Digitalization throughout the process chain
The example of the cold rolling plant

In recent years, digitalization has become increasingly relevant in the industrial sector, including at AMAG. This is driven by growing demand for higher-quality products, greater sustainability, optimized processes and reduced energy consumption. Companies that seize on these opportunities and systematically implement corresponding measures at an early stage are in a position to make their production processes more efficient, more sustainable and, therefore, more competitive.


A metal strip wound onto a mandrill is called a coil. After rolling, metal strips are rolled up into coils to make them easier to transport, store and handle in downstream processing. In metal processing, flat, elongated strips of metal are known as sheets. A sheet is a piece of a coil that has been unwound and cut or already processed.
“Digitalization throughout the process chain” is a multifaceted concept pursued at AMAG. It includes many different aspects, such as digital production chain control, in which sensors monitor machinery and production processes, artificial intelligence (AI) systems to conduct automatic quality checks, or predictive maintenance processes to reduce machine downtime. Another aspect that is particularly important in this context is digital product tracking. This means ensuring that a product remains traceable throughout the entire production chain. It is also linked to the recording of all process data generated during product passes through different machines in the production chain. Further key aspects of digitalization include data-driven decision-making, for example in the form of real-time process control dashboards and the optimization of production processes using statistical hypothesis tests, machine learning and AI. In this context, a key aspect is the digital recording of all process data throughout the production chain, from scrap processing to smelting and casting ingots, and finally producing the end products, such as rolled plates and finished coils. This article concentrates in particular on the cold rolling processes. Examples from practice demonstrate how digital data acquisition can create added value and support the optimization of production processes. At the same time, they highlight the challenges to be overcome and underline the benefits to AMAG and its customers.
Digital recording of all process data throughout the production chain at AMAG
provides a schematic representation of the production process, looking at the example of manufacturing automotive sheets. The production process starts with the input material and scrap sampling to determine the material composition. This means identifying which chemical elements are contained in the scrap. In addition, impurities are identified by detecting foreign substances such as plastics, lacquers and other metals.
The next step is melting the input material and casting it into slabs. A chemical analysis is conducted in parallel to determine the alloy’s composition. This is followed by the rolling process, which involves various steps including homogenization, hot rolling, cold rolling, intermediate annealing, solution heat treatment, passivation, slitting and dry lubrication, and can last several weeks. [1] Quality checks are then conducted to ensure that the material meets the customers’ specifications and requirements, which vary depending on the product’s intended purpose. [2] In the course of its processing with different equipment and machinery, a vast amount of data is recorded by hundreds of sensors and stored as time series. The sensors measure parameters such as the casting temperature, the rolling pressure and the rolling speed in real time. Recording this data creates a comprehensive image of the production process, which forms the basis of a data-driven understanding of the process.
A stable data link with the cloud has been established to counteract challenges. This ensures that the vast quantities of data are automatically processed on an ongoing basis, guaranteeing that up-to-date data is available every day. Data is analyzed on a daily basis so that metadata can be allocated to casting batches, slabs, strip batches, plates and coils, etc. This includes a time-length transformation, in which time-stamped sensor data recorded during casting or rolling of a given slab is allocated to a specific sheet position. Cold rolling represents a particular challenge.It requires structured and efficient infrastructure to store the high-frequency process data and make it available for analyses. This data recording relies on special sensor technology, while an IoT platform and data pipelines are needed to record and transfer sensor data from individual machines into a cloud environment. When it comes to data storage, relational databases can be used for structured data (e.g. production parameters and product quality data), while NoSQL databases are suitable for unstructured data (e.g. images and text files). In addition, cloud-based storage systems (e.g. AWS, Azure) serve as scalable storage options along with extract-transform-load (ETL) processes, which extract data from different sources, transform (i.e. process) them and save them in a standardized format. [3, 4, 5]The necessary AI infrastructure includes high-performance computers, edge computing devices and cloud platforms (e.g. AWS, Azure and Google Cloud) for flexible computing power and storage. Ultimately, it also requires AI tools and analysis platforms; including business intelligence (BI) tools (e.g. Power BI) for visualization, machine learning (ML) platforms and AI services (e.g. Azure Machine Learning). [6, 7]In this context, security and governance structures are also important elements for any company. [8] It is also important to consider the fact that tools and software develop and change on almost monthly basis.As noted previously, process data undergoes a time-length transformation at all machines. This is a complex data transformation step in which time-stamped process data is allocated to a specific position on a slab, plate or sheet. This article uses the example of cold rolling to illustrate the complexity of this step and the associated challenges.


