The paper presents a machine learning-based system aimed at improving the homogeneity of tensile properties of steel strips for automotive applications over their strip length in the annealing and hot dip galvanizing lines. A novel modular approach is proposed exploiting process and product data and combining smart data pre-processing and cleansing algorithms, an ensemble of neural networks targeted to specific product classes and an ad-hoc developed iterative procedure for identifying the variability ranges of the most relevant process variables. A decision support concept is implemented through a software tool, which facilitates exploitation by plant managers and operators. The system has been tested on site. The results show its effectiveness in improving the control of the thermal evolution of the strip with respect to the standard operating practice.

Within the design of a machine learning-based solution for classification or regression problems, variable selection techniques are often applied to identify the input variables, which mainly affect the considered target. The selection of such variables provides very interesting advantages, such as lower complexity of the model and of the learning algorithm, reduction of computational time and improvement of performances. Moreover, variable selection is useful to gain a profound knowledge of the considered problem. High correlation in variables often produces multiple subsets of equally optimal variables, which makes the traditional method of variable selection unstable, leading to instability and reducing the confidence of selected variables. Stability identifies the reproducibility power of the variable selection method. Therefore, having a high stability is as important as the high precision of the developed model. The paper presents an automatic procedure for variable selection in classification (binary and multi-class) and regression tasks, which provides an optimal stability index without requiring any a priori information on data. The proposed approach has been tested on different small datasets, which are unstable by nature, and has achieved satisfactory results.

Water is a fundamental steelworks additional resource; its efficient management is crucial for process reliability, product quality and environmental sustainability. Within steelworks, water is exploited mainly for direct or indirect cooling and is usually reused and recycled after cooling and treatments to eliminate contaminants. However, bottlenecks often exist, limiting water management efficiency and increasing water consumption. These issues are mainly related to water treatments efficiency, lack of water parameters monitoring and the manual/semi-manual management of water networks. Furthermore, these aspects are generally associated with the plant’s service life; brownfield sites are mostly affected. In these cases, improving sensor circuits coupled with decision support tools can support human decisions and lead to significant advantages. The paper discusses a potential application of such tools after new sensors installation in a use case concerning the minimization of the use of high-quality make-up-water for the indirect cooling system of a wire-rod mill in electric steelworks. The effectiveness of the described tool is shown, and the advantages are highlighted in terms of potential savings that can reach 95% and 4% of the current consumption of well and osmotic water in the considered circuit, respectively, corresponding to a saving of about 9400 m³/year of high-quality water.

Authors: Silvia Cateni and Valentina Colla

In this chapter a survey on the problem of classification tasks in unbalanced datasets is presented. The effect of the imbalance of the distribution of target classes in databases is analyzed with respect to the performance of standard classifiers such as decision trees and support vector machines, and the main approaches to improve the generally not satisfactory results obtained by such methods are described. Finally, two typical applications coming from real world frameworks are introduced, and the uses of the techniques employed for the related classification tasks are shown in practice.