Artificial Intelligence and Machine Learning for Industrial Production

In production processes, we use machine learning to generate “knowledge“ from “experience” in a very general sense. Learning algorithms develop a complex model from sample data with the largest possible degree of representation. Subsequently, this model can be applied to new and potentially unknown data of the same kind. Machine learning is an appropriate method whenever processes are too complicated to describe in an analytic way, but there is a sufficient amount of sample data such as sensor data or images. [1] The models are matched with the data flow from operational business and ultimately enable forecasts or recommendations and decisions.

Examples of how machine learning can improve quality and reduce time or costs:

• Discovering anomalies in the behavior of machines or components because the procedures reliably discover deviancies from the normal behavior of a process and consequently enable predictive maintenance.
• Making better decisions in complex situations because the models can identify the connections spanning several manufacturing stages to enhance their ability to serve as assistant systems.
• Adapting manufacturing and assembly processes to current situations quickly because clear correlations between measuring results and process parameters allow for automated control.

Further areas of application of machine learning that we are developing for our customers are human-robot cooperation, autonomous intralogistics, and self-organization in manufacturing.

We support you in selecting the right learning and modeling algorithms, defining, editing, and storing representative training data, generating meaningful models from the training data, and then comparing these models with runtime data. All these tasks require appropriate sensor technology, software tools, and architectures. We support you in establishing these issues in a future-proof and sustainable way.

Research on machine learning is proceeding. For example, the relevant issues are machine learning with extremely large or very small amounts of data, the combination of machine learning with physical or expert knowledge as well as security and transparency of machine learning models.

[1] Fraunhofer Gesellschaft (Hrsg.): Machine learning - an analysis of skills, research and application. Munich, 2018

In production, data always has to be interpreted in the context of the product or the processes. If this is the case, data forms precious resources to improve the value-added process or to de-velop new business models. [1] This also means that each practical case requires its specific data. For this very reason, it is so important to analyze the right, high-quality data. It is only under these preconditions that we can make efficient use of AI. 

Our experience has shown that one of the major obstacles in the use of AI today is to extract the very data from the processes. We support you in gathering AI-relevant data from your machinery and equipment as well as their components. This data is either collected from the machine controls, the existing sensor systems of the machine and/or smart sensors that have been added later [2], which we will chose and install in cooperation with you. Together with you, we will specify the level of granularity required for a specific task and define how data from multiple sources can be integrated smoothly and in what format the data is transferred and stored. In this context, we make sure to select the right data for practical use, to sort existing data sets and to edit them for subsequent modelling.

Using our proven PLUGandWORK-solution components, we also turn components and ma-chinery into data suppliers which have not been networked yet.

In addition, we provide know-how in the fields of data security and data protection because a larger degree of networking includes a higher risk of cyber attacks. Today, however, we have a lot of technologies that ensure, when used in the right way, that you remain the owner of your data remains with you. The decisive factor is a tailor-made and secure IT architecture for collecting, storing and evaluating the data. 

[1] World Manufacturing Forum: The 2018 World Manufacturing Forum Report – Recommendations for the Future of Manufacturing.

[2] Werthschützky, R. (Hrsg.): Sensor Technologien 2022. AMA Verband für Sensorik und Messtechnik e.V., 2018

Studies show that targeted cooperation between companies and research institutions leads faster to new products, services and processes. Within our 'Enterprise Labs', employees from your company work together with Fraunhofer scientists and developers in a team on a daily basis to create concrete product and process innovations. Your employees contribute specific product and process know-how and knowledge of your markets. Your employees will have a workplace at the IOSB, but can also work on defined project tasks at their 'home workplace' in the company. In this way they also act as know-how multipliers at their headquarters. Our scientists have extensive technology know-how and application knowledge from various industries. In this way, cooperation produces targeted results, including jointly developed business cases with measurable customer benefits. Due to our proximity to the Karlsruhe Institute of Technology (KIT), we also integrate master students or doctoral candidates into your tasks.

Our research factories and laboratories are an ideal environment for production-related AI projects. They are equipped with state-of-the-art industrial components ranging from sensors to cloud infrastructure, so that later applications and products can be tested and improved under realistic conditions.

If you also want to work on your own machines and systems, we offer you another AI-related highlight: the Karlsruhe research factory (in German language only). On an area of about 4,500 m², we will instrument manufacturing processes together with industrial partners, evaluate data and develop new AI-related solutions, e.g. to quickly bring new manufacturing and assembly processes to series production readiness. "Immature processes" refer to manufacturing processes that are not yet fully understood and mastered because they are either new, involve new materials or one does not fully understand which process parameters are actually responsible for product quality. At the Karlsruhe research factory, we work with you to find out which screws in the process need to be 'turned' to ensure that the quality of the products is and remains consistently high. Based on ML processes and measurement and control technology, the machines and systems should ultimately adjust their process parameters themselves, e.g. if the quality of the products deteriorates gradually or the ambient conditions change.

Below you will find a selection of further information on the topic (in German language only):