Businesses in the steel and aluminum industry generate huge amounts of data in their production lines, often only available in a heterogeneous form and scattered across multiple systems and databases. Only by tearing down these data silos and extracting the crucial information for value creation can data be turned into information and information into added value. However, extracting the information needed is a time-consuming task. The SMS DataFactory is the basis of the Learning Steel Plant and makes the data from the plant automation available for planning, maintenance, or quality assurance applications. In this way, fully comprehensive preparation and analysis of all plant data take place.
The learning steel plant has one goal: using data to enable production that is as sustainable and resource-efficient as possible. Software collects data from a plant, transforms it into information, and finally into added value using artificial intelligence and machine learning. This allows essential findings to be gained for later implementation in practice – thus saving costs and resources. Predictive algorithms help to detect a plant's condition, predict the product quality, redirect production routes and minimize downtimes. That is why companies need solutions that process large volumes of data and analyze it in real-time to identify new correlations. In the era of digitalization and Industry 4.0, it is crucial to obtain maximum performance from both plant and processing routes. But how do you first bring the enormous amounts of data into an orderly structure?
Key ingredients in contributing added value to a business are the advanced analytics teams comprising data engineers and data scientists. A large proportion of their work is often spent on data engineering tasks to clean, gather, and combine the data from various data sources into a consumable format. The actual problem to be solved loses prominence to the boilerplate work (work that is repeated in an unchanged form) that needs to be done over and over again. In many cases, only the automation experts can select and understand the available data. This is due to the fact that the operational technology (OT) systems store their data in a format that meets their own requirements but not the purpose of analyzing and merging it with other data sources. A lack of documentation of the data sources creates an additional barrier to utilizing the data. Depending on the problem that needs to be solved, the data must be available in different formats. The data are typically viewed from a condition, quality, and planning perspective. For the maintenance view, the data are often retrieved using certain events in time. The data must therefore be addressable and selectable via a time range.
Quality engineers have a strong focus on the product. They track the material through the various intermediate steps of the process route. The problem of data silos is only one of the challenges to be overcome here. Another is tracking changes in the metal during the production process. Metal is converted from a liquid phase into a solid state. Slabs, blooms, and billets are transformed with every roll pass into new geometrical shapes and dimensions. Coils are cross-cut, welded and rolled. This makes it almost impossible to trace causes through the upstream or downstream processing steps. Genealogy is the term used to denote those parent-child relations of the material and is a key piece of information that enables the quality to be tracked across lines.
Within a platform like the SMS DataFactory, genealogy makes it easier to identify where the data come from and how to transform them. Data analysts can immediately start working on the data and select any available signal from an arbitrary line along the process route. The SMS DataFactory ensures that any available sensor- and time-series data are mapped to their position on the slab, bloom, billet, coil, tube, or wire with minimal errors if the automation system has not already performed this. This offers new insight into the transformation and enables the semi-finished products to be analyzed across lines by anyone within a concise time.
For planning purposes, the condition and quality data must be considered to allow systems to decide whether certain products are able to be produced with the current status of input material and maintenance.
Having the data available in the appropriate format enables the learning steel plant to analyze data and process it via IT systems. The data is then enriched with metadata: it becomes discoverable and meaningful. Here, it is now necessary to combine technology know-how with domain know-how. Within the SMS DataFactory, the Data Dictionary enables data analysts to gain deep insights into the available data through interlinked data elements based on their origin, purpose and type. Within the DataFactory, the Data Dictionary provides information on domain-specific correlations like the interaction between mold level and stopper rod position in the continuous casting machine. Furthermore, it is crucial to distinguish between the calculated nominal and actual values of a physical quantity. A standardized naming scheme supports those differences even more, regardless of the various naming schemes offered by different vendors and found in a steel plant. The filterable, full-text search of a Data Dictionary cuts the time required to find the appropriate information among tens of thousands of data. This helps to ensure a faster implementation by data analysts and thus to the successful realization of the learning steel plant.
The need for an overall data platform that speaks the metals industry's language has never been greater. It allows unified access to the process, production, quality, planning, and maintenance data. The SMS DataFactory tackles the problems mentioned by combining homogenized and standardized data access for data scientists, algorithms and other data consumers. That makes it one of the many key enablers of the Industry 4.0 revolution.