A neural network prediction of drill string axial torsional vibration propagation
Abstract
Drill string vibration contributes largely to the premature failure of drill pipes and is crucial to ensure the safety and integrity of offshore drilling. However, a fast and accurate drill string vibration prediction is still challenging. This paper begins with a physical model of coupled axial torsional drill string vibrations using the lumped mass method. Based on the fact that vibration propagation can be represented by the modal decomposition, a neural network model is constructed and trained. The dynamic responses on certain locations on the drill string are utilized as the input of neural network model to predict the responses of each mode shapes; and the vibration propagation along the whole drill string can be reconstructed by the super-position of different mode shapes and their corresponding time-variant parameters. Finally, the performance of the neural network model on predicting dynamic responses of the drill string is discussed. The numbers of sensors and the deployment location are critical on the performance of the neural network model and are intensively discussed. This study provides a novel approach available for the real time prediction of drill string vibration and casts light on the future applications of artificial intelligence in offshore drilling engineering.