A Novel Machine Learning Method for Processing of P and S Suspension Logging Data

1. Home
2. Publications
3. Online Library
4. A Novel Machine Learning Method for Processing of P and S Suspension Logging Data

P-wave and S-wave velocities are used as input for calculating key parameters for offshore foundation design, such as small strain shear modulus, and as input for enhancement of seismic reflection data. P and S suspension logging (PSSL) is a common borehole geophysical logging technique for deriving these velocities in soil and rock. The PSSL tool generates acoustic waves in the borehole and records them, as a trace pair, at two receivers with different spacings from the source. P- and S-wave velocities are derived from the difference in arrival times of the acoustic waves at both receivers. Typically, a competent human user picks the arrival times, i.e. manual picking, which is a time-consuming exercise. This paper presents a machine learning algorithm, known as P and S Interpretation Network (PSINET), that automates the picking process. The algorithm output consists of picked arrival times for both P- and S-wave data, from which the acoustic velocities are derived. PSINET is a deep neural network, trained with a dataset of more than 100,000 manually picked and reviewed PSSL trace pairs. The use of PSINET has been successfully piloted on a public domain dataset (blind data). Results of this campaign show that PSINET predicts correct outputs for 86% of the P-wave data and 44% of the S-wave data. Review by a competent user is still required. PSINET reduces deliverables turnaround time and decreases the influence of interpreter bias. Further training of PSINET is planned to cover a wider range of ground conditions.

5th International Symposium on Frontiers in Offshore Geotechnics (ISFOG2025)

5 - Data Analytics and Machine Learning