Causation Correlation Analysis of Aviation Accidents: A Knowledge Graph-Based Approach (2024)

1. Introduction

As a high-speed, high-frequency, and efficient mode of transportation, aviation flight plays an irreplaceable role in human progress and social life. However, the increase in aviation activities has led to more accidents. According to the ICAO’s yearly accident statistics of global scheduled commercial operations (with a certified maximum take-off weight of over 5700 kg), between 2013 and 2022, the number of global aviation accidents increased from 90 in 2013 to 114 in 2019, though accidents declined after 2020 as a result of the great pandemic, which reduced aviation activity [1,2]. In addition, despite the infrequency of aviation accidents, their occurrence can inflict substantial damage on fatalities and property, thereby having a profound negative societal impact and posing a considerable impediment to the continued progression of the aviation industry. In view of the number of aviation accidents and the harmful consequences of these accidents, it is necessary to ensure the safety of aviation flights, and improving aviation safety has been increasingly emphasized [3].

Therefore, scholars have conducted extensive research on aviation accidents. Accident causation analysis is a traditional research hotspot, and a large number of scholars have constructed index systems, applied different causation models, and improved and combined the models to analyze, evaluate, and prevent accidents. Based on three causation models, namely, HFACS, ATSB, and IATA, Chia-Fen Chi et al. mined the root causes of accidents and proposed a human–machine–environment–procedure classification scheme [4]. Abiodun Brimmo Yusuf et al. integrated the HFACS framework and Fuzzy Cognitive Mapping for in-flight startle causality analysis [5]. Lijie Cui et al. constructed an aircraft safety assessment index under uncertain conditions by analyzing the uncertainties of aviation maneuvers [6]. Human causes are among the main causes of accidents and the focus of analysis [7,8,9,10]. However, traditional methods usually conduct a safety analysis in a qualitative way through causal models and indicator systems in which it is difficult to quantitatively analyze the intrinsic connections among accident causal factors.

Other common dimensions of accident research are analyzing the historical data of accidents and accident signs quantitatively; obtaining the inherent patterns of accident occurrence from the dimensions of data statistics and mining, causality analysis, etc.; and analyzing the trends in accident occurrence and prediction. Husam Kharoufah et al. investigated a random sample of more than 200 commercial air transportation accidents and incidents in total over 16 years and found that the most significant human factors contributing to aviation accidents and incidents are situational awareness and non-adherence to procedures [11]. Peng He et al. fitted the civil aviation accident trends from 1994 to 2020 using the Causal-ARIMA model to improve the accuracy and interpretability of aviation accident trend analysis [12]. Anthony J. Erjavac assessed the quantitative relationship between symptoms and latent causal factors by using multivariate logistic regression to model the relationship among latent causal factors, symptom causal factors, and accident severity [13]. Xiaoge Zhang et al. developed a hybrid model to quantify the risk associated with the consequences of each hazardous cause, thus predicting the risk level for the severity of an unusual aviation event [14].

In summary, aviation accident research focuses on accident causation analysis and the data-driven mining of accident patterns and trends. However, as a systemic issue, aviation accident causation factors are mutually induced and coupled with spatio-temporal randomness and diversity. Therefore, a holistic approach is required for comprehensive interpretation [15]. Consequently, mining the correlation among the internal factors of accident causation and between accident causation and other accident factors from accident data facilitates a deeper understanding of the mechanisms of accidents, providing informational support for aviation safety analysis and reducing the probability of accidents.

The multi-dimensional network model is a tool for multi-dimensional information representation and analysis that is capable of quantitatively analyzing the inherent patterns in network topology structures. It provides a novel perspective for analyzing complex relationships in accident causation. Liu et al. proposed multi-dimensional topological indices and constructed a knowledge graph by analyzing railway operational accidents in 2021 [16]. Subsequently, Ning Wang et al., further improved the multi-dimensional network topology analysis method for railway accidents by adding the time dimension, conducting a more comprehensive study on the knowledge graph of hazard correlation analysis [17]. As for aviation accidents, there are numerous causal factors that contribute to their occurrence, and the inherent patterns within accident causal factors and other accident factors need to be obtained through the analysis of a large number of accident samples. Therefore, this article employs the multi-dimensional network model to quantitatively analyze the correlations among accident causal factors and between accident causal factors and other accident factors.

Additionally, current research on aviation accidents primarily relies on aviation accident reports, which comprehensively reflect information on and the entire process of the accident. Accident reports serve as a direct source of information for a comprehensive and in-depth analysis, providing insights into the whole picture and causes of accidents and playing a crucial role in enhancing aviation safety management capabilities. However, the traditional approach to mining aviation accident reports still relies heavily on manual methods. When dealing with unstructured textual information, manual analysis is inefficient, is prone to missing information, is difficult to conduct on a large number of samples, and has trouble directly retrieving useful information [18]. Therefore, faced with vast and complex aviation accident reports, determining how to efficiently and effectively mine the information within them has become an urgent need for the intelligent development of aviation safety management processes [19].

With the increasing maturity of artificial intelligence methods, technologies represented by knowledge graphs have brought new opportunities to aviation safety research [20]. Unlike the qualitative analysis method used by traditional accident causation analysis models, knowledge graphs describe various relationships between knowledge entities, expressing various types of nodes and their relationships [21]. This approach enables more efficient and comprehensive mining of their associated relationships [22], expresses complex accident causation information [23], and enhances the relationships among accident data [24]. In the field of aviation safety, the application of knowledge graphs in related research is still in its infancy. For instance, Xin Wang [25] and Jintao Wang [26] both proposed different entity extraction methods based on the construction of knowledge graphs.

The sections of this study are structured as follows: In Section 2, we introduce the model of the Aviation Accident Causation Correlation Knowledge Graph (ACAKG). In Section 3, by collecting 437 aviation accident investigation reports globally from 2018 to 2022, we construct the ACAKG using these data as a foundation. We analyze and derive trend patterns, verifying the effectiveness of this method and laying a foundation for further research on accident causal factors. Section 4 discusses the findings, and Section 5 summarizes the content of this study.

2. Methods

2.1. Modeling the ACAKG

Establishing the ACAKG is the first step in conducting aviation accident causation correlation analysis. A knowledge graph is a structured semantic graph composed of vertices (or nodes) and edges, which is used to describe concepts and their relationships. Its basic structure consists of triplets (“entity-relationship-entity”), where vertices represent concepts or entities and edges represent semantic relationships between concepts or entities. Based on the type of knowledge, knowledge graphs are typically classified into general knowledge graphs and domain knowledge graphs. The ACAKG belongs to the latter category, characterized by its specialization, dynamism, and strong contextual correlation [27]. Through the construction of the ACAKG, scattered textual descriptions of aviation accident causes can be linked to entities and relationships, forming a complete, structured, and easily extractable knowledge base [23]. This facilitates the management, retrieval, utilization, understanding, and analysis of aviation accident causal information contained within [22], revealing the potential relationships among the internal causes of accidents as well as those between the causal factors of accidents and other factors of aviation accidents.

Figure 1 demonstrates the basic process of domain knowledge graph construction and application, which is mainly divided into five parts: data acquisition, ontology construction, information extraction, knowledge graph construction, and application. Among them, the ontology module constructs the main concepts in the field of aviation accident causation, and the information extraction module performs preprocessing and extraction on the text. Knowledge storage involves storing the acquired knowledge in a specific structure and visualizing it. In addition, the knowledge graph can be divided into a schema layer and a data layer. The schema layer, with the ontology as its core, can be expressed as classes and attributes. The data layer refers to the entity relationship network composed of all instance data obtained through information extraction [22].

