Model-based anomaly detection in response delay in communication through LTE network

Industrial monitoring systems are developed and deployed to continuously and remotely monitor the status of industrial equipment and detect failures. A gateway node collects status data from sensors attached to machines and then sends them to a server for analysis and evaluation. In this paper, we propose a method to detect anomalies in communication between a gateway and a server over an LTE network, whose failure would bring a serious result such as an operation halt of the whole factory. We model a time series of dynamically and instantaneously changing response delay as sawtooth waves and detect an anomaly based on their statistical characteristics. Through evaluations using real measurements and artificial data, we verified that our method can detect both of constant increase and decrease, rapid increase, long-term increase, and slow decrease in the response delay.