Well logging tool and interpretation framework that employs a system of artificial neural networks for quantifying mud and formation electromagnetic properties
Methods and systems are provided that predict electromagnetic properties of drilling mud and a formation, which involve a logging tool that measures current injected into a measurement zone adjacent a sensor electrode at multiple frequencies. The measured currents at the multiple frequencies are processed to determine complex impedances for the sensor electrode at the multiple frequencies. The complex impedances are used to generate input data, which is supplied to a system of artificial neural networks (ANNs) that is configured to predict and output electromagnetic properties of the drilling mud and the formation within the measurement zone and possibly tool standoff based on the input data. The system of ANNs can employ a cascaded architecture of multiple ANNs. The electromagnetic properties or tool standoff predicted by the system of ANNs can be used to construct a borehole image over varying azimuth and depth.
1. A method of predicting electromagnetic properties of drilling mud and a formation, the method comprising:
configuring a logging tool to i) apply an alternating voltage of multiple frequencies to a sensor electrode that injects current into a measurement zone disposed adjacent the sensor electrode, wherein the measurement zone includes the drilling mud and the formation, and ii) measure values of the current injected into the measurement zone at the multiple frequencies;
processing the measured values of the current at the multiple frequencies to determine complex impedances for the sensor electrode at the multiple frequencies;
generating input data based on the complex impedances for the sensor electrode at the multiple frequencies;
supplying the input data to a system of artificial neural networks (ANNs) that is configured to predict and output the electromagnetic properties of the drilling mud and the formation within the measurement zone based on the input data, wherein the system of ANNs employs a cascaded architecture of multiple ANNs; and
generating a borehole image over varying azimuth and depth using the electromagnetic properties.
2. The method according to claim 1 , wherein:
the electromagnetic properties predicted and output by the system of ANNs include a formation resistivity as well as formation permittivities for the multiple frequencies.
3. The method according to claim 1 , wherein:
the electromagnetic properties predicted and output by the system of ANNs include mud permittivities for the multiple frequencies.
4. The method according to claim 1 , wherein:
the system of ANNs is further configured to predict and output a tool standoff for the sensor electrode based on the input data.
5. The method according to claim 4 , further comprising:
using the tool standoff predicted and output by the system of ANNs to construct the borehole image over varying azimuth and depth.
6. The method according to claim 5 , further comprising:
repeating or replicating operations of the method for multiple sensor electrodes, wherein the borehole image is constructed by combining the electromagnetic properties or the tool standoff predicted and output by the system of artificial neural networks for the multiple sensor electrodes.
7. The method according to claim 1 , wherein:
the system of ANNs employs a cascaded architecture of multiple feedforward ANNs.
8. The method according to claim 7 , wherein:
the cascaded architecture of multiple feedforward ANNs employs i) a first plurality of feedforward ANNs that predict and output mud impedance angles for the multiple frequencies, and ii) a second plurality of feedforward ANNs that predict and output formation permittivities for the multiple frequencies based on outputs of the first plurality of feedforward ANNs.
9. The method according to claim 8 , wherein:
the cascaded architecture of multiple feedforward ANNs further employs i) a third plurality of feedforward ANNs that predict and output ratios of tool standoff and mud permittivity for the multiple frequencies as well as mud permittivity for one frequency of the multiple frequencies, and ii) a computational module that calculates mud permittivity for another frequency of the multiple frequencies based on outputs of the third plurality of feedforward ANNs.
10. The method according to claim 9 , wherein:
the cascaded architecture of multiple feedforward ANNs further employs a fourth feedforward ANN that predicts and outputs formation resistivity and the tool standoff based on the outputs of the first plurality of feedforward ANNs, the outputs of the third plurality of feedforward ANNs and the computational module.
11. The method according to claim 10 , wherein:
the second plurality of feedforward ANNs also uses the formation resistivity and the tool standoff output by the fourth feedforward ANN as input in order to predict and output the formation permittivities for the multiple frequencies.
12. The method according to claim 1 , wherein:
the ANNs of the system of ANNs are trained using synthetic or previously-inverted field data.
13. The method according to claim 1 , wherein:
the ANNs of the system of ANNs are trained independently from one another.
14. The method according to claim 1 , further comprising:
storing in electronic form or outputting the electromagnetic properties predicted and output by the system of ANNs.
15. The method according to claim 1 , wherein:
the drilling mud comprises an oil-based mud.
16. A system for predicting electromagnetic properties of drilling mud and a formation, the system comprising at least one processor that, when executing program instructions stored in memory, is configured to:
obtain measurements of current injected into a measurement zone adjacent a sensor electrode at multiple frequencies, wherein the measurement zone includes the drilling mud and the formation;
process the measurements of the current at the multiple frequencies to determine complex impedances for the sensor electrode at the multiple frequencies;
generate input data based on the complex impedances for the sensor electrode at the multiple frequencies; and
supply the input data to a system of artificial neural networks (ANNs) that is configured to predict and output electromagnetic properties of the drilling mud and the formation within the measurement zone based on the input data, wherein the system of ANNs employs a cascaded architecture of multiple ANNs and
generate a borehole image over varying azimuth and depth using the electromagnetic properties.
17. The system according to claim 16 , wherein: the at least one processor is further configured to use a tool standoff predicted and output by the system of ANNs to construct the borehole image over varying azimuth and depth” to provide appropriate antecedence basis.
18. The system according to claim 17 , wherein:
the at least one processor is further configured to repeat or replicate the program instructions for multiple sensor electrodes, wherein the borehole image is constructed by combining the electromagnetic properties or the tool standoff predicted and output by the system of artificial neural networks for the multiple sensor electrodes.
19. The system according to claim 16 , wherein:
the measurements of the current at the multiple frequencies are performed by a downhole logging tool that applies an alternating voltage to the sensor electrode to inject the current into the measurement zone adjacent the sensor electrode.
20. The system according to claim 19 , wherein:
the downhole logging tool is one of a wireline logging tool or a measurement-while-drilling logging tool.
21. The system according to claim 16 , wherein:
the drilling mud comprises an oil-based mud.