BFI VDEh-Betriebsforschungsinstitut GmbH

Initial situation:

Flatness plays an overwhelming role for both, the process quality and for the quality of strip or plate steel.  Special sensorial equipment was developed to measure flatness and flatness deviations, and, in the past, whole processes have been tailored to deliver high quality flatness characteristics. Yet, for several steel grades of growing importance, good flatness is not reproducibly achieved.

The situation has become worse, as the global market is urging Europe´s steel producers out of their comfort zone. Other countries have caught up with respect to production quality and technology. The number of global competitors with good products is rising, constantly cutting Europe’s market share.

A way out of this predicament is clear: increase product quality even further, introduce new steel grades and customize high-end products to cover niche markets. Driven especially by the demand of the automotive industry, thinner and higher-strength steels can make all the difference here, as automobile manufacturers are forced to use extremely light steel materials for meeting their CO₂ constraints, featuring designed material characteristics like yield strength, ultimate strain and ductility.

Project targets:

HatFlat will apply a unique way of combining existing models based on physical laws – so called first-principal models – and state-of-the-art machine learning (ML) approaches. This so-called physics-informed ML approach allows to
a) improve the prediction of flatness and flatness deficiencies
b) identify the influential factors of greatest impact on flatness and
c) to make proposals for changing the processes (if the product quality specifications are not affected) for achieving a better flatness characteristic.

The project will also utilize proven Industry 4.0 tools, such as digital product twins for hosting the prediction models.

The models developed in HatFlat will be assembled into a holistic assistance tool for flatness prediction. It reflects the overarching red line of our project. Featuring a strong industrial participation of some of Europe´s biggest steel producing companies, plants allocated along the process chain of steel processing with strip and plate products, two highly reputed steel research institutes and a university, all working together to create one new tool for predicting, understanding and improving flatness.

Innovative approaches:

• Application of latest technologies of machine learning to tackle flatness problems and to provide prediction models for the production line.
• Combination of existing first principal models with machine learning for a physics-informed approach.
• Introduction of digital twins as a proven tool to ease management of both data and models.
• Holistic approach, considering process variables over multiple processing steps with highly resolved raw data and not only characteristic values.
• Optimization of the process chain by applying using the prediction models for generating a set of future possibilities, selecting the best one for the given product.
• Joint software tool developed among the partners, not only to simply improve flatness prediction but also to generate knowledge and teach within the plants and complete industry.

Benefits for the industry:

The benefits for the steel industry originated by HatFlat are significant. Since flatness is one of the most important quality features for strip and plates, overall product quality will increase for producers. This will increase customer loyalty and be a competitive advantage over manufacturers in other countries. Additionally, the throughput of plants can be increased because better flatness of products reduces the risk of strip rapture and results in an overall more stable process.

Flatness defects are reduced which aims in directly reduced costs per plant and year. This includes costs for rework, scrapping and complaints of customers. Additionally, first experiences with grades of higher yield strength clearly indicate that with increasing yield strength a higher loss due to flatness must be expected. This shows that flatness defects prevent the plants from producing new steel grades due to persistent flatness problems.