Intelligent detection of thermal damage and mechanical failure in polycrystalline diamond compact bits for deep high-temperature geothermal drilling

Abstract

In deep geothermal resource development, the combination of high temperatures and highly abrasive hard-rock formations causes frequent thermal damage and mechanical failure of polycrystalline diamond compact (PDC) bits, severely limiting drilling efficiency and cost-effectiveness. Traditional manual grading methods struggle to identify subtle early-stage thermal damage under harsh field conditions. To address this challenge, this paper proposes DABD-Net, a deep learning network specifically designed for intelligent damage recognition of PDC bits in geothermal drilling environments. Tailored to the subtle thermal damage and complex corrosive-abrasive wear patterns characteristic of geothermal drilling, the model incorporates a Dynamic Fusion Residual Network (DFRNet) to improve the capture of microscopic thermal damage textures. A dedicated image dataset comprising seven typical damage types was also constructed. The backbone of DABD-Net is an optimized DFRNet designed specifically for extracting damage features from PDC bits. This hybrid architecture efficiently captures local and directional features using PCAConv and EMA modules, while a Transformer module models long-range contextual relationships. In the neck, a Dual-Path Aggregation Module (DPAM) is introduced to enhance multi-scale feature localization by effectively integrating high-level semantic information with low-level spatial details. Experimental results show that DABD-Net achieves significantly higher average precision and precision rates on the test set compared with RT-DETR and YOLO-series models. In particular, for thermal damage (the most critical and difficult-to-detect failure mode in geothermal drilling), the proposed model improves detection precision by approximately 150% over baseline models. Recognition accuracy for corrosion-abrasion damage induced by geothermal fluids also improved significantly. The proposed method enables end-to-end high-precision detection under complex lighting and background conditions at the field site after tripping out, providing an intelligent tool for post-trip bit assessment and bit-retirement decisions in geothermal drilling.