Optimised hammer drilling bit design using artificial neural networks trained by FDEM-generated data

For rapid and cost-effective hammer drilling, accurate prediction of rock impact response is crucial for designing optimal bits and maximising rock fragmentation. Current design optimisation workflows combine numerical simulations and experiments but often require numerous iterations to pinpoint the optimal design. Although physics-based models can potentially reduce experimental expenses, their significant computational demands present challenges when simulating the complex fragmentation dynamics during drill bit-rock interactions. This study introduces a data-driven artificial intelligence (AI) model, employing a multilayer perceptron (MLP) as a surrogate. The model leverages the hybrid finite-discrete element model (FDEM) as a powerful method in rock fracture mechanics to generate a sufficiently large training dataset. An automated workflow has been developed for generating the training data, comprising a pipeline that includes pre-processing, solving, and post-processing modules. Subsequently, the AI models were integrated into an optimisation framework alongside uncertainty quantification to demonstrate their potential in enhancing drilling efficiency through optimised bit design and operations. The MLP exhibits high accuracy in predicting key parameters, including rebound velocity, total crack length, quantities of fragments with different sizes and maximum contact force between rock and insert. Notably, this approach achieves real-time prediction compared to the 5-7 min simulation times of FDEM. Integrating this data-driven model into a design framework enables rapid assessment of different bit designs under various operational conditions. More broadly, this approach has the potential to impact other applications, such as digital twins, serving as a forward and inverse model for predicting rock type and optimising drilling performance.