Sampling rate should be chosen from the shortest behavior that must remain visible, not from a generic preference for more data. Acquisition, analysis, and display can run at different rates. Preserve the raw evidence needed for later investigation, then aggregate each analytical product according to its physical timescale.
Sampling determines what can be observed
A three-second hook-load event appears differently at each sampling rate:
| Sampling rate | Approximate observations inside the event |
|---|---|
| One sample per 10 seconds | 0; the event may be missed entirely |
| 1 Hz | 3 |
| 10 Hz | 30 |
Nothing about the physical event changed. Only the observation system changed.
Events shorter than the sampling interval may never enter the stored dataset.
If a 10-Hz signal is reduced to one sample every ten seconds, later interpolation cannot recreate a two-second torque disturbance that was discarded. It creates estimates between retained points, not new measurements. Raw-data retention is valuable when future analytical questions are not yet known.
Filtering trades noise for event preservation
Consider a signal:
$$x(t) = signal(t) + noise(t)$$
Averaging can suppress noise. That is useful. But averaging also suppresses rapid changes in the signal itself. If the desired information is a slow trend, that is usually acceptable. If the desired information is the rapid change, it is destructive. This is why the correct filter depends on the question.
- Long-term friction trend: Some smoothing may be highly useful.
- Maximum overpull: Averaging can materially understate the value.
- Formation-scale MSE trend: Depth aggregation may clarify the pattern.
- Short torsional disturbance: The same aggregation may remove it.
The objective should define the filtering, not the desire for a visually smooth trace.
Channels and clocks have different requirements
Mud weight rarely needs ten measurements per second; hook load during a mechanical event may. Different information classes operate at different timescales, so a mature architecture does not force every channel to one frequency. SPE-191797 describes an architecture combining real-time streams, contextual information, PLC data, WITS/WITSML, and mudlogging inside one aggregation layer.[1] The requirement is sufficient synchronization for the engineering application, not identical frequency. At 10 Hz, a two-second delay creates a 20-sample offset.
Now imagine investigating whether:
- torque changed before pressure,
- pressure changed before flow,
- WOB changed before ROP.
The event sequence can become incorrect. At higher sampling frequencies, timestamp quality and synchronization therefore become increasingly important. More temporal resolution is useful only if the channels are actually aligned in time.
Separate acquisition, analysis, and display
Torque may arrive at 10 Hz without every model or display running at 10 Hz. Separate:
- Acquisition frequency: How often the original data is captured.
- Analytical frequency: How often a particular KPI or model needs to be evaluated.
- Display frequency: How often the user interface needs to redraw.
Those three rates can be very different.
Match aggregation to the statistic
Hydraulic surveillance provides another useful example. A circulating-system failure can develop:
- rapidly,
- gradually.
IADC/SPE-189700 analyzed pressure behavior on multiple timescales. Slow changes in hydraulic behavior were tracked over longer windows, while more sudden pressure behavior was evaluated over a much shorter window.[2] That illustrates a powerful principle: one event-detection problem can require multiple analytical timescales simultaneously. A useful architecture may therefore run:
- short-window detectors,
- medium-window trends,
- long-horizon state estimates
on the same underlying data. A 1-Hz source can still support:
- 5-second behavior: using five samples.
- 2-minute trend: using 120 samples.
- 30-minute trend: using 1,800 samples.
- Stand-level summary: using all samples associated with the stand.
The source sampling rate sets the maximum retained temporal detail; it does not force every analysis to use the same window. Reducing 10-Hz hook-load data to one value per second also requires choosing which statistic to retain:
- Mean: Good for general level.
- Median: Robust to isolated extremes.
- Maximum: Useful if peak overpull matters.
- Minimum: Useful for certain set-down events.
- Range: Useful for variability.
Different aggregation functions preserve different properties. For an overpull analysis, using only the mean may suppress precisely the quantity of interest. For long-term trend analysis, maximum values may exaggerate isolated spikes. Therefore downsampling involves two decisions:
-
time interval,
-
statistic retained.
One practical visualization technique is to preserve both:
- minimum,
- maximum
within each displayed interval. If 100 raw torque samples must be represented by one screen pixel, a mean shows only central tendency while a min/max envelope preserves the fact that substantial oscillation occurred. This supports long-window visualization without rendering every sample individually.
A practical multiscale architecture
A mature drilling-data system can maintain four paths:
| Level | Representation | Primary use |
|---|---|---|
| Raw | Highest useful available frequency | Forensic review and transient algorithms |
| Real-time operational | Synchronized, quality-checked channels | Rig state, calculations, alerts |
| Event / stand | Connections, overpulls, stand statistics | Workflow and drilling-performance analysis |
| Depth / well / fleet | Depth bins, formation statistics, well KPIs | Offsets, benchmarking, reporting |
No one level is universally better. Each answers a different question.
A stand-level KPI locates the interval; linked one-second traces retain the evidence needed to diagnose it. Aggregation should narrow the search, not sever provenance.
A practical design asks five questions:
- Is the physical behavior a transient, operational state, stand-scale trend, or formation-scale trend?
- Does it last seconds, minutes, hours, or hundreds of feet?
- Must the output preserve an extreme, oscillation, mean level, or cumulative duration?
- Which rig state, mechanical, geological, and BHA context must be synchronized?
- Will a later engineer need enough raw evidence to reconstruct the event?
Related technical resources
- Depth-Based and Time-Based Analysis of Real-Time Drilling Data
- Why Drilling Analytics Need Memory: The Role of Moving Windows and Historical State
- Why Rig-State Classification Is Foundational to Real-Time Drilling Analytics
References
-
Shahri, M., Kutlu, B., Thetford, T., Nelson, B., Wilson, T., Behounek, M., Ambrus, A., and Ashok, P. Adopting Physical Models in Real-Time Drilling Application: Wellbore Hydraulics. SPE-191797-MS, SPE Liquids-Rich Basins Conference, North America, Midland, Texas, 2018.
-
Ambrus, A., Ashok, P., Ramos, D., Chintapalli, A., Susich, A., Thetford, T., Nelson, B., Shahri, M., McNab, J., and Behounek, M. Self-Learning Probabilistic Detection and Alerting of Drillstring Washout and Pump Failure Incidents During Drilling Operations. IADC/SPE-189700-MS, IADC/SPE Drilling Conference and Exhibition, Fort Worth, Texas, 2018.