To address the issues of imbalanced samples and edge deployment for ball screw pairs in engineering, this paper proposes a fault diagnosis method based on mechanism transfer. Firstly, by studying the fault mechanism similarities between rolling components, the comprehensive knowledge of rolling bearings is leveraged to supplement the sample space of ball screw pairs. Then, a dual-layer feature alignment knowledge distillation (DLFA-KD) framework is constructed. Within this framework, a lightweight student model is designed using multi-kernel depthwise separable convolutions (MK-DSC), while maximum mean discrepancy (MMD) and L2 distance are introduced to guide the alignment of features at dual-layer. Experiments demonstrate that this method significantly improves diagnostic accuracy under imbalanced sample conditions, peaking at 99.675%. Furthermore, after deployment on edge devices, the inference speed of the student model is 98.1 times faster than that of the teacher model. This method provides a new approach for edge-based intelligent fault diagnosis in industrial scenarios with imbalanced samples.