Scrap is the most important secondary raw material in the transformation to low carbon dioxide (CO<inf>2</inf>) steel. However, the suitable use of different scrap types for producing high quality steels with the right chemical composition is non-trivial. It requires process control and detailed knowledge of all input materials used. SHapley Additive exPlanations (SHAP), a game-theoretic approach, is often used to interpret machine learning models through visualizations and feature attributions. In this paper, we present a novel application of SHAP values. This enables more precise control of material composition in steel production without the need for additional sensors. This makes it extremely practical for real steel production environments and enables better control of the materials used in the steel production process. As a basis for this approach, various machine learning models were trained and the respective SHAP values computed. To validate the approach, the results were compared with the values from the steel plant. Comparing the calculated values with the historical estimates, the results agree for most input materials and target elements. The key innovation lies in using SHAP values not only for model interpretability, but also as a quantitative tool to estimate the chemical content of input materials (e.g., steel scrap) based on process data. The framework enables chemical composition estimation, relying solely on routinely collected process data. This is a novel application of SHAP and allows the back-calculation of predicted values and can be used in a wide range of applications in industry and academia.

The steel industry is currently in a transformation process in order to be able to produce in a more environment-friendly way in the future. The secondary raw material steel scrap plays a crucial role in this transformation, as recycling scrap in the manufacturing process is environmentally friendly and sustainable. However, the increased use of steel scrap in the steel industry involves new challenges. Processes must be changed, product quality must be maintained and the increased throughput and demand for scrap must be managed. Digitalization and the use of AI technologies can help to optimize and automate the new processes.

When using AI in an industrial environment, there is often the challenge that not enough data of sufficient quality is available. In order to close this gap, a new freely available dataset of European scrap classes, was created and used in this work by applying a novel tilling technique. The creation and even more the annotation of such domain-specific datasets requires a lot of time and expert knowledge. For this reason, a self-supervised approach was implemented using different types of augmentations to extract the fine-grained structures typical for intrinsic disordered objects such as steel scrap. These results were used to control the scrap input as well as the scrap usage and thus automate the process.

The scrap used in the steel production process usually varies in origin and composition, which makes the compilation more difficult. When compiling the scrap mix, steel producers often rely on experience or have to carry out complex trials. A machine learning approach was implemented that can be used to simulate and optimize different scrap compositions. Based on these models, a new approach was developed to estimate the chemical content of the input materials used from standard process parameters without the use of additional sensors.

The integration of AI-models in a heterogeneous industrial environment is a major challenge. Ambient infrastructure needs to be adapted or created as required.To enable the various solutions to be embedded, the different machine learning technologies were combined, required infrastructure was set up as required and online models and interfaces were implemented for productive use. 

In summary, this thesis presents an AI-powered holistic system that combines various technologies, optimizes steel scrap processes, and automates the scrap workflow from scrap entry to the end of the basic oxygen furnace process.

Stålindustrin befinner sig just nu i en omvandlingsprocess för att i framtiden kunna producera på ett mer miljövänligt sätt. Den sekundära råvaran stålskrot spelar en avgörande roll i denna omvandling, eftersom återvinning av skrot i tillverkningsprocessen är både miljövänligt och hållbart. Den ökade användningen av stålskrot i stålindustrin innebär dock nya utmaningar. Processer måste förändras, produktkvalitet måste upprätthållas och det ökade flödet och efterfrågan på skrot måste hanteras. Digitalisering och användning av AI-teknik kan hjälpa till att optimera och automatisera de nya processerna.

När AI används i en industriell miljö finns ofta utmaningen att det inte finns tillräckligt med data av tillräcklig kvalitet. För att täppa till detta gap, presenteras en fritt tillgänglig datauppsättning av europeiska skrotklasser, som skapades med en ny AI-teknik (”tiling”), i denna avhandling. Skapandet och annoteringen av sådana domänspecifika datamängder kräver mycket tid och expertkunskap. Av denna anledning implementerades ett självövervakat tillvägagångssätt med hjälp av olika typer av förstärkningar för att extrahera de finkorniga strukturerna som är typiska för inre oordnade föremål som stålskrot. Dessa resultat användes för att kontrollera skrotinmatningen såväl som skrotanvändningen och på så sätt automatisera processen.

