1. INTRODUCTION
In practice, an oil tanker may either transport oil to refineries, or transport oil from refineries to oil purchasers' locations. The former is known as a crude oil tanker, whereas the latter is a product oil tanker. An oil tanker can carry anywhere from around 1,100 tons for product oil tankers to over 550,000 tons for ultra large crude oil tankers. Due to carrying large amounts of oil, an oil tanker accident may cause large oil spills. In addition to economic losses, this may result in environment pollution, leading to damage to ecosystems. Thus, oil tankers' safety issues require close attention.
To improve the safety of oil tankers, exploring the determining factors of marine accidents is necessary. In the relevant literature, many determining factors have been proposed, including human factors (e.g. Hetherington et al., Reference Hetherington, Flin and Mearns2006), safety management (e.g. Havold, Reference Havold2010), navigation errors (e.g. Ismail and Karim, Reference Ismail and Karim2013) and the natural environment (e.g. Ismail and Karim, Reference Ismail and Karim2013). However, most of those studies focused on post event investigations. In practice, the concept of advance prevention should be more important. An adequate preventative measure can reduce accidents and save losses for organisations (Kontovas and Psaraftis, Reference Kontovas and Psaraftis2009).
For advance prevention, risk assessment of accidents is the initial and often the most important step (Chang et al., Reference Chang, Xu and Song2014). For improving maritime safety, the International Maritime Organization (IMO) developed a Formal Safety Assessment (FSA) framework to reduce the risks of marine accidents (IMO, 2013). FSA is a rational and systematic process for assessing the risks associated with shipping activity and for evaluating the costs and benefits of the IMO's options for reducing these risks (Psaraftis, Reference Psaraftis2012). In the FSA process, a risk matrix is the main analysis tool for risk assessment of accidents. A risk matrix displays the basic properties, “consequence” and “likelihood” of an adverse Risk Factor (RF) and the aggregate notion of risk by means of a graph (Duijm, Reference Duijm2015). In the traditional risk matrix, both the consequence and likelihood are measured by a category scale such as negligible, serious and catastrophic for the consequence measurement and almost impossible, probable and often for the likelihood measurement. In practice, such a discrete scale measurement may limit its applications (e.g. Cox, Reference Cox2008; Smith et al., Reference Smith, Siefert and Drain2009; Levine, Reference Levine2012). Thus, to improve the performance of risk management, a risk matrix with a continuous scale may be considered (Duijm, Reference Duijm2015).
This paper is aimed at the risk assessment of operational safety for oil tankers. Specifically, this paper proposes a revised risk matrix with a continuous scale to assess the risk. In this paper, based on the FSA framework, the RFs of operational safety for oil tankers are first identified. A fuzzy Analytical Hierarchical Process (AHP) model is then conducted to weight those RFs, by which the revised risk matrix is constructed to classify the RFs. Finally, the Chinese Petroleum Corporation (CPC) oil tanker fleet in Taiwan is empirically investigated to validate the research model. The rest of this paper is organized as follows. Section 2 explains the literature reviews. Section 3 describes the research method in this paper. The results are then discussed in Section 4. Finally, some general conclusions and limitations for further research are given.
2. LITERATURE REVIEW
In this section, the traditional risk matrix is first introduced. The relevant literature related to improving the traditional risk matrix is then reviewed. Finally, for identifying the RFs of operational safety for oil tankers, the relevant studies on the safety factors of shipping operations are explored
2.1. Risk Matrix
Risk is often expressed in terms of a combination of the consequences of an event together with the associated likelihood of its occurrence (ISO, 2009). Risk consequence is regarded as the loss or severity to an organisation if a risk event occurs (NPSA, 2008). Traditionally, consequence is generally described by a category scale and rated, such as “insignificant”, “minor”, “moderate”, “major”, and “catastrophic” (Chang et al., Reference Chang, Xu and Song2014). Risk likelihood is defined as the probability of the event occurring and conventionally described by a category scale. such as “rare”, “unlikely”, “possible”, “likely” and “almost certain” (NPSA, 2008; Chang et al., Reference Chang, Xu and Song2014).
