The future of machine vision: AI software designed with users in mind

Machine learning by means of neural networks has developed an indispensable method to solve intricate challenges in optical quality control for manufacturing. When the technology became usable for inline inspection tasks first, neural network architectures themselves were at focus. However, it has become increasingly obvious that the degree of success in implementing vision AI systems is highly dependent on a well-structured and reliable infrastructure. These aspects are commonly summarised under the terms of machine learning operations (MLOps) and human centered design (HCD). Our experiments are conducted using the industrial AI software Neuralyze^®, which has served as a basis for several research projects starting in 2019 to test new approaches to machine learning in manufacturing. In our research, we introduce approaches on how to ideally integrate those methods into AI software concepts to derive an optimum benefit. It is a key goal to retain standardized handling semantics despite the variety of model architectures and use cases.