Multiple clearance robustness optimization of a chain ramming machine based on a data-driven model

This paper provides a feasible scheme for robust optimization of the ramming process of a chain ramming machine with multiple clearances using a data-driven modeling framework based on deep neural networks. The forced ramming process of the chain ramming machine is studied, and a specially tailored experimental platform is demonstrated to validate the multiple clearances combined dynamic model of the ramming machine. In the model, the spatial clearances of the rollers in the track and sprocket teeth grooves and of the projectile in the conveying channel are described in detail for the first time. The displacements, velocities and lateral and vertical swing angles of the projectile when the projectile is forced in place with different combinations of clearances serve as the training dataset for the data-driven-forced ramming model. On this basis, the architecture of the deep neural network for the forced ramming process is designed by an integer optimization method to establish the corresponding data-driven model. Finally, a multiobjective robust optimization study is carried out under the data-driven model, and the optimization results with considering controllable and uncontrollable variance provide a reference for the project to improve the accuracy of ramming projectiles in place.