Anda di halaman 1dari 18

Scheduling rail track maintenance 1

SCHEDULING RAIL TRACK MAINTENANCE TO MINIMISE OVERALL DELAYS


Andrew Higgins, CSIRO, Brisbane, Australia. Luis Ferreira and Maree Lake , School of Civil Engineering, Queensland University of Technology, Australia.
Higgins, A. and Ferreira, L. and Lake, M. (1999) Scheduling rail track maintenance to minimise overall delays. In Proceedings The 14th International Symposium on Transportation and Traffic Theory, Jerusalem, Israel.

ABSTRACT
In Australian freight operations, maintenance costs comprise between 25 -35 percent of total train operating costs. Therefore, it is important that the track maintenance planning function is undertaken in an effective and efficient manner. This paper focuses on the development of a model designed to help resolve the conflicts between train operations and the scheduling of maintenance activities. The model involves scheduling maintenance activities to minimise disruptions to train services and reduce maintenance costs. The main applicability of such a model is as a decision support tool for track maintenance planners and train planners. The track maintenance scheduling problem, which involves the allocation of maintenance activities to time windows and crews to activities, is formulated as an integer programming model. The objective is to minimise a weighted combination of expected interference delays and prioritised finishing time of activities. Minimising the first component will ensure a minimum interference between track maintenance activities and scheduled trains when either are delayed. The heuristic solution is obtained in two steps. Firstly, an initial solution is generated, scheduling each of the activities in turn, where the latter are ordered in terms of the importance of finishing time. Each activity is selected and allocated to an available permissible work crew. If there are no available permissible work crews when the activity is chosen, the earliest finishing crew will be selected. The second stage uses the tabu search heuristic.

The model presented here was applied to an 89 km track corridor on the eastern coast of Australia. The schedule constructed using tabu search has a 7 percent reduction in objective function value as compared to the schedule constructed manually. The model was also used to demonstrate the effects of activity schedule and maintenance resource changes. A four day planning horizon was used for which the model was used to test proposed changes. Increasing the time window by moving less important trains was shown to reduce potential delays significantly.

1. INTRODUCTION Although Australian rail systems have achieved considerable productivity gains in the last decade, unit operating costs are still well below worlds best practice. In 1993/94 track maintenance productivity lagged behind such benchmarks by an estimated $A80 million. This represents 16 percent of potential operating cost savings available if worlds best practice is achieved (Bureau of Industry Economics, 1995). In Australian freight operations, maintenance costs comprise between 25 -35 percent of total train operating costs. Therefore, it is important that the track maintenance planning function is undertaken in an effective and efficient manner. This applies to short-term planning such as daily scheduling of activities; as well as the medium to long-term planning of required maintenance activities. Infrastructure provision is increasingly seen as a separate business to be managed, planned and owned by a different entity. Such vertical separation is seen as providing the vehicle that will improve accountability and profitability within the rail industry. In addition, separate control and management of track is designed to encourage rail transport competition by allowing new train operators to gain access to the right-of-way (Ferreira, 1997). Efficient maintenance planning requires an up-to-date, locally relevant decision support tools. This paper focuses on the development of a model designed to help resolve the conflicts between train operations and the scheduling of maintenance activities. The model involves scheduling maintenance activities to reduce disruptions to train services; maintenance costs; and the amount of time a given track segment has a level of service below a specified benchmark. (e.g. speed restrictions imposed on trains due to poor track condition). This latter aspect is particularly important when contractual obligations exist between track providers and rail carriers relating track performance to access charges. The main applicability of such a model is as a decision support tool for track maintenance planners and train planners. The need for such a model is more pressing under single track train operations, where trains can only pass or overtake each other at specified locations (sidings). A model to optimise a given train schedule with respect to train related operating costs, including the risk of delays, has been put forward by 2

Scheduling rail track maintenance 3 Higgins et al. (1996). However, such a model does not incorporate track maintenance activities in the decision process. The paper is organised as follows: Section 2 provides a brief outline of the types of track maintenace activities which need to be scheduled; this is followed in Section 3 by a statement of the problem and a discussion of past work on the scheduling of track maintenance; Section 4 provides a detailed description of the model developed; Section 5 shows the results of applying the model to a case-study; and finally some conclusions are offered in Section 6.

