(345n) Data Driven Development of Industrial Symbiosis Networks Using Knowledge Graphs

Industrial Symbiosis (IS) establishes industrial networks (Isenmann and Chernykh, 2009) for economic, environmental, and social benefits, bringing together companies from all business sectors through material trading and sharing assets to add value. IS reduces costs for the participating industries and benefits the environment (Lehtoranta et al., 2011). Developing IS networks is a knowledge-intensive practice, where information is necessary to discover potential IS connections between flows of materials, waste and/or other resources related to industrial facilities. Therefore, the efficiency of ICT solutions for decision making in creating IS connections depends heavily on collecting, analysing, and integrating highly diverse quantitative and qualitative data (Grant et al., 2010).

The key objective of our work in the SYMBIOICT project (co-financed by the Operational Program Competitiveness, Entrepreneurship, and Innovation, under the call “RESEARCH-CREATE-INNOVATE”) is to overcome two major barriers for detecting large-scale IS networks: (a) focus on “1-1” connections, (b) lack of modelling and systemic use of existing knowledge.

We propose that knowledge graphs can be effectively used for modelling large and complex IS networks involving heterogeneous industries and a wide range of by-products exchanged. Using the knowledge graph, we accomplished an effective analysis of potential IS networks and synergies that would otherwise be extremely difficult to identify. These results were also visualized on a map, displaying “hot spots” of existing IS synergies as well as cases with high potential of forming new IS connections.

Our approach can be used for supporting decision making in the design process of zero waste Eco-industrial parks, or for identifying potential networks between industrial units at existing industrial parks. The graph can match waste to process, waste to company or company to company and can be further enriched as more technologies are developed or more practices are applied.

2. Background

2.1 Knowledge Graphs

A knowledge graph captures the semantics of a domain using a set of definitions of concepts, their properties, and the relations between them. Knowledge graph has become a term that is recently ubiquitously used yet does not have a well-established definition. Typically, a knowledge graph: (i) mainly describes real world entities and their interrelations, organized in a graph, (ii) defines possible classes and relations of entities in a schema, (iii) allows for potentially interrelating arbitrary entities with each other, and (iv) covers various topical domains (Paulheim, 2017).

2.2 Industrial Symbiosis facilitation

In EU countries, the promotion of IS is usually supported by teams of experts or practitioners that engage in networking sessions with industries and other stakeholders for developing IS projects (Artola et al., 2018). In these sessions, brokers usually apply straightforward techniques (without using ICT tools) to detect the possibility of resource (raw materials, waste, energy) exchange among the participating stakeholders. This approach can be restrictive, leading to serendipitous discovery of opportunities and therefore to IS networks of limited scale and complexity. The success of such efforts relies heavily on the skills and experience of each mediator and not on a systemic use of data and knowledge.

In recent years, a few digital services have been developed focusing on supporting decision making for waste valorisation and reduction. For example, International Synergies has developed SYNERGie, a proprietary ICT resource management database and platform, enabling organizations to identify resource reuse opportunities. Even though these digital solutions incorporate knowledge of best practices for waste valorisation, they often follow the same approach of searching for “1 on 1” connections. In this way they fail to take into consideration data on raw materials, waste and industrial processes, the analysis of which can highlight new possibilities for generating more complex and less obvious IS networks.

3. Methodology

3.1 Data Collection

The challenge of creating a knowledge graph for promoting IS connections, is that the necessary information is usually fragmented and located in many heterogeneous sources. Therefore, our first step was the extraction of all available data related to IS, before moving on to further processing. Data used for the development of the graph database were mainly mined from official repositories of IS best practices as well as research papers and news items. Our main sources were best practices implemented in Sweden: (industrialsymbiosis.se), Finland: (www.industrialsymbiosis.fi/home-en-gb/) and Denmark (Kalundborg Eco-industrial Park). The Nordic countries have a long tradition in sustainable development and thus they offer a breadth of available data and best practises, which could be replicated in other less circular countries. Other novel examples we took into consideration include Guitang group in China as well as IS networks established in Greece (symbiosisproject.eu), which proves that even in countries where circular economy initiatives are still at an early stage, IS networks can be developed, given that the necessary knowledge is available.

