Considering the non-stationary and nonlinear characteristics of on-load tap changer (OLTC) vibration signals, which are also easily affected by background noise, we propose an OLTC fault diagnosis method based on multisensor fusion (MSF) and a deep residual network (ResNet), leveraging the advantages of deep learning (DL). First, principal component analysis (PCA) is employed to transform the one-dimensional vibration signals measured by multiple sensors under different fault conditions into component matrices, and a signal-to-image conversion method is used to generate color images, which are then used as network inputs. Next, ResNet is utilized to extract fault features encompassed within the two-dimensional images. Finally, the proposed method is validated using data from a transformer factory test. The results demonstrate that the proposed method outperforms other DL-based methods in terms of accuracy, exhibits better noise resistance, and more effectively extracts feature information from OLTC vibration signals, achieving higher recognition accuracy.