Framework of industrial networking sensing system based on edge computing and artificial intelligence

Abstract

According to the large amount of and diversity of current industrial manufacturing data, on the basis of analyzed limitations of current industrial networking sensing system, a framework of industrial networking sensing system based on edge computing and artificial intelligence is put forward, including data-getting sensors group, first-level sensor routers, second-level sensor routers and backend severs. The above framework mainly is studied from three perspectives: structure level, workflow and key technologies, and the technology of artificial intelligence is the key to achieve this system with analysis of involved computational intelligence technology, information fusion technology and decision technology. This framework of industrial networking sensing system has important reference value on improving timeliness and certainty of industrial networking communications, largely reduces the entire power consumption and reduce stress of background sever.

Keywords

Industrial networking perceptual layer edge computing artificial intelligence

1 Introduction

The International Telecommunication Union (ITU) formally presented the word “Internet of Things, IoT” in November 2005 [1], which has aroused wide concern in the world, and is considered to be the third wave of the world information industry after the computer and Internet. At the same time, industrial networking technology combined with industrial automation technology and the IoT technology has become more and more valued by national research institutions and standardization organizations [2 –6]. Some important government policies for manufacturing, such as Germany’s “Industry 4.0”, USA’s “Industrial Internet”, the “Made in China 2025” plan, regard the manufacturing industry as the focus of development and the basis of strong power. With the information and industrialization deeply mixed as the main line, the relevant technologies with industrial IoT are becoming the focus of industry research [7 –9]. Industrial IoT means that kind of the network using intelligent terminals with perceptual capabilities, ubiquitous mobile computing patterns, and pervasive mobile network communications in all parts of the industrial production [10 –12].

Fig. 1

Industrial networking architecture based on perceptual data.

The architecture of Industrial IoT, the underlying technology of intelligent manufacturing and intelligent factory, mainly is divided into three levels: perceptual layer, middle layer and application layer [13], as shown in Fig. 1.

The perceptual layer consists of a terminal device and a gateway, such as various types of data acquisition and control module. This layer realizes the information collection and control of the application data via the terminal equipment, as well as information collection and networking controlling of terminal equipment signals via gateway [14]; the middle tier composed of communication networks and various application servers [15] mainly includes the collection and analysis of collected data. It realizes those functions, such as manufacturing process monitoring, real-time monitoring, processing quality diagnosis and MRO of the equipment. Based on the middle tier, the application layer provides the upper-level service of application developments for intelligent home, smart medical and intelligent factory to realize all kinds of applications’ expand in the converged network mode.

With the advent of the interconnection of all things, the data of different systems, devices and applications explodes. The data scale of current enterprise database has developed the level from TB to PB, and forms industrial large data [16]. The strategic significance of industrial large data is not only in how big scales of data the enterprises have, but also in how to analyze, interpret, apply the data reasonably and professionally process the data to realize the added value of the data [17, 18].

Faced with industrial large data, the mode of server background centralized processing of data patterns is currently taken. After a lot of original data collection is completed, the data is transmitted to the application service layer through the network layer. And the service layer integrates with the business logic and security policy to provide the function support for the user layer, after the data cleaning, standardization, preprocessing, storage and analysis of the received raw data are processed, and builds a variety of application servers, relational transactional databases and data warehouses for thematic analysis and decision analysis [19].

At the present stage, the industrial networking uses the cloud computing method of data processing to deal with industrial data. Although it can reduce the cost and increase efficiency of enterprises, it will also bring the following needs and challenges:

The data produced by intelligent equipment equipped with sensors will be huge and diverse, which makes the amount of information from the backstage system large, the data analysis and calculation difficult and the system robustness low.

The massive data transmitted from the field equipment to the background server increases the load of the transmission broadload, causes the network delay, and makes the feedback process of data decision not in time.

The sensor node device does not modularize the data processing and decision, which results in the low dimension of the database processing and decision of the background server.

To solve the above problems, this paper focuses on the perceptual layer of industrial networking, and puts forward a new architecture of industrial IoT sensing system based on edge computing and artificial intelligence technology in order to improve the timeliness and certainty of industrial networking communication, so as to improve the processing efficiency of industrial IoT and the value rate of industrial large data to reduce the pressure on the backend server.

