Machine Learning and Data Analysis

Listening, speaking, analyzing and learning are the main activities of the "Machine Learning and Data Analysis" group. With the methods of sound and signal analysis, acoustic pattern recognition and machine learning, products made of a wide variety of materials can be tested and evaluated in the shortest possible time. This offers advantages in economic and energy efficiency throughout the entire production and product life cycle to industries, such as mechanical engineering, the automotive and transport sectors, the energy sector, agriculture, but also the paper and textile industries.

Self-learning machines for AI-controlled inline testing

Self-learning machines based on artificial intelligence (AI) can test both known and unknown products without great effort. In addition, existing machines can be cost-efficiently upgraded with miniaturized system components to implement AI-controlled inline testing. In this way, the entire production process can be monitored, not just a selection.

Maintenance and wear monitoring: A symbiosis of sensors, algorithms and the knowledge source 'human'

The use of Artificial Intelligence is also essential for machine maintenance and wear monitoring. The "Machine Learning and Data Analysis" group develops sensors and algorithms, and combines them with the human knowledge source to predict the operating life of system components. Based on these trends, maintenance intervals can be optimized, downtimes can be reduced or production cycles can be scheduled. On the basis of machine learning and data analysis, a worn component is neither replaced too early nor too late. This reduces operating costs and prevents damage.

Acoustic pattern recognition for condition monitoring and quality assurance

Acoustic pattern recognition can evaluate a test object on the basis of noises that occur during operation. For this purpose, relevant acoustic information is extracted from the recordings of the operating noises and processed. When the recordings are compared with signals, that occur with an intact test object, a good/bad evaluation of the test object is possible. This allows the condition or quality to be monitored and controlled independently of the underlying measurement method.

Material diagnostics 4.0 for process-integrated and fully automated testing

New is the combination of methods of machine learning and pattern recognition with methods of signal analysis, feature extraction and compression. In this interaction, measurement signal models are automatically created and used to evaluate unknown test objects or test objects in an unknown state. The "Machine Learning and Data Analysis" group aims to raise material diagnostics and non-destructive testing above level 4.0 so that the control of components, assemblies, materials, machines and entire plants can be realized fully automatically and integrated into the process.