A 7,200-ft Permian Basin well. 7/8-3/4 taper rod string, Grade D, running at 7 SPM on a 168-in unit. The simulation predicts a peak polished rod load of 22,400 lbs and 145 BFPD. The field dynamometer card comes back at 25,100 lbs peak load, and the well is making 118 BFPD. That is a 12% load error and an 18% rate error - enough to question whether the rod string is properly designed, whether the unit is undersized, or whether the simulation is useful at all.
That gap is not random. It traces back to a handful of input errors and modeling assumptions that compound through the wave equation. Gibbs and Neely identified this problem in 1966 when they published the diagnostic method for sucker rod pumping systems (SPE-1165-PA) - the simulation is only as honest as the data you feed it. Sixty years later, the math has improved but the input problem has not.
Here are the five root causes we see most often, what the literature says about each one, and how to fix them.
1. Survey Resolution Destroys Friction Accuracy
Rod-on-tubing friction is the dominant source of simulation error in deviated wells, and the friction calculation depends entirely on the wellbore survey. Eickmeier showed in 1966 (PETSOC-66-02-03) that Coulomb friction in deviated wells scales with the normal force at each contact point - which means every change in inclination and azimuth generates a side force that the rods must drag against. Miss the geometry, miss the friction.
The problem is station spacing. A survey with 100-ft stations through a build section interpolates a smooth curve between measurement points. The actual wellbore at 30-ft resolution may have 2-3 deg/100ft doglegs that the coarse survey completely hides. In a 6,800-ft Delaware Basin well we analyzed through PetroBench, switching from a 100-ft MWD survey to a 30-ft gyro survey changed the predicted friction load by 1,850 lbs - which moved the peak polished rod load from 21,200 to 23,050 lbs and brought it within 2% of the field card.
The fix is mechanical: use the highest-resolution survey available. Gyro over MWD. 30-ft stations over 100-ft. If your directional report has fine-resolution data and you have been entering only every third station to save time, stop doing that. The minutes you save on data entry cost you hours of troubleshooting when the card does not match.
For wells with lateral sections, this gets worse. A Midland Basin horizontal with a 6-deg/100ft build rate and a 90-deg lateral has enormous contact forces in the curve. Coarse survey data through that build section can underpredict friction by 20-30%, which cascades into undersized unit selection and premature rod failures.
2. Stale Fluid Properties
Fluid properties drift. A Spraberry well that came online at 15% water cut and 650 scf/bbl GOR may be producing at 55% water cut with 380 scf/bbl GOR eighteen months later. That shift changes the fluid gradient, the gas interference pattern, the pump volumetric efficiency, and the fluid level in the annulus. Every one of those feeds into the wave equation differently.
Consider the fluid load alone. At 15% water cut with 32-degree API oil, the fluid gradient is roughly 0.35 psi/ft. At 55% water cut, it climbs to 0.40 psi/ft. Over a 5,500-ft fluid column, that is an additional 275 psi of load on the plunger - about 1,200 lbs on a 1.75-in pump. The simulation using the original fluid properties will underpredict the minimum rod load on the downstroke and overpredict pump displacement.
The discipline is simple: update fluid properties at least quarterly, or whenever water cut shifts by more than 10 points. Pull the latest well test data. Check the GOR from your separator. If the well has been on the same fluid input card since initial design, the simulation is modeling a well that no longer exists.
3. Rod String Condition vs. Specification
The simulation models new rods at nominal diameter with uniform properties. The rod string in a well that has been running for 14 months is not that. Coupling wear in deviated sections changes the effective contact area. Corrosion pitting on Grade D rods in a high-H2S well reduces the cross-sectional area and changes the stress distribution. A mixed string where the bottom 1,500 ft was replaced with a different heat lot has different elastic properties than the original design.
Everitt and Jennings addressed this in their 1992 paper on rod pump simulation improvements (SPE-23980). They showed that accounting for actual rod condition - particularly wear-related diameter reduction and coupling contact changes - reduced the average simulation error from 11% to under 5% on a 40-well dataset. The model does not need to know everything about the downhole condition, but it needs to know the big items: actual taper configuration, any replaced sections, and approximate wear state.
When you pull rods, document condition. Measure coupling OD at wear flats. Note which joints were replaced. Then update the model. A gradual drift between simulated and actual dynamometer cards over 6-12 months is often the rod string aging, not the simulation breaking. That drift is diagnostic information if you are paying attention to it.
4. Pump Slippage and the Efficiency Assumption
Most simulation tools default to a pump efficiency between 70% and 85% - a single number meant to capture valve leakage, barrel-plunger slippage, gas lock, and incomplete fill. Lea quantified how inadequate this is in his 1991 analysis of pump slippage and its effect on production (SPE-22561). He demonstrated that plunger-barrel clearance, fluid viscosity, and differential pressure interact to produce slippage rates that vary by an order of magnitude across normal operating conditions.
