，，，，

College of Mechanical and Electrical Engineering，Nanjing University of Aeronautics and Astronautics，Nanjing 210016，P.R.China

Abstract:Cloud manufacturing has become a reality.It requires sensing and capturing heterogeneous manufacturing resources and extensive data analysis through the industrial internet.However，the cloud computing and serviceoriented architecture are slightly inadequate in dynamic manufacturing resource management.This paper integrates the technology of edge computing and microservice and develops an intelligent edge gateway for internet of thing（IoT）-based manufacturing.Distributed manufacturing resources can be accessed through the edge gateway，and cloud-edge collaboration can be realized.The intelligent edge gateway provides a solution for complex resource ubiquitous perception in current manufacturing scenarios.Finally，a prototype system is developed to verify the effectiveness of the intelligent edge gateway.

Key words：edge computing；intelligent gateway；microservice architecture；cloud manufacturing

0 Introduction

With the integration of manufacturing technolo?gy，information technology，and network technolo?gy，the manufacturing industry faces a revolution in dynamic resource management.The new scenarios require manufacturing companies react rapidly to meet market and consumer needs.In the novel net?worked hybrid manufacturing，the underlying entity is the execution unit of the manufacturing system and has been dramatically increased.These manu?facturing resources complete networked tasks by sharing processing capabilities.

As a new service-oriented networked manufac?turing paradigm，cloud manufacturing has promoted the continuous development of the manufacturing in?dustry towards intelligence and service.By con?trast，the traditional manufacturing system is divid?ed into equipment，control，workshop，enterprise，and collaboration layers according to physical attri?butes［1］.It is precise that the multi-layer structure has become increasingly flat during the industrial in?ternet of thing（IIoT）development.At the same time，discrete manufacturing resources can be ac?cessed to the cloud manufacturing platform through the network［2］.

Manufacturing resource perception and access is an essential part of cloud manufacturing.In this context，the comprehensive perception and access to various manufacturing resources is the primary support for networking and servicing.Therefore，the original multi-layer architecture divided accord?ing to physical attributes can be simplified into three layers according to its functions：The physical infra?structure layer，control layer，and application lay?er［3-4］.As an aggregation of all manufacturing re?source entities，the physical infrastructure layer pro?vides the hardware foundation for cloud manufactur?ing.The control layer links manufacturing resources and manufacturing services.Finally，the application layer provides the scheduling optimization between consumer needs and manufacturing entities.Rele?vant scholars have carried out much research and practical work around the perception and access to manufacturing resources.

It is also applicable for intelligent manufactur?ing with software-defined manufacturing service（SDMS）as the most apparent feature.However，driven by IIoT and SDMS，the three-level architec?ture requires more practical enabling technologies to satisfy the requirements of mass personalization［5］.This paper combines edge computing and microser?vice，and develops an intelligent edge gateway.The edge gateway can realize the dynamic management of manufacturing resources.

（1）Edge gateway

The role of the edge gateway is highly empha?sized in the white paper on discrete manufacturing edge computing solutions released by the Alliance of Industrial Internet.The edge gateway can con?vert the industrial protocols of various manufactur?ing entities into standard OPC UA and other com?munication protocols［6］.Compared with traditional industrial gateways，edge gateways split each ser?vice component into finer granularity［7］.It enables the gateway upper-layer and the cloud platform to communicate using a unified protocol and informa?tion model.It also reduces the coupling between ser?vice modules and dramatically improves the scalabil?ity of the gateway system.

（2）Microservice architecture

The edge gateway adopts a cloud-native mi?croservice architecture.It integrates lightweight con?tainer management and virtualization technologies to build a unified heterogeneous data integration platform.Microservices can realize distributed de?ployment as a new service type compared with tradi?tional service-oriented architecture（SOA）.More sensor accesses can be supported through the hori?zontal expansion of the edge gateway and solve mas?sive connections in IIoT［8-9］.

In the manufacturing workshop，it is necessary to add data collection and control terminals for each device to realize the perception and access of manu?facturing resources.The SOA model can hardly achieve rapid reconfiguration of the manufacturing system on the edge gateway software architecture.The microservice architecture is a new paradigm for these large-scale entity clusters by contrast.

