Machine Learning use case in manufacturing – an evaluation of the model’s reliability from an IT security perspective

摘要

The use of Machine Learning (ML) solutions for decision automation in manufacturing environments is critical if operators trust ML-predictions without critically questioning them. The vulnerability of ML-applications to data manipulation, data-poisoning and adversarial examples raise concerns about its reliability and security. This paper evaluates an on-edge predictive maintenance solution through an IT security perspective, showing how the model’s forecasting can be affected by intentional data manipulation and thus identifying the system’s vulnerabilities for this particular use case. It concludes with suggestions on how to mitigate threats and manage risks.