For risk assessment of an event, the risk matrix is one of the most popular tools. A risk matrix facilitates assigning a discrete risk category to each combination of consequence and likelihood (Duijm, Reference Duijm2015). In a traditional risk matrix with m consequence categories and n likelihood categories, one can discriminate m × n different risk categories. It is normal to divide the cell of the risk matrix in areas with fewer categories, often by using colours, such as green, yellow and red, to represent Low risk (L), Medium risk (M) and High risk (H), or by deriving a risk score based on the ordinal values of consequence and likelihood, such as the multiplication of the ordinal numbers of the likelihood and consequence category. For example, Figure 1 is a 4 × 4 risk matrix with 16 risk categories which are classified as three types of risk scales by risk scores. The risk score for each category is shown in the parentheses. The risk categories with risk scores 1 ~ 2 are identified as the low risk scale (L); 3 ~ 9 as the Medium risk scale (M); 12 ~ 16 as the High risk scale (H).
Fig. 1. A traditional 4 × 4 risk matrix.
2.2. Limitations of the traditional risk matrix
Although the traditional risk matrix has been applied widely, there are some limitations to its practical applications. Duijm (Reference Duijm2015) reviewed relevant studies and proposed six limitations of the risk matrix in practical applications, of which most studies focus on the following three issues:
-
1 Consistency between the risk matrix and quantitative measures, and, as a consequence, the appropriateness of decisions based on risk matrices (Cox, Reference Cox2008; Levine, Reference Levine2012).
-
2 The subjective classification of consequence and probability (Smith et al., Reference Smith, Siefert and Drain2009).
-
3 The definition of risk scores and its relation to the scaling of the categories (linear or logarithmic) (Franks and Maddison, Reference Franks and Maddison2006; Levine, Reference Levine2012).
To improve the above limitations, some risk matrices based on a continuous probabilityconsequence map were thus proposed (e.g. Meng et al., Reference Meng, Weng and Qu2010; Arunraj et al., Reference Arunraj, Mandal and Maiti2013; Chang et al., Reference Chang, Xu and Song2014). Differing from the traditional risk matrix, these matrices proposed continuous risk scales to identify RFs, and several nonlinear curves based on the product of the risk probability and the risk consequence are employed to divide the risk scale into three types of risks: low risk, medium risk and high risk. These risk matrices improve some limitations of traditional risk matrices. However, in these studies, chosen experts need to score each RF directly based on their subjective perceptions. In practice, this could reduce the measurement validities of the experts, leading to a decrease in the assessment performance of the risk matrix. To improve this limitation, a revised risk matrix with a continual scale based on a relatively comparable scoring system is thus proposed in this paper.
2.3. Shipping operations safety factors
In the relevant studies, the safety factors for shipping operations can be classified into five categories: human factor, machinery condition, ship management, organisation management and natural environment.
2.3.1. Human factor
In the relevant literature, most studies indicated human factors are the most significant determinant for shipping safety. For example, Hetherington et al. (Reference Hetherington, Flin and Mearns2006) investigated the human effects on shipping safety. The result indicated the work safety concept originated from workers' safety knowledge, and the determinants of work safety include fatigue, stress and the work environment. Havold (Reference Havold2010) indicated that safety culture has significant effects on shipping safety for oil tankers which include the management styles of shipping carriers, work stresses, safety knowledge and crews' perceived fatalism. The relevant studies also indicate that crews' work characteristics including professional skills and work attitudes, significantly affect the safety of ship navigation in ports (Hsu, Reference Hsu2012). Crews' safety knowledge and work concentration are the most significant factors affecting the safety of product oil tankers in costal shipping (Hsu, Reference Hsu2015).
2.3.2. Machinery condition
Machinery condition is defined as the condition of a ship's machinery, facilities and equipment for work safety. Relevant studies indicated the maintenance of personal safety equipment may significantly reduce the threats to workers' safety (Gordor et al., Reference Gordor, Flin and Meaens2005). Vessels in poor condition (Liu et al., Reference Liu, Liang, Su and Chu2006) may lead to marine disasters. Improper operation, machinery failure (Hsu, Reference Hsu2012; Reference Hsu2015), and the type, size, age and the condition of a vessel at the time of an accident are significant determinants of ship loss (Kokotos and Smirlis, Reference Kokotos and Smirlis2005). The conditions of the communication equipment and personal safety equipment and maintenance operations significantly affect aviation safety (Chang and Wang, Reference Chang and Wang2010). Furthermore, according to the SOLAS Convention a ship must set up rescue equipment for emergencies, such as fire pumps, generators, air compressor and lifeboats etc., and these devices must be available at any time.
