Producing large workpieces from lead-bronze alloys by centrifugal casting is an energy- and time-intensive process. Together with the Martin Luck metal foundry in Saarland, TH Köln has digitized the casting process and optimized the process parameters using artificial intelligence (AI) to avoid manufacturing errors and thus expensive scrap.
"Among other things, our project partner manufactures plain bearings for machinery in the mining industry. The components, which weigh up to 1.5 tons, are produced in very small batches. Accordingly, the machine parameters have to be reset for almost every part," explains Prof. Dr. Danka Katrakova-Krüger from the Institute for General Mechanical Engineering at TH Köln.
Up to now, both the settings and the documentation have been done manually, which is prone to errors and makes targeted evaluation with high reproducibility impossible. According to the project partners, centrifugal casting could benefit greatly from digitized and (partially) automated production. In the manufacturing process for rings, discs and tubes, molten metal - in this specific case an alloy of copper, tin and lead - is poured into a mold rotating around the central axis, known as an ingot mold. The molten metal is pressed against the mold wall by rotation and hardens there.
"Sometimes it takes three attempts before a product is perfect - the failed attempts have to be melted down again. The associated expenditure of energy and resources, the tying up of capacity and the long delivery times are a burden on the company. AI-supported production systems can help solve these problems," says Prof. Dr. Christian Wolf of the university's :metabolon Institute.
Training the AI and defining process parameters
The project partners first investigated which process parameters have a particularly large influence on the quality and, above all, on the distribution of the lead in the finished workpiece.
"Since lead has a much lower melting point than bronze and is also significantly heavier, inhomogeneous lead distribution can occur during the cooling process, rendering the product unfit for use," says Katrakova-Krüger.
Casting temperature, cooling conditions, rotation speed of the mold or quantity and temperature of the cooling water used are among the relevant parameters. The project team also digitized the existing machinery so that the selected process parameters and machine settings can be automatically recorded and correlated with successful or unsuccessful casting results. An artificial intelligence was then trained with this data.
"We can enter the geometry of the desired component into the resulting system. The AI then relatively reliably suggests parameters that have led to success with the same or similar components in the past," Wolf explains.
The system can additionally evaluate whether a finished component meets quality requirements. The project partners want to build on the results and develop them further to predict component quality even better.
The research project by the Faculty of Computer Science and Engineering at TH Köln and Martin Luck Metallgießerei was funded under the Central Innovation Program for SMEs (ZIM) of the German Federal Ministry of Economics and Climate Protection from September 2020 to June 2023. TH Köln is one of the most innovative universities for applied sciences. Currently, around 25,000 students are enrolled in about 100 bachelor's and master's degree programs.
