1 Towards a Reference Architecture Model Things (IIoT). One driving factor is the motiva- for Industrial Edge Computing tion to operate a converged, more software-based communication infrastructure, built upon open stan- Alexander Willner and Varun Gowtham dards, to enhance flexibility and to allow for data- driven business models in the production industries. This does not only reduce costs for operators but Abstract—In the context of the digital transformation also facilitates a shorter time-to-market span for new of the industry, whole value chains get connected across products. various application domains; as long as economic, ecologic, Specific challenges in this context are require- Chal- or social benefits arise to do so. Under the umbrella lenge: of the Industrial Internet of Things (IIoT), traditional ments related to industrial communication systems. Re- Operational Technology (OT) approaches are replaced or To support determinism, low latency, and autonomy, quire- at least augmented by Information and Communication many technological approaches related to field buses ments Technology (ICT) systems to facilitate this development. and Industrial Control Systems (ICSs) have been To meet industrial requirements, for example, related to evolved in the last 70 years. A significant milestone privacy, determinism, latency, or autonomy, established was the invention of the Programmable Logic Con- Cloud Computing mechanisms are being moved closer to data sources and actuators. Depending on the context, this troller (PLC) in 1968 to easily modify local control distributed Cloud Computing paradigm is named Edge loops. Computing or Fog Computing and various challenges have However, programming PLCs did not evolve Prob- been subject to several publications. However, a proper ref- lem: at the same pace as Internet technologies. Many Limi- erence model that describes the multi-dimensional problem control systems are still programmed using State- ta- space which is being spanned by this paradigm, seems still ment Lists (STLs), Ladder Diagrams (LDs), or a tions to be undefined. Such a model should provide orientation, put work in relation and support the identification of Structured Control Language (SCL), following the current and future research issues. This paper aims to International Electrotechnical Commission (IEC) fill this gap with a focus on industrial automation and 61131 standards. Not only do these approaches follows analog models that have been developed for specific impose significant limitations to the potential com- domains such as the Smart Grid Architecture Model puting capabilities modern endpoints expose, but (SGAM) and the Reference Architecture Model Indus- also younger generations face significant challenges trie 4.0 (RAMI4.0). The proposed Reference Architecture Model Edge Computing (RAMEC) identifies 210 views updating existing code. Further, the maintainability, on the Edge Computing paradigm in the manufacturing scalability, and modularity of these programs are domain. Future iterations of this model might be used for rudimentary. the classification of relevant research, standardization, and To address the latter, the IEC 61499 standards Exist- development activities. ing introduce an event-driven distributed system based Work: Index Terms—Distributed Computing, Cloud Comput- on the IEC 61131. Further, Programmable Automa- IEC 61499, ing, Edge Computing, Fog Computing, IoT, IIoT, Industry tion Controllers (PACs) enable the use of higher- PACs 4.0, Smart Manufacturing level instructions in parallel to software-based PLCs for hard real-time control, for example on Industrial arXiv:2008.04164v1 [cs.DC] 10 Aug 2020 PCs (IPCs). INTRODUCTION These developments are now being augmented Ap- To optimize the efficiency, to establish new by recent developments from within the ICT do- Con-proach: text:Edge business models or to enhance sustainability, various main, more specifically from the fields of network IIoTCom- put- value chains are in the process of getting digi- and system management and distributed systems ing tally connected using Internet of Things (IoT) tech- and communication networks respectively. In the nologies. Within industrial domains, the established 70 years history of computing, we can observe a Operational Technology (OT) is in the process of cyclic alternation between a centralized comput- being augmented or replaced by Information and ing paradigm (1950s: Mainframes; 2000s: Cloud Communication Technology (ICT). This evolution Computing) and a distributed computing paradigm is sometimes denoted as the Industrial Internet of (1980s: Client Server). The current Edge Comput- ing paradigm [9] distributes the Cloud Computing Alexander Willner is with Fraunhofer FOKUS, Berlin, Germany Varun Gowtham