In different stages of product life cycle, including design, simulation, manufacture, test and operation and maintenance, bag filter generates a large amount of data. It excavates the complex, non-linear and coupling internal relationship between big data of product and its operation characteristics, and provides a new way to solve the common problems of design innovation and operation and maintenance optimization in bag filter industry. Aiming at the characteristics of large data of bag filter, a large data mining XGBoost model for on-line monitoring of bag breakage of bag filter is proposed, and the parameter optimization method of XGBoost model based on ant colony algorithm is studied. Compared with Stochastic Forest and BP network mining models, the results show that the XGBoost optimization model method has high accuracy and strong explanability.