Construction of a base ontology to represent accident expertise knowledge

Expertise is an activity carried out by experts that contributes to societal progress, as it helps to elucidate unknown situations. For example, accident expertise eases accident understanding by describing how it happened and by identifying its causes and consequences. As a result, the design of accident expertise in a convenient human–machine structure will enable the querying, reasoning, and reuse of accident knowledge in other tools, such as safety and decision-making systems. However, existing representations of accident knowledge, such as documents, relational databases, or accident ontologies, do not fulfill accident expertise expectations. Moreover, these representations are unlikely to provide the appropriate use of accident expertise knowledge. This study presents a base ontology for accident expertise knowledge representation designed with a model-driven methodology and implemented with semantic web technologies. The study obtained satisfactory results from the evaluation and application of extension and reuse of this ontology with aircraft accident expertise taken from the French bureau of Enquiries and Analysis (BEA) for civil aviation safety.


Introduction
The word "expertise" has two semantic meanings, which can lead to misunderstanding. On the one hand, it could mean "specific tacit knowledge possessed by an expert" Karhu (2002) and could be used to interpret unknown and unexpected problems. Authors such as Wieten (2018), in the same vein, emphasize the fact that when working in a domain, interactions allow the acquisition of expertise. According to these authors, knowledge obtained from books or education is not expertise. In this case, experience is similar to expertise perceived as know-how acquired over time by someone, thus making them an expert (Roventa and Spircu 2009).
On the other hand, expertise could mean "a task carried out methodically by experts to elucidate problems" Farrington- Darby and Wilson (2006). This view brings expertise closer to a specific process individuals or companies use to unlock misunderstandings and depict problems. Figure 1 shows that experts use their expertise (experience) during an expertise activity (process) to bring comprehension to a problem. The outcome of this expertise is later used to make decisions, design safety and feedback systems, or solve problems.
To summarize, expertise can refer to someone's skill acquired over time by studying or carrying out an activity in a domain. It can also mean a process followed to understand a problem.
This study is about the representation of knowledge emanating from expertise, a cognitive activity used to bring interpretation and comprehension to accident problems. However, this process does not exclude the use of experience as the knowledge that experts possess (Navarro and Colbach 2020).
Henceforth, the word "expertise" refers to a process, and precision will be given when referring to knowledge or skills.
In this study, the words task, activity, and process will be used interchangeably to mean a step-by-step method followed by experts to understand a problem.

Motivations
What motivated the proposed study is, first and foremost, the importance of expertise in companies and society acknowledged by French and European standard document NF X50-110 CNS EN 16775 standard "Expertise activities-General requirements for expertise services". These documents set norms on how expertise should be carried out and the roles of different actors involved in the process.
Second, our study was motivated by the lack of appropriate representation of expertise knowledge and the benefit of making it accessible and reusable by humans and machines. Undoubtedly, expertise knowledge helps design safety measures and learn lessons to reduce or avoid risk in the case of accidents. Furthermore, this knowledge is exploited in understanding new problems; therefore, it helps to reduce the burden of this knowledge-intensive task and, as a result, speeds up expertise processes.
Third, enhancing communication among the stakeholders involved in the expertise process would be significant. Figure 2 shows that expertise involves stakeholders such as financial backers, project sponsors, clients, and experts.
Besides this, experts from various fields who must work together because problems are multi-facet or need different domain knowledge face difficulties understanding each order.
Finally, it is worth using expertise knowledge when designing awareness systems to avoid accidents. In addition, it will be rewarding if these tools can also benefit from expertise from different fields.
The contribution of this study is on accident expertise, in particular for the following reasons. First, expertise is a vast field with numerous categories, such as accident, risk, and safety. Each category can be further divided into subcategories. For example, the accident domain can be subdivided into aircraft, automobiles, and housing. As a result, collating all the main categories can be a laborious task and can lead to a meaningless result because each category differs from its counterpart in knowledge, vocabulary, and objectives.
Second, there will be a focus on accident expertise because accidents cause significant material and human loss. For example, a mutual insurance company in Chile registered 625,050 work accidents from 2015 to 2019 with thousands of deaths and disabilities (Bravo et al. 2022), and the World Health Organization (WHO) counted about 1.25 Million deaths in road accidents (Baskara et al. 2019). Another alarming statistic that motivated the choice of accident expertise is the high number of death. In Canada, for example, almost 3000 people are killed yearly in road traffic accidents (Wang and Wang 2011). As accidents are not rare, it will be essential to model accident expertise to make sharing or reusing by humans and systems more effortless. Proper reuse of past accident expertise knowledge can help save time, money, and, most importantly, human lives.
Third, because accident expertise involves interaction between experts from various fields, there is a need for common understanding. For example, in aircraft accident expertise, experts from aviation, weather, safety, and mechanical engineering work together to understand what happened. This aspect calls for the need for a common vocabulary and reasoning mechanism so that experts can understand themselves.
Finally, accident expertise is a field in need of support because it is a knowledge-intensive activity that must be implemented when there is no clear understanding or knowledge to answer questions about accidents. Nevertheless, their outcomes help support new expertise, building safety systems, learning lessons, or making decisions.

