Gaborlet-guided sparse filtering: A novel intelligent method for lithology identification by vibration signals while drilling

Abstract

Traditional lithology identification methods mainly rely on core sampling and well-logging data. Although such methods offer high accuracy, they are expensive, time-consuming, and lack real-time applicability. A large amount of real-time vibration data is generated during the rock drilling. Therefore, a method of vibration signals during drilling was explored to conduct perceptual prediction of rock lithology. However, these signals are often contaminated by noise from other equipment. Therefore, how to effectively extract lithology-related features from noisy signals has become both a fundamental and challenging task in lithology identification. To address this issue, this study proposes a novel Gaborlet-guided sparse filtering (GSF) method to enhance the extraction of informative features from drilling vibration signals. Specifically, a set of Gabor wavelets is first designed to initialize the weight vectors of the sparse filter (SF), enabling the capture of multiscale and multidirectional features. The objective function of the SF is regularized using the L_1/2 norm to mitigate overfitting. The well-designed GSF model is then directly applied to the raw drilling vibration signals for feature extraction. Finally, the extracted features are fed into a Softmax classifier for lithology identification. One experimental case is conducted to validate the effectiveness of the proposed GSF method. The results demonstrate that the proposed approach achieves classification accuracies of 94.07% on the training set and 88.32% on the testing set. This technique enables precise lithology clustering and outperforms the other three lithology identification methods discussed in this article.