Data-driven quality prognostics for automated riveting processes

Abstract

Technologies based in robotics and automatics are reshaping the aerospace industry. Aircraft manufacturers and top-tier suppliers now rely on robotics to perform most of its operational tasks. Over the years, a succession of implemented mobile robots has been developed with the mission of automating important industrial processes such as welding, material handling or assembly procedures. However, despite the progress achieved, a major limitation is that the process still requires human supervision and an extensive quality control process. An approach to address this limitation is to integrate machine learning methods within the quality control process. The idea is to develop algorithms that can direct manufacturing experts towards critical areas requiring human supervision and quality control. In this paper we present an application of machine learning to a concrete industrial problem involving the quality control of a riveting machine. The proposal consists of an intelligent predictive model that can be integrated within the existing real time sensing and pre-processing sub-systems at the equipment level. The framework makes use of several data-driven techniques for pre-processing and feature engineering, combined with the most accurate algorithms, validated through k-folds cross validation technique which also estimates prediction errors. The model is able to classify the manufacturing process of the machine as nominal or anomalous according to a real-world data set of design requirements and operational data. Several machine learning algorithms are compared such as linear regression, nearest neighbor, support vector machines, decision trees, random forests and extreme gradient boost. Results obtained from the case study suggest that the proposed model produces accurate predictions which meet industrial standards.