AI Stick-Slip Detection: Stopping Torsional Vibration Before It Breaks Your String

By Saad Iqbal

The torque gauge climbs past 18,000 ft-lb and the driller’s eyes flick to the RPM readout. Surface RPM reads a steady 120. Downhole, the bit isn’t turning at 120 at all — it’s stopped dead, grinding against the formation while the drill string above it winds up like a torsion spring. Three seconds later the bit breaks free. RPM at the bottom of the hole spikes to 300, maybe 400, for a fraction of a second, and somewhere a few hundred feet up a connection that was rated for steady-state torque just took a hammer blow it was never designed for. Nobody on the rig floor saw it happen. The surface gauge never left 120.

That gap — between what the surface instruments show and what the bit is actually doing three kilometres below — is exactly where AI stick-slip detection earns its keep. It’s not a buzzword bolted onto a driller’s console. It’s a real-time model that infers downhole behaviour from surface data fast enough to act on it before the string pays the price.

What Is Stick-Slip, and Why Does It Keep Eating Your Bit Runs?

Stick-slip is torsional vibration with a split personality. In the “stick” phase, friction at the bit — usually from an aggressive PDC cutting structure, a tight mud weight window, or a build/turn section with high side forces — slows the bit’s rotation or stops it outright. The drill string above keeps being driven at surface RPM, so it winds up, storing torque the way a twisted rubber band stores energy. Once the static friction at the bit is overcome, the string unwinds violently. That’s the “slip”: bit RPM spikes well above surface RPM, sometimes by a factor of two or three, before the cycle resets and starts again.

You rarely get a clean warning light for this. What you get instead are secondary symptoms: a sudden, otherwise unexplained drop in rate of penetration, and — if you’re watching the right curve — a torque signal with a wide, ragged envelope instead of a tight one. In the worst cases the top drive itself stalls. Pason, whose surface rig instrumentation sits on a large share of North American land rigs, has documented this pattern directly: experienced drillers can watch these signs scroll past on a screen for hours without recognising the underlying cause, because none of the individual symptoms screams “torsional vibration” on its own.

Left unchecked, severe stick-slip is not a cosmetic problem. The repeated torque spikes fatigue connections, accelerate PDC cutter damage from the momentary RPM overshoot, and in bad cases drive a downhole motor failure or a twist-off — the kind of NPT event that turns a routine bit run into a fishing job.

The economics are what make this worth automating rather than just training drillers harder. A PDC bit damaged by repeated RPM overshoot doesn’t fail cleanly — it dulls unevenly, drags ROP down for the rest of the run, and still has to come out of the hole on a trip that costs rig time either way. A twist-off from a fatigued connection is worse: a fishing job measured in days, not hours, stacked on top of the rig’s day rate. None of that shows up as a single line item called “stick-slip” on an AFE. It shows up scattered across bit cost, non-productive time, and the quiet erosion of ROP that everyone attributes to “a tough section” instead of a vibration mode nobody caught.

How Do You See It Coming Instead of Explaining It After the Fact?

The textbook response has always been manual and reactive: back off weight on bit, bump rotary speed, and repeat until the drilling smooths out. That works, eventually, but it depends on a human noticing the torque envelope widening against a noisy backdrop of normal drilling variation — while also managing hole cleaning, directional targets, and whatever else is happening on that shift.

The AI stick-slip detection approach flips the order. Instead of waiting for a driller to recognise a pattern, a torsional model runs continuously against the live surface feed — standpipe pressure, torque, surface RPM, hookload — and estimates what the bottom-hole assembly is actually doing in near real time. When the modelled severity crosses a threshold, the system flags it before a human would have pieced the story together from a scrolling log.

What Does an AI System Actually Do Differently From a Driller Watching a Screen?

The clearest production example is Pason’s DAS, an automated drilling optimisation package that explicitly models, measures, and mitigates stick-slip in real time using equipment already installed on most rigs — no extra downhole tool required. DAS runs a proprietary torsional model, developed in collaboration with ExxonMobil’s Upstream Research team, against surface sensor data. When stick-slip severity exceeds a user-defined threshold, the system automatically trims weight on bit and adjusts rotary speed to knock the vibration down, then hands control back to its rate-of-penetration optimisation routine once conditions settle. Pason’s own framing is notable for what it doesn’t claim: that surface-based inference is a free lunch. Instead, the company positions it as a close substitute for far more expensive downhole dynamics tools, close enough to act on for most wells.

The difference from a manual response isn’t just speed, though speed matters when a torque spike is measured in seconds. It’s consistency. A model doesn’t get tired at hour ten of a twelve-hour tour, doesn’t get distracted by a connection or a survey, and applies the same threshold logic on well forty as it did on well one. That consistency is also why operators pair detection with mitigation hardware rather than detection alone — NOV’s SoftSpeed II Stick Slip Mitigator is built specifically to dampen torsional vibration at the top drive once it’s identified, closing the loop between “we see it” and “we stopped it.”

Both of these sit inside a broader shift toward automated rig floor decision-making — the same shift covered in our rundown of AI rig automation and autonomous drilling platforms, where stick-slip mitigation is usually one module among several running on the same control layer.

What Should You Check Before You Trust the Alarm?

Surface-inferred stick-slip detection is a model, not a direct measurement, and it inherits every weakness that implies. Three things are worth checking before you let a threshold-crossing alarm drive automatic WOB and RPM changes unsupervised:

1. Confirm it isn’t whirl or bit bounce wearing a stick-slip costume. Lateral whirl and axial bit bounce can both produce torque and ROP signatures that look similar to torsional stick-slip on a coarse surface read. If the mitigation response for one is applied to the other, you can make the actual dysfunction worse while the dashboard shows green.
2. Re-baseline the torsional model after any major BHA change. A model tuned against one bit, motor, and stabiliser configuration doesn’t automatically transfer to the next run. Treat the first few hundred feet of a new BHA as a calibration window, not a blind-trust window.
3. Watch for threshold fatigue. If the alarm fires constantly in formations where moderate stick-slip is simply the cost of the ROP you’re chasing, crews start ignoring it — exactly the failure mode the automation was meant to fix. Thresholds need periodic review against actual bit condition on trips out, not just left at the vendor default.

This is the same discipline that matters for any sensor-driven early-warning system in the field, from lost circulation detection that reads pit volume and flow trends, to ESP failure prediction that reads motor current and vibration downhole. The pattern recognition only earns trust once someone has checked it against the physical reality on a few real trips.

A Practical Rollout Checklist for Stick-Slip Detection

1. Confirm your rig’s surface sensor suite (torque, RPM, hookload, standpipe pressure) is sampling fast enough for the vendor’s torsional model — most need sub-second resolution to catch a slip event cleanly.
2. Start detection in advisory-only mode for the first several wells so drillers can correlate alarms against what they’re already watching, before handing over automatic WOB/RPM adjustment.
3. Log every threshold crossing with the BHA and formation context, and review it against bit dull grades on the next trip — that’s your ground truth, not the alarm count.
4. Pair detection with a mitigation response (automatic parameter adjustment, or a dedicated damping tool) rather than detection alone; an alarm nobody acts on in time is just a prettier log.
5. Revisit thresholds every time you change bit type, motor, or target formation — don’t run last quarter’s settings on this quarter’s BHA.

Back on that rig floor, the torque gauge that climbed to 18,000 ft-lb and the RPM readout that never budged from 120 would have told two different, true stories to two different systems. A driller watching only the surface RPM saw nothing wrong. A torsional model watching the same feed, built to see what the surface number was hiding, would have flagged the stick phase before the slip ever happened — and that’s the whole argument for putting one on your rig.