2. MAINTENANCE ACTIVITIES Track maintenance covers all the measures for preserving and re-establishing the nominal condition, as well as the measures for determining and assessing the actual condition in a technical system (Kooran, 1992). This section details the following main types of maintenance: rail grinding; rail replacement; tamping; track stabilisation; ballast injection; and sleeper replacement. Rail Grinding: This consists of grinding machines travelling along the track with grinding stones, which are rotating stones or stones oscillating longitudinally, to abrade the rails surface. Rail grinding is conducted to correct rail corrugations, fatigue and metal flow and to re-profile the rail. Rail Replacement: This may be conducted to upgrade the track to a higher gauge rail or to replace the same gauge rail due to defects, wear or derailment damage. Tamping: This is conducted to correct longitudinal profile, cross level and alignment of track. A number of sleepers at a time are lifted to the correct level with vibrating tamping tines inserted into the ballast. Track Stabilisation: Track stabilisers vibrate the track in the lateral direction with a vertical load to give controlled settlement. Tamping and compacting ballast underneath sleepers reduces the lateral resistance of the track. Track stabilisation can restore the lateral resistance to the original level. Ballast Injection (Stoneblowing): Ballast injection, or stoneblowing, is conducted to correct longitudinal profile. The process introduces additional stones to the surface of the existing ballast bed, while leaving the stable compact ballast bed undisturbed (Esvald, 1989).

Sleeper Replacement: In almost all types of sleeper defects, remedial action is not possible and the sleeper requires replacement. Defective sleepers can result in the rail losing the correct gauge, which can cause rollingstock derailments.

3. TRACK MAINTENANCE SCHEDULING This section addresses the short term scheduling of maintenance to minimise the conflict between train operations and track maintenance, once the maintenance to be conducted has been determined. The problem of maintenance planning in the medium to long-term, including track degradation modelling, has been addressed by Ferreira and Murray (1997).

The Problem Planned train schedules under single track operations outside urban areas are likely to be subjected to significant variations on a daily basis. This is due to a number of reasons including missed meets and passes (locations where conflicts between opposing trains are resolved); locomotive and rollingstock failures; and late departures from the origins. Given such uncertainty in the times when specific trains are likely to occupy a given track link, it is necessary to schedule maintenance activities so that any potential conflict with train services is minimised. In addition, the maintenance activities themselves are subjected to uncertainty in duration, due to the longer or shorter times to complete each task. Such uncertainty may lead to disruptions to train services and hence unreliability of train arrivals. A model to estimate such unreliability has been proposed by Ferreira and Higgins (1996). The overall problem consists of scheduling a given set of maintenance activities over a network of rail links such that: (a) Train delays are kept to as low as possible, having regard to the importance of each train. A pre-defined train hierarchy is used to reflect the fact that some trains may carry more time sensitive freight than others. From a market share and business perspective such trains are more important to the operator and therefore they should be free from the risk of maintenance delays as far as possible; (b) For specified long-distance train services, maintenance activities should not be allowed to affect train schedules so as to produce cumulative delays at the destinations above a given level. Such aspect of maintenance scheduling was found to be particularly relevant by Aspebakken et al. (1991) when developing a graphical based system. 4

Scheduling rail track maintenance 5

(c) the direct costs associated with maintenance activities are kept as low as possible; (d) maintenance activities taken place according to an a priori set of priority weights; and (e) maintenance crews are assigned to activities according to a specified area of coverage for each crew.

