Process Monitoring Using Machine Learning for Semi-Automatic Drilling of Rivet Holes in the Aerospace Industry

Abstract

The majority of aircraft rivet holes are drilled with semi-automatic and manually controlled, pneumatically driven machines as full automation is often unsuitable due to workspace restrictions. Lightweight materials of difficult machinability complicate drilling. This is particularly relevant when drilling stack materials, where the machining parameters are determined by the most difficult to machine material layer. To provide reliable rivet connections, drilling in multiple steps, use of minimum quantity lubrication as well as subsequent manual deburring and cleaning are indispensable. Newly developed electrically driven semi-automatic advanced drilling units (ADUs) enable intelligent process layouts and online condition monitoring by evaluating integrated sensor data. Additionally, process parameters can be adapted to suit each material in the stack. In this paper, machine learning is applied to ADU sensor data to predict cutting forces and process conditions based on the internally measured currents of the ADU’s electric motors. The application of machine learning to ADU data is beneficial as drilling in the aerospace industry shows high repeatability and many produced holes, providing a large dataset. The machine learning methods linear regression, artificial neural network and decision tree are applied to force prediction. Furthermore, the k-nearest neighbour method is used to predict material, feed rate and lubrication state. Process monitoring based on the presented results minimizes manual control and rework by the identification of process deviations, resulting in a comprehensive quality assurance as well as optimal tool life exploitation. This leads to a step change in semi-automatic drilling of aircraft structures by overcoming a major productivity limitation.