is with Technische Universitt Berlin, Berlin, paradigm by moving functionalities again closer to Germany data sources and actuators. This allows for the usage 978-1-5386-5541-2/18/$31.00 © 2020 IEEE 2 of Cloud Computing technologies within several ERP ERP critical IoT domains, particularly in industrial use MES MES via MANO cases. For example, complex applications or fed- SCADA SCADA over TSN / 5G erated learning algorithms [kairouz2019advances] PLC vPLC on Edge Computing can be orchestrated towards Edge nodes to either I/O I/O influence local control loops or to provide valuable data to a Manufacturing Execution System (MES). Fig. 1. Future technologies within the current Automation Pyramid However, a review of related work indicated layers Con- tribu- that the multi-dimensional problem space, that the tion: usage of the Edge Computing paradigm in industrial RAMEC fields of application spans, has not been properly Following the Industry 4.0 vision, this pyramid In- dustry described yet. Therefore, the main contribution of will be replaced in the long term by the use of 4.0 this paper is the presentation of a Reference Archi- Cyber-physical Systems (CPSs) or Cyber-physical tecture Model Edge Computing (RAMEC) for the Production Systems (CPPSs). Intelligent units, such manufacturing domain. Analog to the Smart Grid as Autonomous Guided Vehicles (AGVs), will then Architecture Model (SGAM) [11] and the Reference interact directly with each other to enable the most Architecture Model Industrie 4.0 (RAMI4.0) [2] and flexible production at the lowest costs possible. In in contrast to technical architectures such as Multi- this context, the use of an Asset Administration Access Edge Computing (MEC) by the European Shell (AAS) is often mentioned, which, in combi- Telecommunications Standards Institute (ETSI), the nation with an ”asset”, then represents an intelligent RAMEC is supposed to provide orientation and to Industry 4.0 component. put activities such as initiatives, standards, publica- To facilitate this transition, new software-based To- wards tions and implementations into a broader context. industrial infrastructure technologies are needed to software- be introduced into the manufacturing domain (see based in- the right part of Figure 1). The boundaries be- fras- RELATED WORK tween physical PLC and the classical IPC will truc- further be blurred by the use of edge-based and tures Industrial Automation distributed Virtual Programmable Logic Controllers To put the work at hand into context, it is (vPLCs). For example, based on IEC 61499. To Au- toma- important to understand the application domain we meet requirements related to reliable, low-latency, tion are considering, as different use cases entitle dis- and deterministic wired communication, the open Pyra- tinct requirements. In this paper, we focus on IIoT standards within the Institute of Electrical and Elec- mid domains, i.e. the application of IoT technologies tronics Engineers (IEEE) Time-Sensitive Network- in industrial fields of application. More specifi- ing1 (TSN) group are currently being adopted. For cally, the automation within discrete manufacturing. wireless communication, the TSN standards are The most important concept here is the so-called in the process of being included in the next 3rd Automation Pyramid (IEC 62264) that defines the Generation Partnership Project (3GPP) release to different layers which modern industrial automation allow for Ultra-Reliable Low-Latency Communica- is composed of. tion (URLLC) over 5G networks. Further, instead of Although this pyramid is sometimes differently only operating a centralized MES, a Management Fig- defined, in its core, it consists of five layers (see the and Orchestration (MANO) system might dynam- ure 1 left part of Figure 1). It starts with an Input/Output ically install and run applications on Edge nodes (I/O) layer that is controlled by PLCs. These PLCs close to the data source, and, for example, Artifi- are interconnected via a Supervisory Control and cial Intelligence (AI) based pre-processing and data Data Acquisition (SCADA) system with the MES analytics can be executed locally [6]. A handy ex- layer. Finally, the overall production and business ample is an Augmented Reality (AR) based Human systems are exchanging information within an En- Machine Interface (HMI), which requires very short terprise Resource Planning (ERP) system. It is also communication latencies and places high processing important to notice that different layers are con- cerned with different sizes and time frames of data. 1http://ieee802.org/1/pages/tsn.html 3 demands on object recognition. Reference Architecture Model Industry 4.0 Fig. 2. Reference Architecture Model Industry 4.0 (RAMI4.0) [2] Fig. 3. Smart Grid Architecture