Leveraging data potential with AI and streaming

The theoretical potential of production data is huge. When properly linked, it contains important information that AI systems also need. After all, they need big data to be able to learn and make decisions. Compared to data from other departments, however, production data is not so easy to process in a standardized way. This is because the data comes from a wide variety of sources and is therefore heterogeneous. It often has to be stored for a long time so that it can be traced and the company is legally protected. Production data also changes particularly quickly, as it is directly dependent on the machines and processes from which it originates. The challenges are sometimes great. So how well can manufacturing companies currently use their data?

The quality of production data in companies in this country is good, as practice shows. This already fulfills an important criterion for the use of AI. However, there is often still a lack of in-depth understanding of this data. As a result, companies rarely go beyond feasibility studies. The maturity level of the IIoT infrastructure varies greatly. There is often a lack of the right basis for deriving clear added value from the data. At the same time, customers expect precise quality prediction and optimum parameters in production. So what are the key factors in data processing and infrastructure to enable the transition from study to practice - and what is the importance of streaming production data?

Data streaming with the right IT infrastructure

In order to implement AI projects faster, the right IT infrastructure and streaming technologies are key factors. A simple, flexible infrastructure is particularly suitable for processing production data. It should make it possible to store raw data and use it dynamically in real time. An efficient data standard is also particularly important to enable the exchange of data between machines and systems. The OPC-UA and MQTT standards are particularly relevant here. OPC-UA enables 1:1 communication between units, regardless of manufacturer or platform. MQTT, on the other hand, is leaner and offers fast, centrally controlled data distribution - ideal for the digitalization of production.

Another key issue is the question of how data is transferred between systems: Batch processing or stream processing? With batch processing, the system waits until a complete data package is ready, whereas with stream processing, the data is processed continuously and in real time. Streaming is often the better choice for production in particular, as large volumes of data can be used immediately - a basic requirement for AI applications. Once the data has been collected, it is standardized and thus forms a solid basis for further analyses.

With the right infrastructure as a basis, companies can combine AI and data streaming to exploit the full potential of their production data. Production lines and locations can be monitored and controlled centrally, which significantly improves scalability. New production lines can be easily integrated into the AI system as it adapts flexibly. By analyzing data streams in real time, the AI can react directly to changes, optimize processes, rectify errors and even enable predictive maintenance. The combination of streaming and AI therefore offers a high degree of flexibility and saves costs at the same time.

In order for companies to take advantage of these benefits, they not only need the right technology, but also in-depth knowledge of their production processes. Only when the AI has learned exactly which errors are critical and which parameters need to be monitored can it be used effectively. This is where it can make sense to work with external experts who have experience both in production and in dealing with AI.

Smart welded lasts better

One specific use case for the combination of AI and data streaming is stud lift welding. In the past, stops in production had to be checked individually by hand. In order to optimize this, downtimes should be minimized and the optimal welding parameters determined in advance. To do this, machine data is streamed via MQTT, combined with material parameters, information on stud batches and environmental data such as temperature and humidity. Thanks to AI and streaming, the welding quality can be monitored and optimized almost in real time. The AI system also automatically labels the production data.

With precise predictions of the important welding parameters, managers can reduce downtimes and increase the number of fault-free welds. As all processes are monitored remotely thanks to streaming, manual monitoring also becomes obsolete. This solution can also be scaled to different manufacturers and locations, making global implementation faster and safer. As there are no data gaps and the data sources are comprehensively available, pre-processing of the data is made considerably easier.

The future of production is smart

Data streaming forms the basis for AI feasibility studies to make their way into real production. If data from different sources is available at all times, AI systems can learn continuously and are ready for use more quickly. This process therefore reduces the technology's amortization time. Together, data streaming and AI improve each other and thus help to optimize production and make decisions more easily based on sound data. Companies that make targeted use of these synergies generate additional added value from their existing data.

Thomas Pause, Head of IT Infrastructure & Software Engineering at Salt and Pepper

Salt and Pepper, salt-and-pepper.eu