Automotive Intelligent Algorithms: Data Analysis in Real-Time
Casting is now increasingly determined by high-tech at the BMW Group. The light metal foundry in the Landshut plant in Bavaria has recently begun monitoring production using Business Intelligence, Predictive Analytics and Artificial Intelligence - and analyzing all casting processes in real-time with the help of Big Data.
What are the benefits of real-time analysis using intelligent algorithms? The Landshut-based foundry specialists are not only able to generate complete data transparency and data visualization at any time with just one mouse click, they can also make quality predictions. At the same time, the profitability is increased.
Last year, the light metal foundry of the BMW Group's Landshut plant produced 4.3 million cast components with a total weight of 73,000 t. The production scope includes engine components such as cylinder heads and crankcases, components for electric drives or structural components for the vehicle body.
New Opportunity through Artificial Intelligence and Smart Data Analysis
"Artificial intelligence and smart data analysis offer completely new opportunities that go far beyond our existing analysis capabilities. We can use them to manage our foundry intelligently and evaluate huge amounts of data quickly and reliably," says Nelly Apfel, Data Science Officer at the BMW Group's light metal foundry in Landshut, Germany. "This not only ensures the premium quality of our castings, but also provides greater efficiency throughout the entire value-added process. And at the same time it provides an important decision-making aid for process improvements".
Thousands of Parameters Within One Casting Process
This is based on data from various systems, in which thousands of material, condition and process parameters are stored for each casting process and each individual component - starting with the factors influencing the shaping sand cores and the parameters of the individual casting plants, right through to the plants for the subsequent machining of the cast raw parts. In the case of the sand cores alone, this includes a wide range of data, for example the composition of the sand, the room temperature and humidity, the storage time of the sand cores or the time spent in the temperature-controlled high-bay warehouse. In addition, there are all parameters related to the actual casting process, such as the temperature curves of dozens of built-in thermal sensors, pressure curves, vacuum values, cycle times, the data of the respective casting system (such as target parameter specifications), data of the casting tool used (such as the age of the tool or the number of maintenance operations performed) - or the data of the heating and cooling circuits. During the casting process, these control the solidification of the liquid aluminum, which has a temperature of up to 750 degrees.
A clean data basis is required so that root cause analyses can be carried out at all. For this purpose, the machine and process data are linked with quality data and automatically processed so that they can be evaluated in real-time. Quality data includes, for example, the three-dimensional measurement data of castings from the computer tomograph. The 3D measurements are used to determine any defect patterns in the castings - from porosity and blowholes to so-called cold runs during solidification of the metal. Quality data from the BMW Group's vehicle and engine plants, which process components from the Landshut-based light metal foundry, are also used.
Recognition of Cause-and-Effect Relationships Using Intelligent Algorithms
All this linked data is then analyzed using intelligent algorithms and is immediately available to foundry experts in visualized form. "Data transparency helps us to recognize causal relationships. This is important for component quality. And it enables our casting technologists to put together an optimum parameter set for the individual casting systems," explains Nelly Apfel. Parameter value monitoring is used to ensure stable and consistent production. It continuously checks the approved parameters, automatically triggers an alarm in the event of any deviations - and stops casting processes automatically if necessary.
In addition, machine learning can be used to detect recurring patterns or anomalies in the casting processes and to make very precise quality predictions based on possible error patterns (predictive quality). Real errors in the production process are thus reduced to a minimum. "We score the probability of rejects from the parameters with which our components are manufactured," explains Nelly Apfel.
According to Nelly Apfel, there are also other advantages: "The visualization of key process parameters such as flow rates, temperatures or thermal images not only helps those responsible for production, but also enables early intervention by the maintenance department". Two examples: Anomalies in the temperature curves can indicate defects, low flow rates can indicate deposits in the cooling circuits. "This increases the output of our plants and thus the profitability".
No In-Depth IT Competence Required
Pronounced IT competence is not required to operate the intelligent data solution: it can be easily used on a tablet or smartphone via web app. "Previously, such comprehensive data analyses were only possible with time-consuming manual evaluations and test runs," says Nelly Apfel.
She and her team are currently working on a new AI application in the field of deep learning. A neural network is used to evaluate images of cast parts and make quality statements. This is used to automatically determine whether and to what extent a casting needs further processing. The aim is to automatically detect any necessary reworking steps for castings.