Contribution and methodology
This study describes the design and implementation of an upper-level ontology for accident expertise with strengths such as concise vocabulary, precise semantics, and reasoning (Panagiotopoulos et al. 2012). As a result, it aims to build a To ease the BAEO integration and reuse, the unified modeling language (UML) and the model-driven architecture (MDA) were used since they are appreciated in systems designing for their high level of detail representation (De Lope et al. 2021;Malgouyres and Motet 2006). MDA is an Object Management Group (OMG) standard that uses abstract views called models at four (04) different layers of representation to design systems. The most abstract layer is the meta-meta-model, followed by the meta-model, the third layer is the model, and finally, the instance of the model (Paolone et al. 2020). Figure 3 shows the different layers of the MDA architecture.
The meta-meta model is self-defined and describes the meta-model, while the meta-model is used to express a valid model.

Ontology
Ontology can be defined as a formal and explicit artifact used to surrogate a domain knowledge (Fionda and Pirrò 2019;Zhong et al. 2015;Gruber 1993). They are classified into three major groups, as shown in Fig. 4. This classification is based on how they map or abstract domain concepts and the relationship between these concepts.
From top to bottom, in the first place, there are top-level ontologies, also known as upper-level ontologies, which describe primitive concepts such as entities, processes, time, and objects. These upper-level ontologies are made to be extended for specific domains and stand as a base for interoperability. Some examples of these ontologies are the unified foundational ontology (UFO) Guizzardi et al. (2021), the basic formal ontology (BFO), the general formal ontology (GFO), the DOLCE (a Descriptive Ontology for Linguistic and Cognitive Engineering), Cyc and the suggested upper merge ontology (SUMO) Mascardi et al. (2007).
The class in the middle corresponds to the domain and task ontologies. This class of ontologies contains concepts that belong to a particular field and guarantee their reuse in these fields.
Finally, application ontologies correspond to a more specific domain than the previous class.
Furthermore, while domain ontologies capture domain knowledge with a formal and consensual vocabulary (Jean et al. 2007), application ontologies describe components of a domain application and provide semantics to map their integration (Naubourg etal. 2011;Guergour et al. 2006).
The proposed base accident expertise ontology (BAEO) presented in this study belongs to the second layer of the ontology hierarchy. In subsequent sections, we will show its construction, illustrate its utilization from a case of accident expertise taken from the BEA online database, and demonstrate its reuse.
The next section of the work Sect. 2 presents studies of ontologies related to accident and expertise. After this, Sect. 3 describes the design and implementation of the proposed base accident expertise ontology (BAEO), followed by an illustration and reuse. The last section before the conclusion is the discussion Sect. 5, where the gaps covered by the BAEO are highlighted, and its strengths compared to existing similar ontologies are presented.