Cold rolling mill: Time-length transformation of process data, strip separation and multiple passes
Cold rolling plays a decisive role in the aluminium industry because of its ability to modify the product characteristics of aluminium sheets and exert a significant influence on product quality. This process reduces the thickness of a material, which in turn has implications for its length. [9] Thousands of sensors continuously monitor this process, recording data at high frequencies and generating a wealth of data in the millisecond range. Highly sophisticated extract-transform-load (ETL) processes and corresponding infrastructure, as described above, are needed to ensure that this data is processed efficiently.This data processing involves a variety of challenges. First, however, let us examine the term “time-length transformation” in more detail.

This transformation process examines two components that must be linked together: the time component and the length component.High-frequency continuous process data acquisition takes place during the rolling process. For example, data on temperature, speed and force are recorded. These measurements occur at a certain time, which provides the time component.The length component can be described as follows: product quality data - such as defect positions in the finished product (i.e. a plate or sheet) - relates to a specific location along the length of the finished product. In order to link a specific position with the corresponding process data, the time-referenced process data must be allocated to the relevant position along the length of the entire product. This process is known as time-length transformation and must be implemented individually for at each machine, for each product and for every format, for example. Cold rolling is one of the most complex systems when it comes to the time-length transformation. Transforming the time component into a length component involves several steps. The first is taking account of the rolling speed, which may be lower at the beginning than in the middle of the rolling process. Given that the material passes through the rolling mill at different speeds, it is vital to determine the speed of the material at all times along the rolling process. This makes it possible to draw comparisons between data in different process sections.Another challenge is accounting for multiple passes. This is illustrated in Figure 2. [10]Material can pass through a rolling mill several times, with different reductions in material thickness and different throughput speeds on each pass. And, after each pass, the material may also be cut and separated. Depending on the specific product and the customer’s requirements, material may be subject to varying process conditions (e.g. different thickness reduction specifications), which makes the transformation more complex.

Figure 2 provides a schematic outline of a simple example: the material is processed in three rolling steps, with a thickness reduction in each step. After the second rolling pass, the material is cut into two coils. In a third step, the second coil - resulting from the strip separation after the second rolling pass - passes through the rolling mill again and is itself cut into two coils. At the end of the process, a defect that must have occurred during the rolling process was identified at a specific location on the second coil. In order to analyze the causes, the defect must be traced through the individual rolling steps, as shown by the orange dotted line running back through each work step in Figure 2. Hundreds of high-resolution sensor readings are taken in each rolling pass. Figure 2 also shows a sensor signal, with the sections of the strip length that correspond to the defect position after each pass marked in green. [10]Figure 3 illustrates the various processes in the cold rolling mill that are relevant to the time-length transformation, together with the corresponding signal sources. [10] The rolling process spans from the decoiler on the right-hand side to the coiler on the left-hand side. The figure divides the cold rolling mill into the infeed section on the right-hand side of the roll gap and the outfeed section on the left-hand side.Sensors are installed at various measurement points throughout the rolling mill to record production data. A distinction is made between measurement points in the infeed section (before the rolling gap), which record data on the infeed strip, and measurement points in the infeed section, which record data on the outfeed strip. In addition, the cold rolling mill under consideration also has an infeed cutter and an outfeed cutter which must also be considered in the time-length transformation. The location of speed sources is also shown in Figure 3. Reliable position and speed signals for both the infeed and outfeed strips are essential and are therefore coordinated with specialist departments in order to calculate a precise time-length transformation. As Figure 3 shows, the rolling mill features an infeed and an outfeed cutting machine, both of which can slit the strip into different sections and trim scrap from the material (see example in Figure 2) [10] - steps which must also be considered in calculations.