Increasing Importance of Effective Scheduling Contractors have become involved with a broader range of work across the rail engineering spectrum, from design and construction to maintenance and operations (Kramer, 1998). With this increase of outsourcing, it becomes important to effectively schedule the maintenance windows, as contractual arrangements will govern the allowable maintenance times. Penalties are likely to be accrued if the maintenance windows are not as required, resulting in a trade off between customer service and maintenance costs. A maintenance window is a planned pre-organisation of train services for a set time or over a set period, to ensure track possession for prescribed infrastructure works (Griffiths, 1991). The length of maintenance window planned is not necessarily the hours that the maintenance gangs actually receive. The difference between the planned and actual length is train and workblock conflict related (Szymkowiak, 1991). That is, delays to trains reduce the size of the maintenance window. In an attempt to coordinate maintenance and train operations, maintenance has been scheduled at night, outside peak train times or on the weekends. However, this is not always feasible due to other considerations in the scheduling of maintenance, such as noise restrictions, safety aspects or personnel rostering.

Past Work At Burlington Northern, work on coordinating maintenance and train operations was conducted with the aim of notifying customers several weeks in advance when delays due to maintenance could be expected (Aspebakken, et al, 1991). The Service Maintenance Planning System (SMP) was created to allow visualisation of the interaction between trains and maintenance work for specific corridors. Time/distance diagrams and the locations of any conflicts between trains and maintenance are displayed. The system does not schedule the maintenance activities or maintenance crews, instead highlighting the conflicts between the maintenance and trains. Basic priority rules are used to make

adjustments to the timetable. However, this does not necessarily optimise the train and maintenance schedule with respect to overall costs or reliability. Research into the goals, constraints and necessary outputs of a maintenance scheduler has been undertaken, in particular detailing the program Intelligent Maintenance Scheduler (IMS) developed by Ruffing and Marvel (1994). The goal of this system was reported to be maximising the work window size and minimising train delays, with constraints including included labour agreements, train service contracts, physical limitations and maintenance activity characteristics. The required outputs listed were the specific track occupancy times of the maintenance crews and the resultant train delays. Train delays are considered to be the direct and indirect time increases to the train schedule for maintenance to be carried out. This research does not take into account the unforeseen delays of trains or maintenance on the day of operation and maintenance crews are not assigned to activities.

4. PROBLEM FORMULATION

Definition of Terms and Notation Used Expected interference delay:- This is the average delay that both the train schedule and maintenance activities would suffer due to unforeseen events which may occur to a train or activity. For this paper, unforeseen events are of a known discrete distribution. Prioritised finishing time:- It is desirable to have a maintenance activity finished as soon as possible. This is particularly so for those which result in the track being at a higher standard. Minimising the prioritised finishing time requires scheduling the maintenance activities (with different priorities) so as to have the finishing time prior to as many scheduled trains as possible. Time Window: A group of time intervals which are not broken up by a scheduled train.

Objective Function As both trains and maintenance activities are subjected to unexpected delays, the objective function consists of three weighted components, that minimise: expected interference delays (EID) due to delayed trains overlapping the maintenance schedule; EID due to delayed maintenance activities overlapping the train schedule; and prioritised 6

Scheduling rail track maintenance 7 finishing time of maintenance activities. In presenting the analytical objective function and constraints, the following notation and parameters are firstly defined: I K U J Vk Lj Ci ,t = Ak ,t = Bj Fj Maintenance crews which are allocated to the track corridor Set of track links which make up the track corridor Set of scheduled trains on the corridor Set of activities which are to be completed in time horizon T Activities that are scheduled on link k K during time horizon T Crews permissible to work on maintenance activity j J 1 if maintenance crew i I is available at time interval t T 0 otherwise

1 if link k K is available for maintenance at time interval t T 0 otherwise


Number of time intervals required to carry out maintenance activity j J given that the work is continuous Number of extra time intervals required for maintenance activity j J due to discontinuous maintenance Minimum travel time intervals for maintenance crew i I to travel from the track link where activity j1 J is carried out to that link where activity j 2 J is carried out. Cost of maintenance crew i I working on activity j J at time interval
t T Importance of activity j J in terms of finishing time

M i , j1 , j2

C1i , j ,t C2 j SCk ,t t T h

Number of scheduled trains which pass through link k K before time Priority weight of activity finish time relative to EID between the train schedule and maintenance activities

The 0-1 decision variables are defined as follows:

X i, j

1 if track maintenance crew i I is assigned to activity j J 0 otherwise

Y j ,t

1 if work is carried out on activity j J in time interval t T 0 otherwise

The objective function for the model is as follows:

Min Z = h*(
u U k K w j Vk

f (u, w) * Pu , j * Yj ,Ok ,u + w + C 2 j * SCk ,t 1

u U

k K

j Vk

fc( j , Ou ,k t1) * Pu , j )+ (1)

k K

j Vk

where: t1 = max ( t * Yj ,t ) , ie. finishing time of activity j J


t

U = set of scheduled trains Ou ,k = time interval in which train u U is scheduled to enter link k K Pu , j = priority weight when there is an overlap between train u U and activity j J for which either is delayed. This priority is a combination of that
associated with train u U and activity j J .

f(u,w) = probability that train u U is delayed w intervals fc(j,w) = probability that activity j J is delayed w time intervals.
The first component of objective function (1) measures the expected interference delay (or expected amount of weighted overlap) due to delayed trains overlapping scheduled maintenance activities. That is, if train u U was delayed w time intervals on entry to link k K and an activity j J was scheduled (with crew i I ) for this time interval (ie. Yj ,Ok ,u + w =1), then the weighted amount of overlap is equal to Pu , j . For all possible train delays, the expected amount of overlap (ie. EID) between train u U and activity

j J is
w

f (u, w) * Pu , j * Yj ,Ok ,u + w . Therefore, the first component of (1) is achieved by

summing over all track links and scheduled trains. In the second component of (1) which is when delayed maintenance activities overlap scheduled train services, if activity j J is delayed w = Ou ,k t1 > 0, then there is an overlap with train u U with weight Pu , j and probability of occurrence fc( j , Ou ,k t1) . In the third part of objective function (1), that is minimising the weighted completion time of the maintenance activities, C 2 j * SCk ,t 1 is the priority of activity j J multiplied by the number of scheduled trains which are prior to the finishing time of activity j J on link k K .

Constraints
The model is subject to the following constraints:

Yj ,t Ak ,t

j J , t T

(2)

Scheduling rail track maintenance 9


where k is such that j Vk .

t T

Yj ,t = B j +

t T

abs(Y j ,t Y j ,t 1 ) * F j / 2 - F j

j J

(3)

j J

Y j ,t * X i , j Ci ,t

i I , t T

(4)

j J

Yj ,t 1 j Vk

k K , t T

(5)

min(s.t. X i , j1 * Y j1,t + X i , j 2 * Y j 2 ,t + z = 2) M i , j1, j 2


z

if j1 j 2

(6) (7)

i I

j J

t T

C1i , j ,t * X i , j * Y j ,t < Cost budget

max(t * Yj1,t ) < min(t * Yj 2 ,t )


t t

if activity j1 must be completed before the commencement of activity j2

(8) X i , j = 0 if i L j X i , j , Y j ,t = 0 or 1 (9)

i I , j J , t T

(10)

Constraint (2) identifies links which are not allowed to be assigned maintenance during certain times intervals. This applies to when the track is unavailable due to scheduled trains. Constraint (3) enforces the required number of time intervals for which an activity is worked on. The second part of the right hand side on this constraint extends the number of time intervals required for an activity if it is disturbed by a scheduled train. If work on an activity is continuous then this value is equal to 0 (ie.
t T

abs(Y j ,t Y j ,t 1 ) * F j / 2 = Fj ). Some activities will cause the track link to be

unusable until the activity is complete. For these activities which cannot be disturbed by a scheduled train, Fj is set to a large value. The availability of maintenance crews in each time intervals is represented by constraint (4). Activities are prevented from being worked in parallel on the same link using constraint (5). For example, both re-railing and tamping cannot be carried out at the same time and on the same track link. Constraint (6) ensures the minimum number of time intervals between the finishing time of activity j1 and the start time of j2 is M i , j1, j 2 . Such a constraint is required for a maintenance crew to have enough time to travel between links at different parts of the track corridor and to

prepare for the next activity. Constraint (7) ensures that the maximum operating costs for the maintenance crews does not exceed the allocated cost budget. For this formulation, cost is considered as a constraint rather than a component to be minimised. It can be made a soft constraint by attaching a penalty to the amount in which the budget is exceeded. Constraint (8) ensures any necessary ordering between activities. For example, visual inspection may only be carried out after sleepers are replaced. Permissible maintenance crews for individual links are represented by constraint (9). Constraint (10) ensures the decision variables are binary.