Related works
This section presents relevant studies in domains that bridge both ontology and accident expertise. In the domain of accident scenarios, Maalel et al. (2012) proposed an ontology to acquire expert knowledge in analysis and safety assessment processes in railroad accidents and incidents. The ontology includes essential knowledge such as contexts, events, and causes but is strictly limited to the domain of 1 3 railroad accidents. The conceptualization proposed by these authors helps in modeling scenarios that describe actions and consultancies that lead to dangerous situations. Another attempt to model expertise knowledge is the study of safety presented by Kaindl et al. (2016). These authors used an engineering ontology-building approach, ISO 26262 and EN 50126 standard concepts from the railway and automobile industry to construct ontology for safety. Motivated by the harmonization of safety assessment, they provide a taxonomy containing concepts such as risks, harms, and hazards from the railway domain. This ontology cannot capture accident knowledge such as causes, consequences, or events. To make aviation accident reporting easier within the European Coordination Centre for Accident and Incident Reporting Systems (ECCAIRS), Křemen et al. (2017) used the terminology of an existing accident reporting information system to construct an ontology for occurrences, events, factors, and aircraft descriptions. The constructed ontology uses unified foundational ontology (UFO) as its base. Because it was intended for reporting, the ontology does not represent the knowledge of cause and consequences, as a result of this limiting its utilization for accident expertise.
In road accidents, tools such as Mivar Expert System (MES) analyze road accidents and determine optimal parameter values for accident simulations. MES reduces experts' difficulties in reconstructing vehicle accidents and accelerates decision-making (Chuvikov et al. 2019). Nevertheless, the final artifact produced by experts still needs to be explored and considered by the Mivar system.
Similarly, a study from Wu et al. (2020) used ontology in transport. The authors coupled ontology and natural language processing (NLP) techniques to describe subway accidents. This description allowed them to retrieve similar cases of accidents to support decision-making when faced with new cases. Ontology built using this approach formalized the semantics of unstructured and semi-structured documents related to subway accidents and regulations, which are not appropriate for our study.
Another study in the transport sector is the work of  that used accident information and data from the General Estimates System (GES) to design an ontology for vehicle accidents. The ontology allows the sharing, integrating, and reusing of knowledge about vehicles involved in road accidents. Likewise, for an interoperability solution,  proposed an ontology for car accidents through Vehicular Ad hoc Networks (VANETs). Their study defines a shared understanding of a car accident environment such that it can be shared with other vehicles. These ontologies are based on four main concepts: vehicle, accident, occupant, and environment. Furthermore, the study of Vanderhaegen (2021) established a heuristic-based approach using reasoning principles such as induction, abduction, deduction, and counterfact to identify conflicts between humans and autonomous systems in a shared workplace. Even though conflicts can lead to accidents, the proposed study does not question the system's autonomy or the agent that caused the accident. However, it is dedicated to representing knowledge from the cognitive task of accident expertise.
The study of Skalle et al. (2014) designed a model based on the human and technical error to determine offshore accident causes. The authors used case-based reasoning and ontology to represent human organizational and technical errors to spot possible errors that can lead to offshore accidents. Unlike the proposed study, the approach uses a specific domain ontology to diagnose failures in drilling processes while relying on existing cases. Moreover, the study does not focus on accident expertise knowledge representation. The study of emergency healthcare in complex situations such as flight was addressed by Sene et al. (2018). The authors designed a knowledge system by integrating and extending the Systematic Nomenclature Medical Clinical Terms (SNOMED CT) to provide knowledge and advice to isolated medical persons working in an environment with limited resources. Even though this study constructed an ontology to support medical emergencies during flight, the knowledge system aims to propose solutions to medical problems rather than to capture expertise.
In summary, ontologies developed in the above literature have drawbacks for accident expertise despite their contributions to accident-related fields. On the one hand, because these ontologies are domain-specific, it is difficult to use them for different accident expertise knowledge. On the other hand, they only partially capture aspects of accident expertise, such as cause-effects knowledge for some and event for others.
This study overcomes the above limits using an ontology for accident expertise. Furthermore, it focuses on a base ontology to facilitate the construction and integration of specific domain accident expertise knowledge.