2.3.3. Ship management
Ship management is defined as the implementation of the safety management on board a ship. Lu and Tsai (Reference Lu and Tsai2008) examined the effects of safety cultures on container shipping safety, which include management safety practices, supervisor safety practices, safety attitude, safety training, job safety, and co-workers' safety practices. The results indicated that job safety has the most significant effect on vessel accidents, followed by management safety practices and safety training. Furthermore, relevant studies also indicate that educational training has a great effect on safety in the shipping industry (Hetherington et al., Reference Hetherington, Flin and Mearns2006). Operational procedures, regulations and performance assessment of training are significant determinants of occupational accidents (Fabiano et al., Reference Fabiano, Curro, Reverberi and Pastorino2010).
2.3.4. Organisation management
Organisation management is defined as the safety policy and management system of a shipping company. Relevant studies revealed the attitude of the carrier is the most significant factor on safety management, followed by crew work stress and crew safety knowledge (Havold, Reference Havold2005). The carrier's safety policy and safety management have significantly positive effects on crew safety behaviours in container shipping (Lu and Yang, Reference Lu and Yang2010). Organisational culture may influence the performance of safety management in aircraft maintenance technicians (Chang and Wang, Reference Chang and Wang2010). An unfair reward system would result in staff dissatisfaction, leading to decreased work and safety performance (Chang and Wang, Reference Chang and Wang2010).
3. RESESARCH METHOD
In this paper, the Risk Factors (RFs) for operational safety of oil tankers are first identified. A fuzzy AHP approach is then employed to weight those RFs, including both weights of consequence and likelihood. Based on those weights, a revised risk matrix with a continuous scale is finally proposed to assess the risk classes of the RFs.
3.1. Identification of risk factors
Based on the relevant literature about safety factors of shipping operations, four constructs of risk factors (RFs) were identified as follows.
3.1.1. Human factor (HF)
HF is defined as crew safety knowledge and work attitudes, including safety knowledge, work concentration, self-regulation and perceived fatalism etc. (Hetherington et al., Reference Hetherington, Flin and Mearns2006; Havold, Reference Havold2010; Hsu, Reference Hsu2012; Reference Hsu2015).
3.1.2. Machinery condition (MC)
MC is defined as the condition of a ship's machinery, equipment and safety facilities, including the ship's engines, personal safety equipment, safety monitoring systems, warning marking system, etc. (Gordor et al., Reference Gordor, Flin and Meaens2005; Liu et al., Reference Liu, Liang, Su and Chu2006; Chang and Wang, Reference Chang and Wang2010; Hsu, Reference Hsu2012).
3.1.3. Ship management (SM)
SM is defined as the implementation of the safety system and the crew perceived safety climate on board the ship, including the implementation of safety procedures, safety drills, assessment of safety training, safety culture, etc. (Hetherington et al., Reference Hetherington, Flin and Mearns2006; Lu and Tsai, Reference Lu and Tsai2008; Fabiano et al., Reference Fabiano, Curro, Reverberi and Pastorino2010).
3.1.4. Organisational management (OM)
OM is defined as the safety policy and management system of ship carriers, including safety operational procedures, punishment systems, staffing of tasks, and the performance assessment of work safety (Havold, Reference Havold2005; Chang and Wang, Reference Chang and Wang2010; Lu and Yang, Reference Lu and Yang2010).
Based on the above definitions, a two-layer hierarchical structure of RFs for oil tanker operations was first constructed. To improve the practical validity of the RFs, two practical experts were then invited to revise those RFs and check if any important RFs were missed. Further, they also checked the independences between those RFs. The two experts were a shipmaster and a chief engineer. Both came from the CPC oil tank fleet and have over 20 years of experience on board ship. After several rounds of discussion and revision, including combining two items and adding one new item, the final hierarchical structure of the RFs, shown in Table 1, contains four constructs of RFs for the first layer and 16 RFs for the second layer.
Table 1. The risk factors (RFs) and hierarchical structure for oil tankers.