In the knowledge graph, the ontology module can be typically divided into four types of concepts, namely, entities, categories, relationships, and attributes, which are characterized by conceptualization, formalization, domain specificity, and clarity [28]. Entities are the basic unit of the ontology, categories are collections of entities with similar attributes, relationships refer to the connections between categories or entities, and attributes define categories or entities. Ontology construction methods mainly include manual, automatic, and semi-automatic approaches. In the construction of the ontology for aviation accident causation presented in this paper, a semi-automatic approach is adopted [27]. Step 1: An ontology framework encompassing accident causation, flight phases, accident types, and accident consequences is established based on existing research on aviation accident causation and relevant ICAO standards [29,30,31]. Step 2: This framework is refined in response to feedback from experts in aviation safety. Step 3: In the process of semi-automatic triple annotation, when the annotated instances found do not belong to the existing ontology, further improvement and supplementation should be made to the ontology. Figure 2 illustrates the core concepts of the ACAKG ontology.

Information extraction mainly involves three steps. Firstly, manual calibration and refinement are conducted to address any data deficiencies, duplications, or inaccuracies in the report data. Secondly, entity extraction is performed. Thirdly, the relationships between entities are determined and extracted. Entity extraction is the process of extracting information from aviation accident texts. Relationship extraction involves organizing and extracting the relationships between entities. Based on the defined scope of concepts in the ACAKG ontology, the relationships between entities include the relationships among accident causation, causation, and flight phases, and those between causation and accident types and causation and accident consequences. The corresponding entity relationships are “result in”, “phase is”, “type is”, “fatality consequence is”, and “aircraft damage consequence is”. The “result in” relationship can exist among accident causal factors, among accident types, and between causal factors and accident types. In addition to primary causation relationships, accident causal factors can also have parallel relationships, but this paper only discusses causal relationships.

After establishing the ontology and extracting concepts and relationships, the generated knowledge graph and triple relationships serve as the foundational data to build matrices from four dimensions: aviation accident causation, flight phases, accident types, and accident consequences. This process enables the construction of the ACAKG model. The model is represented by seven matrices, each with the following fundamental meanings:

The first matrix is the Causal Adjacency Matrix (CAM), as shown in Equation (1). It determines whether there exists a unidirectional causal relationship between the head entity and the tail entity in the triple relationship. The adjacency matrix is determined by the triples containing the keyword “Causation”. Here, both i and j can represent accident causation or accident types, and KTs represent the set of all triples identified.

$C A M_{i j} = \{\begin{cases} 1 〈 i, c a u s e, j 〉 \in K T s \\ 0 〈 i, c a u s e, j 〉 \notin K T s \end{cases}$

(1)

The second matrix is the Accident Phase Matrix (APM), as shown in Equation (2). It describes the relationship between accident causation or accident types in the triple relationship and the flight phase in which the accident occurs. Specifically, it indicates whether accident causation or type i occurs in flight phase j. The adjacency matrix is determined by the knowledge triples containing the keyword “Phase”.

$A P M_{i j} = \{\begin{cases} 1 〈 i, p h a s e, j 〉 \in K T s \\ 0 〈 i, p h a s e, j 〉 \notin K T s \end{cases}$

(2)

The third matrix is the Causal Relationship Strength Adjacency Matrix (CRSAM), as shown in Equation (3). Building upon Equation (1), this matrix assigns weights to the directed edges, representing the strength of the causal relationship between the head entity and the tail entity. Typically, this weight is related to the frequency of occurrence of a particular causal event in all triples. The higher the frequency, the greater the weight.

$C R S A M_{i j} = \{\begin{cases} ω 〈 i, c a u s e - ω, j 〉 \in K T s \\ 0 〈 i, c a u s e - ω, j 〉 \notin K T s \end{cases}$

(3)

The fourth matrix is the Causation Types Matrix (TM), as shown in Equation (4). It describes the relationship between accident causation and consequences in the triple relationship. Specifically, it indicates whether accident causation i belongs to type j. The adjacency matrix is determined by the knowledge triples containing the keyword “Type”.

$T M_{i j} = \{\begin{cases} 1 〈 i, t y p e, j 〉 \in K T s \\ 0 〈 i, t y p e, j 〉 \notin K T s \end{cases}$

(4)

The fifth matrix is the Accident Consequence Matrix (ACM), as shown in Equation (5). It is defined in the knowledge triples, indicating whether accident causation i leads to accident consequence j. The weight of the directed edge δ represents the severity of the accident consequence. The adjacency matrix is determined by the knowledge triples containing the keyword “Consequence”.

$A C M_{i j} = \{\begin{cases} δ 〈 i, consequence, j 〉 \in K T s \\ 0 〈 i, consequence, j 〉 \notin K T s \end{cases}$

(5)

The sixth matrix is the Shortest Path Matrix (SPM), as shown in Equation (6). The causal relationship adjacency matrix forms a complex network structure mapping all accident causation and accident types. The edges between nodes are unidirectional, representing direct or indirect causal relationships. The shortest path matrix between any two nodes with causal relationships describes the shortest distance between accident causation and accident type in the triples, represented by the number of nodes traversed from node i to node j.

$S P M_{i j} = \sum_{p, q \in N} C A M_{p q}$

(6)

The seventh matrix is the Reachability Matrix (RM), as shown in Equation (7). When the causal relationship edges between nodes in the network are unidirectional, connecting the head and tail nodes, it indicates a causal path from node i to node j. In other words, if there is reachability between two nodes, the shortest path exists between them.

$R M_{i j} = \{\begin{cases} 1, S P M_{i j} > 0 \\ 0, S P M_{i j} = 0 \end{cases}$

(7)

2.2. ACAKG Network Topology Analysis

As a multi-dimensional network graph of aviation accidents, the ACAKG describes various relationships between aviation accidents and accident-causing entities. To explore underlying patterns in aviation accidents, topological indicators can be utilized to conduct network topology analysis on ACAKG. Traditional topological indicators are mainly designed for one-dimensional networks. Given the unique characteristics of the ACAKG, it is essential to establish topological analysis indicators for multi-dimensional networks. These topological indicators can help uncover potential patterns in accidents from a multi-dimensional network perspective, aiding in the development of more targeted safety strategies.

Building upon multi-dimensional topological analysis, this study incorporates the flight phase at the time of the accident as an analytical dimension. While reference [17] augmented the analysis with a time dimension, expanding from a one-dimensional to a multi-dimensional model, it only divided the time dimension into day and night. For aviation accidents, the flight phase dimension can more intricately reflect the correlation between the timing of accidents and their causation. This study further subdivides the time dimension based on the characteristics of the nine phases from take-off to landing. In the ACAKG model, N = {ACN, AN, TN, PN, CN}, where N denotes the set of all nodes in the topological structure, ACN represents the set of all accident causation factors, AN is the set of all accidents, TN is the set of all accident causation types, PN is the set of all accident phases, and CN is the set of all accident consequences.

The strength of the causal relationships between accident causal factors and different flight phases can be measured to assess the risk level of accident causation, divided into Active Risk Level (ARL) and Passive Risk Level (PRL). The Active Risk Level indicates the likelihood that a specific accident causal factor in a certain flight phase leads to other accident causal factors, and it is calculated using Equation (8). A higher value indicates a greater likelihood of the factor to trigger other accident causal factors. Similarly, the Passive Risk Level represents the likelihood that a current accident causal factor in a specific flight phase is caused by other accident causal factors, and it is calculated using Equation (9). A higher value indicates a factor’s higher susceptibility to being triggered by other accident causal factors.

$A R L_{h}^{p} = R M \times A P M / \sum_{i \in H N} S P M_{h i}$

(8)

$P R L_{h}^{p} = R M^{T} \times A P M / \sum_{i \in H N} S P M_{i h}$

(9)

The distribution proportion of accident causation types describes the relationship between the current accident causation and other types of causation at different flight phases. It is divided into Active Causation Type Distribution (ACTD) and Passive Causation Type Distribution (PCTD). Active Causation Type Distribution represents the proportion distribution of other types of accident causal factors caused by the current accident causal factor at different flight phases, as shown in Equation (10). PCTD illustrates the proportion of causation types leading to the current accident causal factor within different flight phases, as shown in Equation (11).