Ontology for accident expertise
We learned the shortcomings of existing ontologies for accident expertise from the previous sections. The main problem is the need for a dedicated ontology for accident expertise knowledge. However, this knowledge is essential for constructing and integrating specific fields of accident expertise, such as railways, aircraft, automobiles, and building construction. This goal will be achieved with the construction of a high-level ontology that can be used to represent accident expertise knowledge from multiple areas. Figure 5 shows the overview architecture intended to be achieved in this study. BAEO will stand as a domain ontology that can be extended for specific domains of accident expertise.
This study relies on a manual approach to ontology construction because, on the one hand, automated techniques can produce redundant and inconsistent conceptualization or even lack semantics due to noisy data (Hur et al. 2021). On the other hand, it is challenging to access all areas of expertise, and if such were the case, it could lead to the construction of isolated ontologies for each area which would be even more challenging.
For this reason, this study will rely on a manual middleout approach for conceptualizing, starting with essential concepts and moving toward a high-level conceptualization. This approach will allow the following (1) reduce the granularity since the goal is to construct a high-level (domain) ontology, (2) avoid inconsistencies due to low-level details, (3) have better control over concepts, (4) have more stable models, and (5) have fewer reworks and efforts (Uschold and Gruninger 1996).
The methodology for ontology construction employed in this study utilizes approaches described by Martínez-García et al. (2020) and Hassan and Mokhtar (2021). The following mechanisms were added to the fundamental steps proposed by these authors: First, an iterative and cyclic process, under which basic steps of the overall methods were defined. Second, the unified modeling language (UML) profile and model-driven architecture as tools for the conceptualization were adopted. These tools will enhance integration from specific areas of accident expertise and increase understanding or usability. Finally, semantic web technology formalizes the ontology and makes it machine-readable.
The UML profile serves as a bridge between UML and OWL, allowing the semantic web to use UML for designing. The UML profile is an extension of classic UML with mechanisms such as stereotypes, tag definition, or tagged values, which makes it an ideal tool for designing ontologies (Djurić et al. 2004) and for which there are rules to transform their representations to OWL (Vo and Hoang 2020). Table 1 below shows the main differences between UML and OWL (Jetlund et al. 2019).
The methodology employed for this study is cyclic and iterative. The following detail describes its phases.  (1) the scope and specifications of the ontology and (2) competency questions that the knowledge from the ontology will be able to answer. 2. Conceptualization: For this phase, the following tasks are to be carried out: • The description of the ontology concepts. • The description of these concepts' relationships. • The ontology design using UML classes.
• Rules and constraints expression.
At this phase, any decision to reuse existing ontologies is made. 3. Formalization: One can easily translate the conceptual model into a formal language from the previous phase.
(1) For example, the designed ontology can be translated into Web Ontology Language (OWL), and (2) the rules in Semantic Web Rule Language (SWRL). SWRL supports OWL for reasoning from an ontology because it can infer knowledge from rules constructed from OWL individuals, classes, properties, and specific built-ins that OWL can not provide. For example, it is possible to build rules with some arithmetic operators with SWRL, while this is not possible with OWL. Tools such as protégé (Musen and Team 2013) are utilized at this step. 4. Evaluation: Ontology evaluation has two main methods, which are validation and verification. They consist in checking, on the one hand, if the ontology structure is coherent and was designed correctly and, on the other hand, if the ontology maps well the real world for which it was designed, respects its characteristics, constraints,  Figure 6 shows the cycle of the methodology for ontology construction used in this study. The construction process iterates on this cycle to produce the desired ontology. This iterative approach helps to refine the ontology as it is constructed.
The sections below illustrate how the BAEO was designed using the abovementioned approach.

