Prodrive Technologies expertise ranges from embedded computing, motion & mechatronics, power conversion, automation control systems, vision & image processing systems and internet of things products, configurable to the customers needs.

In industrial production environments only 53% of all AI/ML applications make it in the long run. Companies often struggle with setting up, integrating, and maintaining stable production environments for machine learning. Defining and implementing a robust MLOps process is key and should address the following elements:

The standardization of data collection, including database structures, communication interfaces and GUIs, automating distributed model training including version control of data and models (Tensorflow/PyTorch models on Kubernetes clusters);
deploying models-as-a-service (for instance with gRPC interfaces);
constant monitoring of model inference performance with dashboards including alerts;
adding model inference feedback with the human-in-the-loop;
and supporting state-of-the-art model explainers to also give a reason why predictions are made.

In this lunch lecture we show this process for the application of automatic visual inspection of electronics using deep learning models as used within Prodrive Technologies.