Design and Testing of a Self-Learning Micromagnetic Anomaly Sensor on a Multidimensional Feature Space

In the context of Industry 4.0, it is crucial to maintain high quality standards and ensure that is consistently monitored . For this task, non-destructive testing
methods are being used. However, the necessary reference samples to identify the normal and rarely occurring defects are continuously needed. In general, producing these reference samples became challenging, inefficient, and costly. In this thesis, using the 3MA-X8 micromagnetic characterization method, significant mechanical properties are obtained and data from different samples is collected. Under the assumption that these defective products occur with low probability, the goal is to develop a method for accurate classification. The algorithm developed trains an autoencoder only on the characteristics of the points that are frequently occurring in the data collection. The method identifies the defects by using the changes in the performance of the autoencoder model as an indicator, when it is asked to reconstruct them. Tests and evaluations of this method were conducted on different samples. A practical application was developed which allows the active use of this method within the MMS software structure that is integrated with the 3MA-X8 system. Using the Anomaly Detector, anomalous behaviors of different samples have been detected, and the introduced method is found to be reliable.