Responding to material properties with data-driven models

In steel processing lines, the challenges associated with the material properties depend heavily on the customer’s requirements for the final product. Some steel producers specializing in high-quality products, such as high-strength steel, want to reduce the downgrade rates for these products based on the material properties. Other companies focus on increasing their line throughput to better exploit their plant’s potential, which simultaneously allows them to achieve a higher order volume. In the past, one significant aspect that had to be considered was how to save energy costs, which is becoming more and more important today. A key concept that SMS has developed is its digital package, the Material Properties Suite, which offers a set of applications to enable manufacturers to respond more effectively to issues relating to material properties.

Consisting of various components, the Material Properties Suite is designed to support metallurgists and process engineers in optimizing the manufacturing process for galvanized or annealed steel. A core element of this product is the data-driven model, which can predict material properties so that producers can respond to deviations in the upstream process by influencing processes further downstream. The preferred connection between the different applications and the process is the SMS DataFactory.

Within the Material Properties Suite, not only the data from the annealing or galvanizing line is used, but also from the upstream production lines. Results from the mechanical testing laboratory are also included. The SMS DataFactory collects data and transforms it into the required format for the digital applications.

Identifying potentials with Material Properties Reporter

In today’s evolving manufacturing world, having information readily available is critical for driving quick decision-making processes, which in turn ensures lasting business success. The Material Properties Reporter offers KPIs and metrics for the mechanical testing laboratory results for the produced coils.

The created reports are presented as a one-page snapshot of the steelmaking process and illustrate how it performs against key metrics. The reports represent a key source of identifying potential for improvement and help the in-charge personnel to make the relevant decisions.

Feeding the prediction model with metallurgical knowledge

The central component of the Material Properties Suite is the prediction model. Primarily, this is a data-driven model based on large quantities of process data that are recorded during several continuous production processes, which are run in series. In the results presented here, more than 20,000 coils produced in two different production facilities were used to train two model instances. Furthermore, the incorporation of metallurgical knowledge into machine learning technologies ensures improved model quality and explainability.

For a wide range of materials with a yield strength in the 200 MPa to 700 MPa range, the mean absolute error of between 10 MPa and 15 MPa was achieved for each material.

The validation of the model results was performed using partial dependence plots and accumulated local effects plots. A partial dependence plot indicates the relationship between an input variable and the prediction. Accumulated local effects describe how features influence the prediction of a machine-learning model locally. The outcome was vetted with metallurgical know-how.

Analyzing process impacts with Material Properties Simulator

The standard way to determine the mechanical properties of a coil is to carry out a destructive test. Samples are analyzed in the laboratory. The results are then typically saved in the laboratory management system and can be shared with other management systems pertaining to the manufacturing process. Combined with the process data from different production units in a factory or metallurgical route, this is the basis for the Material Properties Simulator, where the data-driven model performs simulations using the input data. As a result, the model is able to determine mechanical properties such as tensile strength or yield strength.

The Material Properties Simulator enables user-friendly interaction with the model. The web-based front end offers different filtering options and allows the relevant process parameters to be manipulated. It gives the user the opportunity to simulate the effect of one or more changes to the process parameters on the mechanical properties in an environment that has no impact on ongoing production.

Achieving the best possible performance with Material Properties Optimizer

Once it is clear how certain process parameters influence the material properties, processes are ready to be optimized. The Material Properties Optimizer helps to find appropriate values for key process parameters. By using the data-driven model, one or more sets of parameters that will achieve the best possible performance are suggested. Before making the calculation, the user can define which process parameter(s) should be improved. Furthermore, there is an option that allows constraints to be applied to the model. After setting the boundary conditions, the model is triggered, and as soon as the results are ready, they are displayed to the user. At this point, it is important to mention that the model not only provides one result but a set of results. Based on the final production strategy, the model can recommend a single optimal solution.

The Material Properties Optimizer can make these decisions easier and faster, as it features the data-driven model that delivers predictions with high accuracy and a fast response time.

One strategy for increasing process line throughput while maintaining product quality within specifications is to increase the process speed. Consequently, the dwell time of the strip in the furnace is shorter than if the process were run at a lower process speed. As a result, the soaking temperature of the strip stays below the original value when the furnace zone temperatures are the same.

Example of a mass flow increase strategy

Another consideration could be to reduce the heating section temperature and keep the production speed at the original value. Similar to the previous example, the soaking temperature is reduced. One target here is to reduce energy costs thanks to the lower heat demand for operating the furnace.

Example of a pure energy-saving strategy

Thanks to the model implemented in the Material Properties Suite, the process engineers are able to determine the right balance between process speed/annealing temperature, taking into account the required mechanical properties of the products.

How will quality management evolve in the future?

The idea of predicting the quality of a product by using the process data for achieving that product is a concept that is employed in many industrial sectors. For steel producers, it is crucial to predict steel quality parameters, as the focus of modern manufacturing quality departments has shifted from reactive to proactive methods over the last decade. A key product developed by SMS is the Material Properties Suite, starting with the mechanical properties (yield strength, tensile strength). Looking ahead, expanding this concept to include other quality-relevant parameters is an aspect worth considering. The components of the Material Properties Suite presented here are part of the Intelligent Furnace product family and can be extended with hardware components, thereby making the solution even more valuable to steel makers, e.g., by providing online measurements along the coil length.

The Material Properties Suite aims to reduce the number of product downgrades and increase productivity, lower energy costs, and improve material homogeneity. SMS brings together metallurgical expertise and data science to help steel manufacturers achieve their individual goals.

Contact: Andrea Asaro