Solving the Model


The model presented here, which contains a large number of decision variables for a real life problem, is suited to local search type heuristics due to the ease in constructing an efficient and meaningful search neighbourhood. The heuristic applied to find a near optimal solution is the Tabu Search (TS) heuristic (Glover 1990 and Glover et al. 1993). The TS escapes local optimal solutions by allowing up-hill (non-improving) moves to be performed when no down-hill (improving) moves are available. For the maintenance scheduling problem, a move can be to swap attributes of two maintenance activities (with probability P); or to shift an activity to a different set of (available) time windows (with probability 1-P). At each iteration of the TS, the neighbourhood (or part of it) is searched, for which the best non-tabu move found in the search is applied. A move is tabu if it is one of the L most recent moves applied. The tabu status is over-ridden (aspiration criteria), if the solution is better than any at that stage of the search. Depending on the problem at hand, the general TS can be extended so as to promote various forms of diversification and intensification. For the application of this model, the search is re-initialised after LOCIT iterations by replacing the current solution with the best found so far. The tabu search is terminated after GLOBIT iterations. Before the TS is applied an initial solution is found by scheduling each of the activities in order of prioritised finishing time (third part of objective function (1)). Activities are allocated to available permissible work crews. The budget constraint is ignored at this stage.

5. MODEL TESTING: AN APPLICATION

The Test Problem


The track corridor chosen to test the model is 89km long and stretches between Gympie and Maryborough on the Queensland north coast railway, Australia. On the busiest day of the week, over 30 trains are scheduled to travel along this track. There are four 10

Scheduling rail track maintenance 11 different types of trains scheduled along this track, namely: heavy freight (HF); fast freight (FF); diesel hauled passenger trains (P); and fast passenger trains (FP). The track corridor, which contains 13 sidings, has two sets of maintenance crews are allocated to it. It is assumed that one set is responsible for the track links between Gympie and Thebine, while the other is allocated to links between Thebine and Maryborough. It is assumed that a crew allocated to the Gympie - Thebine corridor may work on the other track links, but at 1.5 times the normal cost. A maintenance activity cannot be scheduled in a time interval which contains a scheduled train. The list of activities for each track link along with the minimum time intervals to complete each are shown in Table 1. For the base problem under study, the priorities of these activities are scaled so that the delay due to train disturbances will be about 5 percent of the entire objective value. This means that the activities will be predominately scheduled so as to minimise activity finishing time. Delays due to disturbances will only be significant if the solution technique is to decide between two schedules with similar total activity finishing time, but vastly different disturbance delays.

Table 1: Number of time intervals required to complete each activity


Number of time intervals Activity Track link 1 2 3 4 5 6 7 8 9 10 11 12 13 1 16 12 6 7 9 6 5 4 10 9 14 6 12 2 4 9 8 7 8 6 10 10 6 10 3

Using the result of Higgins (1996), train delays are found to follow a negative exponential distribution with a mean delay of 0.30 hours. For the purposes of this example, delay distributions for maintenance activities are assumed to be same as that for trains. If an activity was discontinuous due to scheduled trains, the extra time required

was assumed to be 0.2 hours for all activities. That is, if an activity stopped 5 times due to a scheduled train, an extra time interval was added to its duration. It is assumed that maintenance crews 2 hours to travel the length of the entire corridor. Since this corridor contains 12 track links in series, it is estimated that the number of hourly time intervals required for crews to travel between two links is equal to the number of links separating them divided by 5. A change in maintenance crews during a activity is disallowed. It is assumed that work can be carried out on a activity any time of the day with equal costs. The base case contains six maintenance crews (three for Gympie - Thebine, three for Thebine - Maryborough)