2 The architecture of industrial IoT sensing system based on edge computing and artificial intelligence

2.1 Industrial IoT sensing system based on edge computing and artificial intelligence

In this paper, the industrial IoT sensing system based on edge computing and artificial intelligence refers to the introduction of edge computing model and artificial intelligence technology into the perceptual layer of industrial IoT in the present stage, then analysis and management of the data collected by the sensing group of the bottom sensor, which helps the actuator to make rapid decision It aims to establish the sensing system with independent wireless sensor network, and realize the optimization of the perceptual layer of industrial IoT. By adding the sensor router which includes the artificial intelligence algorithm, the sensing layer wireless sensor network has the ability of analysis and decision, and the field equipment is managed with the support of the high-dimensional and more scientific decision of the background server.

Optimization of industrial networking system is to achieve the following two functions through the first-level and second-level sensor routers distributed in the field, combined with the backend server. Firstly, the wireless sensor network which is independent of the perceptual layer can analyze and judge the data and make decisions in time. Secondly, that the background server receives intelligent analysis of the data results by the wireless sensor network can reduce the computational volume of background server, and improve the overall robustness of the industrial networking system. As shown in Fig. 2, it is the sensor router control system based on artificial intelligence algorithm model.

Fig. 2

Sensor router control system based on artificial intelligence algorithm model.

The sensor router in this paper is a kind of entity which can accept the data of the sensor in the environment and act on the environment through the actuator. It integrates self-calibration, self-diagnosis, self-learning, self-decision, adaptive and self-organizing artificial intelligence technologies, and is composed of information fusion module, environment evaluation module, action planning module and monitoring module, respectively, for local information processing and behavioural decision-making.

Information fusion module: for the synthesis of all kinds of environmental perception signals, through comparison and matching, to obtain certain credibility of the environment description.

Environmental assessment module: through a certain priori model, the description of environmental change characteristics is made from perceptual signals.

Action planning module: a corresponding action plan based on environmental assessment. Monitoring module: under normal circumstances to monitor the execution of the operation; encountering abnormal events to make corresponding processing commands.

Fig. 3

Architecture of Industrial IoT perceptual system based on edge computing and artificial intelligence.

2.2 Architecture of Industrial IoT sensing system based on edge computing and artificial intelligence

Based on the above ideas, perceptual layer architecture of industrial IoT based on edge computing and artificial intelligence was constructed. The system consists of four parts, namely sensor sensing Group, actuator group, first-level sensor router and second-level sensor router, as shown in Fig. 3.

The sensor sensing group consists of a field wireless sensor node device, which is responsible for collecting equipment data and transmitting the sensor data to the first-level sensing router. The first-level sensing router collects only the sensor data related to its decision-making process, performs data filtering and cleaning, feature extraction, module classification and decision making. If there is a direct related decision-making process with the actuator, then it sends the instruction directly to the executor to make it do the action. At the same time, the first-level sensor router, in accordance with the requirements of second-level sensor routers, reports the relevant data to the second-level sensor router. Second-level sensor router collects all the data reported by first-level sensor router for higher dimensional decision-making. If there is decision-making process directly correlated with the executor, then it sends instructions directly to the executor and let it to do the action. At the same time, second-level sensor router, in accordance with the requirements of the server background, reports the relevant data to the server background. Each hierarchy communicates through a ZigBee tree network.

The first-level sensor router classifies the collected data and filters the invalid data, and makes a simple calculation and analysis decision to reduce the network load. The second-level sensor router collects all the first-level sensor router’s modular data, and provides a scheme that requires a variety of scene modules (a first-level sensor router represents a scene module). This makes the decision-making dimension of the whole industrial IoT system higher.

Fig. 4

The process of industrial IoT sensing system based on edge computing and artificial intelligence.

2.3 Process analysis

The industrial networking sensing system flow mentioned in this paper is shown in Fig. 4.

(1) Acquisition of data: sensor sensing group collects data, and through wireless way sends to the first-level sensor router.

(2) Primary intelligence analysis decision: data cleaning and filtering, feature extraction, module classification and decision making judgment by first-level sensor router. If data is judged as abnormal data then it sends the relevant instructions to the executor and the executor produces action. The generated primary intelligence decision results and data are passed into the second-level sensor router.

(3) High-dimensional intelligent analysis decision: the higher dimensional intelligent analysis decision to the data by second-level sensor router. If judged as abnormal data, it sends the relevant instructions to the executor and the executor produces action. The resulting high dimensional intelligent decision results, data processing results, and unhandled data are sent to the server background.

