The invention provides a deep-learning-based intelligent assembling system of a product digital model. The system is composed of four levels: a first level being a user model layer, a second level being a key technique layer, a third level being a logic layer, and a fourth level being a data layer. The user module layer includes a parameterized modeling module and an intelligent assembling processing module in a product model design process. The key technique layer consists of a parameterized design module, a deep learning model establishment module and an intelligent assembling technique module. The logic layer includes a parameterized modeling mechanism module, a deep learning-based classification recognition module, and an intelligent assembling algorithm module. The data layer consists of a historical product assembling model library, a parameterized product assembling model library, and a new product assembling model library. According to the system, on the basis of digital model information of parts, an assembling relationship of part models is determined by using the learning property of a deep belief network, so that the three-dimensional assembling design of the product model becomes convenient, the design cycle is reduced, and the design efficiency is improved.