Top-Down Human-Cyber-Physical Data Fusion Based on Reinforcement Learning

Abstract

With the development of industrial Internet and artificial intelligence, data fusion in crossdomains and cross-layers have become an inevitable trend. Most of the data fusion involved in the production process of hot rolling are concentrated on the level of sensors, Internet of Things (IoT) and the Internet; but human data are not well integrated. In order to avoid the human factor from becoming the bottleneck of the entire production schedule, this paper proposes a ternary data fusion model based on reinforcement learning algorithm. The related data source from human-cyber-physical space includes: social network, Internet and IoT. By merging the ternary data, a variety of data (including humans') can be quickly calculated to obtain better and faster decisions. In order to achieve automated fusion from ternary data, this paper proposes a method based on reinforcement learning: firstly, the domain ontology used for associating ternary data is reduced and tessellated (dimension reduction), and then the reinforcement learning model is used to form "the new ontology". Compared with resource-intensive global calculations (which may cost a few days), the new method can complete the calculations in minutes. This means that the new method optimizes the data source required for decision-making and improves the efficiency. Finally, the production scheduling of hot rolled steel is used as an example to verify the feasibility of the proposed method. INDEX TERMS human-cyber-physical data fusion, ternary data fusion, cyber-physical system, domain ontology, reinforcement learning

Data management in digital twins: a systematic literature review

“…As depicted in Figure 5.1, we additionally removed four studies (SHA; ZEADALLY, 2015;DAI et al, 2017;JIANG;CHEN;LIU, 2021;HÄNEL et al, 2021) that obtained a score inferior to five, which in our judgment, denotes irrelevance to our SLR. Thus, our final selection includes 57 studies that are aligned with the primary objective of our SLR.…”

“…Several works (PROPER; BORK; POELS, 2021; GÓMEZ-BERBÍS; AMESCUA-SECO, 2019;SAHLAB et al, 2021; AZZAM et al, 2019;JIRKOVSKY;OBITKO;MARIK, 2017) proposed an ontology, and the use of a RDF graph to relate the conceptual data model and domain data. Based on three ontologies,(CHEN et al, 2020) proposed a data model based on an ontological layer and a method addressing data fusion and entity resolution. We also found four studies (CHEVALLIER; FINANCE;…”

“…According to Table5.2 (column Domain), most selected studies address the domains of I4.0 (59.64%), followed by smart cities (24.56%) and healthcare (7.01%). Five studies (8.77%) do not detail the domain/subdomain, and one study is a domain-agnostic proof-of-concept (POC)(DAO et al, 2014).• ing hot lamination schedule(CHEN et al, 2020), among others. In the specific area of O&G, the DTs address asset management (AGARWAL; MCNEILL, 2019), improvement of drilling operations(ANDIA;ISRAEL, 2018;BRACKEL et al, 2018), and improvement of digital reservoir simulation (AL-ISMAEL; AL-TURKI;…”

Constructing Industrial Knowledge Graph Through Ontology and Link Prediction

“…For instance, a hybrid approach was built to gain an effective representation of sentence semantics and output the maximum probability sequence for chemicalnamed entity recognition [4]. A reinforcement learning approach was deployed to collect knowledge from the ternary space of humans, cyberspace and the Internet of Things (IoT) [21]. The interdependencies between different time series and the before-and-after relations of time series were mined using a CNN and LSTM, respectively [22].…”

Artificial intelligence and internet of things in manufacturing decision processes

“…RL, a branch of ML in which an agent learns to make sequential decisions through interactions with an environment, is considered in [44], [53], [54], [67], [77]. The research focused on the context of planning and control in the business plan and logistics industrial systems.…”