Skrotet som används i stålproduktionsprocessen varierar vanligtvis i ursprung och sammansättning, vilket försvårar sammansättningen. Vid sammanställningen av skrotblandningen förlitar sig stålproducenterna ofta på erfarenhet eller komplexa försök. En maskininlärningsmetod implementerades som kan användas för att simulera och optimera olika skrotsammansättningar. Dessutom utvecklades ett nytt tillvägagångssätt baserat på dessa modeller för att uppskatta det kemiska innehållet i de ingående materialen som används från standardprocessparametrar utan användning av ytterligare sensorer.

Integreringen av AI-modeller i en heterogen industriell miljö är en stor utmaning. Omgivande infrastruktur måste anpassas eller skapas efter behov. För att de olika lösningarna skulle kunna bäddas in, kombinerades de olika maskininlärningsteknologierna, infrastrukturen sattes upp efter behov och online modeller och gränssnitt implementerades för produktivt bruk.Sammanfattningsvis presenterar denna avhandling ett AI-drivet holistiskt system som kombinerar olika teknologier och optimerar och automatiserar skrotprocessen från skrotinträde till slutet av konverterprocessen.

In the blast furnace and basic oxygen furnace route, pig iron and steel scrap are used as resources for steel production. The scrap content can consist of many different types of scrap varying in origin and composition. This makes it difficult to compile the scrap mix and predict the future chemical analysis in the converter. When compiling the scrap mix, steel manufacturers often rely on experience and trials. In this paper, we present a machine learning approach based on XGBoost to predict the chemical element content in the converter. Data from around 115000 heats were analyzed and a model was developed to better predict the content of the tramp elements copper, chromium, molybdenum, phosphorus, nickel, tin and sulphur at the end of the basic oxygen furnace process. The study shows that it is possible to predict the chemical element content for tramp elements in the converter based solely on data available in advance and routinely collected without the necessity of additional sensors or analysis of input material. Given the nature of scrap classifications for (external) scrap types, this is non-trivial. Furthermore, an online model was implemented, accessible via a defined synchronous interface, which allows to optimize the use of different scrap types by predicting the chemical content at the end of the basic oxygen furnace process and simulating with new combinations of input material. Not all types of steel scrap are always available. With the model developed, new scrap input constellations can now be created to ensure that the quality of the melt is maintained. However, for very accurate predictions, the data from the upstream processes must be of high quality and quantity. Efficient scrap management, monitoring of the scrap input and confusion checks.

There has been remarkable progress in the field of Deep Learning and Computer Vision, but there is a lack of freely available labeled data, especially when it comes to data for specific industrial applications. However, large volumes of structured, semi-structured and unstructured data are generated in industrial environments, from which meaningful representations can be learned. The effort required for manual labeling is extremely high and can often only be carried out by domain experts. Self-supervised methods have proven their effectiveness in recent years in a wide variety of areas such as natural language processing or computer vision. In contrast to supervised methods, self-supervised techniques are rarely used in real industrial applications. In this paper, we present a self-supervised contrastive learning approach that outperforms existing supervised approaches on the used scrap dataset. We use different types of augmentations to extract the fine-grained structures that are typical for this type of images of intrinsically unordered items. This extracts a wider range of features and encodes more aspects of the input image. This approach makes it possible to learn characteristics from images that are common for applications in the industry, such as quality control. In addition, we show that this self-supervised learning approach can be successfully applied to scene-like images for classification.

DOES - Dataset of European scrap classes. Today, scrap is already an important raw material for industry. Due to the transformation to green steel, the secondary raw material scrap will become increasingly important in the coming years. With DOES a free dataset is presented, which represents common non-alloyed European scrap classes. Two important points were considered in this dataset. First, scrap oxidizes under normal external conditions and the visual appearance changes, which plays an important role in visual inspections. Therefore, DOES includes scrap images of different degrees of corrosion attack. Second, images of scrap metal (mostly scrap piles) usually have no intrinsic order. For this reason, a technique to extract many overlapping rectangles from raw images was used, which can be used to train deep learning algorithms without any disadvantage. This dataset is very suitable to develop industrial applications or to research classification algorithms. The dataset was validated by experts and through machine learning models.