• 沒有找到結果。

EMPIRICAL STUDY FOR THE EFFECT OF CONSECUTIVE DRIVING ON

ACCIDENT RISK FOR TRAIN DRIVING

5.1 Train operations and work shift regulations of the TRA

The TRA is the only institution in Taiwan providing 24 hour service for both passenger

and freight railway operations with 219 stations and 1,097.2 kilometers of track. The TRA has

1,250 drivers who alternately drive both passenger and freight trains, and each is assigned to

one of five dispatching units (Taiwan Railways Annual Report, 2006). Drivers are assigned to

freight trains for at least two consecutive weeks after finishing a specific number of work

shifts driving passenger trains. A driver’s work schedule is arranged and strictly controlled

under regulations issued by TRA. The work shift regulations include:

(1) driving distance for each shift must be less than 300 kilometers;

(2) each shift must not exceed 6 hours from 6 a.m. to 10 p.m.;

(3) each shift must not exceed 5 hours from 10 p.m. to 6 a.m.;

(4) drivers’ rest duration between consecutive shifts must be longer than 6 hours; and

(5) drivers must have at least one off-duty day a week (duration must exceed 24 hours).

5.2 Data collection

5.2.1 Driver responsible accidents

According to the TRA’s operation rules, a train accident is defined as an event causing

more than 10 minutes delay in operations, and the related personnel are responsible for

reporting it to the Accident Investigation Prevention Committee (AIPC). Details of the

accident report include, among other things, the characteristics of train driver(s) and his/their

corresponding work shift information. The AIPC then investigates possible causes of the

accident and determines whether the personnel are responsible for its occurrence; thus, they

are further classified into human- or non-human-error accidents. Given this, a driver’s

consecutive driving hours before the accident can be determined by combining his work shift

record with the accident report.

The records of accidents that occurred during 1996-2006 were collected and used in this

study. Among the total of 10,990 TRA accidents, 10,371 were reported as non-human errors

and 619 were human error accidents. Furthermore, among those human error accidents, 193

accidents were attributed to train driver errors, which accounted for 31% of all human error

accidents. Based on the study purpose, only the driver-responsible accidents are counted in

the study. According to the statistics of accident occurrence time, 172 driver-responsible

accidents occurred at the on-board driving stage, 12 at the pre-starting stage, and only 9 at the

post-arrival stage. For the purpose of studying the effect of consecutive driving on accident

risk, a total of 172 on-board driving accidents were examined in this study, which included

122 passenger train accidents and 50 freight train accidents.

5.2.2 Driving exposure to the risk of accident

As mentioned in Chapter 2.3.2, the on-board driving stage is defined as the operating

duration from the starting station to the ending station for each train driver. The driving task

for a train driver at this stage is relatively continuous and a driver needs more concentration

and alertness to operate safely. In addition, “trunk line driving and shunting” at the on-board

driving stage is a relatively continuous job for train drivers during their work shifts, and

usually influences train operations, occupies the main tracks, and leaves available driving

records for train drivers. Therefore, only the on-board driving time was calculated and was

selected as the exposure measure in this study.

On average, there were 1,072 working shifts per day and approximately 4.3 million

working shifts were collected to calculate the on-board driving hours during 1996-2006. A

passenger train driver had a mean of 2 hours and 45 minutes of on-board driving, while a

freight train driver had a higher average of 3 hours and 20 minutes. In addition, the variation

in on-board driving time for passenger train drivers (SD = 89 min) was larger than that for

freight train drivers (SD = 42 min). This demonstrated that the length of on-board driving for

Distributing all drivers’ work shifts from the eleven observed years into different time

slots, the accumulated on-board driving hours in each time slot for both types of train driving

are shown in Figure 5-1. The total on-board driving hours for passenger trains was 9.88

million hours while the total on-board driving hours for freight trains was 2.57 million hours,

which comprised only 20.6% of total on-board driving hours. The distribution of driving

exposures for different time slots also showed that passenger train drivers had greater

variability in their on-board driving hours as the pattern deviates more from a uniform

distribution than that of freight train drivers.

