Prototyping Machine-Learning-Supported Lead Time Prediction Using AutoML
Janek Bender, Jivka Ovtcharova
Abstract
Many Small and Medium Enterprises in the domain of Make-To-Order- and Small-Series-Production struggle with accurately predicting lead times of highly customisable orders. This paper investigates an approach using AutoML integrated into existing enterprise systems in order to enable Lead Time Prediction based on Machine Learning models. This prediction is based on both order data from an ERP system as well as real-time factory state informed by an IIoT platform. We used simulation data to feed the AutoML model generation and developed a lightweight web-based microservice around it to infer lead times of incoming orders during live production. Using industry standards, this microservice can be seamlessly integrated into existing system landscapes. The simplicity of AutoML systems allows for swift (re)training and benchmarking of models but potentially comes at the cost of overall lower model quality.