This paper implements an intelligent edge gate?way that combines edge computing and microser?vice architecture.First，the problem of data block?ing caused by the heterogeneity of IIoT is solved by the standard protocol.Second，the implementation of microservice architecture solves significant infor?mation interaction delays［10］.

The rest of this article is organized as follows.The system architecture of the edge gateway is first outlined in the cloud customization environment.The authors’group describes the critical technology for the physical manufacturing entities to generate virtual manufacturing resources in the edge gate?way.Then the authors’group proposes an edge gateway operating mechanism based on microser?vice architecture.Finally，a prototype system of the edge gateway application in the production site is in?troduced.

1 System Architecture

A gateway is essential for realizing edge com?puting-oriented distributed manufacturing resource access in the cloud manufacturing scenario.Com?pared with monolithic architecture and SOA，mi?croservices are optimized in system complexity.Therefore，a gateway based on microservice archi?tecture and edge computing is more suitable for the fragmentation control of distributed manufacturing resources than SOA and cloud computing.

The proposed intelligent edge gateway has a four-layer system architecture as a middleware plat?form linking manufacturing resources and the cloud platform.After analyzing the foundation of the exist?ing edge gateway architecture，we propose a system architecture based on the microservice concept，as shown in Fig.1.Specifically，the first layer is the protocol adaptation layer，which is mainly responsi?ble for accessing heterogeneous control systems as the essential service.Secondly，the basic service layer mainly completes service management（e.g.，service registration，service configuration，and ser?vice tracking）and API service isolation.Thirdly，the dynamic routing layer realizes real-time informa?tion interaction between the horizontal devices and the vertical cloud platform.Finally，the analysis and decision-making layer completes resource allocation and scheduling optimization by analyzing the realtime status of underlying physical entities.

Fig.1 Architecture of the edge gateway based on microservice

The granularity is further refined based on these four essential services［11］.Each basic service can be split into a series of smaller microservices.

（1）Protocol adaptation layer

The protocol adaptation layer is mainly respon?sible for linking the heterogeneous manufacturing en?tities.These entities usually use different industrial protocols for data transmission.The protocol adapta?tion layer can be compatible with standard industrial protocols such as OPC UA，MTConnect，and NC_Link［12-13］.When the analysis and decision-mak?ing layer request is sent to the protocol adaptation layer，this layer converts the request into a format that can be recognized by the driver to control the underlying sensors.In order to improve the system’s stability，the protocol adaptation layer adopts the de?sign of redundancy and overtime disconnection.

（2）Basic service layer

The basic service layer is essential to provide necessary support services for upper-layer applica?tions.The perfection of basic services premises the stable and efficient operation of upper-level applica?tions.In the edge gateway architecture，the basic service layer mainly includes four parts：Service reg?istration，service configuration，service tracking，and yellow pages.Specifically，as the first and pri?mary roles，service registration refers to registering the service at the edge gateway for subsequent task scheduling when the device is powered on or restart?ed after a breakdown.Service configuration com?pletes the standardized description of primary attri?butes such as processing capabilities and physical at?tributes.Service tracking realizes real-time feedback of current processing progress，equipment status，and other parameters.The yellow pages service pro?vides all resources connected to the gateway，and the device can be quickly found based on the IP ad?dress.

（3）Dynamic routing layer

The function of the dynamic routing layer is to realize information interactions within the edge gate?way.The purpose of the dynamic routing layer is to achieve edge-cloud collaboration.The dynamic rout?ing layer mainly includes two services.One is the in?formation interaction between devices in the gate?way，which can be directly communicated by query?ing the microservice of yellow page.The other is the information interaction between the edge gate?way and the cloud platform，which is implemented by the API gateway.

Furthermore，the API gateway is the entrance to access the entire edge gateway system.The cloud platform requests the API gateway，and then the API gateway transmits the request to the corre?sponding microservice.After the microservice re?sponds，the API gateway returns the result to the cloud platform finally.

（4）Analysis and decision-making layer

The analysis and decision-making layer is main?ly to run some intelligent edge gateway applications.Those operations such as data cleaning and feature extraction are realized by defining some commonly used data processing methods on the edge gateway.Resource allocation and scheduling optimization are the core part of applying computing tasks to the edge gateway.The scheduling optimization is achieved through the interactive negotiation of the dynamic routing layer in the gateway［14］.In the scheduling optimization，the best resource alloca?tion is determined through continuous information interaction between microservices.Through self-or?ganization and negotiation，this paper designs a global task scheduling decision mechanism at the edge gateway.The negotiation between the mi?croservices completes the dynamic scheduling of manufacturing resources.

