Open Source OPC UA Data Traffic Characteristic and Anomaly Detection using Image-Encoding based Convolutional Neural Network Helmy Rahadian(a*), Steven Bandong (a**), Augie Widyotriatmo (b*,d), Endra Joelianto (b**,c,d)
a) Engineering Physics Graduate Program, Institut Teknologi Bandung, Indonesia
*33321001[at]mahasiswa.itb.ac.id, **bandong.steven[at]gmail.com
b) Instrumentation and Control Research Group, Institut Teknologi Bandung, Indonesia
*augie[at]tf.itb.ac.id, **ejoel[at]tf.itb.ac.id
c) National Center for Sustainable Transportation Technology (NCSTT), Institut Teknologi Bandung, Indonesia
d) University Center of Excellence Artificial Intelligence on Vision, NLP & Big Data Analytics (U-CoE AI-VLB), Institut Teknologi Bandung, Indonesia
Abstract
The development of OT and IT technology and industrial growth require the design of automation systems that can accommodate scalability and interoperability between devices. OPC UA is a communication protocol that bridges data exchange between devices using different communication platforms and protocols. OPC UA can also connect devices between levels in automation pyramids. As an open platform, open-source OPC UAs such as open62541, OpenOPCUA, and FreeOpcUa are currently being developed by several developer communities. Implementing open-source OPC UA is an attractive option if cost is a significant consideration. However, the primary purpose of implementing OPC UA is to communicate or exchange information effectively and reliably- information about the characteristics and performance of open-source OPC is needed before designing a particular open-source OPC-based automation system platform. This paper utilized FreeOpcUa, a Python OPC UA library, to determine communication traffic features between client and server and perform anomaly detection on the traffic. The results showed that when all clients read server data simultaneously, there was duplication (up to 9%) and loss (up to 5%) of some data packets. Otherwise, the server could read all clients^ transmitted data. Anomaly detection testing with an image-encoding CNN also showed promising results, with accuracy, precision, recall, and F-score values approaching one.