Scope and specification
This study aims to construct a base ontology to represent accident expertise knowledge. This ontology will be reused in various fields of accident expertise, such as aircraft, automobiles, and railways.
As a result, the proposed ontology must capture highlevel knowledge common to accident expertise.
Competency questions for the proposed ontology are the following.
• How did the accident occur? In accident expertise reports, experts always try to describe the sequence of events that precede the accident. Knowing how the accident occurred can help define safety measures. • When did the accident occur? The time the accident occurred allows experts to verify events and conditions, such as weather, that could have provoked the accident. • Where did the accident occur? The place of the accident is used to define the environment in which the accident happened and can be exploited to identify other agents involved in the accident. • Which stakeholders are involved in this accident? Stakeholders are organizations or groups that are involved in the accident. • How many victims were in the accident? Victims are agents that suffer from an accident, and knowing the number aids in evaluating the damages of the accident. • What were the causes of the accident? This question aims at identifying everything that triggers the accident. Pre-sented in the reports, it can be used to prevent similar accidents. • What is the list of victims of the accident? Victims are agents that suffer from an accident. They can be humans or non-humans agent. • Who are potential witnesses of the accident? Witnesses are those who saw or were present when the accident occurred. They can testify of the sequence of events they assisted. • What are the consequences? Reported consequences are damages provoked by accidents. This information can be used to reimburse victims or stakeholders. • Which vehicle/equipment/asset/part is involved in the accident? This question concerns assets damaged during the accident. It helps to evaluate the accident.

Conceptualization
We can identify the following knowledge containers from these competency questions and accident expertise reports: context, activity, cause, and consequence. These knowledge containers cover essential knowledge of accident expertise, without which it will be considered as incomplete.
For ontology designing, this work uses the object management group (OMG) MOF (meta-model object facility), which offers an independent platform framework and facilitates ontology interoperability concerning the model-driven architecture (MDA) MOF (2015). UML offers diagrams, extensions, and an Object Constraint Language (OCL), that makes it flexible and suitable for modeling ontologies' class/ subclass hierarchy, relations, and axioms (Kogut et al. 2002).
The knowledge containers identified earlier are considered the main concepts of accident expertise. They form the building blocks of the proposed base accident expertise ontology (BAEO) taxonomy completed with additional sub-concepts.  In other words, a context presents an environment where an accident occurred. For this study, context is designed from concepts found in the work of Cabrera et al. (2019). This concept includes the sub-classes: location, agent, resource, and environment.
-Agent: Corresponds to an actor actively or passively participating in an accident. -Resource: Its instances are any tool involved in an accident. Resource instances are domain-dependent, and specifying them will vary from one accident expertise field to another. For example, one may have resources such as helicopters or airplanes in an aircraft accident expertise, vehicles in a road acci-dent expertise, and cranes in construction accident expertise. -Location: instances of the location concept identify the place of an accident. -Environment: Instances of this concept are resources that are not specific to the domain of an accident, such as natural or physical phenomena.
• Activity It describes anything that happened in the context of an accident under expertise at a specific length or interval of time. This concept adds dynamic and temporal characteristics to the represented expertise. An activity can have a beginning and end event. • Consequence This concept describes the outcome and damages yielded by an entity in the accident context. In general, these consequences can involve material damage, including human death. Sub-classes under consequences are disruption, destruction, loss, harm, and failure. These classes of consequences are divided into two groups which are material and living consequences. • Cause These elements are identified as triggers of the consequences of an accident. They are the factors of the accident and are grouped into sub-classes: humancause for human causes, naturalcause for natural causes like wind, fog, earthquake, or weather conditions, to name a few, and systemcause for system causes.
Figure7 shows the above-named main concepts of the BAEO and the relations that exist between them.
Rule: An expertise is completes only if these five knowledge containers are not empty; otherwise, it is considered incomplete.
Rule: The number of victims equals the number of agents affected by the accident consequences.
Constraint: An accident expertise must have at least a consequence.
Constraint: An event must have a specific time or period. Constraint: An activity must have a beginning and ending event or duration.
Constraint: An accident expertise must have a context. The proposed modeling approach is based on modeldriven architecture (MDA) and UML, as shown in Fig. 8. This approach uses the UML profile to adapt UML for ontology designing.
The first layer (M3) comprises the meta-object facility, a language and self-defining framework derived from the core UML. This language contains basic concepts such as class, association, or data type for describing other meta-models (Gaševic et al. 2009).
The second layer (M2) of this modeling architecture is based on the ontology definition meta-model (ODM), in which there is the OWL meta-model and the RDFS metamodel (Odm 2007).
The third layer (M1) comprises the ontology model built from components of M2. This layer is followed by the last layer (M0), which is the real-world representation itself.
In terms of reuse, as shown in Fig. 9 this study uses the time ontology to assign temporal instances to event or activity entities of the BAEO.
For further hierarchy levels, Fig. 10 presents the consequence class hierarchy, Fig. 11 the context class hierarchy, and Fig. 12 the cause class hierarchy.

