Gains:
- Ability to understand the physical meaning of well logs (gamma, resistance, density, neutron, sonic) and their construction and interpretation with AI
- Ability to carry out AI-supported calculation of petrophysical quantities such as porosity, water saturation and clay volume and facies preliminary classification
- Ability to verify petrophysical interpretations produced by AI with core data, Archi equation and physical boundaries
When a well is drilled, we cannot see directly below the ground; Instead, we get well logs. A well log is a continuous record of physical properties measured against depth throughout the well. An instrument (log tool) is lowered into the well and, as it is pulled up, measures the radioactivity, electrical resistivity, density, and more of the surrounding rock. Petrophysics is the science of extracting the properties of the rock and the fluids in it (how many voids are there, is there water or oil in the voids, can the fluid flow) by reading these logs. In this unit, we use AI in two powerful places in petrophysics: structuring dispersed log data and producing fast, auditable blueprints for petrophysical interpretation. The unchanging rule: the log is an indirect measurement, the core tells the truth, the engineer confirms the interpretation.
Let's Get to Know Basic Logs
Each log has a physical quantity it measures and what it says:
- Gamma ray (GR): Natural radioactivity of the rock. Clay and shale give high values, clean sand and limestone give low values. It is the basis of clay volume (Vshale) estimation.
- Resistivity: The resistance of the rock to electricity. Water conducts electricity (low resistance), oil/gas does not (high resistance). Resistance is therefore the main clue to the presence of hydrocarbons.
- Density (RHOB): The bulk density of the rock. It is used in porosity calculation; As space increases, density decreases.
- Neutron (NPHI): Sensitive to the amount of hydrogen in the rock; It is an indicator of porosity because it contains water and hydrocarbon hydrogen. It reveals gas zones with density ("neutron-density crossover").
- Sonic (DT): The time it takes for a sound wave to travel through the rock. It provides information about porosity and rock strength.
Tip: A single log is never interpreted in isolation. In petrophysics, the power is in reading the logs together: gamma clay, resistivity fluid, density-neutron porosity and gas are described together.
Configuring Log Data
Log data mostly comes in LAS (Log ASCII Standard) files; but curve names (RHOB or DEN?), units (API or gr/cc), and depth references vary from site to site. AI is very powerful at standardizing this mess.
Critical principle: seek transformation, not inference. The model should match the curve explicitly written in the file to the standard name, and not "guess" and produce a curve that does not exist. If there is a missing log, it should be written as "none" and not made up.
Step by step: log comment workflow
- Take out the crooked inventory. Which logs are present, which are missing, what are the units? Have AI tabulate inventory, you verify.
- Do quality control. In zones with enlarged hole diameter (caliper), the density log is distorted; Mark these zones. AI can scan for outliers/impossible values (negative porosity, unlimited resistance).
- Calculate the volume of clay. Vshale from gamma log; You determine the clean and clay line references.
- Calculate the porosity. From density and/or neutron log, matrix and fluid densities to local values.
- Calculate water saturation. With the Archi equation; then hydrocarbon saturation = 1 − Sw.
- Connect with core. If there is core porosity/permeability at the same depth, calibrate the calculation with that.
Archival Equation: Verification Anchor
At the heart of petrophysics stands the Archi equation. It simply physically connects that water saturation decreases as the resistance of the rock increases (water decreases and hydrocarbon increases). The inputs to the equation are porosity, formation water resistance, and measured resistance; Its output is water saturation Sw.
When the AI suggests a Sw value, don't blindly accept it: put the inputs into the Archi equation and supply them manually. The result must be between 0–1; local coefficients (a, m, n) must be within reasonable range; must be calibrated by core. This is exactly an example of how to verify a non-safety-critical but economically critical output.
Caution: The arch equation does not work as it does in clayey/shale formations; Clay brings additional conductivity, making water saturation higher than it actually is. In clayey zones, clay-corrected models such as Simandoux or Waxman-Smits are required. If AI says "Archi with Sw=0.3", be sure to question whether the formation is clayey.
Three Mini Cases: By the Numbers
Case 1 — Caliper trap. In one well, the model marked 2,310–2,325 m as a “good reservoir” with high porosity (32%). The petrophysicist looked at the caliper log: in this interval the hole had collapsed, its diameter had expanded; The density log was therefore giving falsely low density (high false porosity). Core porosity was 14%. AI mistook corrupted data for real; Caliper control solved the trap.