Model Results
The best solution found for the base case, using the tabu search is illustrated in Figure 1 for the allocation of activities to time windows. This is a time-distance train scheduling diagram for which the horizontal dashed lines represent the intermediate sidings. The 72 to 96 hour section of planning horizon in this figure contains a large number of scheduled trains, making the available time windows very few in number and short. It would not be practical to allocate activities during this period unless they are urgent. Table 2 shows the start and finish times for each maintenance activity on each track link. A schedule was constructed manually by ordering activities in terms of priority and allocating them in turn to the cheapest available work crew. The schedule constructed using tabu search has a 7 percent reduction in objective function value as compared to the schedule constructed manually. The model was also used to demonstrate the effects of activity schedule and maintenance resource changes. The first sensitivity test involved changing the train schedule as shown in Figure 2. Trains marked 1 and 2 (Figure 1) are rescheduled to positions 1a and 2a (Figure 2) respectively, so as to increase the time windows earlier in the schedule. This allows less train interruptions for activities, lower expected delay due to disruptions and earlier completion time of activities. Illustrated in Figure 2 is the schedule obtained using the tabu search heuristic. There is a 2 percent reduction in prioritised maximum finishing time and an 18 percent reduction in expected delays due to train and activity disturbances. Maintenance planners may also wish to know how many crews should be available at a given time. Too few crews will prevent activities from being completed in the allocated time horizon. Too many will cause a large increase in maintenance costs with very little improvement in the maintenance schedule. The costs and benefits of modifying the 12

Scheduling rail track maintenance 13 number of work crews has been analysed and the results are shown in Table 3. Results shown are in terms of the objective function as a percentage of the base case of 6 crews. When reducing the number of crews to below 6, there is a major increase in the prioritised finishing time of activities, while more than six crews results in only minor improvements. Less than 5 crews gives an infeasible solution for a four day planning horizon. A test was carried out to assess the effects of modifying the priority of activity finishing time with respect to expected interference delay. The results are displayed in Table 4. The first column represents the importance ratio of prioritised activity finishing time with respect to expected interference delay for which the base problem has a ratio of 1. This ratio is halved at each row of Table 4. The greatest fluctuation in prioritised finishing time of activities and expected disturbance delay is when the importance ratio is between 0.125 and 0.0312. When the ratio is less than 0.675, expected disturbance delay is very dominant in the objective function. This is an important result since it is difficult for maintenance planners to determine the cost of having activities finish early relative to delays due to train and maintenance activity disturbances. Planners can use such a result to obtain the desired balance in the objective function.

Table 2: Maintenance Activity Schedule - Base Case


Start and Finish Times (hrs in 24 hr clock) Activity 1 Track link 1 2 3 4 5 6 7 8 9 10 11 12 13 S 34 16 11 11 18 16 24 7 2 24 39 33 39 F 67 32 21 22 29 22 31 10 14 36 70 40 60 S 1 1 25 30 3 3 40 42 28 16 2 F 5 11 35 41 10 8 59 64 36 27 S 3 F

16

S = Start time; F = Finish time

Table 3: Sensitivity Analysis: Changing the Number of Maintenance Crews


Objective Function as Percent of Base Case Number of crews 5 6 (Base Case) 7 8 Prioritised activity finishing time 113.0 100.0 98.8 96.1 Overall objective function 111.1 100.0 98.5 95.7

Table 4: Results: Changes to the Importance of Finishing Time and EID


Objective Function As Percent of Base Case Importance Ratio 1 (Base Case) 0.5 0.25 0.125 0.0625 0.0312 0.0156 0.0078 Prioritised Finishing Time 100 100.5 100.8 102.7 108.1 114.5 116.3 117.1 Expected Disturbance Delay 100 97.1 87.1 81.2 72.8 66.0 64.4 63.8

