Research on Drill String Vibration Identification Based on Hybrid Convolutional Neural Network

Drill string vibration is a key factor affecting the safety and efficiency of oil and gas drilling. Its multimodal vibration characteristics under complex conditions (such as the coupling of lateral, longitudinal, and torsional vibrations) are difficult to classify accurately using traditional signal processing methods. This paper proposes an intelligent recognition method for drill string vibration status based on a hybrid model of Convolutional Neural Network (CNN) and Support Vector Machine (SVM). First, triaxial vibration acceleration time-series data is collected through a downhole measurement-while-drilling system to construct a dataset containing four typical working conditions (general vibration, whirl, initial stick-slip, and complete stick-slip). A CNN architecture is designed to address the non-stationary characteristics of vibration signals: multi-scale convolutional kernels are used in parallel to extract local time-domain features. Subsequently, the deep features extracted by CNN are fed into SVM for classification, and a global max-pooling layer is introduced to suppress noise interference, completing the final vibration status recognition. Experimental results show that this hybrid model method achieves an average classification accuracy of 94.2% on the test set. This study provides a new data-driven approach for real-time diagnosis of complex downhole conditions, which has significant engineering application value for reducing unplanned drilling interruptions and optimizing drilling parameters.