Steelmaking is full of high-end equipment and complex chemical and mechanical processes. Although computer-controlled and monitored by thousands of sensors, certain disruptions are just inevitable due to the high complexity and the vast amount of operational data. aivis helps you to understand and avoid those disruptions.
AI in Steelmaking
An autonomous AI for operations data like aivis enables you to realize tremendous savings and improvements along the entire production chain – be it by preventing all kinds of process disruptions, by monitoring critical components with anomaly detection, or by improving the information base through soft sensors.
Prevent known process disruptions
Understand and prevent all kinds of process disruptions that show up regularly and whose occurrences can be found in the historical data of the process.
Monitor the health of critical components
Monitor the health of critical components with anomaly detection to receive timely warnings of emerging malfunctions and avoid unplanned downtimes.
Measure critical parameters with soft sensors
Create virtual sensors for critical process parameters that are hard or even impossible to measure and increase transparency and safety.
Preventing known process disruptions
Whether electric arc furnaces or basic oxygen furnaces, hot rolling or cold rolling – steel making is full of high-end equipment and complex chemical and mechanical processes. Although being computer-controlled and monitored by hundreds, sometimes thousands of sensors, certain kinds of disruptions just keep on occurring. The high complexity and large volumes of data make it hard to maintain an overview and initiate necessary countermeasures in good time.
This is where aivis comes into play: The powerful AI for operational data can handle even the most extensive amounts of data of thousands of sensors and autonomously look for root causes and warn in time about arising disruptions.
Just by looking into your data, aivis finds root causes that are yet unknown.
Screenshot of an aivis prevention report about a tap hole blockage indicating two independent segments identified as root causes. Per segment, all blockages being part of this segment are listed (Events), and the signal misbehaviors which are characteristic to this segment (Segment Signals) including a signal visualization.
How it works
All aivis requires to start is a hint to recognize all occurrences of the disruption in the past. It is somewhat similar to giving a scent sample to a tracking dog to pick up the scent.
aivis then automated, unbiased, and autonomously searches all historical operational data to identify critical dynamics that usually were present right before the disruption occurred.
aivis then creates a report that allows to understand those dynamics and find suitable countermeasures. Furthermore, it makes a model which detects those dynamics to apply the countermeasures in time.
Avoiding tap hole blockage
The free opening of a tap hole is of critical importance. A regularly blocking tap hole significantly decreases productivity and endangers the engineers who have to do the lancing to reopen the tap hole. There might be multiple reasons for a blocked tap hole; some are known, some are not.
With aivis, you can figure out all the destructive dynamics that lead to the blockage. Furthermore, you can receive warnings about arising blockages, including the current reasons, to timely apply suitable countermeasures.
During basic oxygen steelmaking (BOS), the critical problem of slopping occurs when foaming slag exceeds the vessel’s height and overflows, causing metal loss, process disruption, environmental pollution, and endangering people.
The complexities that lead to slopping pose challenges even for experienced operators. Because although the problem is well researched in theory, and there are even white box prediction models for the slag level, in reality, each converter is a little different, following its own dynamics.
With aivis, a customized slopping prevention model can be created for each converter just by the press of a button. Autonomously and unbiased, aivis analyzes each converters historian revealing all critical dynamics that led to slopping in the past. Then, it creates a model for the converter, which detects those bad dynamics and warns about them in time to apply suitable countermeasures. Also, aivis creates a root cause report providing full transparency and explainability of these dynamics, which helps the operators to find countermeasures and improve their understanding.
Avoiding metal strip breaks
The breakage of strip steel during rolling is a disruptive event that can be very costly. Besides loosing some of the product, it is dangerous to people and can damage the equipment. Thus, avoiding strip breaks is of vital importance.
Due to the high complexity of the process, which is monitored by several thousand sensors, even experienced operators have only limited knowledge of when countermeasure must be initiated to avoid strip breaks, resulting in significantly lowered productivity.
With aivis, a live guard can be created, that monitors the current process and warns the operator if a dangerous process state occurs and suggests suitable countermeasures.
These are just some examples out of the wide range of applications for aivis in steelmaking. In general, aivis can do the same for any regularly occurring problem recorded in the historian of the asset in question.
Scalable health monitoring of critical components
Critical components like e.g. proportional valves can be found in large numbers. So, even if one component fails only rarely, their large number makes unplanned process interruptions due to component failure inevitable. This has a negative impact on productivity.
With a suitable monitoring solution based on anomaly detection, impending component failures could be detected and dealt with in a timely manner before they actually occur. But usually creating such monitoring models is difficult and expensive, because a lot of pre-knowledge has to be built into the model to reflect the experiences of skilled operators and monitor the relevant relations. This made the generation of monitor models in large numbers impossible.
With aivis, it get’s a lot easier. The only relevant input for aivis is the signal, that reflects the loss in performance, usually a target-actual comparison. aivis then finds out on its own all relevant relationships to monitor for watching the components health, enabling a predictable and plannable maintenance of the components during planned downtimes.
With aivis, watching critical components is so simple, that it can easily be scaled up.
Watching proportional valves
Proportional hydraulic valves are sensitive for unexpected maintenance which causes unplanned downtimes and lowers productivity. However, an anomaly detection model watching the valves can significantly lower these downtimes. Because, luckily, only in the rarest of cases does the valve fail from one moment to the next. Usually, impending failure looms hours or even days before the actual breakdown enabling timely maintenance during planned downtimes and maintenance windows.
For creating the anomaly detection model, to watch a valve, aivis only requires the target-actual comparison signal. It then automatically searches all available signals of the valve for critical relationships that make the unhealthy behavior of the valve visible. The model can be easily deployed into the components live data streams.
In general, aivis can do the same for any critical components or system that provides enough data.
Creating soft sensors for critical quantities
In steel production, not every process parameter can just be measured. Due to the complex chemical processes under intense heat, some parameters are very difficult or even impossible to measure directly, such as e.g. the homogeneity of the gas flow through a blast furnace.
As a result, operators have to derive these variables from their experience and other data. But as processes get more complex and data volumes more extensive, this becomes more and more challenging. So operators sometimes have to literally fly blind, which lowers productivity and can even be dangerous.
This is where soft sensors can make a huge difference. These sensors use a mathematical model to deduce the critical process parameters from more accessible process parameters. The model is gained by analyzing and learning from the relations of the target signal to the other signals in the historical data.
As soft sensors have been known and used for several decades now, their creation has always been a struggle. Data scientists and process engineers had to work hand in hand to work out critical relations between the signals and cast them into a model. Furthermore, the soft sensors required continuous maintenance since their models tended to drift and had to be realigned.
This changes with aivis: You just have to define your target parameter and press start. aivis then figures out all relevant relationships on its own, quickly creating an accurate and durable soft sensor.