3.2. Questionnaire design
In this paper, an AHP questionnaire with a nine-point rating scale was designed to measure the subject's perceived likelihood and consequence on each RF respectively. Based on the hierarchical structure of RFs in Table 1, an AHP survey with five criteria and 16 sub-criteria was created. To validate the scale, the survey was then pre-tested by the previous two experts who checked the survey. Based on the results of pre-tests, some statements in the survey were revised.
3.3. Research sample
Since this paper employs the oil tanker fleet of CPC (Chinese Petroleum Corporation) as an empirical study to validate the proposed model, the top-four crews working in CPC's oil tanker fleet were surveyed, including shipmasters, chief officers, chief and second engineers. To improve the survey validity, an assistant was assigned to help the subjects to fill out the questionnaire. Currently, CPC's fleet has seven oil tankers with a total of 28 top-four crews, in which six vessels with 24 top-four crews were surveyed in this study. For each of the surveyed sample, a Consistency Index (CI) was first employed to test its consistency. The results indicated four samples with CI > 0.1 were highly inconsistent (Saaty, Reference Saaty1980). Thus, those four respondents were asked to revise their survey. This step was performed repeatedly until all surveys were consistent. The profiles of the validated 24 respondents' features are shown in Table 2. The result shows that all subjects have at least five years of work experience (with 75% of subjects over ten years) in their company. Note, the remarkable qualifications of the respondents endorse the reliability of the survey findings.
Table 2. Profiles of the respondents.
3.4. The weights of risk factors
In this paper, 24 pairwise comparison matrices are obtained for each comparison of RFs in each layer. In the past, most relevant studies employed arithmetic mean or geometric mean to present multiple subjects' opinions. However, those two means are sensitive to extreme values. Thus, a fuzzy number is considered to integrate the 24 subjects' perceptions in this paper. Firstly, the geometric mean was employed to represent the consensus of respondents (Saaty, Reference Saaty1980; Buckley, Reference Buckley1985). A triangular fuzzy number characterised with minimum, geometric mean and maximum of the measuring scores was then used to integrate the 24 pairwise comparison matrices into a fuzzy positive reciprocal matrix. Then, based on this matrix, a fuzzy AHP approach was employed to weight the RFs for both of the measurements of the respondents' perceived “likelihood” and “consequence”.
3.4.1. The fuzzy positive reciprocal matrix
Suppose
$\tilde{A}=[\tilde{a}_{ij}\rsqb _{n\times n}$
is a fuzzy positive reciprocal matrix where
$\tilde{a}_{ij} =[l_{ij}\comma \; m_{ij}\comma \; u_{ij}\rsqb $
is a triangular fuzzy number with
$$[l_{ij}, m_{ij}, u_{ij}] = \left\{\matrix{[1,1,1], \hfill &{\rm if}\,i = j \hfill; \cr [1/u_{ji},\,1/m_{ji},\,1/l_{ji}] \hfill &{\rm if}\,i \ne j. \hfill}\right. $$
For ease of exposition, let
$A^{\lpar k\rpar }=[a_{ij}^{\lpar k\rpar }\rsqb _{n\times n}$
denote the pairwise comparison matrix with n RFs for the kth subject. Then, according to the above integration procedure, the 24 pairwise comparison matrices A
(k),
$k=1\comma \; 2\comma \; \ldots\comma \; 24$
, can be integrated into the following fuzzy positive reciprocal matrix:
$$\tilde{A}=[\tilde{a}_{ij}\rsqb _{n\times n}$$
where
$\tilde{a}_{ij} = \left[\min_{1\le k\le 30} \lcub a_{ij}^{\lpar k\rpar }\rcub \comma \; \lpar \Pi_{k=1}^{30} a_{ij}^{\lpar k\rpar } \rpar^{1/30}\!\comma \; \ \max_{1\le k\le 30} \lcub a_{ij}^{\lpar k\rpar }\rcub \right \rsqb $
is a triangular fuzzy number, i = 1, 2, …, n and j = 1, 2, …, n. According to the arithmetic operations of fuzzy numbers (Kaufinami and Gupta (Reference Kaufinami and Gupta1991)), the fuzzy positive reciprocal matrix
$\tilde{A}=[\tilde{a}_{ij} \rsqb _{n\times n} $
can be expressed as follows:
$$\tilde{a}_{ij} =\left\{{\matrix{{[1,1,1],} & {{\rm if}\,i = j;}\cr{{({\tilde a}_{ji})}^{ - 1},} & {{\rm if}\,i \ne j.}\cr}}\right.$$
