Advancing Downhole State Detection in Coiled Tubing Operations with a Novel Deep Learning Approach

Abstract Coiled tubing offers efficient solutions for various well intervention and completion operations, significantly enhancing productivity and safety in the oil and gas industry. Within this modern coiled tubing process, plugs are crucial for achieving zonal isolation in multi-stage hydraulic fracturing operations, ensuring that treatment fluids and proppants are precisely delivered to targeted intervals. After the fracturing process, these plugs are milled away for subsequent process and production. In this milling operation, identifying downhole states, such as the bottom-hole assembly (BHA) contacting and milling through the plugs (e.g. tag plug and through plug), is crucial for optimizing performance, protecting equipment, and reducing non-productive time (NPT). Decisions regarding tag plug and through plug processes were traditionally made based on human interpretation, which varied significantly due to differences in the experience levels of operators and organizational policies. This approach was inherently inconsistent and susceptible to human error. This paper presents a transformer-based deep learning framework, leveraging its attention mechanism to dynamically focus on crucial time-series features within real-time operational states in coiled tubing plug milling jobs data. By preprocessing surface weight time series data to reduce noise and enhance signal quality, the framework segments data into moving windows and predicts binary indicators (0 or 1) for each time frame. The model was trained on one set of coiled tubing data and tested on entirely new, unseen datasets from a different operation, demonstrating exceptional accuracy in detection compared to field engineers and strong generalization in real-world deployment. This approach provides robust real-time statistical decision support for coiled tubing operators, enhances operational efficiency, and minimizes disruptions. It highlights the potential of deep-learning transformer models to address the complexities of coiled tubing operations and improve industry practices through AI-driven automation and precision.