$A C T D_{h, t}^{p} = \sum_{i \in H N} (C R S A M_{h i} \times C T M_{i t} \times A P M_{i}^{p}) / (C R S A M \times A P M)$

(10)

$P C T D_{t, h}^{p} = \sum_{i \in H N} (C R S A M_{i h} \times C T M_{i t} \times A P M_{i}^{p}) / (C R S A M^{T} \times A P M)$

3. Aviation Accident Case Research

3.1. Aviation Accident Data Collection

Based on the constructed ACAKG model, this paper selects a total of 437 accident cases from 2018 to 2022. The accident dataset mainly comes from the Aviation Safety Network (ASN), which is a statistical website of aviation accident data belonging to the Flight Safety Foundation, providing complete information on aircraft accidents, and the sources of information mainly come from government agencies, aviation accident investigation committees, and civil aviation agencies [32]. The NTSB, UKAB, JTSB, BFU, etc., are the governmental agencies of the United States, the United Kingdom, Japan, Germany, and other countries, which are responsible for publishing their own national aviation accident investigation reports. In response to missing data, duplications, and errors in the collection, the data were calibrated against the original data published by national organizations. The final aviation accident database was formed, including structured data fields such as accident time, location, type, consequence (including fatalities and aircraft damage), nature of the flight, flight phase, etc., and unstructured data fields such as accident description, causation, consequence, improvement measure, etc.

According to the research needs, only four data fields, namely, accident consequence, flight phase, accident type, and accident causation, are extracted for the causation-related analysis of aviation accidents. Among them, accident consequence and flight phase correspond to only one data point in each accident; accident causation and accident type correspond to one data point or multiple data points, and there is a causal or juxtaposition relationship between multiple data points, though only the causal relationship is analyzed in this paper.

The ACAKG consists of 31 accident causation nodes (897 causal factors were identified, including 497 human factors, 119 aircraft factors, 100 environment factors, and 166 management factors), nine flight phase nodes from P01–P9, eight accident-type nodes from T01–T08, and 12 accident consequence nodes, including fatality nodes from RH01–05 and aircraft damage nodes from RA01–07.

In addition, the operating type distribution of 437 accidents is shown in Figure 3, and it should be noted that the aircraft are capable of carrying at least 12 passengers.

3.2. Modeling the ACAKG

Based on the aviation accident database that is described above, a total of 437 aviation accident reports from the five-year period of 2018–2022 were used for entity identification to construct the ACAKG ontology framework from four entity dimensions, including accident causation, flight phase, accident type, and accident consequence. Based on the information extracted from the database, the ACAKG constructed the schema layer and data layer, as shown in Figure 4.

Among them, 31 accident causal factors are numbered as H01–H13, A01–A06, E01–E09, and M01–03 according to the four major categories of “human–aircraft–environment–management”, and the specific classifications and frequencies are shown in Table 1.

From the time dimension, flight phases were selected as the research object. For a more focused follow-up analysis, nine phases were merged into four phases according to the flight characteristics, i.e., the ground phase, take-off/landing phase, level flight phase, and maneuvering phase. The number and the frequency of each phase of the accident are shown in Table 2.

The consequences of accidents are divided into two components: fatalities and aircraft damage. In order to facilitate the measurement of the consequences of accident types, the accident consequences need to be quantified. The method of affiliation function is used to represent the weights of accident consequences with different granularities between 0 and 1, as shown in Table 3. Based on the principle that the consequences of injury are more serious than those of aircraft damage, this paper sets the weights of fatalities and aircraft damage to 0.7 and 0.3, respectively, so that the quantified consequences of different accident types can be obtained.

Aviation accident types numbered T01–T08 were also identified as accident knowledge entities, as shown in Table 4.

The essence of a knowledge graph is that it is a dataset that expresses factual and semantic relationships in the form of triples [23]. In this paper, we mainly constructed four types of relationships, namely, “causation result in”, “phase is”, “type is”, and “consequence is”. Taking the three aviation accidents with accident numbers 2018-12, 2021-104, and 2022-74 as examples, the relationship descriptions and triples were obtained and are shown in Table 5.

3.3. Statistical Strength Analysis (t-Test)

To accurately reflect the true nature of the research objectives, this paper conducts t-tests on the fundamental data, including accident causation, accident phase, and accident type, to infer whether these differences are statistically significant, thereby enhancing the statistical strength of subsequent conclusions.

Table 6 systematically evaluates whether the 31 causal factors significantly deviate from the hypothesized value of 0.0 through one-sample t-tests. The results indicate that the averages of 23 factors (H01, H02, H03, H07, H09, H11, H12, H13, A04, A05, A06, E01 to E09, M01, M02, and M03) do not exhibit statistically significant differences from 0.0 (p > 0.05), suggesting that the averages of these items closely fluctuate around 0.0 without substantial deviation. By contrast, the averages of the remaining eight factors (H04, H05, H06, H08, H10, A01, A02, and A03) significantly differ from 0.0, specifically manifesting as averages notably higher than 0.0, a finding with important statistical significance. Further assessment of the effect size using Cohen’s d reveals that three factors exhibit medium effects (with effect sizes between 0.50 and 0.80), which not only reinforces the reliability of the data but also indicates that our research findings possess a certain degree of inferential statistical strength.

Table 7 shows that the values of all four indicators, PC01, PC02, PC03, and PC04, are significantly different from the value of 0.0 (p < 0.05), indicating that their mean values exhibit statistically significant differences from the value of 0.0. Specifically, the mean values of these four indicators are all higher than 0.0, suggesting positive and significant performance in the studied field. In summary, there are statistically significant differences between the mean values of PC01, PC02, PC03, and PC04 and the value of 0.0. Furthermore, the assessment using Cohen’s d values reveals that three of these four indicators fall within the moderate effect range, while one reaches the high effect range.

Table 8 shows that the results indicate that the mean values of the four indicators T01, T06, T07, and T08 do not exhibit statistically significant differences from 0.0 (p > 0.05), suggesting that the mean values of these indicators are close to 0.0 and lack statistically meaningful distinctions. Conversely, the mean values of the four indicators T02, T03, T04, and T05 are significantly different from 0.0 (p < 0.05), specifically manifesting as mean values that are notably higher than 0.0.

3.4. Analysis of ACAKG Model Results

3.4.1. Causal Relationship Intimacy Analysis of Aviation Accidents

Equations (8) and (9) are used to calculate the active causal relationship and passive causal relationship of an accident causal factor; they are called Active Risk Level and Passive Risk Level.

Figure 5 shows the Active Risk Level and Passive Risk Level for all causal factors. For different flight phases, the Passive Risk Level of an accident causal factor is generally higher than its Active Risk Level. Horizontally, it can be concluded from Figure 5a that among the H-type, the Active Risk Levels of H06 and H08 are significantly higher than those of the others; among the A-type, the Active Risk Levels of A02 and A05 are significantly higher than those of the others; among the E-type, the Active Risk Levels of E01 and E04 are significantly higher than those of the others; and among the M-type, the Active Risk Level of M01 is significantly higher than that of M02. From Figure 5b, it can be concluded that among the H-type, the Passive Risk Levels of H03 and H07 are significantly higher than those of the others; among the E-type, the Passive Risk Level of E04 is significantly higher than that of the others; and among the M-type causal factors, the Passive Risk Level of M01 is significantly higher than that of M02. This indicates that these causal factors are more likely to be caused by others and require special attention.

3.4.2. Type Distribution Proportion Analysis of Aviation Accident Causation

The active type distribution ratio and passive type distribution ratio of accident causal factors can be obtained using Equations (10) and (11), and the active type distribution ratio and passive type distribution ratio of the four phases are shown in Figure 6. In Figure 6, (1) represents the active type distribution ratio in various flight phases, and (2) represents the passive type distribution ratio.

