Physics-driven feature alignment combined with dynamic distribution adaptation for three-cylinder drilling pump cross-speed fault diagnosis

Abstract

Monitoring the health of drilling pumps and diagnosing faults is crucial for the smooth operation of oil drilling activities. However, existing deep learning algorithms struggle with varying speed conditions. To address this, we propose a method that combines physics-driven feature alignment with Dynamic Distribution Adaptation (DDA) for diagnosing drilling pump faults across varying speeds. Our physics-driven feature alignment method reduces discrepancies between samples by aligning amplitude values, angular sampling frequencies, and impulse phases of signals. DDA further improves cross-domain feature matching by dynamically adjusting marginal and conditional distributions during domain adaptation. Experiments on a three-cylinder drilling pump platform under varying speed conditions demonstrate the superior performance of our method compared to state-of-the-art approaches.