Research on Human-Error Factors of Civil Aircraft Pilots Based On Grey Relational Analysis

In consideration of the situation that civil aviation accidents involve many human-error factors and show the features of typical grey systems, an index system of civil aviation accident human-error factors is built using human factor analysis and classification system model. With the data of accidents happened worldwide between 2008 and 2011, the correlation between human-error factors can be analyzed quantitatively using the method of grey relational analysis. Research results show that the order of main factors affecting pilot human-error factors is preconditions for unsafe acts, unsafe supervision, organization and unsafe acts. The factor related most closely with second-level indexes and pilot human-error factors is the physical/mental limitations of pilots, followed by supervisory violations. The relevancy between the first-level indexes and the corresponding second-level indexes and the relevancy between second-level indexes can also be analyzed quantitatively.


Introduction
With the rapid technological development, the number of civil aviation accidents caused by the faults of the electromechanical equipment has declined sharply. To the contrary, aviation accidents caused by inappropriate pilot operation is increasing year by year. Human-error factors have accounted for 60-80% of the whole accident causes, becoming one of the main causes of civil aviation accidents [1]. By now, the research on human-error factors mostly concentrates on qualitative analysis, such as cause & effect analysis, event tree analysis, simulated investigation analysis, fishbone diagram analysis [2][3][4][5], and little improvement in the research on quantitative analysis has been realized. Quantitative research on cockpit human-error factors needs a great number of samples for mathematical statistics analysis, with high requirements for data integrity and data accuracy, and has not shown the relationship between human error and safety factors clearly. To reduce the aviation accident rate in China, it is of great importance to use the big data of aviation accidents to analyze the occurrence rules of human-error factors for higher pilot efficiency.
Based on the method of grey relational analysis, this paper researches the human-error factors of pilots in civil aircraft cockpits. There are many human-error factors that affect the task execution, the interactive environment and the interactive process are very complex and the relationship between human-error factors and other factors is still in an unclear grey system. Using grey relational analysis for the human-error factor research can remedy the limitations of traditional statistics analysis and provide theoretical basis for prohibiting and reducing human errors and mitigating flight risks.

grey relational analysis
Grey correlation means uncertain correlation between objects, factors or between factors and main acts. The main idea is to judge how closely those objects are related based on geometry similarity between referential data sequences and comparative data sequences, which shows the relevance between curves. Finding out the main factor that affects the target values by calculating the relational degree is an easy and practical system analysis method, which can make up the shortages of traditional mathematical statistics method, such as demand of large samples, large amount of calculation and inconsistence between the results of quantitative analysis and qualitative analysis. The calculation process of grey relational analysis is as follows: 1)Determine the referential sequence, denoted as: 2)Determine the comparative sequence, denoted as: 3)Calculate the correlation coefficient between the referential sequence and the comparative sequence, denoted as: ρ  (0,1) means the identification coefficient, and normally we take the value of 0.5. 4)Calculate the relational degree, denoted as: i  means the relational degree between the sequence i X and the sequence 0 X , and the higher value means the closer relationship. 5)Rank the relevancies according to their values, and find out the factors that affect the main act.

Construction of human error index system
Contingency exists in the occurrence of civil aviation accidents, and both dominant factors and recessive factors can result in human error, so it is hard to quantify the relationship between human-error factors. Based on the Human Factors Analysis and Classification System (HFACS) [6], we can classify the factors that affect the piloterror incidents as 4 first-level indexes (organizational influences, unsafe supervision, preconditions for unsafe acts and unsafe acts) and 18 second-level indexes. According to the cognitive process of aircraft operation, we can build the human error index system (see Fig 1) by analyzing the degree to which these second-level factors interact with each other.   [7]. Based on HFACS model, analyze the human factors in those 88 accidents and find out that there are 347 human-error events, then summarize the analysis results for every year (see Table 1).

Grey relational analysis of human errors and first-level indexes
Set the amount of human error events (X0) in civil aviation accidents as the referential sequence, and set organizational influences (X1), unsafe supervision factors (X2), precondition for unsafe acts (X3) and unsafe acts (X4) as comparative sequences. Use the initial value method to nondimensionalize these sequences, and the results are as follows : T he results are shown in Table 2. Apparently, the precondition for unsafe acts is the main factor that affects human error in civil aviation accidents, rather than unsafe supervision or unsafe organization. Notably, inadequate organizational management typically involves operational processes such as inadequate or nonexistent procedures, directives, standards, and requirements, or in the case of commuter/on-demand operations, inadequate surveillance of operations.

Grey relational analysis of first-level indexes and second-level indexes
Set 'organizational influences' X1 as the referential sequence, and set 3 second-level indexes of the first-level index X1 as the comparative sequence. Use the same method as above and we can get the relational degrees, respectively, between 3 second-level indexes (inappropriate resource management, bad organizational environments, mistakes in operational process) and the organizational influences: The data shows that inappropriate resource management is the factor that has the most important effects on organizational influences. The reason why bad organizational environment has less impact may be that the accident reports are relatively simple. Some involved departments only include immediate causes and do not explore the root causes reflected by accidents. Organizational culture plays a key role in any flight accident.
The relational degrees between second-level indexes (inadequate supervision, planned inappropriate operations, failure to correct known problems and supervisory violations) and the first-level index 'unsafe supervision' are When considering pilot personal conditions, the factor 'skill-based errors' has the highest relational degree with unsafe acts, such as unintentionally touching control device that is irrelevant to the current task or operating the process in wrong order. The followed is violations which include not conducting instructions from air traffic control department, unauthorized approaches and so on. Decision errors and perceptual errors are also important factors that can lead to accidents, for example, making risky decisions when time is limited.

Grey relational analysis of human errors and second-level indexes
Set X0 as the referential sequence, and set all the secondlevel indexes as comparative sequences. Use the initial value method to calculate relational degrees and we can get the ranking: