Abstract: Due to the complexity of assembly and testing processes in aero-engine manufacturing, the collected assembly data encompass a large number of assembly features, which seriously interferes with the accurate prediction of assembly quality. Selecting the key quality features of aero-engine assembly to achieve quality prediction becomes a highly challenging task. Therefore, to address this issue, a two-stage feature selection method for aero-engine assembly data based on multi-objective optimization is proposed. Firstly, the optimization objectives of feature selection are defined. In the first stage, the relevant features are selected based on the max relevance and min redundancy (MRMR) algorithm to calculate the mutual information of assembly features and testing indicators. This process filters out the most relevant features related to testing indicators while removing redundant features with interference effects. In the second stage, by introducing a population initialization strategy and adaptive genetic operators, a key quality feature selection process based on the improved non-dominated sorting genetic algorithm-Ⅱ (NSGA-Ⅱ) is proposed to obtain the Pareto front of key quality feature subsets for aero-engine assembly. Finally, experimental results demonstrate that the proposed two-stage feature selection method has better applicability and effectiveness than traditional methods, which enhances the feature selection performance and improves the accuracy of quality prediction for aero-engine assembly.

Key words: aero-engine assembly, multi-objective optimization, high-dimensional data, key quality features