An ontology-based methodology for hazard identification and causation analysis

Highlights

• A generic hazard identification model that can be used for most fire/explosion/toxicity scenarios.

• Ontology has been used to develop a graphical representation.

• Hazards are identified with probabilities of occurrence.

• Probability information from expert knowledge and historical data can be used.

Abstract

This article presents a dynamic hazard identification methodology founded on an ontology-based knowledge modeling framework coupled with probabilistic assessment. The objective is to develop an efficient and effective knowledge-based tool for process industries to screen hazards and conduct rapid risk estimation. The proposed generic model can translate an undesired process event (state of the process) into a graphical model, demonstrating potential pathways to the process event, linking causation to the transition of states. The Semantic web-based Web Ontology Language (OWL) is used to capture knowledge about unwanted process events. The resulting knowledge model is then transformed into Probabilistic-OWL (PR-OWL) based Multi-Entity Bayesian Network (MEBN). Upon queries, the MEBNs produce Situation Specific Bayesian Networks (SSBN) to identify hazards and their pathways along with probabilities. Two open-source software programs, Protégé and UnBBayes, are used. The developed model is validated against 45 industrial accidental events extracted from the U.S. Chemical Safety Board's (CSB) database. The model is further extended to conduct causality analysis.

Introduction

Hazard identification is the first and possibly the most important step in process risk assessment and management. Dynamic hazard identification is a concept that captures system variations with time and offers mechanisms to use updated process knowledge and information (Paltrinier et al., 2015). Methodologies for dynamic hazard identification include the Dynamic Procedure for Atypical Scenarios Identification (DyPASI) (Paltrinier et al., 2013), dynamic risk assessment (DRA) (Paltrinier et al., 2013), the bow-tie method in process hazard identification (Nakayama et al., 2016; Saud et al., 2014) and the risk barometer (Knegtering and Pasman, 2013). Applications of these approaches have been documented in the literature (Villa et al., 2016).

Tools for automated hazard identification are critical for industrial implementation. Computational tools e.g. Hazard Identification and Ranking (HIRA) (Khan and Abbasi, 1998) have been developed and applied to fire, explosion and toxic release scenarios. HAZOPExpert for chemical process systems (Venkatasubramanian and Vaidhyanathan, 1994) and software tool HAZID (McCoy et al., 1999) have been proposed. Blended Hazard Identification (BLHAZID) (Seligmann et al., 2012) is another automated technique that combines the function-goal-relationship with FMEA and FTA. Multilevel flow modeling with reasoning has been proposed for an automated HAZOP analysis in offshore installations (Wu et al., 2013b).

A hazard scenario initiates with one or more abnormal events and gradually evolves under the influence of operational and environmental conditions. The scenario-based model should provide the assessment of hazards along with hazard propagation pathways. HAZOP might be the most widely used scenario-based hazard identification method. However, traditional HAZOP has its limitations regarding textual definitions, which may not be standardized for computational purposes. Information sharing among different technical and management teams is a key to proper risk management; standardization of language and information is a prerequisite for effective information sharing. Standardization of hazard information is also the key to develop computer-aided hazard identification techniques.

This work presents a dynamic automated scenario-based hazard identification methodology with quantitative reasoning capabilities. The developed framework uses an ontology for knowledge modeling. The use of ontology facilitates standardization of hazard information and greatly enhances its reusability. Also, an ontological framework helps produce a systematic analysis of knowledge and facilitates its computerized processing and sharing between humans and computers. The generic nature of the developed tool allows its use across process industries and for a wide range of hazards. Incorporating the probabilistic analysis tool overcomes the qualitative nature of traditional hazard identification methods. Use of the Bayesian networks facilitates updating of the model using new process knowledge and data; this further enhances the applicability of the model throughout the life-cycle of a plant.

Ontology (Sánchez et al., 2007) is widely used as a basis for knowledge engineering to develop artificial intelligence and to share common understanding and information among people or software agents, enable the reuse of domain knowledge, make domain assumptions explicit, separate domain knowledge from the operational knowledge and analyze it (Noy and McGuinness, 2001). Ontology engineering has found uses in the fields of process design (Brandt et al., 2008), chemical process systems management (Wiesner et al., 2008), safety analysis (Daramola et al., 2011), HAZOP (Zhao et al., 2009) and operational risk management (Lykourentzou et al., 2011). It is effective for incident databases (Batres et al., 2009, 2014), semantic web based control (Elhdad et al., 2013), fault diagnosis based on anomaly detection (Hieb et al., 2009), and fault diagnosis (Bernaras et al., 1996; Pradeep et al., 2012; Zhou et al., 2015). It is valuable for failure mode effect analytical studies (Ebrahimipour et al., 2010), process control systems (Melik-Merkumians et al., 2010), fault diagnosis based on FMEA (Ebrahimipour and Yacout, 2015a), physical asset integrity management (Ebrahimipour and Yacout, 2015b) and hazard identification in construction safety (Zhang et al., 2015). There is an example of ontology based process hazard evaluation (Wu et al., 2013a), which is more qualitative in nature; whereas the current work utilizes a different methodology and quantitative approach.

To share a common ontology-based platform, Web Ontology Language (OWL) was developed by the World Wide Web Consortium (W3C), which is designed for machine interpret-ability of information, instead of only presenting it to humans (McGuinness et al., 2004). Probabilistic Ontology (PR-OWL) adds an extension to OWL, incorporating uncertainty using Bayesian statistics (Costa et al., 2005; Da Costaet al., 2008; Fenz et al., 2009) and has been used effectively in application with uncertainty associated complex intelligent systems (Paulo C G Costa et al., 2006; Laskey et al., 2010). Multi-Entity Bayesian Network (MEBN) utilizes first-order Bayesian logic in the PR-OWL environment. Similar to Bayesian Networks, MEBN theories use directed graphs to specify joint probability distributions for a collection of related random variables (Laskey, 2008). MEBN theories represent knowledge as a collection of MEBN Fragments (MFrags) which contains uncertainty information (Carvalho et al., 2009).

OWL is used in this work to capture process knowledge to develop a knowledge model. Hazard identification is one of the uses of the model. OWL allows modeling knowledge in usable form which can be further used for other purposes as well. OWL provides the ease of use through open-source software, without extensive knowledge on programing. In our consideration, OWL might be an appropriate platform for process engineers who can easily use the model, tailor it to their needs and use for hazard identification.

An alternative platform that could be used for knowledge modeling is the Unified Modeling Language (UML). The differences and comparative advantages of UML and OWL have been the subject of extensive research in the AI community. We, as users of the tools, considered UML and OWL as two different examples of system AI development. In this work, the objective is hazard identification and in this context, the use of OWL is limited to knowledge representation. However, a long term objective is to use the developed model for other purposes related to risk assessment. From a user perspective, the authors decided to utilize OWL for few distinct reasons: (1) OWL has greater interpretability over UML; (2) OWL based PR-OWL incorporate the probabilistic reasoning; (3) OWL provides the versatility to utilize GUI tools (e.g. Protégé) without extensive programing; and (4) development opportunities in future by easier modification.

The proposed dynamic hazard scenario with a step-by-step procedure to develop an ontology-based model is described in Sec. 2, followed by case studies with results in Sec. 3, the discussion in Sec. 4 and conclusions in Sec. 5. Additional information regarding the accident scenarios is included in the appendices.