3 Key technologies of industrial IoT sensing system based on edge computing and artificial intelligence

In this paper, the industrial networking sensing system includes the theories, methods and models of information, computer, automation, industrial management and intelligent decision making. This paper mainly talks the realization of the idea and method of the edge computing and artificial intelligence two key technologies involved in the field equipment sensor active sensing and sensor router acquisition, calculation, decision, transmission data to form the sensing layer of the unique wireless sensor network.

3.1 Edge computing

Edge computing [20, 21] is a new computational model of computing on the edge of the network. The network edge device has enough computing power to realize the local processing of the source data and send the results to the cloud computing center [22]. In this paper, the structure of the system is based on the edge computing model, on the basis of the existing server model as the core of the centralized large data processing technology, and complemented with the edge of large edge data processing technology taking edge computing model as the core, applies itself to the industrial networking system to solve the problem of the background system dealing with massive data. The first-level sensor router and the second-level sensor router are the nodes of edge computing, and the premise and foundation of the wireless sensor network that the field equipment relies on the perceptual layer, the purpose of which is to modularize the complex mass of data and to ensure the timeliness and reliability of the data. Edge computing enables the application and infrastructure of the factory itself to be built between the backend server and the sensor router to improve data transmission performance, ensure real-time processing, and reduce cloud computing load.

3.2 Artificial intelligence

In recent years, artificial intelligence technology has been widely concerned about how to deal with the massive data in time because of its intelligence characteristics which are not possessed by traditional methods, including cognitive computing, modular classification decision support and other technologies [23, 24]. The aim of artificial intelligence is to create intelligent machines with certain intelligence level on the basis of understanding natural intelligence (especially human intelligence), so that these artificial intelligence machines can have much better operation speed, work precision, work intensity and endurance than human beings [25 –27].

In the architecture of the system studied in this paper, it mainly puts a large number of artificial intelligence algorithm models background server used in processing large data on the environment of the sensor router [28 –30], so that the perceptual layer of industrial networking system through the distribution of the two-level sensor routers of in the field environment can achieve the classification, screening, extraction of data detected from the environment, and therefore the three functions of task prediction, anomaly detection and timely decision are realized by the data obtained. The key is to realize the technology of large data analysis of intelligent factory on the basis of acquisition of multiple sources of data, including computational intelligence technology, information fusion technology, decision technology and other artificial intelligence analysis technologies.

Fig. 5

Flow chart of BP algorithm.

(1) Computational intelligence technology

In this paper, the computational intelligence technology is defined as the application of artificial intelligence model and technology in the process of information processing. According to the definition, computational intelligence in artificial intelligence involves in a wide range of content. This paper mainly discusses the artificial neural network reverse propagation (Back Propagation, BP) model, proposed by D.Rumelhart [28] in 1988, which is used for forward multilayer inversion of the algorithm. This algorithm solves the problem of the connection right of the hidden unit layer in multilayer neuron network, and overcomes many problems which cannot be solved by the simple perceptron machine, which makes the BP model become one of the important models of neural network and be widely used. The process based on the B-P algorithm is shown in Fig. 5.

The first step: to preprocess the original data collected by sensor sensing group, select the training sample set, and take out one example from the sample set;

The second step: to set up the initial values of the weights and thresholds ω_ji (0) , θ_j (0) as the small random number;

The third step: to provide training sample, input vector: X_k (k = 1, 2, …, p); expected output: D_k (k = 1, 2, …, p);

An iterative calculation for each value through step fourth to step sixth below;

The fourth step: to compute the actual output of the network and the state of the hidden layer unit: $O_{kj} = f_{j} (\sum_{i} ω_{ji} O_{ki} + θ_{j})$ (2-1)

The fifth step: to calculate input and output layer training error: $Output layer : σ_{kj} = O_{kj} (1 - O_{kj}) (t_{kj} - O_{kj})$ (2-2) $Input layer : σ_{kj} = O_{kj} (1 - O_{kj}) \sum_{m} σ_{km} ω_{mj}$ (2-3)

The sixth step: to correct weights and thresholds: $ω_{ji} (t + 1) = ω_{ji} (t) + η σ_{j} O_{kj} + σ [ω_{ji} (t) - ω_{ji} (t - 1)]$ (2-4) $θ_{i} (t + 1) = θ_{j} (t) + η σ_{j} + α [θ_{i} (t) - θ_{i} (t - 1)]$ (2-5)

The seventh step: to judge whether the target satisfies the accuracy request e ⩽ ɛ when K experiences 1-p each time. Here, ɛ represents the precision;

The eighth step: End.