BAEO formalization
Formalization is transforming a model diagram into a machine-readable and reasonable format.
Ontology languages offer constructs to formalize models. However, each language has its level of expressivity and limitations.
Examples of ontology languages are RDF, RDFS, OWL 1, and OWL 2.
Under the proposed methodology, a subset of web ontology language (OWL) called OWL 2 Description Logic (OWL 2 DL) was adopted for its expressiveness and decidability. This language is built from specific components of the Description Logic (DL), is oriented toward expressive ontologies, and offers semantics, interoperability, and reasoning benefits. OWL 2 DL provides constructs for ontology formalization such as classes and subclasses, property hierarchies, property chain, object-properties for defining relationships between individuals, data-properties for their value properties, and inverse of properties (Gayo et al. 2017;Horrocks 2005). The advantage offered by OWL 2 DL can not be achieved by its counterparts because it provides other vocabularies that they do not possess.
For reasoning, this work relies on Pellet reasoner because it is free and open-source software that supports the SROIQ DL as OWL 2 DL. In addition, it is optimized for standard DL reasoning (Bock et al. 2008).
The following formalization in OWL 2 DL is obtained from the design presented in the previous sections.
Code 1 encodes the main concepts of the BAEO, which are OWL classes as declared with the RDF type property. Code 2 encodes the consequence class hierarchy of the BAEO. As shown in Fig. 10 the consequence class has two sub-classes: the living and material consequences. These classes, in return, have sub-classes, too: Injury, fatality, failure, disruption, and destruction. Code 3 encodes the cause class hierarchy of the BAEO and specifies its sub-classes: the system, natural and human causes. Code 4 encodes the context class hierarchy of the BAEO. The first level of sub-classes comprises resource, location, environment, and agent. The agent class has two sub-classes which are the living and non-living agents. In terms of rules, Code 5 shows the rule from which one can infer the number of victims.

Property chains
Property chains are mechanisms provided by OWL 2 DL to infer new knowledge from chains of properties in the ontology.
From the given design, the following chains were covered.  Figure 13 shows the agent-activity-cause chain. In other words, if an agent partakes in an activity that induces a cause, then the cause is said to have been done by the agent. This chain identifies the agent at the root of a cause. Figure 14 shows the activity-cause-consequence chain. This chain means that activity through a cause can provoke consequences. This chain links a consequence to the activity that produces it. Figure 15 shows the agent-cause-consequence chain, which means an agent, through its causes, can be at the origin of consequences. This chain links an agent to the consequence it causes. Figure 16 shows the cause-activity-consequence chain. This chain relies on the chain in Fig. 14 to show that a cause can yield an activity that will provoke consequences. This chain links a cause to a consequence it provokes through an activity.

BAEO evaluation
For this study, the following evaluation and validation were carried out: • The external evaluation consists in querying the knowledge stored by the ontology regarding the competency questions. • The internal evaluation checks the ontology consistency with a reasoner. We used Pellet reasoning in this study (Jain 2021). • The validation was done with the SHApe Constraint Language (SHACL).

