Automated cementing quality detection using a domain-specific, multi-scale convolutional neural network.

Area of Science:

• Petroleum Engineering
• Artificial Intelligence in Geosciences

Background:

• Cementing quality is crucial for safe, long-term oil and gas well production.
• Traditional manual analysis of acoustic logging data (VDL images, amplitude curves) is inefficient and experience-dependent.

Purpose of the Study:

• To develop an automated, efficient, and accurate method for cementing quality detection.
• To introduce the CemQ-CNN, a multimodal Convolutional Neural Network model for classifying cementing quality.

Main Methods:

• A multimodal CNN (CemQ-CNN) was designed to process both VDL images and acoustic logging curves simultaneously.
• A diverse dataset of 5,000 labeled logging samples from 150 wells across three geological blocks was utilized.
• Data was split into training (70%), validation (15%), and testing (15%) sets for model evaluation.

Main Results:

• The CemQ-CNN model achieved an overall classification accuracy of 95.7% on the test set.
• The model demonstrated robust performance across 'good,' 'medium,' and 'poor' cementing quality classes, with macro-average recall of 95.6% and precision of 95.5%.
• The multimodal approach outperformed models using single data sources.

Conclusions:

• The proposed CemQ-CNN offers an intelligent, automated solution for cementing quality evaluation, enhancing efficiency and consistency.
• This CNN-based method provides a reliable and innovative paradigm, assisting and standardizing traditional manual interpretation in oil and gas well operations.