The health status of On-Load Tap Changers (OLTC) is critical for high-voltage direct current (HVDC) transmission systems. With the development of artificial intelligence, various machine learning techniques have been employed for OLTC fault diagnosis. Currently, these studies mainly focus on establishing richer fault feature vectors or improving the performance of fault diagnosis models. However, the limitations of relying on single vibration information are often neglected. Consequently, this paper introduces a diagnosis method that combines multiple vibration information, utilizing the Light Gradient Boosting Machine (LightGBM). Firstly, data of normal and faulty conditions of UCG-type OLTC are collected, and time-frequency features are extracted. Secondly, a multi-sensor fusion decision-making method is designed based on LightGBM. Finally, this method is compared with five conventional diagnostic techniques, and the fusion results from different sensors are analyzed. The findings indicate that the proposed multi-sensor feature fusion approach significantly enhances the accuracy and reliability of fault diagnosis, offering a novel solution for early OLTC fault detection.