Under CopyrightMayer, DirkDirkMayerBeyer, VolkhardVolkhardBeyerLehmann, MartinMartinLehmann2023-11-062023-11-062023-07-07https://publica.fraunhofer.de/handle/publica/456503https://doi.org/10.24406/publica-211710.24406/publica-2117In industrial applications, sensors are a key element to enable added value of data analytics. However, sensors are no longer just measuring transducers, but communicating computing platforms, able to implement complex algorithms in order to merge and analyze data directly at the transducer. Due to the evolution of microelectronics, discrete sensor designs have been replaced by microelectromechanical systems (MEMS) for years. This enables sensor systems, that are compact, energy-efficient and less expensive, even a full integration into mechanical subsystems and components of machines has become possible. Still, industrial sensor applications, particularly those in a dynamic closed-loop control for robotics, production systems or autonomous vehicles place certain requirements on the reliability of the sensors. Signals acquired by the sensors have to be fully trustworthy in order to avoid accidents, damage to the equipment or deterioration of process quality. Integration of further sensing elements as canary devices and machine learning algorithms allows for self-monitoring of the sensor systems and the sensor electronics, i.e. to trigger warnings when signals show anomalies, operating conditions are out of a defined range, or disturbances of the sensor signals occur. Since sensor systems have become complex, micro-mechatronic systems, also development and test methods have to be expanded. In case of integration of machine learning, new training and test procedures for sensor-embedded AI have to established to ensure a reliable and error-free operation under harsh industrial operation conditions. Therefore, techniques from machine learning, hardware-in-the-loop validation and environmental testing have to be merged to develop laboratory environments that allow for the simulation of a wide range of operational conditions and injected faults. A high-fidelity, accelerated test of prototypes of smart sensors in the laboratory reduces the amount of costly field tests at real machines or vehicles, but also enables rapid feasibility studies when exploring new sensor applications. The contribution gives an overview of recent developments in the projects Trust-E and the AI application and test center at Fraunhofer EAS.enDDC::000 Informatik, Informationswissenschaft, allgemeine Werkepresentation