Digital control of casting processes

In order to avoid errors in manufacture while producing large cast pieces from lead-bronze-alloys by centrifugal casting, the University of Applied Sciences TH Köln and the metal foundry Martin Luck Metallgießerei from Saarland/Germany have together digitalised the casting process and optimised the process parameters using Artificial Intelligence (AI).

“Our project partner among other things produces plain bearings for machines used in mining. Those components weigh up to 1.5 tons and are produced in very small volumes. This means that the machine parameters need to be readjusted almost for every part,” says Prof. Dr. Danka Katrakova-Krüger from the institute for General Mechanical Engineering at TH Köln. The usual problem: Adjustment and documentation up to now had to be done manually. This leads to errors and makes focused, highly reproducible evaluation impossible.

“Sometimes it take three attempts until a product is perfect – and the failed attempts need to be melted down. The resulting need for energy and resources, the demand on capacities and the long delivery times put pressure on the company. AI based production systems can help to solve this problem.” says Prof. Dr. Christian Wolf from the :metabolon Institute at the university.

The first step was research into which process parameters have an especially large influence on the quality and especially the distribution of lead in the finished workpiece. “As lead has a much lower melting point than bronze and is also significantly heavier, during the cool-down process this can result in an inhomogeneous distribution of lead, which renders the product useless,” says Katrakova-Krüger. Relevant parameters were revealed to be, e.g., the casting temperature, cooling conditions, the rotation speed of the casting die or the amount and temperature of the water used for cooling.

The existing machine park was digitalised so that the chosen process parameters and machine settings can be determined automatically and put into relation with successful or unsuccessful casting results. These data are the base for training an AI. “We can enter the geometry of the desired component into the system that has been created. The AI then rather reliably suggests parameters which have led to success for identical or similar components in the past,” says Wolf. The AI can also evaluate whether a finished component meets the quality requirements. The results so far are to become the base for further development to improve predicting a component’s quality even further.

The research of the Faculty for Computer Science and Engineering Science at TH Köln and the metal foundry Martin Luck Metallgießerei was financially supported through the central innovation programme for mid-size enterprises (Zentrales Innovationsprogramm Mittelstand ZIM) by the Federal Ministry for Economy and Climate Protection from September 2020 to June 2023.