Development of process safety knowledge graph: A Case study on delayed coking process

Abstract

Process safety is one of the essential preconditions for the achievement of green manufacturing. The improvement of process safety management requires a comprehensive risk analysis based on the collection of almost all safety related information, which are usually unstructured knowledge and experience. To handle the information and support the risk analysis, a process safety knowledge graph is prompted and the development of domain ontology on delayed coking process is elaborated. The combined top-down and bottom-up approaches are used in defining the process safety schema on ontology level. Several multi-structured data sources are introduced in establishing the process safety knowledge, in which the hazard and operability analysis (HAZOP) reports and process diagrams are most important. The ontology design and data extraction are demonstrated in the manuscript while various related applications are discussed. This process safety knowledge graph might empower the knowledge-based analysis abilities in discovering the hidden relationships between possible risk causes and consequences in an emergency situation, and could provide a foundation for more application related to process safety.

Introduction

As the basic industry of the national economy, in recent years, the chemical industry in China has developed rapidly. Due to stricter safety policies, environmental policies and public appeal all over the world, green manufacturing, which defines the environmentally-friendly operations and process safety (Qian, Zhong, Du, 2017, Yuan, Qin, Zhao, 2017, Zhou, Li, Zhou, Wang, Zang, Meng, 2018, Cernansky, 2015), has been increasingly attractive. The achievement of green manufacturing in the chemical industry is more urgent because of the frequent alarms in the production process and recent major accidents (Tauseef, Abbasi, Abbasi, 2011, Huang, Zhang, 2015, Wang, Wu, Reniers, Huang, Kang, Zhang, 2018). To improve the process safety (Crowl and Louvar, 2001), models of different production processes in the chemical industry are the foundation. However, due to the complex relations between different types of equipment, chemicals, process indexes, and accidents in the production processes of the chemical industry, it is extremely difficult to accurately model these production processes. Thus, though a few data-based methods have been proposed to approximate some practical production processes, the modeling of the process safety is still not fully achieved due to the incomplete application of knowledge and information in these processes (Zahedi, Lohi, Karami, 2009, Chen, Wang, 2010, Tian, Shen, Liu, 2012).

To improve the process safety, there are several system-based hazard analysis methods promisingly used to derive potential accident scenarios. such as Functional Resonance Analysis Method (FRAM) and Process Resilience Analysis Framework (PRAF). FRAM was first proposed by Hollnagel in 2012 (Hollnagel, 2012) and focuses on the understanding of interactions and emergence phenomena in complex sociotechnical systems. In recent year, FRAM has been applied in oil process units to identify emergent risks (Shirali et al., 2013). In order to incorporate both technical and social factors into a comprehensive risk analysis, Jain et al. have proposed PRAF. PRAF emphasizes dynamic, unpredictable or even unknown types of threats, uncertainty, system degradation and complex interactions. In Jain et al. (2019b) and Jain et al. (2019a), PRAF is applied in cooling tower operations and a batch reactor to improve risk and safety management, respectively. Although these two system-based hazard analysis methods can effectively improve risk and safety management, they all need require quantitative relationship data between various units in the process industry as support, and cannot characterize and use the empirical quantitative relationship in the actual process, which makes the modeling of the process incomplete. In addition, due to the inability in automatical construction, the implementation and subsequent improvement of these two methods require a lot of human resources.

As a new powerful tool of knowledge representation, knowledge graph was first proposed by Google in 2012 Steiner et al. (2012). Knowledge graph (Wang, Zhang, Feng, Chen, 2014, Jia, Qi, Shang, Jiang, Li, 2018) focuses on describing various entities and concepts in the real world. The knowledge graph can be regarded as a new semantic web, where the ontology is extended at the entity level. With this special form, numerous applications in different fields can be realized with knowledge graph technologies (Qiao et al., 2016), including the semantic search, intelligent question-answering, and intelligent recommendation. Fu et al. extract the public business transaction information of some chosen companies and summarize information of major transaction news on the Internet to build an enterprise information knowledge graph (Chen et al., 2018). Based on this developed knowledge graph, the statistical characteristics of financial laws are tried to be analyzed. Similarly, Bean et al. construct a knowledge graph containing four types of nodes: drugs, protein targets, indications, and adverse reactions (Bean et al., 2017). Based on this graph, the prediction for adverse reactions that were not observed during randomized trials is realized by a machine learning algorithm to improve patient safety. These successful applications in other industries indicate that the knowledge graph technology is a potential solution to promote the process safety by making full use of knowledge and information in the production process of the chemical industry.

A semi-automatic knowledge graph development solution for process safety in the chemical industry is designed in this paper. Each part of the entire development process can be run programmatically with manual supervision and modification. The main contributions of this paper are listed as follows:

• Different from FRAM and PRAF, a semi-automatic knowledge graph development solution for improved process safety is proposed in this paper, which can qualitatively and comprehensively model various processes in the form of semantic networks. This construction framework makes full use of the knowledge of process safety in the chemical industry and provides the foundation for improving process safety level. And it can be promoted to other production processes in the chemical industry.

• We construct a schema representing the process technologies and concepts in the chemical industry, which is the foundation of the knowledge graph. The main challenge of constructing this schema is how to properly define classes, relations, and properties to describe the process safety analysis process, while minimizing ambiguity and ensuring inference.

• To semi-automatically extract useful data from different data sources, such as the existing relational databases and text documents, different extraction solutions are designed based on natural language processing (NLP) technologies.

• Based on constructed knowledge graphs, several useful applications, including knowledge visualization, automatic information retrieval and extraction of causal chains, are realized in the case of delayed coking process.