0

Driving exposure (million hours)

Passenger Freight

Fig. 5-1 passenger and freight train driving exposure for different consecutive driving hours

5.3 Accident rates for different time slots of consecutive train driving hours

Dividing the number of total accidents that occurred in the observation period by the

total driving hours, we find that TRA experienced an average accident rate of 13.82 accidents

per million driving hours from 1996 through 2006. If we further investigate the accident rates

for freight and passenger train driving separately, we will find the freight and passenger train

driving experienced 19.45 and 12.35 accidents per million driving hours, respectively, over

the observation period. Freight train driving had a 58% higher accident risk than passenger

train driving, which is consistent with the expectancy discussed in the previous section.

Furthermore, based on the collected data, the accident rates for each time slot could be

measured by dividing the number of accidents that occurred in one time slot by its

corresponding driving exposure (i.e., the on-board driving hours in the same time slot),

indicating the accident risk for that time slot. That is, the accident rate for the ith time slot can

be expressed as

ARi=Ai/Hi (1)

where ARi is the accident rate for the ith time slot, Ai is the number of accidents that occurred

in the ith time slot, and Hi is the accumulated driving hours in the ith time slot.

According to the above definition, the accident rates for different time slots for the

consecutive driving of all trains, passenger trains and freight trains, are computed and

illustrated in Figure 5-2. We find the accident rates for all train driving had a distinct peak

after one hour of consecutive driving, followed by a relatively low level, which subsequently

increased again. This study result is consistent with the findings of both Wharf (1993) and

Kecklund et al’s (1999) studies.

However, if the accident rates for different time slots are investigated separately for

freight and passenger train driving, a significant difference in accident rates over time will be

found between the two types of train driving, as can be seen in Figure 5-2. That is, the

accident rates for freight train driving were significantly higher than those for passenger train

driving during the first hour of driving, but this phenomenon disappeared after one hour of

driving.

A further investigation indicates the average accident rate during the first hour of freight

train driving was 34.04 accidents per million driving hours, which was about 3.2 times the

average accident rate for passenger train driving (10.78 accidents per million driving hours)

during the first hour. But these statistically significant differences in accident rates between

the two types of train driving were not found at α= 0.05 after one hour of driving. Further

investigation indicates that, among the 24 freight train accidents that occurred in the first hour

of initial driving, 20 accidents (83%) occurred in the marshalling yards. That result is

consistent with the expectancy that freight train driving will experience higher accident risk

for shunting in the marshalling yards than running on the main lines.

Fig. 5-2 The accident rate over consecutive driving hours for train driving

0 10 20 30 40 50 60

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

Consecutive driving hours

Accidents per million driving hours

Passenger train driving Freight train driving All train driving

Figure 5-2 also indicates that the accident rate for freight train driving reached its peak

after one hour of driving (i.e., the 4th time slot) and then dropped to the lowest accident rate

for all time slots. Afterward, the accident rate increased again across consecutive driving

hours. This result is similar to the findings of previous railway studies in that train drivers had

a distinct peak of accident risk during the second and third hours of duty followed by a

relatively low level, and then an increase again (Kecklund et al., 1999; Wharf, 1993). The

only difference with the current study is the time of occurrence of this early peak of accident

risk, which reflects the fact that pre-driving hours were included in the on-duty hours in

previous studies, but not included in this study.

Interestingly, the peak accident rate in the early driving hours was not found for

passenger trains. The accident rates for passenger train driving seemed to increase gradually

as the driving hours were accumulated, which is consistent with the hypothesis that prolonged

driving induces fatigue and then increases the accident risk for train driving. This

phenomenon could also be found for freight train driving if we neglect the early peak of

accident risk during the first hour of on-board driving. Therefore, the study results shown in

Figure 5-2 seem to indicate the effect of consecutive driving on accident risk actually exists

for both passenger and freight train driving, even within a short mission of 4.5 on-board

driving hours. However, an extra accident risk was found for freight train driving in the first

hour, which reveals a distinct early peak accident rate. This extra accident risk for freight train

driving in the first hour could be explained by volume of shunting in the marshalling yards,

which is expected to have a higher accident risk based on the complicated track layout and

semi-automatic traffic guidance for TRA.

5.4 Modeling the accident risk for consecutive driving

Given the accident rates for different time slots over consecutive train driving hours, it is

possible to explore the effect of continuous driving on the accident risk for train operations.

