Productivity prediction in aircraft final assembly lines: Comparisons and insights in different productivity ranges

This paper investigates the productivity prediction for aircraft final assembly lines. Various approaches are proposed and the performance of each approach is compared and discussed. The predictive models established in this paper can be divided into three groups. The first group is a simulation model, built by practitioners with extensive insights into the plant layout, material handling, manufacturing process and resource allocation. The second group consists of three representative regression models, including linear, polynomial and exponential regression. The last one is based on machine learning, including multilayer perception, gradient boost regression tree and random forest.

Compared with typical light industrial products, aircraft product has a large number of modified types in order to meet the customers’ personalized requirements. In addition, the management of aircraft final assembly lines depends on different productivity ranges and managers often need to verify within a given tolerance the performance of the predictive models, which is not covered by previous studies. In light of the above, this paper provides a comprehensive comparison among different modelling approaches. A real aircraft final assembly line including three aircraft types is adopted to illustrate the feasibility of each approach.

The advantages and limitations of different approaches in terms of their efficiency, precision and generalization ability are presented. The aim of this paper is to provide a practical guidance to the choice of suitable approaches and datasets, as well as data processing techniques. The conclusion drawn in this paper can be applicable to other customer order dependent and multi-product assembly problems.