Efficient Deployment of Machine Learning Models in Manufacturing and Industrial Environments using ROS

This paper presents a deployment concept that aims to overcome the challenges in the implementation of Machine Learning (ML) models in manufacturing and industrial environments. In these contexts, robots are not typically viewed as production machines. However, the potential for applying advanced techniques such as condition monitoring extends beyond production lines to encompass robotic systems. As a result, there arises a need for a modular solution that integrates into the existing ecosystem while accommodating the requirements of robotic environments. By embracing modularity and interoperability, our proposed deployment concept not only addresses the challenges specific to industrial robotics but also fosters a holistic approach to enhancing operational efficiency and performance in diverse manufacturing settings.

For this, an easily customizable and adjustable system that handles both data acquisition and data transfer is needed. By using the Robot Operating System (ROS) for all necessary data handling, we achieve a highly modular, efficient, and easy-to-use low-code deployment pipeline. Our approach splits the different processing steps into separate nodes and automatically sets up all necessary communication channels, achieving high interchangeability and a quick time-to-deploy. The approach is explained in detail and demonstrated for the real use case of deploying models to monitor handling robots.