Some previous studies indicated the relationship between accident rates and consecutive truck

(or automobile) driving hours fit an exponential model (Chang and Hwang, 1991; FMCSA,

2000; Folkard 1997) or a quadratic model (Kecklund et al., 1999; Wharf, 1993). Therefore,

four different regression models are considered to formulate the relationship between accident

rates and consecutive driving hours for train driving, according to the trends of accident rates

over different time slots demonstrated in Figure 5-3. These four models are expressed as

follows:

Model 1 (Linear): AR = a + bt (2)

Model 2 (Log linear): ln(AR)= a + bt or AR=exp(a +bt ) (3)

Model 3 (Quadratic): AR= a+ bt + c t2 (4)

Model 4 (Modified Quadratic): AR= a+ c t2 (5)

where a , b and c are the parameters to be estimated and t is the cumulative on-board driving

hours. Models 1, 2, and 4 are used to formulate the increasing trend of accident risk for

consecutive train driving hours, while Model 3 is especially considered to catch the trend

shown in Figure 5-2, which had a distinct peak of accident risk during the first hour of driving

followed by a relatively low level, which then increased again

Table 5-1 Accident rates and relevant statistics for different types of regression models

Model types Accident rate (Accidents

per million driving hours) a (p-value) b (p-value) c (p-value) R2 All trains

* The best model among the four candidate models.

** It is suggested to be the best model in terms of its explanatory power, though the parameter b is only marginally significant with a p-value of 0.16.

These four candidate models were applied to model the trends of accident risk resulting

from consecutive driving for all trains, passenger trains, and freight trains separately.

According to the model estimation results summarized in Table 5-1, Model 3 is suggested to

be the best model for all train driving in terms of its explanatory power, though the parameter

b is only marginally significant with a p-value of 0.16. For passenger train driving, Model 4 is

the best model to describe the increasing accident risk over time for consecutive driving. As

to freight train driving, Model 3 is obviously better than the other three models to catch the

trend of accident risk over time for consecutive driving. However, the explanatory ability of

any of the four candidate models is not good enough for freight train driving.

Apparently, the models for passenger train driving had much better fit than those of

freight train driving due to the lack of a distinct risk peak in the first hour of driving. It was

found that the pattern of increasing accident risk caused by consecutive driving for passenger

trains is similar to that for automobiles or trucks. Conversely, although the accident rates over

time for consecutive freight train driving are not significantly different from those of

consecutive passenger train driving after one hour of driving, the abnormally high accident

risk in the first hour of driving makes the continuous models unable to clearly depict the

pattern of accident risk over time for freight train operation. Therefore, a combined model,

which handles the additional accident risk for freight train driving in the first hour through a

dummy variable and combines the accident rates of both passenger and freight train driving, is

formulated as follows:

AR = a+ bt + c t2 + D (dt + et2) (6)

where D is the dummy variable and D = 1 for freight train driving in the first hour, and D = 0

otherwise. A stepwise regression procedure using backward-elimination was employed to find

the best model for Equation 6, and the estimated results are summarized in Table 5-2. The

explanatory ability of the best model, with R2 = 0.842, was significantly improved as

compared to the single models for passenger and freight train driving, respectively.

Table 5-2 The estimated results for the combined model of accident risk over time

Parameter a (p-value) b (p-value) c (p-value) d (p-value) e (p-value) R2

Step 1 13.08(0.00) -3.48(0.11) 1.37(0.00) 33.83(0.04) 14.16(0.47) 0.854 Step 2 12.37(0.00) -2.91(0.15) 1.27(0.00) 44.85(0.00) 0.852

*Step 3 9.73(0.00) 0.67(0.00) 46.40(0.00) 0.842

* The best model estimated by the stepwise regression procedure.

According to the estimation results of the best combined model shown in Table 5.2,

the accident rates over consecutive driving hours for both passenger and freight trains are

illustrated in Figure 5-3. It indicates the accident risk for passenger train driving increases

with the accumulated driving hours, and shows the accident risk will double after four

consecutive hours of driving, as compared with the accident risk of driving during the first

hour. As to the extra accident risk for freight train driving, it is found to increase sharply with

accumulated driving time during the first hour. That is, the accident rates for freight train

driving were 3.3 and 5.5 times of those of passenger train driving after half an hour and one

hour, respectively, of consecutive driving. This might be a function of the increasing train

length accompanied by the accumulated driving hours in the marshalling yard that increases

the difficulty of shunting and, therefore, increases the risk of accident. In addition, the

accident rates for freight train driving went sharply down to the risk levels of passenger train

driving after one hour of on-board driving in the marshalling yards for shutting.