3.4.2. The consistency tests
Before calculating the weights of the RFs via the matrix Ã, an immediate problem is how to test the consistency of such a fuzzy positive reciprocal matrix. Buckley (Reference Buckley1985) conducted the consistency test for a fuzzy positive reciprocal matrix whose entries are trapezoid fuzzy numbers. He used the geometric means to defuzzify the fuzzy numbers and thus convert the fuzzy positive reciprocal matrix into a crisp matrix. Then the consistency test can be undertaken for the crisp matrix by the same method in an AHP. In this paper, the method of Buckley (Reference Buckley1985) is used to defuzzify the Ã. Consequently, the fuzzy entries
$\tilde{a}_{ij} =[l_{ij} \comma \; m_{ij} \comma \; u_{ij}\rsqb $
in the à can be defuzzified as:
$$a_{ij} =(l_{ij}\cdot m_{ij}\cdot m_{ij}\cdot u_{ij})^{1/4},\quad i=1,2,\ldots,n,\enspace j=1,2,\ldots,n$$
Generally, the following Consistency Index (CI) and Consistency Ratio (CR) are two indices used to test the consistency of a positive reciprocal matrix (Saaty, Reference Saaty1980):
$$\hbox{CI}=\displaystyle{{\lambda_{{\rm max}} -n}\over{n-1}}$$
and
$$\hbox{CR}=\displaystyle{{\hbox{CI}}\over{\hbox{RI}}}$$
where
$\lambda_{{\rm max}}$
is the maximum eigenvalue of the positive reciprocal matrix and n is the number of sub-criteria of the matrix. RI represents a Randomized Index shown in Table 3 (Sahin and Senol, Reference Sahin and Senol2015). Saaty (Reference Saaty1980) suggested that the
$\hbox{CR}\le 0.1$
is an acceptable range.
Table 3. The Randomized Index (RI).
The consistency tests of the positive reciprocal matrices for likelihood measurements are shown in the second row of Table 4. The third row in Table 4 shows the results of the consistency tests for consequence measurements. Since all the CI and CR indices in Table 4 are less than 0.1, all the positive reciprocal matrices in the sample data are consistent.
Table 4. The results of the consistency tests.
3.4.3. The weights of RFs
In this paper, we adopted the NGMR (Normalisation of the Geometric Mean of the Rows) method (Saaty, Reference Saaty1980) to determine the local weights of RFs in Ã, which is elaborated in the Appendix. Further, the global weights of the RFs can then be found by multiplying the low level of local weights of the RFs by their corresponding high level of global weights. The results are shown in Table 5 (for likelihood measurement) and Table 6 (for frequency measurement). For example, in Table 5, the local weight of HF1 is 20.25%, and its corresponding high level of global weight (i.e the HF construct's weight) is 12.00%. Then, the global weight of HF1 should be: 20.25% × 12.00% = 2.44%, shown in the last field of Table 5.
Table 5. The likelihood weights of risk factors (RFs).
Note: The boldfaced numbers present the RFs with higher weights.
Table 6. The consequence weights of risk factors (RFs).
Note: The boldfaced numbers present the RFs with higher weights.
3.5. The revised risk matrix
Obviously, a Risk Factor (RF) with higher likelihood weight and consequence weight should be a RF with higher risk. Based on this concept, a Risk Index (RI) is thus constructed by the product of consequence weight and likelihood weight (Cox, Reference Cox2008; Levine, Reference Levine2012; Montewka et al., Reference Montewka, Goerlandt and Kujala2014). Let
$\omega_{i}^{L} $
and
$\omega_{i}^{F} $
be the consequence weight and likelihood weight of ith RF respectively. Then, the Risk Index of ith RF is defined as
$$RI_{i} =\omega_{i}^{L} \times \omega_{i}^{F} ,\quad i=1,2,\ldots,n$$
Finally, the RI can be normalised as:
$$RI_{i} =\displaystyle{{\omega_{i}^{L} \times \omega_{i}^{F} }\over{\sum_{i=1}^n (\omega_{i}^{L} \times \omega_{i}^{F})}}\times 100\% ,\quad i=1,2,\ldots,n$$
Based on Equation (5) and the RFs' likelihood and consequence weights in Table 5 and Table 6, the RIs (Risk Indices) for each RF can be found in the last field of Table 7. The result indicates the RF with the highest risk is OM4 (8.82%), followed by HF2 (8.34%) and SM2 (8.22%).