A 1.75-in, 20-ft pump with a nominal 0.003-in plunger clearance in a well producing 28-degree API oil at 180F bottomhole temperature has a very different slippage rate than the same pump in a 40-degree API oil at 140F. The lower-viscosity fluid at higher temperature slips past the plunger faster during each stroke. Lea's work showed that a 0.001-in increase in clearance - common after 8-12 months of wear - can increase slippage losses from 5% to 18% depending on fluid properties.
Speed matters too. A pump achieving 80% volumetric efficiency at 6 SPM may drop to 62% at 11 SPM because the valve response time becomes a meaningful fraction of the stroke cycle. If you change pump speed in the simulation but leave efficiency fixed, you will overpredict production at higher speeds.
The best practice is to calibrate: run the simulation at current operating conditions, compare predicted rate to actual measured rate, and back-calculate the effective pump efficiency. Use that calibrated value as your baseline. When you change pump speed or stroke length in the simulation, adjust efficiency directionally based on Lea's slippage relationships rather than holding it constant.
5. Dynamometer Card Comparison - A Worked Example
The dynamometer card is your ground truth. When the simulated surface card diverges from the measured card, the shape of the divergence tells you which input is wrong. Here is how that works in practice.
Take a 8,400-ft Eagle Ford well. Rod string: 1-in top section (3,200 ft), 7/8-in middle (2,800 ft), 3/4-in bottom (2,400 ft), all Grade D. Pump: 1.50-in, 25-ft barrel. Unit: 228-in stroke, 6.5 SPM. Simulation predicts peak load of 28,900 lbs, minimum load of 5,200 lbs, and 95 BFPD.
The field card shows peak load at 31,600 lbs - 2,700 lbs high. Minimum load is 4,100 lbs - 1,100 lbs low. Production is 78 BFPD - 18% below prediction. The card shape shows a pronounced upstroke friction hump and a narrow downstroke, suggesting both high friction and poor pump fillage.
Step one: the survey. The original model used a 90-ft station MWD survey. Re-importing the gyro data at 30-ft stations through the build section reveals a 4.5 deg/100ft dogleg at 3,800 ft that the MWD survey smoothed over. Correcting this adds 1,900 lbs of friction to the upstroke prediction and brings peak load to 30,800 lbs. Closer, but still 800 lbs short.
Step two: fluid properties. The model was built at initial completion - 8% water cut, 720 scf/bbl GOR. Current well test shows 34% water cut, 510 scf/bbl GOR. Updating these changes the fluid load calculation and adjusts the minimum load prediction from 5,200 down to 4,400 lbs. Peak load moves to 31,300 lbs as the heavier fluid column adds to the rod weight on the upstroke.
Step three: pump efficiency. Calibrating against actual production (78 BFPD) gives an effective pump efficiency of 68% rather than the default 80%. The well has been on pump for 11 months without a workover, so wear-related slippage is the likely cause. The final simulation - corrected survey, updated fluids, calibrated efficiency - matches the field card within 1% on peak load and 7% on minimum load. That is a model you can design from.
This is the kind of iterative calibration workflow that PetroBench's engine is built around - overlay the simulated card on the field card, adjust inputs systematically, and converge on a model that reflects the actual well. It takes 15-20 minutes per well when the tools make comparison easy. It takes hours when they do not.
Closing the Gap
Perfect simulation accuracy is not achievable and chasing it wastes time. The physics of coupled rod-fluid-pump interaction in a deviated wellbore are complex enough that some deviation between model and reality is inherent. But there is a difference between 2-3% error from irreducible model uncertainty and 15-20% error from bad inputs. The first is engineering. The second is guessing with extra steps.
The pattern across every mismatch we have diagnosed through PetroBench support is the same: the simulation math is fine, the inputs are stale. Survey data from a different vendor than the one who drilled the lateral. Fluid properties from initial completion. Pump efficiency left at the software default. Each one adds 3-5% error, and they compound multiplicatively, not additively.
The question is not whether your simulation tool can produce accurate results. Any wave-equation solver built on Gibbs' method can. The question is whether your workflow makes it easy to keep inputs current across 200 or 2,000 wells - or whether it quietly lets them go stale while you trust the output. That is a tooling problem, not a physics problem, and it is solvable.
References
Gibbs, S.G. and Neely, A.B. 1966. Computer Diagnosis of Down-Hole Conditions in Sucker Rod Pumping Wells. SPE-1165-PA. Journal of Petroleum Technology 18(1): 91-98.
Eickmeier, J.R. 1966. Applications of the Delta II Dynamometer to Sucker Rod Pumped Wells. PETSOC-66-02-03. Journal of Canadian Petroleum Technology 5(2): 65-71.
Lea, J.F. 1991. Plunger Slippage and Its Effect on Production. SPE-22561. Presented at the SPE Annual Technical Conference and Exhibition, Dallas, Texas, 6-9 October.
Everitt, T.A. and Jennings, J.W. 1992. An Improved Finite-Difference Calculation of Downhole Dynamometer Cards for Sucker-Rod Pumps. SPE-23980. SPE Production Engineering 7(1): 121-127.