To ensure that data can be evaluated correctlyand that the time-length transformation is properly implemented, even when the strip is separated or passes through the rolling mill multiple times, several steps are required and briefly summarized below. The rolling mill is capable of processing the material in several passes with different thickness reduction and different process conditions in each pass. The first step is therefore to identify the individual material segments after each rolling pass. To take several passes into account, each segment must be tracked separately and merged into a complete time- / length-curve. Linking the data from different passes requires a dedicated system.A particular challenge arises when the material is separated into individual segments. The data must therefore also be segmented, with each segment representing a piece of the material subjected for further processing after being cut. The relevant process data, on parameters such as temperature, speed and elongation, is then determined for each segment. The process signals are recorded separately for each material segment that passes through the further rolling process. Seamless tracking of the exact position and the corresponding process data for each piece during cutting is required to ensure consistent allocation of the process data to the material. [10]
Summary
The time-length transformation of process signals in rolling mills requires precise modeling of the rolling process, taking account of dynamic conditions (e.g. rolling speed and scrap trimming). Advanced data transformation methods are needed to take account of multiple passes along the rolling mill and material cutting processes. The deployment of modern data processing tools, a cloud platform and mathematical methods facilitate this transformation and enable precise and efficient data processing. The overarching objective is digital recording of the entire process chain.
This makes it possible to analyze data on the entire production process, from the scrap input material to casting, rolling, heat treatment and other steps through to the final product quality checks. Defect tracking and analysis can be conducted for the entire process chain as shown in Figure 2, which depicts defect tracking and analysis within a process step (in this case, cold rolling). This makes it possible to identify correlations between process parameters and product quality - and, as a result, identify variables that may influence the process and lead to defects.
Customer benefits
More precise process control and optimized planning result in consistently high quality, an improved process understanding and higher reliability, all which mean that customers benefit from more consistent products. More time-efficient processes also translate to shorter delivery times.
Sources:
[1] A. K. Vasudevan, R. D. Doherty. Aluminum Alloys- Contemporary Research and Applications: Contemporary Research and Applications. Elsevier, 2012. [2] B. Prillhofer, H. Antrekowitsch, H. Böttcher, P. Enright. Nonmetallic inclusions in the secondary aluminium industry for the production of aerospace alloys. In Light Metals-Warrendale-Proceedings, TMS, p. 603, 2008.[3] R. Elmasri, S. B. Navathe, S. B. Fundamentals of Database Systems. 7. Aufl., Pearson, 2015[4] J. Mineraud, O. Mazhelis, X. Su, S. Tarkoma. A gap analysis of Internet-of-Things platforms. Computer Communications, Bd. 89-90, S. 5-16, 2016.[5] M. Armbrust, A. Fox, R. Griffith, A. D. Joseph, R. Katz, A. Konwinski, et al. A view of cloud computing. Communications of the ACM, Bd. 53, Nr. 4, S. 50-58, 2010.[6] Microsoft. Azure Machine Learning Documentation. Abgerufen von: https://docs.microsoft.com/en-us/azure/machine-learning/, 2023.[7] Microsoft. Power BI Documentation. Abgerufen von: https://docs.microsoft.com/en-us/power-bi/, 2023.[8] Microsoft. Microsoft Security & Compliance Documentation. Abgerufen von: https://docs.microsoft.com/en-us/microsoft-365/security/, 2023.[9] R. V. Singh. Aluminum Rolling: Process Principles and Their Applications. McGraw-Hill Education LLC, New Delhi, 2011.[10] A. Haidenthaler, M. Schreyer, P. Pfeiffer, W. Fragner. Insights into a cold rolling mill: A survey through length-related process data, strip splitting & multiple passes", Manuskript eingereicht und akzeptiert zur Veröffentlichung. The International Journal of Advanced Manufacturing Technology