Abstract: Automated mechanical assembly represents a critical and highly challenging task in modern industrial manufacturing. The core difficulty stems from the geometric complexity of components and the diversity of contact constraints, which lead to the theoretical search space of assembly process growing exponentially, thus obtaining feasible assembly sequences under such constraints become highly difficult. To address this challenge, we introduces a physics-based recursive algorithm for automatic assembly sequence generation, designed to produce physically-valid sequences within acceptable computation time. The proposed method first employs a recursive assembly strategy to enhance the probability of discovering feasible sequences. It then incorporates an efficient greedy search mechanism to accelerate the iterative search process. Finally, physical constraints are integrated to prune the solution space, thereby reducing computational complexity and further improving efficiency. Experimental evaluation on public mechanical assembly datasets demonstrates that our approach achieves improvements in both success rate and computational efficiency, outperforming baseline methods by an average of 20% to 40%.