Table 7. The classification of risk factors (RFs).
Note: The boldfaced numbers present the RFs with higher weights.
Based on the Risk Indices (RIs), a revised risk matrix with continuous curves is constructed to classify the RFs in this paper. The matrix is shown in Figure 2, in which the consequence weight is depicted on the x-axis and the likelihood weight on the y-axis. Based on Equation (6), three decreasing curves (C1, C2, and C3) with different RI means are created to divide the matrix into four quadrants which are named as E (Extreme risk), H (High risk), M (Medium risk) and L (Low risk). The first curve C1 with RI mean = 6.25% are obtained by averaging all the 16 RFs' RIs. This divides all the RIs into two groups. Group one contains five RFs: OM4, HF2, SM2, OM2 and OM3, by which the second curve C2 with RI mean = 8.19% is found by averaging their RIs. Likewise, averaging the rest of 11 RFs' RIs in the other group, we have the third curve C3 with RI mean = 5.37%. The results, shown in Figure 2, indicate three RFs (OM4, HF2 and SM2) are classified as: E, two RFs (OM2 and OM3) as: H, six RFs as: M, and five RFs as: L. In practice, for the RFs in the first two classes E and H, CPC's managers should pay more attention to improving their safety.
Fig. 2. The revised risk matrix.
4. DISCUSSION
4.1. The result of revised risk matrix
In the relevant literature the traditional risk matrix with a discrete scale measurement may limit its applications. Thus, some revised risk matrices with continuous scale were developed to improve on the limitations. However, in those studies, respondents need to score each RF directly based on their subjective perceptions. In practice, it may be difficult for respondents to score a RF precisely in such a directly scoring measurement. Relatively comparable scoring could be more objective for evaluating the RF's consequence and likelihood degrees. For example, it could be easier for respondents to compare which RF is more likely to occur in two RFs rather than to score each of the two RF's likelihoods directly. In this paper, based on a fuzzy AHP approach a relatively comparable scoring is proposed to weight the RFs, by which a revised risk matrix with continuous curves is then yielded to classify the RFs. Compared to the previous studies, the proposed risk matrix could increase the measurement validity of the respondents, leading to improved assessment performance of the risk matrix.
4.2. The result of empirical study
The empirical result shows that three RFs are classified as extreme risk: OM4 (Inadequate staffing for tasks), HF2 (Insufficient self-regulation) and SM2 (Failing to implement safety drills on board ship); and two RFs are classified as high risk: OM2 (Inadequate assessment system for safety performance) and OM3 (Inadequate reward-penalty system for safety performance). This result indicates Organisation management (OM), which contains three E-H risk classes of RFs (OM2, OM3 and OM4), is the RF construct with the highest risk. Based on the results, this paper further conducted post-interviews with the previous two experts who checked and pre-tested the survey, and proposes suggestions for improving the safety of oil tanker operations as follows.
4.2.1. Improving organisational management
The result indicates Organisational Management (OM) with three E-H class of RFs is the highest risk construct, including staffing of tasks (OM4), assessment system (OM2) and reward-penalty system (OM3) for safety performance. For OM4, this paper suggests CPC should perform a work study to measure crews' work contents regularly, including workload and work balance. In practice, overload is one of the main determinants of crews' fatigue, leading to increased work accidents. For OM2, this paper suggests the assessment system for safety performance should be practical and realistic, and the system should be revised regularly. As for OM3, the reward-penalty for crews' safety performance should be significant. In practice, an adequate reward-penalty system could effectively discipline crews' safety behaviour.
4.2.2. Improving crew literacy
Practically, for improving HF2 (crew self-regulation), carrier managers may focus on enhancing crew literacy. The post-interview indicates that crew motivation to work on board ships is weak in Taiwan, leading to ship carriers needing to employ more foreign crews. Due to differences in language, culture and lifestyle, those foreign crews have poorer self-regulation. This may result in decreasing their work and safety performance, meaning HF2 has a higher improvement priority. This paper suggests ship carriers may make a policy to cooperate with maritime schools to train students for recruit crews, by which ship carriers may decrease the number of foreign employees, and improve crew literacy.
