Assembly connects functional modules and components of products. The efficient and accurate assembly can improve performance of the product operation and maintenance. It is therefore essential to have an effective method for product assembly. Existing methods of the mechanical product assembly use mainly manual processes that rely on experience of operators. This paper proposes a reinforcement learning method to enable an automatic operation for improved efficiency and accuracy of the mechanical product assembly. A representation of the product assembly is proposed to build a machine learning model. The automatic assembly of product operations is planned by reinforcement learning agents. Constraints of assembly operations are considered to develop searching strategies of the maximum reward for the optimal solution of assembly operations. A quantitative method is proposed to measure efficiency of assembly operations based on the operation time. The proposed method has been applied in the assembly improvement of function modules of an industrial machine.

装配连接产品的功能模块和组件。高效，准确的组装可以提高产品运行和维护的性能。因此，必须有一种有效的产品组装方法。机械产品组装的现有方法主要使用依赖于操作员经验的手动过程。本文提出了一种强化学习方法，以实现自动操作，从而提高机械产品组装的效率和准确性。建议使用产品组装的表示形式来构建机器学习模型。产品操作的自动组装是由强化学习代理计划的。考虑组装作业的约束条件，以开发最大回报的搜索策略，以寻求组装作业的最佳解决方案。提出了一种基于作业时间来衡量装配作业效率的定量方法。所提出的方法已经应用于工业机器的功能模块的装配改进中。