Case 2 — Correct capture of the gas zone. In two adjacent wells, AI suggested a gas zone from neutron-density crossover (neutron low, density high porosity—classic gas effect). When the engineer linked this with production testing, it was confirmed that the zone was indeed producing gas. Here the AI pointed out the correct physical signature and was confirmed by well testing.
Case 3 — Exaggerated hydrocarbon in clayey zone. The model assigned Sw=0.25 (i.e. 75% hydrocarbon) to a zone with flat Archi. The gamma log showed that this zone was clayey; clay should have lowered the resistance and actually increased the Sw. With Simandoux correction, Sw = 0.55; The zone was much wetter than expected and had lower commercial value. Blind acceptance could lead to investment mistakes.
Weak Prompt / Strong Prompt
Weak prompt:
Calculate porosity and water saturation from this log.[log data]
Powerful prompt:
Generate petrophysical interpretation DRAFT from the following (anonymized) log data. Rules:- First list the curve inventory and missing/suspicious logs. Mark zones with caliper expansion as "density questionable". - Calculate Vshale from gamma log; Obtain clean/clay line references by asking. - Calculate porosity with density-neutron; EXPRESSLY write matrix and fluid density assumptions.- Calculate water saturation with Archi; If Vshale is high, suggest Simandoux and write a reason. Specify the coefficients a, m, n, if hypothetical. - Check each result with the physical limit (0<=Sw<=1, 0<=porosity<=0.4). - Mark the steps that require calibration with core. Don't present any results as "definitive".Data: [log data]
Four Copiable Templates
1) Curve inventory and QC:
Map this list of LAS curves to standard names (GR, RESD, RHOB, NPHI, DT, CALI). Write "none" for the missing ones, do not make them up. If there is unit discrepancy and impossible value (negative porosity, excessive resistance) give "row | problem | suggestion" table. List: [curve/data]
2) Vshale account draft:
Calculate Vshale from gamma log. Ask me for clean line (GRmin) and clay line (GRmax) references, don't assume. Give linear and non-linear (Larionov) methods separately and explain the difference. Data: [GR]
3) Water saturation and model selection:
Calculate water saturation with porosity, resistivity and Vshale below. Justify whether Archi or Simandoux is suitable according to the Vshale threshold. Write the coefficient assumptions a, m, n. Check the result with 0-1 limit. Data: [values]
4) Core-log calibration control:
Below are the log-derived porosity and core porosity at the same depths. Summarize the systematic bias and scatter; Give suggestions for calibrating log interpretation. Mark extreme outliers for data quality.Data: [log vs core]
Log Comment Validation Hierarchy
evidence
directness
what does it say
Usage
Core (core)
most direct
True porosity/permeability
Calibration, critical zone
Well test (pressure-flow rate)
direct
True flow, permeability
Productivity confirmation
Log (density, resistance...)
indirect
Continuous profile
Main comment, fill in the blank
AI prediction
derived
Hypothesis, screening
outline, marking
Common mistakes
- Bypassing caliper/data quality. Mistaking the distorted density in ruined hole zones as real porosity.
- Single log comment. Looking only at porosity and neglecting the context of resistivity and gamma.
- Straight Archi in the clayey zone. Overestimating hydrocarbon saturation without clay correction.
- Not calibrating with core. Moving the log comment to the signature without directly connecting it to the measurement.
- Hiding the coefficients. Presenting results without writing down a, m, n and matrix/fluid assumptions.
In summary
- Well logs are indirect measurements; are read together and calibrated by core.
- AI is powerful in structuring log data and drafting petrophysical interpretation; but it requires transformation, not inference/fabrication.
- The Archi equation is the physical verification anchor for water saturation; Clay-corrected models are required in clayey zones.
- Caliper and data quality control prevents spurious porosity traps.
- Core and well testing are at the top of the evidence hierarchy; At the bottom, AI prediction alone is insufficient.
Application task
Select a 20–30 meter interval from a (representative) log set of a well. Generate curve inventory, Vshale, porosity and water saturation draft with the powerful prompt. Then: (1) check if there is a zone with caliper expansion, (2) manually provide the Sw result by putting inputs into the Archi equation, (3) discuss whether Simandoux is needed if one of the zones is clayey. Question the model's assumption (e.g. matrix density) at at least one point.
checklist
- [ ] I know the physical meaning and reading together of basic logs (GR, resistance, density, neutron, sonic).
- [ ] I want transformation while structuring the log data, I do not make up the missing log.
- [ ] I control porosity and water saturation with physical boundaries and Archi/Simandoux.
- [ ] I mark caliper/data quality traps (ruined hole).
- [ ] I calibrate and verify the log interpretation with core and well testing.