2 Kernel Microservices of Edge Gateway

After completing the architecture design ac?cording to the functional attributes of the edge gate?way in cloud manufacturing，it is necessary to imple?ment the enabling technology of each layer.The goal of these techniques is to combine the microser?vices with the intelligent edge gateways.

The most critical role of edge gateway in cloud manufacturing is data transformation，ontology modeling，information interaction，and resource al?location.A manufacturing entity can generate a selforganizing virtual network resource in the edge gate?way through protocol conversion，ontology informa?tion model，and information interaction.This virtual manufacturing resource can realize the service encap?sulation and virtualization access of physical enti?ties［15］.

2.1 Data transmission microservice

In cloud manufacturing，the ubiquitous percep?tion of distributed manufacturing resources is the first issue to be solved.However，manufacturing re?sources cannot directly interact with each other due to the heterogeneity of communication protocols.Therefore，the protocol adaptation microservice adopts the adapter/agent mode to solve the multiprotocol adaptation problem.The adapter/agent mode configures special adapters for heterogeneous control systems to achieve resource information col?lection and standard format conversion.And then，the data transmission microservice releases these standard data through the agent to realize workshop entities’information interconnection and monitor?ing［16］.Finally，the adapter/agent program will be encapsulated into a protocol adaptation microser?vice，which runs in the edge gateway.The microser?vice of protocol adaptation is composed of the prima?ry function and various functional modules.These functional modules include general control，monitor?ing，database，and special modules such as NC code and AGV path planning.The model of data transmission microservice in the protocol adaptation layer is shown in Fig.2.

Fig.2 Model of data transmission microservice in protocol adaptation layer

（1）Main module

The main module of the data transmission mi?croservice contains the adapter’s overall working logic and exception handling mechanism.The main module is responsible for the connection，communi?cation，and management of the interfaces of the re?source adaptation model.In addition，it also moni?tors and responds to the manufacturing system inter?face.

（2）Control module

The control module in the data transmission mi?croservice contains some essential functions for con?necting and controlling the corresponding underlying entities.The control module can realize the critical functions in the traditional mode，such as program start and reset.

（3）Monitoring module

The monitoring module contains functions such as information collection of the underlying manufac?turing entity and status monitoring.Some of these information collection functions can also be obtained through the equipment manufacturer’s dynamic link library.

（4）Database module

The database module contains standard data?base operation functions，such as the connection be?tween the device adaptation model and the underly?ing device.In addition，the database module also has functional operations such as data adding，delet?ing，modifying，and querying.

（5）Other dedicated modules

With the exception of the three basic modules mentioned above，the data transmission microser?vices of the edge gateway also have other special?ized modules，such as NC code modules and AGV path modules.

2.2 Ontology information microservice

In order to achieve unified control of the under?lying physical manufacturing resources by the edge gateway，these manufacturing entities need to be modeled based on standardization.In this paper，the ontology-based resource modeling enables the un?derlying equipment to access the IoT manufacturing system and it is efficiently invoked by the upper-lay?er applications.It is necessary to abstract the under?lying equipment and establish its ontology model［17］.These devices can be controlled uniformly by estab?lishing a standard ontology information model.

The upper-layer application of edge gateway is concerned with the functional attributes of the under?lying equipment.Therefore，this paper takes the functional model of the underlying equipment as the main discussion object and ignores the knowledge of other fields such as structural models and behavioral models［18］.Finally，the ontology information mi?croservice establishes the ontology information of available equipment according to the functional attri?butes of manufacturing resources.