Fig. 6

General process of information fusion processing based on neural network.

Fig. 7

The basic structure of neural network ES.

(2) Information fusion technology

The fusion of sensor information refers to the use of information from different sensors to complete the description of a scene module. The key of it is to use certain methods to effectively deal with or infer the various obtained information. In this paper, the neural network technology is applied to multi-sensor information fusion. First of all, according to the requirements of the system and the characteristics of the sensor, the appropriate neural network model is to be selected, and also the mapping relationship between input and sensor information output and system decision is to be established. Then based on the existing sensing information and system decision, it studies the data, determine the allocation of power intelligence, complete network training, makes the well-trained neural network participate in the actual fusion process. As shown in Fig. 6, the obtained information passes through the appropriate process 1, as the input to the first-level sensor router, which utilizes the neural network to process the data and outputs the related results, and then interprets them as the system specific decision behavior through the process 2.

In this paper, the second-level sensor router is like a local server. Through the integration and fusion of the information obtained by a variety of first-level sensor routers, the second-level sensor router can complete data description and actuator decision-making more reliably, accurately and rapidly.

(3) Decision technology. The function of the system is finally expressed in the output of the behavior.

All the environmental perception and data analysis are for the executor to take appropriate action in time, so the decision technology of artificial intelligence is indispensable. As the decision-making process generally needs high-performance model, this paper mainly discusses neural network ES (Expert System, ES) expert system.

Neural network ES is a neural network with self-learning, adaptive, distributed storage, associative memory, parallel processing, as well as robustness and fault-tolerant characteristics to realize the ES function. By automatically acquiring the module input, organizing and storing the learning example provided by the experts, selecting the structure of the neural network, and calling the learning algorithm of the neural network, the knowledge acquisition is realized for the knowledge base. When the new learning instance is entered, the knowledge acquisition module automatically obtains the new network weight distribution through the learning of the new instance, thus updating the knowledge base. The basic structure of the neural network ES is shown in Fig. 7.

4 Conclusions

Industrial networking is moving towards the direction of “high precision, high efficiency and high reliability", which makes the collection of multi-source information produced in production process, the classification of modules, and dynamic monitoring, analysis, prediction and rapid decision-making of production process with the higher requirements. From the perspective of system implementation, this paper, based on edge computing and artificial intelligence, puts forward the architecture of industrial IoT sensing system, and probes into its key technology. The above mentioned method and technology have important reference value for the application in the field of IoT remanufacturing, which lays an important research foundation for the development of industrial IoT.

Footnotes

Acknowledgments

This work was sponsored by the innovation method project of Ministry of Science and Technology of China (2016IM020100), the Key R & D project of Zhejiang Province (No. 2018C01074), and 521 talent project of ZSTU. All these are gratefully appreciated.

References

International Telecommunication Union (ITU), ITU Internet Reports: The Internet of Things, Tunis: World Summit on the Information Society (WSIS), (2005), 2005.

Min

, Heng

and Hong

P.Q.

, Channel estimation method for industrial internet of Things based on deterministic scheduling, Journal of Chongqing University of Posts and Telecommunications (Natural Science Edition) 25(2) (2013), 154–160 (in Chinese).

Leigh

and Holly

, GE moves machines to the cloud, (2013), DOI: www.marketwatch.com/story/ge-moves-machines-to-the-cloud-2013-06-18.

Xiaojiang

, Jianli

and Mingdong

, Services and key technologies of the internet of things, ZTE Communications 8(2) (2010), 26–29.

Yoo

and Chang

, The IT convergence framework design in the internet of things environment, Eurasip Journal on Wireless Communications & Networking 2013(1) (2013), 53.

Gierej

, The framework of business model in the context of industrial internet of things, Procedia Engineering 182 (2017), 206–212.

Pushpavalli

and Kamarajapandiyan

, Scalable and protected distribution of delicate fitness proceedings, International Journal of Innovative Research in Computer and Communication Engineering (IJIRCCE) 6(4) (2018).

