Machine Learning-Driven Maintenance Order Generation in Assembly Lines

Abstract

The automatic generation of maintenance orders facilitates the prompt detection and root cause analysis of deviations or failures in the assembly process. The aim of this project is to use supervised learning models to recognise deviations in the throughput times of a fully automated assembly process carried out by robots. The model identifies errors, categorises their causes and transmits the information back to the Enterprise Resource Planning (ERP) system. The data collected comes from a development and test assembly station with an industrial robot. The data set from the assembly station was expanded using agent-based simulation in order to train the four models Support Vector Machine, K-Neares Neighbour, Naive Bayes and Decision Tree. The SVM model proved to be the most suitable model for automatic fault detection with an accuracy of 99.51 %. The model was integrated into the assembly station and an algorithm was developed to automatically generate maintenance messages to transmit the failure code to the ERP system.