Fig. 5-3 The estimated accident rate model over consecutive driving hours for passenger and freight train driving

0 10 20 30 40 50 60

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5

Consecutive driving hours

Accidents per million driving hours Estimated model for passenger train driving

Estimated model for freight train driving during the first hour

5.5 Discussion

This study investigated train-driver-responsible accidents by examining their accumulated

on-board driving hours and the associated increasing trend of accident risk over time caused

by consecutive driving. Differentiation of accident risk between passenger and freight train

driving helps us to investigate the distinct early peak problem of accident risk for rail

operation raised by previous literature. Some findings and their implications follow.

5.5.1 Accelerating accident risk of train driving compared with truck driving

The accident risk for train driving was found to double after four hours of consecutive

accident risk for train driving seems to occur earlier than that of automobile driving

(Amundsen and Sagberg, 2003; Chang and Hwang, 1991; Elvik et al., 1997; Mackie and

Miller, 1978). Mackie and Miller (1978) investigated 750 truck crashes and found crash

occurrences began to increase after five hours of driving, and the risk during the second five

hours was twice that of the first five. Elvik et al. (1997) reported truck accident risk

significantly increases after eight hours of consecutive driving and there is a tendency for

increasing risk when driving more than 9-11 hours (Amundsen and Sagberg, 2003). In

addition, Chang and Hwang (1991) studied the effect of prolonged driving on accident risk for

a U.S. trucking company and found the risk after five hours driving was double that of the

first hour. Greater fatigue generated by higher working pressure and a more monotonous

driving environment are considered to be two important reasons for an accelerated accident

risk for consecutive driving during train operations.

Train driving is a dynamic control and decision-making task (Kecklund et al., 2001;

Reinach and Raslear, 2001). The complexity of the operating environment and the work

requirements (i.e., higher density of switches and signals, stations, track works, and grade

crossings speed restrictions) affect the degree of salient environmental information that must

be identified, processed, committed to memory, and used to take appropriate actions.

Especially important is the fact that the train driver’s job is largely governed by timetables

and technical conditions (e.g., type of train and track layouts) that restrict the driver’s ability

to decide how the job should be done. Therefore, these harsher working requirements for train

driving may result in accelerated fatigue for train drivers and, thus, increase train accident risk

faster than it would for automobile driving.

Furthermore, highly automated duties, such as automatic train controls, can be perceived

as boring and monotonous tasks. In particular, when driving during the night everything is

dark and the driver cannot see anything but the signals. The working environment increases

the monotony of train driving except when near signals or stations, and a monotonous and

non-stimulating environment is likely to provoke sleepiness. Johnson (1982) found that tasks

that are more monotonous and lacking in interest are more likely to make people fall asleep.

In addition, performing monotonous tasks may gradually cause a decline in behavior

performance (Thiffault, and Bergern, 2003), reduce levels of alertness, and increase crash risk

(Horne and Reyner, 1995). The monotonous driving environment is, therefore, expected to be

another reason for accelerating the accident risk of consecutive train driving.

5.5.2 The early peak of accident risk for freight train driving

In this study, freight train driving was found to be associated with a risk peak within the

first hour of initial driving. This early peak phenomenon of accident risk was also found in

previous railway studies, but lacked further investigation for its possible causes. Compared

with passenger train driving, freight train driving is usually associated with higher working

complexity. These working characteristics might lead to an earlier peak of accident risk for

freight train driving.

Shunting in the marshalling yard is an unsafe activity, not only because the yard’s track

layout is more complicated than that of the main lines but also because circulation of freight

trains in the yard is guided by a semi-automatic interlock system for TRA. Differing from

passenger train driving, which is directed by an automatic interlock system in the main lines,

the operation of shunting requires more attention by yard staff and freight train drivers. These

operation characteristics for freight train driving in the marshalling yards are thought to be the

reasons for TRA to experience higher accident risk for shunting in the marshalling yards.

A higher proportion of night work shifts for freight train driving is also expected to

increase the accident risk for shunting in the marshalling yards. However, antithetically,

increase the accident risk for shunting in the marshalling yards. However, antithetically,

相關文件