4.2.3. Improving the performance of training
For improving SM2 (safety drills on board ship), carrier managers may construct a complete training system. For ships' crews, there are two types of training: license training and shipboard drills. The former, which is held at safety organisations on land, may increase the crew's advanced knowledge of safety, such as tanker accident features, identification, prevention, rescue etc, while the latter (drills) is held on board the ship. For example, each crew member needs to participate in an abandon ship drill and a fire drill on board each month. However, those drills must be implemented regularly and in a realistic manner, so they do not become a mere formality.
5. CONCLUSION
This article is aimed at the risk assessment of operational safety for oil tankers. In this paper, sixteen Risk Factors (RFs) were constructed for oil tanker operations. The result can provide a reference for relevant studies on oil tanker safety. Further, based on a fuzzy AHP approach, a revised risk matrix with a continuous scale was proposed to assess the RFs' risk classes. The revised risk matrix may improve the traditional risk matrix and provide a theoretical reference for methodological researches in risk assessment of accident.
For validating the practical application of the proposed model, CPC's oil tanker fleet in Taiwan was empirically investigated. The result identifies five Extreme (E) or High (H) Risk Factors (RFs). Based on this result, some management implications and suggestions are proposed for CPC. CPC is the biggest oil tanker carrier in Taiwan. The results may provide practical information for other oil tanker operators to make policies in improving their operational safety performance. In practice, the transport of product oil tankers is used in countries with coastlines, island countries, and even landlocked countries with wide rivers. Compared to the transport of tanker trucks, it could decrease operational cost significantly. However, operational safety is vulnerable. The result may provide useful information for those countries' oil companies in operating the transports of product oil tankers
In this paper, 24 crews from CPC's oil tanker fleet in Taiwan were empirically surveyed to validate the proposed model. For enhancing the validity of the questionnaire investigation, this paper adopted an interview survey instead of a mailed survey. Thus, the validity and reliability of the findings in this paper could be endorsed. However, for better confirmation of the empirical results, more representative samples may be necessary in future research. Furthermore, due to socio-cultural differences, the results of empirical study may not be applicable to other areas. However, the research model may offer a theoretical base from which to develop a new one to fit different cultures.
APPENDIX
Let
$\tilde{A}=[\tilde{a}_{ij}\rsqb _{n\times n}$
is a fuzzy positive reciprocal matrix for n RFs, where
$\tilde{a}_{ij} =[l_{ij}\comma \; m_{ij}\comma \; u_{ij}\rsqb $
is a triangular fuzzy number. Based on the arithmetic operations of fuzzy numbers (Kaufinami and Gupta, Reference Kaufinami and Gupta1991), we have the geometric means (
$\tilde{{w}}_{i}\rpar $
for the ith RF (