Through the analysis, it is found that in the ground phase, take-off/landing phase, and leveling phase, the active causal type distribution ratio of some causal factors such as H02, H09, A05, A06, E01~E03, E05~E09, and M03 is 0, which indicates that these causal factors cannot lead to the generation of other causal types; the passive causal type distribution ratio of some causal factors such as A06 and E05~E09 is 0, which indicates that these causal factors cannot lead to the generation of other causal types; and the passive causal type distribution ratio of some causal factors such as A06 and E05~E09 is 0. The proportion of the passive causal type distribution of some causal factors, such as A06 and E05~E09, is 0, which means that these causal factors also cannot be caused by other causal types, i.e., these causal factors will not appear in the ground phase, take-off/landing phase, and leveling phase. For the maneuvering phase, in addition to the above causal factors, the proportion of the distribution of active causal types for H11 and M01 is 0, indicating that the causal factors of H11 and M01 cannot be caused by other causal types as active causal factors. In addition, most of the active causal factors are able to lead to the H-type, E-type, A-type, and M-type causations, while most of the passive causal factors are generally caused by the H-type and A-type causations. Among them, the active causal distribution types of H03, H11, H12, E04, and M01 are concentrated in one causal type, indicating that the causal factors caused by these causal factors are relatively fixed, and the passive causal type distributions of H01~H03, H10~H13, A02~A04, and E03 are concentrated in one causal type, indicating that they can only be caused by one of these types of causal factors.

By analyzing the distribution proportion of causal factor types, we can better understand the extent to which different causal factors influence the occurrence of accidents, thereby facilitating the subsequent development of targeted prevention strategies for aviation accidents.

3.4.3. Correlation Analysis of Accident Causation Types

The correlation of an accident causation type is categorized into the direct correlation type or the indirect correlation type, which indicates the strength of direct or indirect causal relationships between accident causation types at different flight phases, and it is calculated using Equations (12) and (13), respectively.

Figure 7 shows the direct and indirect correlations for different types of accident causation. Figure 7a shows the direct and indirect correlations of accident causal types during the ground phase. Overall, the H-type and M-type causations have the highest direct accessibility, followed by higher direct accessibility between the H-type and E-type and the A-type and H-type causations, indicating a higher direct correlation between these types of accident causal factors, and there is higher indirect accessibility between the H-type and H-type, A-type and H-type, and H-type and A-type causations, indicating a higher indirect correlation between these types of accident causal factors. In addition, there is no direct or indirect accessibility of accident causal factors between the E-type and A-type, E-type and E-type, E-type and M-type, and M-type and H-type causations. The direct and indirect correlations of accident causation types in the take-off/landing phase and leveling phase are shown in Figure 7b,c, respectively. The high direct accessibility between the H-type and H-type, H-type and A-type, H-type and E-type, H-type and M-type, and A-type and H-type causations indicates that the direct correlation of accident causal factors is higher for these types, including the H-type and H-type, A-type and H-type, H-type and A-type, and H-type and E-type causations. The high indirect accessibility of the E-type and A-type causations indicates that the direct correlation of accident causal factors is higher for these types, and the H-type and M-type causations have a higher indirect accessibility, indicating a higher indirect correlation between these types of accident causal factors. Similarly, there is no direct or indirect accessibility of accident causal factors for the E-type and A-type, E-type and E-type, E-type and M-type, and M-type and H-type causations. Figure 7d shows the direct and indirect correlations of accident causal factor types for the maneuvering phase of the aircraft. For the maneuvering phase, the direct accessibility between the H-type and H-type and the H-type and M-type causations is high, indicating that these types of accident causal factors have high direct correlations with respect to the other types of factors, and the indirect accessibility between the H-type and H-type, A-type and H-type, H-type and A-type, H-type and E-type, and H-type and M-type causations is high. In addition, there is no direct or indirect correlation between the E-type and the other types of accident causal factors, nor is there with the M-type and H-type and M-type and A-type causations.

Exploring the accessibility among different types of accident causal factors helps us to understand the correlations among different types of accident causal factors both locally and holistically, which can help us to develop preventive strategies for aviation accidents.

3.4.4. Correlation Analysis between Aviation Accident Causation Types and Flight Phase

The correlation between accident causation type and flight phase is also categorized into direct and indirect correlations (DCTP and ICTP), which indicate the degree of correlation between the current accident causal factor type and the flight phase in which future accidents will occur as a direct or indirect result. The direct and indirect correlations between causation type and time can be calculated using Equations (14) and (15).

Figure 8 shows the statistical results of the direct and indirect correlations between the causal type and the different flight phases. It can be seen from the figure that the indirect correlation between each causal type and the flight phase is higher than the direct correlation in different phases. Specifically, the indirect correlation of the H-type causal factors is significantly higher than their direct correlation in all phases; the direct correlation of the H-type causal factors is higher than the direct correlation of the other causal factor types; the comparative results of the direct correlation of the causal factor types in different phases are H-type > E-type > M-type > A-type, with the indirect correlations being H-type > A-type > E-type > M-type; and the correlations of different causal factors in the take-off/landing and leveling phases are higher than those in the ground and maneuvering phases.

3.4.5. Correlation Analysis of Aviation Accident Consequence

The direct and indirect consequences of accidents can be obtained through Equations (16) and (17). Referring to Table 3, appropriate weights can be assigned to these accident consequences. Assuming that only accidents with the identifiers 2018-12, 2021-104, and 2022-74 occurred within the past 5 years, the consequences of these accidents can be calculated based on their respective accident reports. From Table 5, it is determined that the accident with the identifier 2018-12 involved runway and abnormal landing accidents, resulting in no fatalities (weight 0) and substantial aircraft damage (weight 0.7). The accident with the identifier 2021-104 involved a collision accident, resulting in six fatalities (weight 0.7) and the destruction of the aircraft (weight 1). Similarly, the accident with the 2022-74 identifier involved runway and abnormal landing accidents and resulted in no fatalities (weight 0) and substantial aircraft damage (weight 0.7). Based on the data, the average consequences of different accident types within the past 5 years can be calculated. The average fatalities for runway accidents are represented as (0 + 0)/5 = 0, and the average aircraft damage is as follows: (0.7 + 0.7)/5 = 0.28. Consequently, the total average consequence of runway accidents is 0.7 × 0 + 0.3 × 0.28 = 0.084. Similarly, the total average consequence of abnormal landing accidents is 0.7 × 0 + 0.3 × 0.28 = 0.084, and the total average consequence of collision accidents is 0.7 × 0.14 + 0.3 × 0.2 = 0.158. Using the aforementioned computation method, the average consequences of different accident types within the past 5 years can be determined.

Figure 9 illustrates the aggregation of weighted fatalities and aircraft damage to determine the level of accident consequences directly and indirectly caused by causal factors at different phases. In Figure 9a, it is evident that during the take-off/landing and level flight phases, factors H07, A05, and E01 have higher direct impacts on accident consequences, while during the maneuvering phase, factors H07, E01, and M01 pose greater direct consequences. Overall, during the ground phase, factors H07, A05, E01, and M01 have more significant direct impacts on accident consequences. Figure 9b shows that factors H01, H04–H06, H08, H10–H13, and A01–A04, whether directly or indirectly contributing to accidents, have a higher severity compared to other factors. The indirect impact of the H-type and A-type factors on accident consequences is greater than that of the E-type and M-type factors. The average severity of the consequences caused by causal factors during the take-off/landing and level flight phases is the highest, followed by those caused by causal factors during the ground phase, with the maneuvering phase having the lowest average severity of consequences indirectly caused by causal factors.

4. Discussion

4.1. Identifying Key Risks

Based on the correlation analysis between the direct and indirect severity of accident consequences caused by causal factors shown in Figure 9, it is possible to identify the causal factors that have a greater impact on the severity of accident consequences and their corresponding numerical values. By systematically controlling factors that have a significant influence during different flight phases, it is possible to effectively reduce the severity and impact of accident consequences, thereby enhancing the safety of the aircraft and onboard personnel. According to Equations (16) and (17), the causal factors are ranked in descending order based on their impact on the severity of accident consequences, and causative elimination is carried out. The resulting curves illustrating the impact of mitigating causal factors on the severity of accident consequences at different phases are shown in Figure 10.

