Industrial AI – The Dividing Line Between Winners and Losers?

Europe criticises, China invests: Chinese firms receive funding either from OEM suppliers or in the form of government grants. In contrast, European companies tend to be somewhat more sceptical towards the new technology. A survey conducted by Johannes Messer Consulting amongst companies in the European foundry industry in 2025 found that 50 per cent of decision-makers consider the introduction of AI applications to be premature and too costly.

And yet die casting stands to benefit greatly from the advantages that AI applications bring.

AI can see what you cannot

Die casting is widely regarded as a highly complex process, with a significantly greater number of influencing parameters than many other metalworking manufacturing methods. This is precisely why AI systems offer a competitive advantage. Many of these parameters interact with one another, further complicating the processes involved. Due to this complexity, certain outcomes cannot be predicted using conventional algorithms.

By contrast, artificial intelligence does not require pre-defined models. Instead, it can learn autonomously and thereby derive non-linear relationships from large volumes of data from a variety of sources. The AI improves automatically with every data point and requires no reprogramming.

A game changer for inspection and maintenance

Image-based quality inspection is one classic area of application in die casting. Deep learning methods, such as convolutional neural networks (CNNs), can learn to distinguish the characteristics of conforming parts from those with defects when trained on sufficiently large and accurately labelled image datasets. Depending on the inspection method used, these systems can detect surface cracks, cold shuts and other visible flaws. For internal defects such as porosity, X-ray procedures are often employed alongside AI. AI-assisted inspection systems can accelerate and standardise testing processes.

Predictive maintenance is another area of considerable potential. AI-based models can analyze large volumes of time-dependent process and machine data simultaneously, detect anomalies at an earlier stage, and identify complex patterns that simple threshold values fail to capture. What often matters is not a single pressure reading in isolation, but the interplay of subtly shifting pressure peaks, marginally longer switchover times, rising oil temperatures, and increased power draw. Looking ahead, AI will also be capable of supporting the diagnosis and prioritization of maintenance measures.

Artificial factory advisers for process optimization

As with predictive maintenance, artificial intelligence can assist with reducing scrap and optimising processes. By using linked machine, process and quality data, it is possible to train models to estimate the probability of quality deviations or scrap at an early stage for individual shots.

Furthermore, AI can reveal correlations between process parameters, temperature control and defect formation, thereby supporting the optimisation of tool tempering. AI also has potential applications in energy efficiency: models can forecast energy consumption, identify inefficiencies, and provide data-driven optimisation recommendations.

Trash in, trash out

However, for all areas of application, the following holds true: the performance of Industrial AI systems depends heavily on the quality of available data. The more precise, complete, consistent and up to date the data is, the more robust the results will be. Incomplete or poor-quality data, on the other hand, increases the risk of misclassification, misleading correlations and unreliable predictions.

Changes to sensor configurations or other altered conditions, as well as new machinery, can lead to data and concept drift, thereby degrading model quality. Therefore, data quality, continuous monitoring and validation against real production data are among the most important prerequisites for reliably deploying Industrial AI. Models should be monitored regularly during operation and adapted or retrained using new, validated data where necessary.

Another issue is the so-called 'black box' phenomenon. This refers to the fact that, while modern AI models deliver results, they do not explain how they arrived at them. This is particularly problematic for Industrial AI, where decisions must be traceable for safety, liability and regulatory compliance reasons. Furthermore, a lack of transparency hinders fault diagnosis and the transferability of models to new situations.

Europe’s only chance lies in data quality

According to the Johannes Messer Consulting study, data quality and completeness could be the key to the European die casting industry's success in competing with China. One possible solution is a European platform for data exchange, where all process participants collaborate. Such an approach may be the only way for Europe to reclaim its international competitiveness.