CNC machine tools are the most common man?ufacturing service providers as the lowest manufac?turing entity in cloud manufacturing.Although het?erogeneous CNC systems have different control log?ics，they all play the same role in the manufacturing system.The structure of an ontology-based informa?tion model of a simple CNC machine tool is shown in Fig.3.For CNC machine tools，the ontology in?formation model can be simplified as essential equip?ment with four functions：Workpiece processing，processing time estimation，fixture closing，and fix?ture opening.Among them，the processing time es?timation provides important decision-making param?eters for the resource allocation microservice.Each CNC ontology information model has the same ef?fect as the physical operation through control instruc?tions and feedback messages.Taking the interactive command of the access control model based on the ontology as an example，the interactive command consists of three parts，including an object，an ac?tion，and enabling information（if needed，F(xiàn)ig.3）.Each interactive instruction must return the execu?tion status and determines whether the execution is successful.Then，it can be used as a link to access specific manufacturing resources，such as executing machining programs（machining_workpiece），esti?mating makespan（evaluate_machining_time），and releasing and clamping workpieces（release_work?piece，grab_workpiece），etc.In addition，if a local manufacturing resource is activated，it is known that the edge gateway has requested the job to be ma?chined（run）from the first step until the end of all task steps.

Fig.3 Structure of an ontology-based information model of a simple CNC machine tool

2.3 Information interconnection microservice

In the intelligent edge gateway，the unification of the typical industry protocol is realized through the protocol adaptation layer.On this basis，dynam?ic manufacturing resources can realize real-time in?formation interaction between devices through the dynamic routing layer of the edge gateway.

The underlying device establishes a connection with the cloud platform through the edge gateway.All kinds of interaction information generated by the manufacturing resources are directly sent to the edge gateway.Therefore，the interactive information in the edge gateway mainly includes device-to-device and edge-to-cloud interaction information.The inter?action information contains the dialog initiator ID，the executor ID，and the information content.

For device-to-device communication，the edge gateway will complete the following steps after re?ceiving the interaction request.First，it is necessary to unpack the interaction information，obtain the ID of the conversation initiator，and record the instruc?tion number in the message content.Then，the in?formation interaction finds the IP address and port number in the yellow pages microservice through the receiver ID and forwards the message content to the receiver.

For edge-cloud information interaction，the edge gateway adopts a direct connection.The for?warding of crucial tasks no longer depends on the service terminal through the emerging network com?munication technology.Due to the near-field gain and single-hop gain between cloud edges，the endto-end interaction provides lower latency time and higher reliability communication services in the cel?lular network environment［17］.

The end-to-end information interaction in cloud manufacturing is shown in Fig.4.The cloud plat?form usually does not directly interact with the un?derlying physical entity in the actual execution pro?cess.Considering the requirements of security and other aspects，the cloud platform and the underlying resource need to use edge gateways to perform oper?ations such as message encryption and forwarding.

Fig.4 End-to-end communication in cloud manufacturing

2.4 Resource allocation microservice

Those resources can only passively receive scheduling decisions from the cloud platform as an independent individual in the traditional centralized manufacturing mode.It is a great challenge to cope with frequent production disturbances in an increas?ingly variable and large-scale personalized custom?ization environment.Therefore，the proposed edge gateway adopts the negotiation-based active schedul?ing method to complete the dynamic manufacturing resource allocation.With the help of IIoT，the phys?ical manufacturing entity generates a virtual network resource through the edge gateway（e.g.，protocol adaptation，ontology information modeling，and ser?vice registration）.The virtual network resource plays a crucial role in improving the interconnection of manufacturing equipment in the workshop and adapting the manufacturing process promptly.

Fig.5 shows a flowchart of resource allocation.There are two primary sources of decision-making information for virtual network resources.One is the physical manufacturing entity status obtained through the protocol adaptation layer.The other comes from device-to-device communication.That information is passed to the edge gateway analysis and decision-making layer.In detail，the informa?tion is classified and mainly divided into the process?ing task information related to the physical manufac?turing entity’s action information and the task infor?mation.Second，according to the information classi?fication results，the resource allocation microservice analyzes the processing status and workshop envi?ronment.Finally，the resource allocation microser?vices make scheduling optimization based on optimi?zation goals such as production cost and resource consumption.Thus，the coordinated allocation and dynamic scheduling of resources are realized，and the manufacturing efficiency is improved.

Fig.5 Negotiation based scheduling mechanism

3 Case Study

The information interaction of large-scale man?ufacturing resources has brought enormous network communication pressure in cloud manufacturing.During the interaction，different data schemas，in?terfaces，and communication protocols complicate these transfers.Therefore，this paper proposes the real-time ubiquitous perception and access of manu?facturing resources and finally implements an intelli?gent edge gateway based on the microservice archi?tecture.