We eliminate causal factors in descending order of the severity of their impact on the consequences of the accident. If two causal factors are of equal severity, the causal factors are eliminated first in the order of types H, A, E, and M and then in increasing order of number. In Figure 10, the successive elimination of 31 causal factors significantly reduces the severity of indirect accident consequences. The most notable effects in reducing the harm caused by aviation accidents are observed by eliminating causal factors during the take-off/landing and level flight phases.

When four main causal factors (H7, A5, E1, and M1) are eliminated, the severity index of direct accident consequences is reduced to 76.5% in the ground phase, and the severity index of direct accident consequences is reduced to 81.9% in the take-off/landing and level flight phases. When three main causal factors (H7, E1, and M1) are eliminated, the severity index of direct accident consequences is reduced to 79.2% in the maneuvering phase. When four causal factors (H10~H12 and A2) are eliminated, the severity index of indirect accident consequences is reduced to 60.4% in the ground phase. When four causal factors (H1 and H4~H6) are eliminated, the severity index of indirect accident consequences is reduced to 59.9% in the take-off/landing and level flight phases. When four causal factors (H8, H12, A3, and A4) are eliminated, the severity index of indirect accident consequences is reduced to 69% in the maneuvering phase.

During the ground phase, when 15 causal factors (H1, H2, H4~H9, H13, A1, A2~A4, E1, and M1) are eliminated, the severity index of direct accident consequences is reduced by 80%; when 10 causal factors (H1, H4~H6, H8, H10~H13, and A2) are eliminated, the severity index of indirect accident consequences is reduced by 80%. During the take-off/landing and level flight phases, when 20 causal factors (H1, H2, H4~H11, H13, A1~A5, and E1~E4) are eliminated, the severity index of direct accident consequences is reduced by 80%; when 10 causal factors (H1, H4~H6, H8, H10~H13, and A1) are eliminated, the severity index of indirect accident consequences is reduced by 80%. In the maneuvering phase, when 16 causal factors (H1, H2, H4~H7, H9~H11, H13, A1~A2, E1, E3, and M1~M2) are eliminated, the severity index of direct accident consequences is reduced by 80%; when 10 causal factors (H1, H4~H6, H8, H10~H12, and A3~A4) are eliminated, the indirect severity index of accident consequences is reduced by 80%. Overall, eliminating the indirect causal factors that affect accident consequences, as opposed to direct causal factors, is more effective in reducing the probability of accidents and their severity. Focusing on the key identified factors is beneficial for efficiently reducing the likelihood of accidents.

Furthermore, to verify the significance of the data, PC01 was used as the dependent variable for regression analysis (the analysis for PC02, PC03, and PC04 was conducted similarly to PC01). The obtained p-values for both direct impact and indirect impact were 0.000. This highly significant statistical result indicates a negative relationship between the number of causative factors and PC01, which is consistent with the conclusions of this paper.

4.2. Preventive Measures

Based on the case analysis from the previous section, the correlations among accident causal factors, flight phases, accident types, and accident consequences are derived. Combining these findings allows for the development of corresponding preventive strategies, typically formulated in the following three steps. The first step involves identifying the key factors contributing to accident consequences based on the correlation between causal factors in different flight phases and the severity of the accident consequence. The second step is to analyze the causal correlation and correlation of causal types between the key causal factors and other causal factors through modeling, so as to clarify the main causes of the key causal factors and the chain reactions caused by the key causal factors, as well as the causal types leading to the frequent occurrence of accidents. The third step involves analyzing the occurrence frequency of other causal factors that lead to key causal factors and their frequency during different flight phases. By formulating specific preventive strategies based on this three-step analysis, it is possible to mitigate key accident causal factors and other critical factors that result in high consequences. To optimize preventive measures, adjusting preventive strategies in a timely manner throughout the entire flight process based on the frequency of causal factors arising in different flight phases allows for effective targeted mitigation of risks.

5. Conclusions

This article proposes an aviation accident causation correlation analysis model based on a semi-structured aviation accident report as the research foundation. By constructing a knowledge graph, this article enhances the aviation accident domain’s knowledge base from a causal perspective, establishing a solid foundation for applying aviation accident knowledge graphs from a big data perspective in the future. Additionally, this article determines the strength of causal relationships among various accident causal factors in aviation accidents. This analysis identifies the types of causation that are prone to occur, the main types of causation affecting accident occurrence, the strength of correlations among causation types, and the correlations between causation types and different flight phases, and further investigates the impacts of different causation types on the severity of accident consequences. By combining the analysis results, the critical causal factors that significantly influence accidents during different flight phases are identified, allowing for a reduction in the probability of accidents through further prevention and control of these key accident causal factors. Through a case analysis, the feasibility and effectiveness of the proposed model and methodology are validated. Overall, the approach presented in this article is beneficial for uncovering the underlying patterns of aviation accident causation and strategically preventing the risk of aviation accidents.

Furthermore, as a semantic network with relational connections, knowledge graphs possess certain advantages in mining, extracting, and associating textual data, and hold significant value in aviation safety. In response to the current research deficiencies, further improvements will be made to the aviation accident knowledge graph in terms of enriching the ontology construction and expanding the sample size. The ontology of accident causations will be refined, and a larger, more scientific aviation knowledge graph database will be established to enhance data credibility. This approach aims to extract hidden patterns from a larger dataset, validate conclusions’ accuracy and evolutionary nature over time, and continuously update and expand the aviation accident knowledge base.

Author Contributions

Conceptualization, J.X.; methodology, L.C. and H.X.; software, H.X.; validation, H.X.; formal analysis, J.X.; investigation, L.C.; resources, W.T.; data curation, H.X.; writing—original draft preparation, L.C.; writing—review and editing, W.T.; visualization, W.T.; supervision, J.X.; project administration, L.C.; funding acquisition, J.X. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China, grant number 52074309.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data are contained within the article.

Conflicts of Interest

The authors declare no conflicts of interest.