Validation
The ontology hierarchical structure and relationships were verified using the PROTEGE 1 built-in interaction debugging tool DEBUGGER, which checks the coherence and consistency of ontologies using PELLET reasoner in the case of this study. Figure 17 shows that the built ontology is consistent and coherent. Furthermore, the BAEO understanding, logical consistency, modeling issues, and ontology language specification were evaluated with no critical level error from the OntOlogy Pitfall Scanner, which contains over 40 common pitfalls in ontology development. In other words, the established ontology is documented for human understanding, was designed correctly with primitive from the OWL, and complies with the specification of the OWL (Özacar 2022;Poveda-Villalón et al. 2012).

Constrains verification
Although OWL 2 DL is suitable for domain description because of its high expressivity, it lacks the means for autovalidation. For this purpose, the W3C community developed constraint languages such as SHApe Constraint Language (SHACL) or the Shape Expression (ShEx), which allows RDF data validations. Significantly, these languages and 1 http:// prote ge. stanf ord. edu/. 200T accident from the ontology and the report SHACL, in particular, do not only enhance RDF graph understanding, but they also help to detect problems on data graphs and therefore provide guarantees for better interoperability (Pareti and Konstantinidis 2021). In this paper, SHACL was used as a language to build validation graphs since it offers basic inference and the possibility of supporting open-world assumptions (OWA) over its counterpart ShEx (Martínez-Costa and Schulz 2017). SHACL provides a way to define the data model, and value restrictions called shape, which an RDF knowledge graph most respect (Das and Hussey 2021;Cimmino et al. 2020).
Code 6 describes the validation of Event instances specifying that an event must have a temporal property in the RDF graph. Code 7 describes the validation of Activity instances specifying that an activity has at least a beginning or ending event.

Structural verification
We used the ontology pitfall scanner (OOPS!) to verify the BAEO structure. This tool is used for ontology diagnosis 1 3 and repair because it identifies common mistakes such as domain and range classes intersections, naming conventions issues, and cycles in taxonomy hierarchy (Poveda-Villalón et al. 2014).
The OOPS tool validation revealed that the BAEO has 01 critical alert in multiple ranges in a property. This alert on BAEO:operatesIn object property is because it was designed that an agent can operate in an event or activity as well as a particular resource. Figure 18 describes the structure of the BAEO obtained after the design process. Figure 18a shows the class hierarchy, Fig. 18b and c the property hierarchy. Figure 19 shows the metrics of BAEO structure in terms of classes, axions, and properties.

Illustration
The case of aircraft accident expertise was addressed as the application BAEO and its reuse. For this purpose, the accident expertise report of a Piper PA34-200T aircraft that occurred on the 7th of December 2016 at Bale-Mulhouse was treated.
• How did the accident occur? This Code 8 is the query that describes events of the accident and when these events took place. Figure 20a shows the result of this query, compared with the result from the report as shown in Fig. 20. Figure 20b shows the summary of the main events of the accident from the report. This result is included in the output of the events from the ontology as shown in Fig. 20a. • What are the causes of the problem? This query in Code 9 shows the accident's main causes and consequences in Fig. 21. Figure 21a is the output of this query. We cannot capture technical causes as shown in Fig. 21b because of the need for specific domain knowledge. This is normal since BAEO is designed to be a high-level ontology that can be extended into various domains.  Causes from ontology and the report quences that they suffered. As in the report, the victim is the Pilot who died in the accident and whose aircraft was destroyed. Figure 22a shows the result of this query and its comparison with the one in the report. Figure 22 shows the similarity between these results. The result from the ontology is consistent with the report but lacks some detail about the domain as the report because it is still a high-level representation.
We hosted the base accident expertise ontology on git for easy download. The BAEO is available at ENIT, 2 and Fig. 23 illustrates its graphical representation.