Take the self-organizing IoT manufacturing system as an example.The prototype system and edge gateway architecture are shown in Fig.6.This production line has typical features of cloud manu?facturing，including service encapsulation and virtu?alization access of physical entities.We use the IoT manufacturing line to prove the effectiveness of the proposed edge gateway system.

Fig.6 Intelligent manufacturing prototype system

We have developed two modes of edge gate?way systems.The original one is developed with a service-oriented architecture，and the software sys?tem is particularly complex.Currently，the gateway is reconstructed with a microservice architecture，which significantly improves the system’s stability.

In the microservice-based edge gateway，each manufacturing entity will register a computing node in the pool online and inherit a series of microser?vices from the parent master node.Then，the map?ping of manufacturing entities to manufacturing ser?vices is completed through operations such as proto?col adaptation，service registration，and activation of heartbeat packets.

The node pool is similar to a virtual manufac?turing line，and each node corresponds to a manufac?turing entity.In contrast，that the nodes interact and communicate with each other completes the self-or?ganizing production of task orders through real-time information interaction［19-20］.

3.1 Prototype system

The basic workflow of the self-organizing IoT manufacturing system is as follows.First，con?sumes customized machined parts online.Then the order data is sent to the edge gateway system after resource allocation in the cloud platform.In our pro?totype system，all manufacturing resources are taskdriven and can complete tasks in a self-organizing negotiation manner.Due to the heterogeneity of the manufacturing element control system and the com?plexity of scheduling tasks，edge gateways in smart factories are vital.Those physical manufacturing en?tities such as CNC machine tools generate corre?sponding virtual network resources through the edge gateway.These virtual units run in the edge gate?way in microservices and finally realize negotiation and scheduling through information interaction.

The interaction between the manufacturing re?sources and the cloud is carried out in the dynamic routing layer of the edge gateway，which contains two parts of the horizontal device-to-device commu?nication and the vertical devices-to-cloud one.Parts of the information exchange process are as follows.

（1）Device-to-device communication：After each device is online，the cloud manufacturing sys?tem registers services，and the edge gateway as?signs an IP address and adds it to the Yellow Page service.During the manufacturing execution stage，the interaction initiator will access the gateway’s Yellow Page service to query the object entity’s IP address，package the information and send it through the gateway.

（2）Device-to-cloud communication：For the security of the industrial control network，the manu?facturing resource cannot directly access the cloud.The edge gateway encapsulates and encrypts the in?teraction information of the manufacturing entity and sends it to the cloud platform.

3.2 Improvement of microservice based edge gatewate

In the self-organized IoT manufacturing system built above，the relevant functions of the edge gate?way are verified.These functions mainly include the interconnection and interoperability between devices and the dynamic reconfiguration of the manufactur?ing system.By comparing the edge computing and microservice architecture models proposed in this paper with SOA，the improvement of the edge intel?ligent gateway in actual production is verified.The main features of the proposed edge gateway are sum?marized as follows.

（1）System scale

The advantage of microservice-based edge gateways is program scale.Compared with the origi?nal service-oriented architecture，the number of codes for the edge gateway of the microservice archi?tecture has been reduced from more than 50 000 lines to 10 453 lines.

（2）Communication latency

During end-to-end information interaction in the manufacturing system，communication latency has always been a concern.The proposed microser?vice-based edge gateway generates virtual network resources by protocol adaption and ontology infor?mation modeling.The information interaction be?tween devices is converted from the physical layer to the network layer interface.This kind of edge gateway based on microservice architecture has dra?matically improved network bandwidth，latency，re?dundancy，and security.

（3）Dynamic reconfiguration

Another improvement of microservice architec?ture compared with SOA is the efficiency of system reconstruction.In the traditional SOA architecture，the offline and re-online of a single device can easily cause a system breakdown.In contrast，the edge gateway based on microservice architecture can real?ize the dynamic management of manufacturing re?sources.

4 Conclusions

The information interaction of large-scale man?ufacturing resources has brought enormous network communication pressure in cloud manufacturing.During the information interaction，different data schemas，interfaces，and communication techniques complicate these transfers.Therefore，this paper proposes the real-time perception and access of man?ufacturing resources and finally implements an intel?ligent edge gateway based on the microservice archi?tecture.Without a large-scale upgrade of automation equipment，digitization can be effectively improved by adding edge gateways and edge data collection terminals.