Predictive Maintenance · Rod Pump

Rod pump failures. Caught days early.

Per-well dynamometer ML detects pump fillage degradation, gas interference, parted rod, and fluid pound. Each signature flagged before the production rate moves enough to cross a fixed SCADA threshold.

Illustrative industry figures put a reactive workover at $15-30K plus $5-15K/day of deferred production. Catching it early turns both into a scheduled job.

Four rod-pump signatures.

Four rod-pump signatures.
Pump fillage degradationML detectsCard area shrinking >15% over 7 daysRoot causeWear, gas interference, formation issueActionSchedule pump pull at 14-21 days, not after parted-rod event
Gas interferenceML detectsCard shape: filled → gas-interfered at top-of-strokeRoot causeGOR increase, separator failure, formation gas breakthroughActionInvestigate separator before scheduling pump work
Parted rodML detectsSudden bottom-load drop with maintained top-loadRoot causeRod failure, typically at coupling or transition jointActionPull immediately; production at risk every hour
Fluid poundML detectsCharacteristic late-stroke spike patternRoot causePump-off / over-pumping (rate exceeds inflow)ActionAdjust SPM via controller before mechanical damage compounds

All Equipment Signatures 

A beam pumping unit on a fenced gravel lease, with a coated pipe spool on the ground and low hills behind under an overcast sky
Beam pumping unit on a fenced lease · Photograph

Frequently asked

What rod-pump foremen ask.

What rod pump failures does WorkSync detect?

Four primary failure modes. (1) Pump fillage degradation: dynamometer card area shrinking more than 15% over 7 days. (2) Gas interference: card shape shifting from filled to gas-interfered at top-of-stroke. (3) Parted rod: sudden bottom-load drop with maintained top-load.

(4) Fluid pound: the characteristic late-stroke spike pattern. Each gets flagged with dollar impact, intervention cost, and a rank in the work plan.

What is the lead time versus fixed-threshold SCADA?

Days of advance warning are typical, because the card-shape signatures change before the production rate does. As an illustrative industry figure: for an at-risk well producing 50 BO/d at $75/bbl, every 24 hours of early intervention is roughly $3,750 of deferred-production avoidance.

Workover avoidance versus scheduled workover, what is the math?

Illustrative industry figures: a reactive workover on a failed pump runs $15-30K typical cost, plus days of deferred production at $5-15K/day depending on well rate. A scheduled workover caught days pre-failure carries the same workover cost, near-zero deferred production, and no emergency rig premium.

The catch time is the whole difference.

What data does it need?

Dynamometer card data from the pump-off controller, production rate from SCADA or run tickets, and well-test data from your accounting system. The model trains on 90-180 days of historical card data per well. SCADA gaps shorten the predictive lead time; deployment proceeds either way.

Does it work with all pump-off controllers?

Yes. Any controller exporting standard dynamometer card formats is in scope. The card is the substrate; the controller brand is incidental.

How does this connect to the broader anomaly detection system?

Rod-pump signatures are one equipment class in the broader anomaly detection capability. See /capabilities/anomaly-detection for the equipment-class signatures across rod pump, ESP, gas lift, plunger lift, and surface equipment. The flagged anomalies feed the ranked work plan via Work Engine.

See rod-pump ML on your wells.

4-week pilot on the stack you already own. Pick one field. Day 14 we show the at-risk pumps. Day 28 you decide, under the Impact Guarantee: we charge when your number moves.