Energy consumption of wireless sensors and large amounts of data critical
Machines and industrial plants are often monitored and analyzed in real time using wireless sensors in order to detect potential problems or deviations from normal operating conditions at an early stage and predict maintenance requirements. However, this can lead to limitations: If the sensors are located in many places or in places that are difficult to access and are expected to work reliably for years, their energy consumption is critical in terms of battery life. The sensors that are of interest for monitoring also generate a large amount of data. To date, this can only be transmitted to the central monitoring device with a high energy consumption. In addition, defects on machines occur very rarely or not at all during normal operation and certain error states, such as defective bearings and machine parts, cannot simply be triggered.
New algorithm learns good states – wireless sensor only transmits deviations
With the patented edge AI solution, data is processed directly on the sensor node and reduced for wireless transmission, which reduces its energy requirements and therefore increases its service life. Interference data and error states also no longer need to be recorded in advance.
The much-cited term edge AI combines edge computing and artificial intelligence. This allows algorithms and machine learning to be executed directly on interconnected wireless sensors – locally or with an internet connection. While edge computing allows data to be stored in the vicinity of machines and systems, AI algorithms process the data directly at the edge of a network or at the interface to the higher-level network and can provide feedback on the machine status in real time.
The new algorithm developed at IMMS collects data and learns the normal machine behavior patterns directly on the sensor, largely unsupervised. The regular behavior patterns of a system and certain novelty criteria, which stand for deviating behavior, are learned from the recorded data on the wireless sensor. Data is only sent to a central monitoring system if the calculated novelty value deviates from the normal criteria for “healthy behavior”. This completely eliminates the need to transmit raw data to a central monitoring system. This on-device learning and subsequent retraining make the algorithm generalizable for different industrial scenarios and robust against possible data drift. The method for recognizing the novelty value of data combines mathematical procedures in a specific sequence for ongoing condition monitoring and predictive maintenance over years.
On-device data reduction through singular value decomposition and correlation
For this purpose, the proper operation or good condition of a bearing or machine, for example, is first recorded, its characteristics are automatically determined by singular value decomposition and a threshold value is determined on this basis in order to be able to detect deviations. This threshold value is then further improved by learning certain anomalies and states.
In order to reduce the data set on the sensor node so that as little data as possible has to be transmitted to the monitoring system, only a certain number of the dominant singular values with their associated information is used. In the application, newly recorded vibration values are analyzed using canonical correlation analysis. The differences between the learned characteristics of the good state and the newly recorded ones are used to qualitatively record changes in state.
The method can be implemented in various applications for monitoring machines and can be licensed.
German patent: DE 10 2024 100 703 B3, IP available, patent applicant/owner: IMMS Institut für Mikroelektronik- und Mechatronik-Systeme gemeinnützige GmbH (IMMS GmbH), inventor: Rick Pandey.
– – – – –
Further links
👉 www.imms.de
Photo: IMMS
