Abstract

MLOps, or Machine Learning Operations, play a significant role in streamlining production deployment, monitoring, and management of machine learning models. Integrating MLOps with edge devices poses unique challenges that require customised deployment strategies and efficient model optimisation techniques. This paper introduces LinkEdge, a set of tools that enable the integration of MLOps practices with edge devices. LinkEdge consists of two sets of tools: one for setting up infrastructure within edge devices to be able to receive, monitor, and run inference on ML models and another for MLOps pipelines to package models to be compatible with the inference and monitoring components of the respective edge devices. The LinkEdge platform is evaluated by obtaining a public dataset for predicting the breakdown of Air Pressure Systems in trucks. Additionally, the platform is compared against a commercial tool that serves similar purposes. The overall performance of LinkEdge matches that of already existing tools and services while allowing end users setting up Edge-MLOps infrastructure the complete freedom to set up their system without entirely relying on third-party licensed software.