$i=1\comma \; 2\comma \; \ldots\comma \; n\rpar $
:
$$\tilde{{w}}_{i} =\lpar {\mathop \Pi\limits_{j=1}^{n} \tilde{{a}}_{ij} } \rpar ^{1/n}=[ {\lpar {\mathop \Pi\limits_{j=1}^{n} l_{ij} } \rpar^{1/n}\!,\ \lpar {\mathop \Pi\limits_{j=1}^{n} m_{ij} } \rpar^{1/n}\!,\ \lpar {\mathop \Pi\limits_{j=1}^{n} u_{ij} } \rpar^{1/n}} \rsqb ,i=1,2,\ldots,n.$$
Summing up the
$w_{i}\comma \; i=1\comma \; 2\comma \; \ldots\comma \; n$
, yields:
$$\sum\limits_{i=1}^n {\tilde{{w}}_{i} } =\left[ {\sum\limits_{i=1}^n {\lpar {\mathop \Pi\limits_{j=1}^{n} l_{ij} } \rpar^{1/n}} ,\sum\limits_{i=1}^n {\lpar {\mathop \Pi \limits_{j=1}^{n} m_{ij} } \rpar^{1/n}} ,\sum\limits_{i=1}^n {\lpar {\mathop \Pi\limits_{j=1}^{n} u_{ij} } \rpar^{1/n}} } \right].$$
Then, the fuzzy weight for the ith RFs (
$i=1\comma \; 2\comma \; \ldots\comma \; n\rpar $
can be obtained as:
$$\eqalign{\tilde{{W}}_{i} & = \tilde{{w}}_{i} /\sum\limits_{i=1}^n {\tilde{{w}}_{i} } =\left[ {\left( {\displaystyle{{\lpar {\mathop \Pi\limits_{j=1}^{n} l_{ij} } \rpar^{1/n}}\over{\sum\limits_{i=1}^n {\lpar {\mathop \Pi\limits_{j=1}^{n} u_{ij} } \rpar^{1/n}} }}} \right)\!,\left( {\displaystyle{{\lpar {\mathop \Pi \limits_{j=1}^{n} m_{ij} } \rpar^{1/n}}\over{\sum\limits_{i=1}^n {\lpar {\mathop \Pi\limits_{j=1}^{n} m_{ij} } \rpar^{1/n}} }}} \right)\!,\left( {\displaystyle{{\lpar {\mathop \Pi\limits_{j=1}^{n} u_{ij} } \rpar^{1/n}}\over{\sum\limits_{i=1}^n {\lpar {\mathop \Pi\limits_{j=1}^{n} l_{ij} } \rpar^{1/n}} }}} \right) } \right],\cr & \qquad i=1,2,\ldots,n}$$
Since the weight
$\tilde W_i$
of the ith RF (
$i=1\comma \; 2\comma \; \ldots\comma \; n\rpar $
is fuzzy, this paper adopted Yager's index (1981) to defuzzify the
$\tilde W_i$
into a crisp number W
i
, i = 1, 2, …, n. Yager's index is defined based on an area measurement. Suppose the fuzzy number
$\tilde W_i$
with a α cut function:
$W_{i}\lpar \alpha\rpar =[W_{i\alpha}^{L} \comma \; W_{i\alpha}^{R}\rsqb $
, by Yager's index, the
$\tilde W_i$
can be defuzzified as:
$$W_{i} = \int_0^1 {\displaystyle{{1}\over{2}}\lpar {W_{i\alpha }^{L} +W_{i\alpha }^{R}}\rpar} d\alpha.$$
In Equation (A3), for convenience of explanation, let
$\tilde{{W}}_{i} =[l_{i}^{W}\comma \; m_{i}^{W}\comma \; u_{i}^{W}\rsqb $
, where
$$\eqalign{[l_{i}^{W} , m_{i}^{W} , u_{i}^{W} ] & = \left[ {\left( {\displaystyle{{\lpar {\mathop \Pi\limits_{j=1}^{n} l_{ij} } \rpar^{1/n}}\over{\sum\limits_{i=1}^n {\lpar {\mathop \Pi\limits_{j=1}^{n} u_{ij} } \rpar^{1/n}} }}} \right)\!,\left( {\displaystyle{{\lpar {\mathop \Pi\limits_{j=1}^{n} m_{ij} } \rpar^{1/n}}\over{\sum\limits_{i=1}^n {\lpar {\mathop \Pi\limits_{j=1}^{n} m_{ij} } \rpar^{1/n}} }}} \right)\!, \left( {\displaystyle{{\lpar {\mathop \Pi\limits_{j=1}^{n} u_{ij} } \rpar^{1/n}}\over{\sum\limits_{i=1}^n {\lpar {\mathop \Pi\limits_{j=1}^{n} l_{ij} } \rpar^{1/n}} }}} \right)} \right],\cr & \qquad i=1,2,\ldots,n.}$$
Based on Equation (A4), the
$\tilde W_i$
(
$i=1\comma \; 2\comma \; \ldots\comma \; n\rpar $
can be defuzzified as (Yager, Reference Yager1981):
$$W_{i} =(l_{i}^{W} +2 m_{i}^{W} + u_{i}^{W} )/4,\quad i=1,2,\ldots,n.$$
Finally, normalising the W i (i = 1, 2.., n), then we have the crisp weight of the ith RFs as:
$$\omega_{i} =W_{i} /\sum\limits_{i=1}^n {W_{i} } ,\quad i=1,2,\ldots,n$$