References

International Civil Aviation Organization. ICAO Safety Report 2014 Edition. 2022. Available online: https://www.icao.int/safety/Documents/ICAO_2014%20Safety%20Report_final_02042014_web.pdf (accessed on 1 March 2024).
International Civil Aviation Organization. ICAO Safety Report 2023 Edition. 2023. Available online: https://www.icao.int/safety/Documents/ICAO_SR_2023_20230823.pdf (accessed on 1 March 2024).
Huesler, J.; Strobl, E. Predicting the Number of Fatalities in Extreme Civil Aviation Accidents. J. Air Transp. 2023, 31, 150–160. [Google Scholar] [CrossRef]
Chi, C.-F.; Sigmund, D.; Lin, Y.-C.; Drury, C.G. The Development of a Scenario-Based Human-Machine-Environment-Procedure (HMEP) Classification Scheme for the Root Cause Analysis of Helicopter Accidents. Appl. Ergon. 2022, 103, 103771. [Google Scholar] [CrossRef] [PubMed]
Yusuf, A.B.; Kor, A.-L.; Tawfik, H. Integrating the HFACS Framework and Fuzzy Cognitive Mapping for In-Flight Startle Causality Analysis. Sensors 2022, 22, 1068. [Google Scholar] [CrossRef] [PubMed]
Cui, L.; Zhang, J.; Ren, B.; Chen, H. Research on a New Aviation Safety Index and Its Solution under Uncertainty Conditions. Saf. Sci. 2018, 107, 55–61. [Google Scholar] [CrossRef]
Wiegmann, D.A.; Shappell, S.A. A Human Error Approach to Aviation Accident Analysis: The Human Factors Analysis and Classification System; Ashgate: Hants, UK; Burlington, VT, USA, 2003. [Google Scholar]
Lyu, T.; Song, W.; Du, K. Human Factors Analysis of Air Traffic Safety Based on HFACS-BN Model. Appl. Sci. 2019, 9, 5049. [Google Scholar] [CrossRef]
Lázaro, F.L.; Nogueira, R.P.R.; Melicio, R.; Valério, D.; Santos, L.F.F.M. Human Factors as Predictor of Fatalities in Aviation Accidents: A Neural Network Analysis. Appl. Sci. 2024, 14, 640. [Google Scholar] [CrossRef]
Leite Martins de Sant’Anna, D.A.; Garibaldi de Hilal, A.V. The Impact of Human Factors on Pilots’ Safety Behavior in Offshore Aviation Companies: A Brazilian Case. Saf. Sci. 2021, 140, 105272. [Google Scholar] [CrossRef]
Kharoufah, H.; Murray, J.; Baxter, G.; Wild, G. A Review of Human Factors Causations in Commercial Air Transport Accidents and Incidents: From to 2000–2016. Prog. Aerosp. Sci. 2018, 99, 1–13. [Google Scholar] [CrossRef]
He, P.; Sun, R. Trend Analysis of Civil Aviation Incidents Based on Causal Inference and Statistical Inference. Aerospace 2023, 10, 822. [Google Scholar] [CrossRef]
Erjavac, A.J.; Iammartino, R.; Fossaceca, J.M. Evaluation of Preconditions Affecting Symptomatic Human Error in General Aviation and Air Carrier Aviation Accidents. Reliab. Eng. Syst. Saf. 2018, 178, 156–163. [Google Scholar] [CrossRef]
Zhang, X.; Mahadevan, S. Ensemble Machine Learning Models for Aviation Incident Risk Prediction. Decis. Support Syst. 2019, 116, 48–63. [Google Scholar] [CrossRef]
Hulme, A.; Stanton, N.A.; Walker, G.H.; Waterson, P.; Salmon, P.M. What Do Applications of Systems Thinking Accident Analysis Methods Tell Us about Accident Causation? A Systematic Review of Applications between 1990 and 2018. Saf. Sci. 2019, 117, 164–183. [Google Scholar] [CrossRef]
Liu, J.; Schmid, F.; Li, K.; Zheng, W. A Knowledge Graph-Based Approach for Exploring Railway Operational Accidents. Reliab. Eng. Syst. Saf. 2021, 207, 107352. [Google Scholar] [CrossRef]
Wang, N.; Yang, X.; Chen, J.; Wang, H.; Wu, J. Hazards Correlation Analysis of Railway Accidents: A Real-World Case Study Based on the Decade-Long UK Railway Accident Data. Saf. Sci. 2023, 166, 106238. [Google Scholar] [CrossRef]
Tang, L.; Meng, Y. Data Analytics and Optimization for Smart Industry. Front. Eng. Manag. 2021, 8, 157–171. [Google Scholar] [CrossRef]
Jiao, Y.; Dong, J.; Han, J.; Sun, H. Classification and Causes Identification of Chinese Civil Aviation Incident Reports. Appl. Sci. 2022, 12, 10765. [Google Scholar] [CrossRef]
Tamašauskaitė, G.; Groth, P. Defining a Knowledge Graph Development Process through a Systematic Review. ACM Trans. Softw. Eng. Methodol. 2023, 32, 27. [Google Scholar] [CrossRef]
Zhang, X.; Srinivasan, P.; Mahadevan, S. Sequential Deep Learning from NTSB Reports for Aviation Safety Prognosis. Saf. Sci. 2021, 142, 105390. [Google Scholar] [CrossRef]
Gan, L.; Ye, B.; Huang, Z.; Xu, Y.; Chen, Q.; Shu, Y. Knowledge Graph Construction Based on Ship Collision Accident Reports to Improve Maritime Traffic Safety. Ocean Coast. Manag. 2023, 240, 106660. [Google Scholar] [CrossRef]
Peng, C.; Xia, F.; Naseriparsa, M.; Osborne, F. Knowledge Graphs: Opportunities and Challenges. Artif. Intell. Rev. 2023, 56, 13071–13102. [Google Scholar] [CrossRef]
Li, G.; Chen, Y.; Liao, Q.; He, Z. Potential Destination Discovery for Low Predictability Individuals Based on Knowledge Graph. Transp. Res. Part C Emerg. Technol. 2022, 145, 103928. [Google Scholar] [CrossRef]
Wang, X.; Gan, Z.; Xu, Y.; Liu, B.; Zheng, T. Extracting Domain-Specific Chinese Named Entities for Aviation Safety Reports: A Case Study. Appl. Sci. 2023, 13, 11003. [Google Scholar] [CrossRef]
Wang, J.; Qu, J.; Zhao, Z.; Dong, X. SMAAMA: A Named Entity Alignment Method Based on Siamese Network Character Feature and Multi-Attribute Importance Feature for Chinese Civil Aviation. J. King Saud Univ.-Comput. Inf. Sci. 2023, 35, 101856. [Google Scholar] [CrossRef]
Abu-Salih, B. Domain-Specific Knowledge Graphs: A Survey. J. Netw. Comput. Appl. 2021, 185, 103076. [Google Scholar] [CrossRef]
Yan, J.; Wang, C.; Cheng, W.; Gao, M.; Zhou, A. A Retrospective of Knowledge Graphs. Front. Comput. Sci. 2018, 12, 55–74. [Google Scholar] [CrossRef]
Wiegmann, D.A.; Shappell, S.A. The Human Factors Analysis and Classification System (HFACS). In A Human Error Approach to Aviation Accident Analysis; Routledge: London, UK, 2017; pp. 45–71. [Google Scholar] [CrossRef]
Commercial Aviation Safety Team/International Civil Aviation Organization (CAST/ICAO) Common Taxonomy Team (CICTT). ICAO Phase of Flight Definition and Usage Notes (1.3). 2013. Available online: http://www.intlaviationstandards.org/Documents/PhaseofFlightDefinitions.pdf (accessed on 15 December 2023).
Commercial Aviation Safety Team/International Civil Aviation Organization (CAST/ICAO) Common Taxonomy Team (CICTT). Aviation Occurrence Categories Definition and Usage Notes (4.8). 2021. Available online: https://www.wyvernltd.com/wp-content/uploads/2022/04/OccurrenceCategoryDefinitions2021.pdf (accessed on 15 December 2023).
Aviation Safety Network. Available online: https://aviation-safety.net/about/ (accessed on 1 October 2023).

Figure 1. Process of knowledge graph building and application.

Figure 2. Concepts in ACAKG ontology.

Figure 3. Operating type distribution of accidents.

Figure 4. Schema layer and data layer of aviation accident mapping.

Figure 5. Distribution of causal relationship intimacy. (a) Active Risk Level. (b) Passive Risk Level.

Figure 6. The type distribution proportions of aviation accident causation. (1) Active type distribution ratio. (2) Passive type distribution ratio. (a1) Ground phase. (a2) Ground phase. (b1) Take-off/landing phase. (b2) Take-off/landing phase. (c1) Level flight phase. (c2) Level flight phase. (d1) Maneuvering phase. (d2) Maneuvering phase.

Figure 7. Correlation types of accident causal factors. (a) Ground phase. (b) Take-off/landing phase. (c) Level flight phase. (d) Maneuvering phase.

Figure 8. Correlation between aviation accident causation type and flight phase. (a) Ground phase. (b) Take-off/landing phase. (c) Level flight phase. (d) Maneuvering phase.

Figure 9. Consequences of accidents at different phases. (a) Direct accident consequence. (b) Indirect accident consequence.

Figure 10. The impact curves of mitigating causal factors at different phases on the severity of accident consequences. (a) Direct impact. (b) Indirect impact.

Table 1. Causation and frequency.

