Anomaly detection for logistics system

Abstract

This dissertation address anomaly detection in logistic systems. A logistic system is a system that manages the flow from the origin to the destination of the logistics. Logistics are carried out by logistics robots, such as Overhead Hoist Transport system, Autonomous Mobile Robot(AMR), and Automated Guided Vehicle(AGV). This dissertation covered the following three topics. (1) Calculation of driving stability when switching control in an autonomous vehicle. (2) Anomaly detection of the OHT system, which is the backbone of semiconductor logistics. (3) Operational states aware anomaly detection of the logistics robot. The first topic, calculation of driving stability, suggests a method to learn driving stability, which is difficult to define and collect. Calculated driving stability of the proposed model shows similar changes to driver readiness from actual data. In addition, it was confirmed that the characteristics of stable driving were learned for data channels not considered for training data. Proposed model calculate the driving stability based on various driver, vehicle, and environmental data. In the second topic, the anomaly detection of the OHT system, the study was conducted based on data collected through the IoT board from the test bed same as the actual semiconductor fab. In previous anomaly detection studies, little attention has been paid on large facilities such as logistics robots. IoT board used in the study can collect data and detect anomalies without interruption of factory operation. Based on the actual data collected from the test bed, we confirm that the proposed model has sufficient anomaly detection performance. In the third topic, an anomaly detection considering the characteristics of logistics robots that are constantly moving and exposed to changing environments. Logistics robots conduct various tasks and represent different data patterns at each task. By considering these characteristics, we tried to achieve more robust and better anomaly detection performance. Experiments have confirmed that the proposed model has a better understanding of the operational states of the logistics robot and more reliable anomaly detection.