When the available evidence cannot separate plausible causes, the correct output is not normal and not fault; it is unable to determine. That state protects the distinction between missing evidence and evidence of absence, and it prevents probability or fluent language from turning an underdetermined problem into a false diagnosis.
Use a three-state decision
Instead of:
$$Output \in { Normal,\ Fault }$$
consider:
$$Output \in { Normal,\ Fault,\ Unable\ to\ Determine }$$
Now the model can distinguish:
- Evidence supports normal: The available observations are sufficiently consistent with expected behavior.
- Evidence supports fault: The observations provide adequate evidence for the fault interpretation.
- Unable to determine: The system does not have enough reliable information to distinguish competing explanations.
This better represents engineering reality. If the hook-load sensor stops transmitting, the system can no longer calculate:
- hook-load trend,
- hook-load deviation,
- several derived mechanical indicators.
Assigning missing features a healthy default makes the model appear more confident even though it only lost information. In the real-time tripping-dysfunction model described by SPE-191426, unavailable hook-load data makes the trend feature unknown, not constant or erratic. The network therefore cannot place weight on evidence that does not exist.
This is a fundamental analytical principle: Absence of evidence is not evidence of normal operation.
Match specificity to evidence
The hydraulic example makes this clear. Available evidence shows:
- measured pressure is below model,
- flow in is stable,
- pressure trend is persistently decreasing.
That may strongly support a circulating-system abnormality, while washout and pump failure still share too many signatures to separate.
Now add flow-out behavior. Perhaps the two failure modes become easier to separate. So we can think of diagnostic specificity as a hierarchy.
- Level 1: Something unusual is happening.
- Level 2: The abnormality appears hydraulic.
- Level 3: The evidence supports a washout-or-pump-failure family.
- Level 4: The evidence distinguishes one specific failure mode.
Each step requires more evidence.
Each higher level of diagnostic specificity requires more discriminating evidence.
If the evidence supports washout or pump failure with strong confidence, a system that reports:
Hydraulic system abnormality: washout or pump failure should be investigated.
has provided useful information. A more specific system might report:
Drillstring washout detected.
That is more specific, but potentially wrong. Specificity is valuable only when the evidence supports it. The same boundary applies elsewhere:
| Domain | Evidence may support | Evidence may not prove |
|---|---|---|
| Torque and drag | Abnormal drag | A cuttings bed at a specific depth |
| Bit performance | Deteriorating bit effectiveness | Shoulder-cutter thermal damage |
| Hole cleaning | Poor hole condition is likely | Exact cuttings-bed geometry |
Engineering analytics should preserve the distinction between observation and inference.
Confidence cannot replace missing information
A probability can make uncertainty look precise. A result of 52% washout and 48% pump failure does not meaningfully distinguish the two. A useful system should recognize:
$$|P_1-P_2| \approx 0$$
and report specific root cause unresolved. With pressure, flow in, flow out, model prediction, and rig state, a specific diagnosis may become possible. Without flow out, abnormal hydraulic behavior may still be evident, but the missing discriminator must broaden the conclusion. Diagnostic specificity should decrease when discriminating evidence disappears.
Represent model and evidence conflict
Missing data is not the only reason a system should abstain. Suppose every sensor is functioning. But the physical model itself has large uncertainty under the current conditions. Examples might include:
- uncertain mud rheology,
- uncertain hole geometry,
- unknown washout,
- poor friction-factor calibration,
- uncertain bit condition.
Now the measurements may be precise while the expected model response is uncertain. A measurement-model residual is less meaningful when the model uncertainty band is broad. SPE-181076 explicitly discusses uncertainty in both the sensor reading and the model prediction. A hypothetical stuck-pipe surveillance problem could contain conflicting evidence:
| Evidence | Interpretation |
|---|---|
| Torque increasing | Supports worsening hole condition |
| Hook-load behavior normal | Does not support worsening condition |
| Hole-cleaning activity extensive | Argues against severe accumulation |
| Flow below historical norm | May support concern |
What should the system do? A poor implementation may choose whichever signal has the largest numerical score. A stronger implementation represents the conflict. Possible output:
Evidence is mixed. Current data does not support a confident hole-condition diagnosis.
This is not indecision. It is an accurate representation of the state of knowledge.
Explain the limit
A bare Unable to determine is not very useful. A better result explains the limitation:
- Unable to distinguish washout from pump failure because flow-out data is unavailable.
- Unable to assess hook-load trend because the channel has not updated for 45 seconds.
- Unable to compare current MSE with the historical baseline because the formation has not been resolved.
The explanation converts uncertainty into an actionable next step by identifying the missing information. In a hypothetical real-time event, the measured evidence is:
- flow in = stable,
- standpipe pressure falls 300 psi,
- modeled SPP remains approximately stable,
- rig state = rotary drilling.
At first, both washout and pump failure are plausible. If flow out is available and remains stable while pressure falls, the model may become more specific. If it is missing, the discriminating evidence disappears and the output should broaden:
Hydraulic pressure-loss anomaly detected. Current evidence cannot distinguish washout from pump failure because flow-out data is unavailable.
That answer tells the engineer:
- what changed,
- what the model knows,
- what it does not know,
- why.
Loss of a discriminating sensor should reduce diagnostic specificity rather than leave confidence unchanged.
Make abstention a data product
A practical abstention framework records the reason:
| Reason | Meaning |
|---|---|
| Required sensor missing | A key discriminating signal cannot be evaluated. |
| Context unresolved | The BHA, wellbore, formation, or other required context is unknown. |
| Outside validated domain | The current state falls outside the model's intended conditions. |
| Causes indistinguishable | Several explanations fit the evidence almost equally well. |
| Evidence conflict | Reliable sources point in materially different directions. |
| Data quality inadequate | Sensor confidence is too low. |
| History insufficient | A trend or persistence calculation lacks enough data. |
Any of these can justify Unable to Determine or a deliberately broader conclusion. The resulting engineering event can still record the event family, unresolved diagnosis, family-level confidence, missing and observed evidence, and the next check.
The important design principle is separation. Confidence in: something is wrong does not have to equal confidence in: exactly what is wrong. The same principle applies to assistant workflows. When supporting reports or notes are unavailable, the response should identify the missing evidence and stop at the most specific conclusion supported by the available telemetry.
This reconstructed AIDE status card is an explanatory figure, not a native application capture. It separates what telemetry supports from the run objective that cannot be verified without daily-report and notes context.
Related technical resources
- Distinguishing Sensor Faults from Real Drilling Process Changes
- Probabilistic Reasoning for Drilling Dysfunction Detection
- Evaluating the Reliability of an AI Drilling Assistant
References
-
Ashok, P., Ambrus, A., Ramos, D., Lutteringer, J., Behounek, M., Yang, Y. L., Thetford, T., and Weaver, T. A Step by Step Approach to Improving Data Quality in Drilling Operations: Field Trials in North America. SPE-181076-MS, SPE Intelligent Energy International Conference and Exhibition, Aberdeen, Scotland, 2016.
-
Shahri, M., Wilson, T., Thetford, T., Nelson, B., Behounek, M., Ambrus, A., D'Angelo, J., and Ashok, P. Implementation of a Fully Automated Real-Time Torque and Drag Model for Improving Drilling Performance: Case Study. SPE-191426-MS, SPE Annual Technical Conference and Exhibition, Dallas, Texas, 2018.