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

Cementing quality is a key factor in ensuring the long-term safe production of oil and gas wells and preventing defects. Traditional cementing quality evaluation mainly relies on logging interpreters manually analyzing acoustic logging data, such as Variable Density Logging (VDL) images and acoustic amplitude curves. This process is highly dependent on personal experience, labor-intensive, and inefficient. To address these issues, this paper proposes an automated cementing quality detection method, CemQ-CNN, based on a Convolutional Neural Network (CNN). In this context, "intelligent" refers to the model's ability to perform automatic classification from raw data, thereby increasing efficiency and consistency. This method constructs a multimodal input CNN model that can simultaneously process VDL images and acoustic logging curve data, achieving automatic, fast, and accurate classification of cementing quality. We collected and labeled 5,000 logging samples from 150 different wells across three distinct geological blocks, ensuring dataset diversity, categorizing them into three cementing quality levels: "good," "medium," and "poor." By allocating 70% of the data for training, 15% for validation, and 15% for testing, our model demonstrated Good performance on the test set. Experimental results show that the proposed method achieves an overall classification accuracy of 95.7%, demonstrating robust performance across all three quality classes ('Good', 'Medium', and 'Poor'), with a macro-average recall rate of 95.6% and a precision rate of 95.5%. Compared to models using a single data source, this multimodal model performs better. The study demonstrates that an effective intelligent method based on CNN can assist and standardize traditional manual interpretation, providing a reliable and innovative paradigm for cementing quality evaluation.