3.2 Knowledge Graph Model

By applying data analytics technologies, we produced a knowledge graph of existing and potential IS connections, matching input/output flows of industrial facilities. Nodes and edges (or relationships) were used to represent entities involved and their in between connections. In the proposed model nodes can have one of three different labels: Company, Material or Process which connect to each other with relationships (edges) having one of the following three labels: Is_Input, Has_Output or Has_Process. Each connection can be weighted using parameters such as the distance of the facilities (transport cost), the facility size and the cost savings from the proposed waste prevention.

4. Results

4.1 Implementation

In our approach we utilize the neo4j platform (neo4j.com) to create a graph database for storage, automation, and visualization of IS networks. Using the data collected we created a Knowledge Graph containing: 49 industrial facilities, 99 industrial processes, and 154 different materials. The resulting Knowledge Graph can suggest connections between companies, waste, and processing technologies.

Regarding nodes, three different labels were used: [a] Nodes with label: “Company” represent the geographic location where the various processes take place, [b] Nodes with label: “Material” represent materials in any state (solid, gaseous, liquid) and can be waste materials, raw materials, byproducts etc, [c] Nodes with label: “Process” describes actual industrial processes that transform raw materials into useful products, but also activities such as farming etc.

Regarding edges, three different labels were used: [a] Relationships with type: “Is_Input” relate Material nodes with Process nodes, [b] Relationships with type: “Has_Output” relate Process nodes with Material nodes, [c] Relationships with type: “Has_Process” relate Company nodes with Process nodes.

The node labeled Company is connected with the node labeled Process through the edge labeled Has_Process, whereas the node Process and Material can connect with edged labeled Is_Input or Has_Output depending on the direction. Both edges and nodes have attributes which vary depending on the label. Properties regarding materials can be physical, chemical, etc. whereas properties regarding processes can be financial, operational etc.

4.2 Query Example

Using our graph, we can perform a variety of queries for facilitating IS planning and networking, offering matches between (a) materials to process, (b) materials to company, (c) company to company. E.g., a question such as: "My facility produces a biowaste with the following characteristics: 18% lignin and 80% moisture. In which other processes is this material being used as an input?" is expressed as follows in neo4j:

match (n)-[:Is_Input]->(a:Process)

where n.Lignin=18 and n.Moisture=80

return a

union

match (n)-[:Is_Input]->(a:Upgrade)

where n.Lignin=18 and n.Moisture=80

return a

By executing the above query, neo4j presents a result set with processes that can utilize such a material (originally waste) as an input, based on the data of the Knowledge Graph.

4.3 Visualization of IS potential

Combining information extracted from the knowledge graph with data regarding industrial facilities in Greece (e.g., location, industrial NACE codes), we detected cases with high potential of forming new IS connections. We visualized these proposed IS connections on a digital platform, where industrial facilities are connected based on the potential exchange of materials and/or other resources. Each IS connection is weighted based on the distance between the industrial facilities involved combined with the size of the facility. Results are visualised on an interactive map, indicating proposed IS connections as well as “hot spots” with high potential of IS.

5. Conclusions

In conclusion, the developed model allows multi-faceted decision support for establishing IS networks, consisting of several different partners and synergies that would otherwise be extremely difficult to identify. In this way, a user of the knowledge graph can use even limited information as parameters to obtain suggestions on relative applied industrial symbiosis practices, proposed technologies that can process their waste, and generally identify possible IS opportunities. Finally, we demonstrate a digital platform that exploits IS knowledge graphs present IS hotspots and potential new connections.

Our future work will focus on the data collection process, which is not automated, as well as incorporating data regarding the economic value of the exchanges. Finally, at this stage, we do not provide narrow searches involving constraints in equipment and processes, or even involving company location information, but such type of data has already been collected and will be soon analysed to provide such type of searches.