Development of Automatic Recognition and Recording System for Rig Jobs

Abstract

Abstract Keeping a record of drilling process automatically in real time is a requisite for KPI studies and getting support from ROCs experts. There are five categories of technical data describing a well's drilling process, which are rig jobs in time sequence, job operational parameters, unit lithology by depth, wellbore configuration by depth and job appraisal respectively. Among them, the data of rig jobs in time sequence covers whole process of well construction and is collected manually via Daily Drilling Reports or Shift forms with little data quality control and is easy to be filled by mistakes or omissions. The other four types of data are mainly collected via mud logger, MWD/LWD tools, wire-line logging tools, etc and mostly realized automatic recording with strict data quality controls. In order to solve the shortcomings of data collecting and recording of rig jobs in time sequence, an automatic recognition and recording system for rig jobs is developed. Previously, the team had tried to identify 25 common rig jobs such as on-bottom drilling, tripping-in, tripping-out, circulating, reaming, etc by developing a judgment algorithm based on real-time operating parameters from mud logger and is proved to be inaccuracy and incompleteness due to insufficient data resolutions and limited data points especially when drill string is out of hole. This time, the team takes advantages of video data from on-site surveillance cameras and develops an automatic recognition and recording system for rig jobs by integration of AI video recognition algorithm with previous method. The AI video recognition is realized by applying U-Net, Attentioned, GAN, etc. and the AI model trained by a mass of labeled data. Finally, the system is deployed in sync with Wellsite Data Platform (WDP) and tested on line with real-time data from both cameras and mud logger. The online tests showed that the system could recognize various rig jobs from spud-in to end of well with correctness rate of more than 97% and improved the correctness rate of previous method by more than 20%. In May to November of 2020, the system was first deployed in sync with WDP in a deep exploration well GQ-6. The video and mud logging data of its six drilled offset wells were loaded, replayed, and the rig jobs in time sequence of each well were recognized and recorded in the WDP database. With share of the accurate rig jobs data, the ROC experts could easily find out the causes of NPT and ILT and optimize drilling operations to realize the well GQ-6 the fastest well in the block The system also save much of manual work for data record of rig jobs.