We present a real-time machine learning framework to verify ultrasonic waveforms quality, a critical automation feature in the acquisition system orchestrating well integrity operations. The article describes two machine learning approaches to address the specific challenges of the problem, namely the high diversity of patterns and a limited number of wrongly recorded waveform examples present in the field. The first method uses physics simulations and a simple yet effective domain adaptation algorithm to fit field reality. The second method augments machine learning model with unsupervised deep learning and only uses field data. Tests have been carried out on several field datasets capturing different environments.
Presentation Date: Monday, October 12, 2020
Session Start Time: 1:50 PM
Presentation Time: 3:55 PM
Location: Poster Station 1
Presentation Type: Poster