Type of Accident Causation	Number	Accident Causation	Definition of Accident Causation	Frequency of Each Causation
H	H01	Pilot was under-skilled	Errors resulting from the pilot’s lack of skills and experience.	18
H	H02	Crew were under-skilled	Errors resulting from the crew’s insufficient skills and experience or related factors. (In this article, “crew” refers specifically to other members of the aircraft’s flight team, excluding the pilot.)	16
H	H03	Pilot’s fault	Errors resulting from the pilot during the execution of proficient tasks, which are associated with procedures, training, or proficiency levels.	153
H	H04	Crew’s fault	Errors resulting from the crew during the execution of proficient tasks, which are associated with procedures, training, or proficiency levels.	92
H	H05	Ground personnel’s fault	Errors resulting from ground personnel during the execution of proficient tasks, which are associated with procedures, training, or proficiency levels.	33
H	H06	Pilot’s violation	Deliberate non-compliance by the pilot with regulations and procedures designed to ensure flight safety.	39
H	H07	Crew’s violation	Deliberate non-compliance by the crew with regulations and procedures designed to ensure flight safety.	36
H	H08	Pilot lost situational awareness	Errors arising from misjudgments of distance, altitude, visibility, and other factors when the pilot’s perception does not align with the actual conditions.	12
H	H09	Crew lost situational awareness	Errors arising from misjudgments of distance, altitude, visibility, and other factors when the crew’s perception does not align with the actual conditions.	11
H	H10	Pilot had obstructed vision	The pilot’s line of sight is obstructed due to factors such as weather conditions and terrain features.	13
H	H11	Pilot lacked good physical and mental performance	The pilot’s performance is adversely affected due to suboptimal physical or psychological conditions.	27
H	H12	Crew lacked good physical and mental performance	The crew’s performance is adversely affected due to suboptimal physical or psychological conditions.	16
H	H13	Ground personnel lacked good physical and mental performance	The ground personnel’s performance is adversely affected due to suboptimal physical or psychological conditions.	0
A	A01	Airframe failure	The occurrence of failures in specific components of the airframe.	47
A	A02	Engine failure	The occurrence of failures in the engine.	27
A	A03	Airborne equipment system failure	The occurrence of failures in the airborne equipment system.	27
A	A04	Lack of fuel	A deficiency in fuel supply due to exhaustion, leakage, contamination, or other contributing factors.	6
A	A05	Aircraft design defect	A deficiency in the design of the aircraft.	12
A	A06	Mechanical problem	Other mechanical problems exist.	0
E	E01	Weather condition	Adverse weather conditions, such as high winds, hailstorms, lightning strikes, etc.	55
E	E02	Geographical condition	Geographical conditions, including mountainous terrain, high altitudes, etc.	9
E	E03	External environment—animal	The appearance of wild animals.	11
E	E04	External environment—insufficient light	Insufficient illumination caused by various factors.	1
E	E05	Airport environment	Issues arising in airport infrastructure, runway environments, etc.	24
E	E06	Air environment anomaly	Abnormalities in the aerial environment.	1
E	E07	Social relations	Issues pertaining to social relationships, including family, colleagues, etc.	0
E	E08	Political environment	Unfavorable political environment.	0
E	E09	Economic environment	Unfavorable economic environment.	0
M	M01	People management	Problems with people management.	52
M	M02	Work organization	Problems with work organization.	58
M	M03	Safety fundamentals	Problems with safety fundamentals.	56

Table 2. Accident phase and frequency.

Initial Number	Phase	Combined Number	Combined Phase	Frequency of Each Phase
P01	Standing	PC01	Ground phase	30
P02	Pushback/towing	PC01	Ground phase	30
P03	Taxi	PC02	Take-off/landing phase	292
P04	Take-off
P09	Landing
P06	En route	PC03	Level flight phase	92
P08	Approach	PC03	Level flight phase	92
P05	Initial climb	PC04	Maneuvering phase	21
P07	Maneuvering	PC04	Maneuvering phase	21

Table 3. Mortality and weighted injury quantification.

Number	Injury Degree	Weight	Number	Damage Degree	Weighting
RH01	0	0	RA01	None	0
RH02	1	0.1	RA02	Minor	0.3
RH03	2–5	0.3	RA03	Substantial	0.7
RH04	6–14	0.7	RA04	Destroyed	1
RH05	more than 15	0.9	RA05	Damaged Beyond Repair	1
			RA06	Missing	1
			RA07	Unknown	0.1

Table 4. Aviation accident types and frequency.

Number	Accident Type	Frequency of Each Type
T01	Runway mishap	208
T02	Collision mishap	164
T03	Abnormal landing	149
T04	Aircraft fire	2
T05	Loss of control	57
T06	Security-related accident	2
T07	Rejected take-off	10
T08	System failure	68

Table 5. Description of aviation accidents.

Accident Number	Probable Cause	Phase	Type	Fatalities	Aircraft Damage	Knowledge Triples
2018-12	CAUSES: Metal fatigue leading to the breakage and dislocation of the hydraulic fitting that gave rise to loss of hydraulic fluid, resulting in the loss of steering wheel control and braking during taxing of the aircraft. CONTRIBUTING FACTORS Human factor: Delay in detecting the four hydraulic failure lights, especially by the pilot monitoring, during the landing roll and early part of taxing into the apron, owing to the conspicuous location of the lights when the pilot’s eyes usually remain outside the co*ckpit.	Taxi	Runway mishap	0	Substantial	(A01, result-in, C08); (H03, result-in, C08); (A01, and, H03); (C08, result-in, C01); (A01, phase-is, P03); (H03, phase-is, P03); (A01, type-is, T01); (H04, type-is, T01); (A01, fatality consequence-is, RH01); (H03, fatality consequence-is, RH01); (A01, aircraft consequence-is, RA03); (H03, aircraft consequence-is, RA03).
2021-104	The first officer’s (FO’s) improper decision to attempt to salvage an unstabilized approach by executing a steep left turn to realign the airplane with the runway centerline, and the captain’s failure to intervene after recognizing the FO’s erroneous action, while both ignored stall protection system warnings, which resulted in a left-wing stall and an impact with terrain. Contributing to the accident was the FO’s improper deployment of the flight spoilers, which decreased the airplane’s stall margin; the captain’s improper setup of the circling approach; and the flight crew’s self-induced pressure to perform and poor crew resource management, which degraded its decision-making ability.	Approach	Collision	6	Destroyed	(H11, result-in, H03); (M02, result-in, H04); (H11, and, M02); (H03, result-in, T02); (H11, phase-is, P08); (H03, phase-is, P08); (M02, phase-is, P08); (H11, type-is, T02); (H03, type-is, T02); (M02, type-is, T02); (H11, fatality consequence-is, RH04); (H03, fatality consequence-is, RH04); (M02, fatality consequence-is, RH04); (H11, aircraft consequence-is, RA04); (H03, aircraft consequence-is, RA04); (M02, aircraft consequence-is, RA04).
2022-74	The pilot’s improper fuel planning, which resulted in a total loss of engine power due to fuel exhaustion, an emergency landing, and runway excursion.	Landing	Abnormal landing; Runway mishap	0	Substantial	(H03, result-in, A04); (A04, result-in, A02); (A02, result-in, C03); (C03, result-in, C01); (H03, phase-is, P09); (A04, phase-is, P09); (A02, phase-is, P09); (H03, type-is, T01); (A04, type-is, T01); (A02, type-is, T01); (H03, type-is, T03); (A04, type-is, T03); (A02, type-is, T03); (H03, fatality consequence-is, RH01); (A04, fatality consequence-is, RH01); (A02, fatality consequence-is, RH01); (H03, aircraft consequence-is, RA03); (A04, aircraft consequence-is, RA03); (A02, aircraft consequence-is, RA03).

Table 6. Effect size index of accident causation.

