Abstract
Hidden violations are prevalent in complex activities on onshore drilling platforms, posing significant risks to safe operations. Despite ongoing QHSE compliance efforts, unsafe behaviors persist among personnel. This study presents a deep learning-based classification model that automatically categorizes hidden hazard and violation descriptions. Each classification corresponds to a specific negative score, which is ultimately reflected in employees' job performance. The goal is to enhance safety awareness and foster a stronger QHSE culture among well personnel.
This study utilizes over 20,000 data points, with the bge-large-zh semantic vector model fine-tuned to classify hidden defects into eight categories, including defects in well control facilities and special equipment. Violations are categorized into seven major groups and 20 subcategories, such as illegal operations, unauthorized commands, and labor discipline violations. The model achieves a classification accuracy of 95%, significantly reducing the need for manual inspection. By automating the identification of hidden hazards, the model provides technical support for the QHSE safety supervision platform and offers engineers new insights into applying fine-tuned semantic vector models for real-time risk assessment in drilling operations.
