Unsupervised constrained discord detection in IoT-based online crane monitoring

Abstract

Maritime transport is an indispensable element of the global logistics network. Most maritime loading-unloading operations are supported by quay cranes, making their availability and condition critical to port operations. This work identifies discordant trends arising during the vibration-based condition monitoring of these quay cranes in one of the busiest container ports in the United Kingdom. This work proposes an unsupervised and constrained discord detection approach for irregular but near real-time time series data obtained from multi-modal IoT-based condition monitoring sensors installed on these cranes and transmitted over a 5G network. Due to the live nature of the seaport’s operations, the development of controlled anomaly signatures for a baseline reference was not possible. To address the challenges of incomplete asset health information, irregular and batched time-series sensor data, massive data volumes, and the lack of the assets’ vibration signature baselines, this paper proposes an unsupervised, robust, and fast discord detection mechanism that can rapidly highlight discordant time-series chunks in the received vibration data at a central server. A Support Vector Machine based One-Class Classifier (OCC-SVM) is used to identify the discordant vibration signatures in the time series data. During the development of this approach, the timestamped data chunks from the add-on IoT-based vibration sensors were clustered into two weight classes (loaded and unloaded) based on the crane’s default Programmable Logic Controller (PLC) sensors. The efficacy of this method is checked against the crane maintenance logs and data from a separate crane’s vibration and PLC data. Further, a method for generating synthetic noise-embedded vibration signatures to test the effectiveness of the discord detection method has been devised. Finally, the practicality of the proposed OCC-SVM approach for discord detection was inspected in a constrained environment setting.