, Yuan

, Ma

, et al., Fast and parallel trust computing scheme based on big data analysis for collaboration cloud service, IEEE Transactions on Information Forensics & Security (2018), 1–1.

Hatzivasilis

, Fysarakis

, Soultatos

, et al., The industrial internet of things as an enabler for a circular economy Hy-LP: A novel IIoT protocol, evaluated on a Wind Park’s SDN/NFV-enabled 5G industrial network, Computer Communications (2018).

10.

Gubbi

, Buyya

, Marusic

, et al., Internet of things (IoT): A vision, architectural elements, and future directions, Future Generation Computer Systems 29(7) (2013), 1645–1660.

11.

Institute of Strategic and Regulatory Studies, Machinery Industry Information Institute. Implementation proposal of german industry 4.0 strategic plan (excerpt), World Manufacturing Technology and Equipment Market 3 (2014), 42–48 (in Chinese).

12.

Holdren

J.P.

, Power

, Tassey

, et al., A national strategic plan for advanced manufacturing, Washington, D.C., USA: US National Science and Technology Council, 2012.

13.

Youlong

and Jie

, Big-data-based technical framework of smart factory, Computer Integrated Manufacturing System 22(11) (2016), 2691–2697 (in Chinese).

14.

Sha

L.T.

, Xiao

, Sun

and Wang

R.C.

, A leakage perception method for backdoor privacy in industry Internet of things environment, Journal of Software (2018), (in Chinese). DOI: http://www.jos.org.cn/1000-9825/5356.htm.

15.

Nan

, Min

and Juanwei

, Framework for industrial internet of things oriented to steel continuous casting plant MRO, Computer Integrated Manufacturing System 17(2) (2011), 413–418 (in Chinese).

16.

Khare

and Khurana Prof. A., “Framework for trust management” cloud environments, International Journal of Innovative Research in Computer and Communication Engineering (IJIRCCE) 6(3) (2018).

17.

Floerkemeier

, Roduner

and Lampe

, RFID application development with the accada middleware platform, IEEE Systems Journal 1(2) (2007), 82–94.

18.

Dorksen

, Monks

and Lohweg

, Fast classification in industrial big data environments, Emerging Technology and Factory Automation IEEE (2014), 1–7.

19.

Pengfei

, Knowledge Big Data Integration and Application Research of Life Cycle Design Complex Product, Zhejiang University, 2016 (in Chinese).

20.

Jie

, Liang

, Wei

, Youlong

and Xinyu

, Big-data-driven operation analysis and decision-making methodology in intelligent Workshop, Computer Integrated Manufacturing System 22(5) (2016), 1220–1228 (in Chinese).

21.

Tan

K.H.

and Zhan

, Improving new product development using big data: A case study of an electronics company, R & D Management 47(4) (2016), 570–582.

22.

Liu

, Mao

and Zhang

, et al., Delay-optimal computation task scheduling for mobile-edge computing systems, IEEE International Symposium on Information Theory (ISIT) (2016), 1451–1455.

23.

Dhana Praveena

and Selvi

, A study on association rule mining algorithms used in web usage mining, International Journal of Innovative Research in Computer and Communication Engineering (IJIRCCE) 6(2) (2018).

24.

Shi

and Dustdar

, The promise of edge computing, Computer 49(5) (2016), 78–81.

25.

Weisong

, Hui

, Jie

, Quan

and Wei

, Edge computing-A emerging model for the internet of everything era, Journal of Computer Research and Development 54(5) (2017), 907–924 (in Chinese).

26.

Hutton

D.M.

, The quest for artificial intelligence: A history of ideas and achievements, The Quest for Artificial Intelligence: A History of Ideas and Achievements (2011), 1553–1553.

27.

Chrisley

and Chrisley

D.R.

, Artificial intelligence: Critical concepts, Intelligence 11(2) (2000), 15–16.

28.

Zhong

Y.X.

, Artificial intelligence: Concept, approach and opportunity, Chin Sci Bull 62 (2479), 2473–2479 (in Chinese).

29.

Paul

, Pinjari

, Hong

W.H.

, Seo

H.C.

and Rho

, Fog computing-based IoT for health monitoring system, Journal of Sensors (2018).

30.

Rumelhart

D.E.

, Hinton

G.E.

and Williams

R.J.

, Learning representations by back-propagating errors, Neurocomputing: Foundations of research. MIT Press, 1988, pp. 533–536.