Anti-dive explained: why torches crash on holes and how to prevent it

Ask anyone who runs a DIY plasma table what their worst crash was, and odds are the story involves a hole. Not a crash into a hole — a crash because of one. The torch crosses a void it cut ten seconds ago, the controller does exactly what it was designed to do, and half a second later your torch is embedded in the workpiece.

This failure mode has a name — diving — and the logic that prevents it is anti-dive. It’s the single most important feature in a torch height controller, and the least understood. Let’s fix that.

The physics: why voltage lies to you

A THC tracks sheet height by reading arc voltage. More gap means more voltage, so the controller chases a setpoint: too much voltage, go down; too little, go up. Simple and effective — as long as the relationship between voltage and height holds.

Now picture the arc crossing a hole. The plasma jet still conducts — it’s a supersonic column of ionized gas, it doesn’t need metal directly under the nozzle. But the arc root has to travel much farther to find metal: down through the hole, sideways to its walls, or across to the far edge. Longer arc path, higher resistance, higher voltage.

Your controller sees that voltage spike and reaches the only conclusion its logic allows: the sheet dropped away, I must descend. It drives the Z axis down at full speed. The sheet, of course, hasn’t gone anywhere — it’s right there on the far side of the hole, waiting to meet your nozzle at 200 mm/s.

That’s a dive. On a good day it costs you a nozzle. On a bad day, a torch, a plate, or your fingers’ proximity to all of the above.

Why naive fixes don’t work

Every THC designer goes through the same stages of grief with anti-dive:

“Just lower the correction speed.” Now the dive is slower, but so is every legitimate correction. On warped sheet, the torch can’t keep up and cut quality dies. You’ve traded crashes for dross.

“Ignore corrections above a voltage threshold.” Better — but warped thick plate can legitimately exceed your threshold, and suddenly the THC freezes exactly when you need it most.

“Pause THC during corners.” Helps for kerf-crossing slowdowns, does nothing for holes in straight lines.

The fundamental problem: voltage alone can’t tell you why it spiked. A void and a genuinely dipping sheet can produce identical readings. Any robust solution has to bring in more information.

How SmartTHC approaches it

SmartTHC uses several layers, because in the workshop, single-layer defenses get breached:

1. Dual voltage filtering. The firmware maintains two views of the arc voltage: a fast EMA (exponential moving average) that feeds the PID for responsive tracking, and a slow 200-sample average that represents the “normal” cutting voltage. The fast signal reacts; the slow signal remembers.

2. An explicit anti-dive state. When the fast voltage spikes far above the slow reference — the signature of a void — the controller doesn’t just freeze. It triggers an active Z lift: a fast, parametrizable retract, 3 mm in about 120 ms by default. The torch hops over the void instead of pausing and hoping. When voltage settles back to normal, cutting height control resumes.

I chose an active lift over a simple freeze after one too many incidents where “frozen” still wasn’t high enough for a curled-up part waiting on the far side. Explicit is better than implicit — you can tune the lift height and speed, and you can watch it work.

3. Motion gating. The THC only activates when the table is actually moving at cutting speed. This kills a whole class of false triggers during pierce and lead-in, where voltage is wildly unstable by nature. A voltage spike while standing still after piercing isn’t a hole — it’s just physics settling down.

4. Plasma-gated filtering. The slow reference average only updates while the arc is confirmed stable. Transients from pierce, arc transfer and voids never contaminate the baseline the anti-dive compares against.

Tuning anti-dive in practice

Out of the box, the defaults are deliberately conservative. On your bench, before any real cut, verify the behavior: jump the input voltage high (a bench supply through the divider works fine) and confirm the Z axis lifts — the right direction, by roughly the configured height. This check takes two minutes and is non-negotiable: an inverted anti-dive doesn’t save torches, it drives them into the plate.

On real cuts, watch the serial log. SmartTHC emits structured key=value status lines and an event when anti-dive triggers. If it fires constantly on clean sheet, your threshold is too tight or your reference filter hasn’t stabilized. If it doesn’t fire on obvious voids, loosen it. Tune on scrap before you trust it on a paid job.

The bottom line

Height control gets you good cuts. Anti-dive keeps you in business long enough to enjoy them. When you evaluate any THC — commercial box, LinuxCNC setup, or DIY project — the first question shouldn’t be “how fast is the PID” but “what exactly does it do when the arc crosses a hole?” If the answer is vague, so is your torch’s life expectancy.

SmartTHC’s entire anti-dive logic is open source, documented, and tunable from the LCD menu or the serial console. The build guide walks through the wiring, the bench-test procedure, and the tuning workflow step by step — so your next hole crossing is a non-event instead of a repair bill.