14

Scheduling rail track maintenance 15

6. CONCLUSIONS AND FURTHER RESEARCH


There has been little published research which aims to optimise both train and track maintenance operations simultaneously. Past work has either altered the train schedule to accommodate track maintenance using priority rules or held the train schedule as a fixed input and assigned the maintenance crews to the activities. The scheduling of maintenance activities and allocation of crews to these are traditionally performed using the experience and knowledge of planners. Since maintenance costs are such a large proportion of total operating costs, there is a need for the development of operations research tools aimed at assisting the maintenance planners. This paper has presented a model aimed at improving rail maintenance decisions. The objective was a combination of minimising the prioritised finishing time of each activity, as well expected interference to (and from) scheduled trains. The tabu search heuristic technique was used to find a solution to the large 0-1 integer program. Such a technique was suitable since the neighbourhood was easily defined by swapping the order of activities or maintenance crews; or by shifting an activity to a different time window. The model was applied to a 89 km track corridor on the eastern coast of Australia. A four day planning horizon was used for which the model was used to test proposed changes including rescheduling trains and changing the number of maintenance crews. Increasing the time window by moving less important trains was shown to reduce potential delays significantly. The model is mainly aimed at providing for an off-line planning function. However, the same model could be used by local track managers and train planners in real-time so that adjustments could be made to a planned schedule of activities in the light of unplanned train services or train cancellations. Such a system would need to be integrated into a train dispatching real-time database. As a part of further research, an ideal planning process would be to schedule maintenance activities at the same time as constructing the train schedule. As proposed by Ruffing and Marvel (1994), instead of constructing and optimising a train plan to minimise delays due to train conflicts and unplanned events, a train plan can be constructed to maximise time windows available for possible maintenance activities. A long continuous time window is less likely to be disrupted by trains (or vice versa) than many short time windows split by scheduled trains. 16

Scheduling rail track maintenance 17

REFERENCES
Aspebakken, J. I., Galen, G. L. and Stroot, R. E. (1991). Service maintenance planning on high densities, heavy haul traffic routes. Heavy Haul Workshop, International Heavy Haul Conference, Vancouver, Canada. Bureau of Industry Economics (1995). Rail freight 1995: International Benchmarking. Report 95/22. AGPS, Canberra. Esvald, C. (1989) Modern Railway Track, MRT-Productions, West Germany. Ferreira, L. (1997). Rail track infrastructure ownership: Investment and operational issues. Transportation, 24 (2), 183-200. Ferreira, L. and Higgins, A. (1996). Modelling reliability of train arrival times. Journal of Transportation Engineering, American Society Civil Engineers (ASCE), 122 (6),414-420. Ferreira, L. and Murray, M. (1997). Modelling rail track deterioration and maintenance: current practices and future needs. Transport Reviews, 17 (3), 207-221. Higgins, A., Kozan, E. and Ferreira, L. (1996). Optimal scheduling of trains on a single line track. Transportation Research - Part B, 30B (2),147-161. Higgins, A, (1996) Optimisation of train schedules to minimise transit time and maximise reliability, PhD Thesis, School of Mathematical Sciences, Queensland University of Technology, Brisbane, Australia. Glover F. (1990). Tabu Search: A Tutorial, Interfaces 20: 74-79. Glover, F., Taillard, E. and de Werra, D. (1993) A Users Guide to Tabu Search, Annuals of Operations Research, Vol 41 (1), pp. 3-28. Griffiths, B. (1991) Implementation of a Maintenance Window Programme in an Operating Suburban Network, Preprints of the 1991 International Heavy Haul Workshop, International Heavy Haul Association, Vancouver, Canada, pp.195-202. Koogan, N. A. (1992) Track Maintenance at the Netherlands Railways, Ninth International Rail Track Conference, Perth, Australia.

Kramer, J. (1998) Outsourcing in Rail Engineering: Contracting' s Role, Railway Track and Structures, April 1998, pp. 23. Ruffing J. A. and Marvel B. (1994). An Analysis of the Scheduling of Work Windows for Railroad Track Maintenance Gangs. Proceedings of the 1994 International Heavy Haul Workshop, pp. 1 - 4. Szymkowiak, J. A. (1991) Proactive Management of Service and Maintenance Conflicts, Preprints of the 1991 International Heavy Haul Workshop, International Heavy Haul Association, Vancouver, Canada, pp.121-131.

18

Anda mungkin juga menyukai