Problem: 

Decision-making processes are the key for efficient production and sustainability in almost every manufacturing process. Many decisions are made until a final product gets delivered to a customer. This especially holds for complex manufacturing processes with many production steps such as steel production. Maybe the most crucial decision in this context is the assessment of product quality.

Growing customer expectations together with increasing availability of relevant information and high flexibility of final product features are taking established Decision Support Systems (DSS) performing automatic release decisions continuously to their limits.

Solution:

Emerging machine learning technologies could solve this problem, but concepts for their robust industrial application performing high-stakes decisions are missing.

DeepQuality aims to improve the automatic quality assessment of steel products by means of a holistic approach combining deep learning technology with sophisticated management of underlying training data to enable the optimal use of all available data sources and simultaneously simplify the configurability and maintainability of previous DSS.

The project consists of the following concepts realizing a human-centered lifecycle for the robust industrial application of deep learning quality models.

Use cases:

In the first use case, a commercially viable implementation is delivered into an existing automatic coil release software tool provided by QuinLogic GmbH, who is part of the SMS group. The implementation is carried out at the wire rod manufacturer Arlenico S.p.A.

In the second use case, the most extensive use of the developed DeepQuality methods, an implementation based on available open source tools is envisaged. The applicability of this Open DeepQuality framework is demonstrated by means of a prototypical installation in flat steel production at ArcelorMittal Bremen GmbH.