Virtual PLC in Industrial Edge Platform: Performance Evaluation of Supervision and Control Communication

Abstract

Edge computing allows for data processing at reduced latency since the computational power is moved close to the data sources. Traditionally, edge computing has been often used in industrial scenarios for implementing gateways between the OT (operational technology) worlds and the IT (cloud) world. Recently, big manufacturers of industrial PLC (programmable logic controller) started promoting the use of containerized virtual PLC hosted inside edge computing platforms. They foresee an innovative integration of container based applications, including automation control, with all the data centric services and application already available for edge ecosystems. Even if a clear advantage from the scalability and maintainability could be expected, would virtual PLCs meet the stringent requirements of industrial automation? This paper is part of a multistage research work, and, as a first step, it is focused on the evaluation of the performance of virtual PLC when exchanging data with other machines, controllers, supervisors, and data acquisition systems in a machine-to-machine scenario. After a brief overview of the involved technology, the design of a methodology for comparing real PLC and virtual PLC is described. Then, performance metrics, and an experimental setup for the evaluation of existing devices are defined taking care of the sources of uncertainty. The effectiveness of the proposed methodology is demonstrated considering a real use case. Through the use of the suggested methodology, important insights of the use case are revealed: for instance, the considered virtual PLC could work as fast as a real PLC with minimum communication latency in the order of 3 ms but, currently, there is a random delay with an average of 50ms whose source has been identified to be the IP stack implementation of the virtual PLC. Finally, the proposed methodology allows for the creation and the validation of analytical models of the use case.