Number	Mean	Difference	Difference 95% Confidence Interval	df	Standard Deviation	Cohen’s d
H01	0.097	0.097	−0.013~0.207	30	0.301	0.322
H02	0.000	0.000	0.000~0.000	30	0.000	0.000
H03	0.032	0.032	−0.034~0.098	30	0.180	0.180
H04	1.258	1.258	0.513~2.004	30	2.033	0.619
H05	1.097	1.097	0.320~1.874	30	2.119	0.518
H06	0.258	0.258	0.027~0.489	30	0.000	0.000
H07	0.065	0.065	−0.027~0.156	30	0.250	0.258
H08	0.129	0.129	0.004~0.254	30	0.341	0.379
H09	0.000	0.000	0.000~0.000	30	0.000	0.000
H10	0.194	0.194	0.018~0.369	30	0.477	0.405
H11	0.032	0.032	−0.034~0.098	30	0.180	0.180
H12	0.065	0.065	−0.067~0.196	30	0.359	0.180
H13	0.194	0.194	−0.006~0.393	30	0.543	0.357
A01	0.968	0.968	0.322~1.613	30	1.760	0.550
A02	0.194	0.194	0.018~0.369	30	0.477	0.405
A03	0.161	0.161	0.024~0.298	30	0.374	0.431
A04	0.097	0.097	−0.101~0.294	30	0.539	0.180
A05	0.000	0.000	0.000~0.000	30	0.000	0.000
A06	0.000	0.000	0.000~0.000	30	0.000	0.000
E01	0.000	0.000	0.000~0.000	30	0.000	0.000
E02	0.000	0.000	0.000~0.000	30	0.000	0.000
E03	0.000	0.000	0.000~0.000	30	0.000	0.000
E04	0.032	0.032	−0.034~0.098	30	0.180	0.180
E05	0.000	0.000	0.000~0.000	30	0.000	0.000
E06	0.000	0.000	0.000~0.000	30	0.000	0.000
E07	0.000	0.000	0.000~0.000	30	0.000	0.000
E08	0.000	0.000	0.000~0.000	30	0.000	0.000
E09	0.000	0.000	0.000~0.000	30	0.000	0.000
M01	0.032	0.032	−0.034~0.098	30	0.180	0.180
M02	0.129	0.129	−0.028~0.286	30	0.428	0.302
M03	0.000	0.000	0.000~0.000	30	0.000	0.000

Table 7. Effect size index of accident phase.

Number	Mean	Difference	Difference 95% Confidence Interval	df	Standard Deviation	Cohen’s d
PC01	1.774	1.774	0.934~2.614	30	2.291	0.775
PC02	28.677	28.677	13.013~44.342	30	42.705	0.672
PC03	11.161	11.161	6.510~15.813	30	12.681	0.880
PC04	1.516	1.516	0.796~2.237	30	1.964	0.772

Table 8. Effect size index of accident.

Number	Mean	Difference	Difference 95% Confidence Interval	df	Standard Deviation	Cohen’s d
T01	1.871	1.871	−0.398~4.140	30	6.185	0.303
T02	2.129	2.129	0.641~3.617	30	4.056	0.525
T03	1.710	1.710	0.276~3.143	30	3.909	0.437
T04	0.194	0.194	0.018~0.369	30	0.477	0.405
T05	1.452	1.452	0.424~2.480	30	2.803	0.518
T06	0.032	0.032	−0.034~0.098	30	0.180	0.180
T07	0.000	0.000	0.000~0.000	30	0.000	0.000
T08	0.161	0.161	−0.005~0.328	30	0.454	0.355

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Causation Correlation Analysis of Aviation Accidents: A Knowledge Graph-Based Approach (2024)

FAQs

What is the accident causation theory in aviation? ›

The Swiss cheese model of accident causation illustrates that, although many layers of defense lie between hazards and accidents, there are flaws in each layer that, if aligned, can allow the accident to occur.

Discover More Details ›

What are the causal factors of aviation accident? ›

Accident Cause Factors

Failure to obtain and/or maintain flying speed. Failure to maintain direction control. Improper level off. Failure to see and avoid objects or obstructions.

Know More ›

Which of the following is a causal factor in many aviation accidents? ›

Pilot error is the number one cause of aviation accidents.

Learn More ›

How many aviation accidents are caused by human error? ›

“Statistics show that up to 80 percent of all aviation accidents can be attributed to human error.” Statistics show that up to 80 percent of all aviation accidents can be attributed to human error. The most dangerous times include takeoff and landing and the time periods before and after these events.

Learn More Now ›

What is an example of accident causation theory? ›

Unsafe acts by individuals, coupled with mechanical or physical hazards present in the environment, create a precarious situation. Examples include failure to follow proper procedures and the presence of equipment malfunctions or hazardous conditions.

Keep Reading ›

What are the three levels of accident causation? ›

Abdelhamid and Everett (2000) developed the Accident Root Causes Tracing Model (ARCTM), aimed at supporting the investigation of the causes of accidents considering three classes of sources: (i) failure to identify unsafe conditions previous to the task starting; (ii) decision to proceed with the task despite ...

What are 3 of the most common causes of airplane accidents? ›

Six Most Common Causes of Airplane Crashes

Pilot Error. Pilot error is often noted as the leading cause of aviation accidents. ...
Mechanical Failure. ...
Weather Conditions. ...
Air Traffic Control Errors. ...
Bird Strikes. ...
Runway Incursions.

Feb 1, 2024

Read The Full Story ›

What is the most common cause of general aviation accidents? ›

The main cause of fatal general aircraft accidents include: The Pilot's loss of control of the aircraft while in close proximity to the runway. Yes, the takeoff and landing have always been and remains the most dangerous parts of the flight.

Tell Me More ›

What are the 3 factors that cause accidents? ›

Negligence---Failure to observe basic safety rules of instructions or to maintain equipment. Anger/Temper--Causes a person to become irrational and to disregard common sense. Hasty Decision--Acting before thinking can lead people to make hazardous “Shortcuts.”

What is one of the most common causes of accidents? ›

Various national and international researches have found these as most common behavior of Road drivers, which leads to accidents.

Over Speeding: Most of the fatal accidents occur due to over speeding. ...
Drunken Driving: Consumption of alcohol to celebrate any occasion is common. ...
Distraction to Driver: ...
Red Light jumping:

Read On ›

What are the three key factors that define an aircraft accident? ›

Key Takeaways. Powerplant (engine) failure, loss of control on the ground, and loss of control in the air account for the vast majority of general aviation accidents. The largest cause of fatal accidents is loss of control in-flight.

View Details ›

What is the biggest killer in aviation? ›

The top 10 leading causes of fatal general aviation accidents

Loss of control in-flight (LOC-I) ...
Controlled flight into terrain (CFIT) ...
System component failure – power plant (SCF-PP) ...
Fuel-related problems. ...
Unknown or undetermined. ...
System component failure – non-power plant (SCF-NP) ...
Unintended flight in IMC (UIMC)

More items...

Mar 11, 2020

Tell Me More ›

What percentage of aviation accident causation do human errors generally account for? ›

Research by the National Aeronautics and Space Administration into aviation accidents has found that 70% involve human error.

Are 90% of crashes caused by human error? ›

Human Error in Car Crashes

According to a study by the National Highway Traffic Safety Administration (NHTSA), an estimated 94% of motor vehicle accidents were caused by driver error.

View Details ›

What is the principle of accident causation? ›

Human factors theory in accident causation suggests that accidents happen due to human error or behavior, and that typically these can be attributed to factors such as skill deficiencies, violations of rules or procedures, poor decision-making, and lack of situational awareness.

Show Me More ›

What is Heinrich's theory of accident causation? ›

Heinrich's law is based on probability and assumes that the number of accidents is inversely proportional to the severity of those accidents. It leads to the conclusion that minimising the number of minor incidents will lead to a reduction in major accidents, which is not necessarily the case.

Discover More ›

What are the three components of the systems theory of accident causation? ›

The systems theory of accident causation views a situation in which an accident may occur as a system comprised of the following components: oPerson (host); Machine (agency); Environment. Likelihood of an accident occurring is determined by how these components interact.

See Details ›

What is the reason's model of accident causation? ›

Reason's (1997) model classifies factors con- tributing to accidents into three domains: Organisational/systems, local workplace and unsafe acts. In doing so, the model moves the blame from human error to the environment in which humans work.

Read The Full Story ›