Automatic Assembly and Control Method of Automotive Transmission Gear Shaft Based on Improved SAM Model and SSD Algorithm

Abstract

Automotive transmission gear shaft assembly demands high precision, stability, and efficiency. Traditional methodsstruggle with occlusions, lighting, and interference. This research proposes a novel joint assembly and control method usingenhanced "segmentation anything" and "single-stage multi-box detector" models. The detection stage employs a dual backboneand feature pyramid for multi-scale fusion, boosting accuracy via an attention mechanism. The segmentation stage improvesfeature representation to compute assembly pose from centroids and spindle direction. Impedance control enables force-displacement coordination. Results show the improved detector achieves 91.5% average precision, with 90.1% IoU and 93.5%Dice score, surpassing mainstream models. In end-to-end tests, the method attained a 98.6% success rate, 0.43 mm alignmenterror, 49.7 N peak insertion force, 5.4 s cycle time, and a 1.8% repair rate. Optimal tolerance tests at 50 μm clearance yielded a98.8% success rate, 1.4% mismatch alarm, and 103 ms adjustment time. The method demonstrates high accuracy, stability, androbustness, significantly advancing automation in complex industrial assembly.