Discussion
To the best of our knowledge, the task of accident expertise representation using ontology has not been investigated. However, some research has been done in domains similar to accident expertise, such as accident, risk, or safety ontology.
The proposed study is dedicated to the field of accident expertise. It describes the construction of a high-level ontology (BAEO) to represent knowledge of accident expertise.
This study bases its methodology on an iterative, cyclic, and model-driven architecture (MDA) approach that complements the approaches proposed by Mokhtar (2021) andDe Nicola et al. (2009) for building autism ontology and a software engineering approach for ontology construction.
These approaches are themselves based on well-known methods such as METHONTOLOGY (Uschold and King 1995) and ontology development 101 (Noy and McGuinness 2001), which define essential steps to consider when designing ontology.
Because the proposed ontology is domain-independent and dedicated to accident expertise knowledge representation, there is a need to have an understandable design. This approach based on model-driven development provides a systematic engineering process, and interoperability and reusability (Silva et al. 2021). These characteristics will allow developers and knowledge engineers to easily extend this foundation to their domain of concern.
Furthermore, the proposed study distinguishes itself from others by its ability to capture essential features of accident expertise such as cause, consequence, event, context, and domain. In contrast, existing ontologies in the field close to accident capture only some of these essential concepts. Table 2 shows a benchmark of this study compared to existing projects.

BAEO reuse
BAEO, as a base ontology, was designed with reusability in mind. Its concepts were made general to facilitate its reuse in various fields of accident expertise. Bringing more specializations to these concepts makes extending BAEO to specific domains possible. For example, a road accident expertise ontology can be obtained by (1) extending the baeo:Human concept with the rao:Person concept from the road accident ontology as shown in Fig. 24, (2) extending the baeo:MaterialResource concept with the rao:Vehicle concept. rao stands for road accident ontology. 3 Another reused of the BAEO is in the domain of construction accidents expertise.

3
The construction industry expertise semantics can be obtained from the cifying its concepts in the following ways (Nowobilski and Hoła 2023;Rafindadi et al. 2022): • The causes of accidents in the construction industry are due to winds, technical or organizational mistakes, or human behavior. From these specific concepts, the BAEO can be extended from the baeo:Cause concept as shown in Fig. 25. • Similarly to the cause concept, the baeo:Human agent concept can be extended with more specific and construction domain agents as in Fig. 26. Specific concepts such as surveyor, architect, and engineer can be attached in this case. • Like the above case the remaining concepts can be specified with concepts of the construction industry.
These illustrations of the reuse of the BAEO in other domains show that it is domain independent and can be extended in other specific accident domains.

Conclusion
Accident expertise is an expensive and knowledge-intensive activity carried out by human experts to understand accidents, and the representation of this knowledge will be wealthy for the integration, search, and designing of expertise aid systems. This study elaborates on the development of a domain ontology to represent accident expertise knowledge using a model-driven methodology and semantic web technologies. This ontology has advantages over its counterparts because it  is high-level and general-purpose, therefore, can be extended to specific domains of accident expertise. Furthermore, it overcomes the principal concepts of accident expertise not present in existing accident ontologies.
Evaluations and illustrations on specific domains such as aircraft, roads, and construction yielded satisfactory results establishing this ontology's accuracy, completeness, and adaptability. However, this ontology suffers from its strengths (domain ontology as opposed to application ontology) because it cannot represent details about specific domains; therefore, its extension to a specific domain is recommended if one requires a certain level of clarity of accident expertise.
This study stands as a base prospect for future research activities on expertise processes. Thereupon, the next step is the development of knowledge-intensive casebased reasoning to assist experts in carrying out Expertise Processes as described in the NF X50 110 French standard for Expertise Processes. The extraction of patterns from existing cases through selection, preprocessing, transformation, and mining (Nwagu et al. 2017) will be coupled to the BAEO knowledge to enhance case retrieval and adaptation and simplify hypotheses generation for the approach.

Conflict of interest
The authors declare no competing interests.