can feed is the ceiling. On a 9 inch radius the finest $12 that holds top speed is 0.0005 mm; at a lower feed a finer value holds in proportion, and a smaller radius reaches the same chord rate sooner.

$398, the planner buffer depth, runs over its whole 30 to 1000 range.