Visualization of 1D CNN Lithology Identification Model from Rotary Percussion Drilling Vibration Signals Using Explainable Artificial Intelligence Grad-CAM

In recent years deep learning has gained a lot of popularity because its ability to work on complex tasks. It been used in many industries optimize operations and help decision-making. Deep neural networks have often referred as ‘Black boxes’, that is they take inputs give outputs with high accuracies without giving an insight into how work. important demystify verify are looking at the correct patterns. This paper proposes use Gradient-Weighted Class Activation Mapping (Grad-CAM) visualize behavior lithology identification models drill vibrations input one- dimensional convolutional network (1D CNN). The models, time acceleration, frequency model had 99.8% 99.0% classification accuracy. could distinguish between granite marble rock based vibration signatures. With Grad-CAM, it was possible make 1D CNN transparent by visualizing regions were for predictions. Grad-CAM results indicated successfully learned significant frequencies contained each rock's signal.