Aart van Harten

Comparison of agent-based scheduling to look-ahead heuristics for real-time transportation problems ☆

We consider the real-time scheduling of full truckload transportation orders with time windows that arrive during schedule execution. Because a fast scheduling method is required, look-ahead heuristics are traditionally used to solve these kinds of problems. As an alternative, we introduce an agent-based approach where intelligent vehicle agents schedule their own routes. They interact with job agents, who strive for minimum transportation costs, using a Vickrey auction for each incoming order. This approach offers several advantages: it is fast, requires relatively little information and facilitates easy schedule adjustments in reaction to information updates. We compare the agent-based approach to more traditional hierarchical heuristics in an extensive simulation experiment. We find that a properly designed multi-agent approach performs as good as or even better than traditional methods. Particularly, the multi-agent approach yields less empty miles and a more stable service level.

On the availability of a k-out-of-N system given limited spares and repair capacity under a condition based maintenance strategy

This paper considers a k-out-of-N system with identical, repairable components. Maintenance is initiated when the number of failed components exceeds some critical level. After a possible set-up time, all failed components are replaced by spares. A multi-server repair shop repairs the failed components. The system availability depends on the spare part stock level, the maintenance policy and the repair capacity. We present a mathematical model supporting the trade-off between these three parameters. We present both an exact and an approximate approach to analyse our model. In some numerical experiments, we provide insight on the impact of repair capacity, number of spares and preventive maintenance policy on the availability.

On the interaction between maintenance, spare part inventories and repair capacity for a k-out-of-N system with wear-out

In this paper we consider a k-out-of-N system with identical, repairable components under a condition-based maintenance policy. Maintenance consists of replacing all failed and/or aged components. Next, the replaced components have to be repaired. The system availability can be controlled by the maintenance policy, the spare part inventory level, the repair capacity and repair job priority setting. We present two approximate methods to analyse the relation between these control variables and the system availability. Comparison with simulation results shows that we can generate accurate approximations using one of these models, depending on the system size.

On-line scheduling of multi-server batch operations

The batching of jobs in a manufacturing system is a very common policy in many industries. The main reasons for batching are the avoidance of setups and/or facilitation of material handling. Good examples of batch-wise production systems are the ovens that are found in the aircraft industry and in the manufacture of semiconductors. These systems often consist of multiple machines of different types for the range and volumes of products that have to be handled. Building on earlier research in the aircraft industry, where the process of hardening synthetic aircraft parts was studied, we propose a new heuristic for the dynamic scheduling of these types of systems. Our so-called look-ahead strategy bases its decision to schedule a job on a certain machine on the availability of information on a limited number of near future arrivals. The new control strategy distinguishes itself from existing heuristics by an integrated approach that involves all machines in the scheduling decision, instead of only considering idle machines. It is shown by an extensive series of simulation experiments that the new heuristic outperforms existing heuristics for most system configurations. Especially in the case of complex systems, where multiple products have to be handled by non-identical machines, the new heuristic proves its value as a practical scheduling tool. Important insight is obtained with regard to the relation between the system is configuration and its performance.

An Exact Solution for the State Probabilities of the Multi-Class, Multi-Server Queue with Preemptive Priorities

We consider a multi-class, multi-server queueing system with preemptive priorities. We distinguish two groups of priority classes that consist of multiple customer types, each having their own arrival and service rate. We assume Poisson arrival processes and exponentially distributed service times. We derive an exact method to estimate the steady state probabilities. Because we need iterations to calculate the steady state probabilities, the only error arises from choosing a finite number of matrix iterations. Based on these probabilities, we can derive approximations for a wide range of relevant performance characteristics, such as the moments of the number of customers of a certain type in the system en the expected postponement time for each customer class. We illustrate our method with some numerical examples. Numerical results show that in most cases we need only a moderate number of matrix iterations (∼20) to obtain an error less than 1% when estimating key performance characteristics.

Scheduling vehicles in automated transportation systems : algorithms and case study

Abstract. One of the major planning issues in large scale automated transportation systems is so-called empty vehicle management, the timely supply of vehicles to terminals in order to reduce cargo waiting times. Motivated by a Dutch pilot project on an underground cargo transportation system using Automated Guided Vehicles (AGVs), we developed several rules and algorithms for empty vehicle management, varying from trivial First-Come, First-Served (FCFS) via look-ahead rules to integral planning. For our application, we focus on attaining customer service levels in the presence of varying order priorities, taking into account resource capacities and the relation to other planning decisions, such as terminal management. We show how the various rules are embedded in a framework for logistics control of automated transportation networks. Using simulation, the planning options are evaluated on their performance in terms of customer service levels, AGV requirements and empty travel distances. Based on our experiments, we conclude that look-ahead rules have significant advantages above FCFS. A more advanced so-called serial scheduling method outperforms the look-ahead rules if the peak demand quickly moves amongst routes in the system.

Optimal clustering of frequency-constrained maintenance jobs with shared set-ups

Since maintenance jobs often require one or more set-up activities, joint execution or clustering of maintenance jobs is a powerful instrument to reduce shut-down costs. We consider a clustering problem for frequency-constrained maintenance jobs, i.e. maintenance jobs that must be carried out with a prescribed (or higher) frequency. For the clustering of maintenance jobs with identical, so-called common set-ups, several strong dominance rules are provided. These dominance rules are used in an efficient dynamic programming algorithm which solves the problem in polynomial time. For the clustering of maintenance jobs with partially identical, so-called shared set-ups, similar but less strong dominance rules are available. Nevertheless, a surprisingly well-performing greedy heuristic and a branch and bound procedure have been developed to solve this problem. For randomly generated test problems with 10 set-ups and 30 maintenance jobs, the heuristic was optimal in 47 out of 100 test problems, with an average deviation of 0.24% from the optimal solution. In addition, the branch and bound method found an optimal solution in only a few seconds computation time on average.

Two-stage generalized age maintenance of a queue-like production system

In general, the initiation of preventive maintenance should be based on the technical state as well as the operating state of a production system. Since the operating state of a production system is often subject to fluctuations in time, the planning of preventive maintenance at preset points in time (e.g. age/block replacement) cannot be optimal. Therefore, we propose a so-called two-stage maintenance policy, which - in a first stage - uses the technical state of the production system to determine a finite interval [t, t + At] during which preventive maintenance must be carried out, and - in a second stage - uses the operating state of the production system to determine the optimal starting time for preventive maintenance within that interval. A generalized age maintenance policy optimizing both t and At is formulated in the first stage. To this end, the actual starting time of preventive maintenance is modelled in terms of a uniform distribution over the maintenance interval. Moreover, the expected costs of preventive maintenance are modelled as a decreasing function of the interval size. An efficient algorithm is developed to demonstrate the optimal strategy for a queue-like production system, via numerical results that offer useful insights.

Approximations for Markovian multi-class queues with preemptive priorities

We discuss the approximation of performance measures in multi-class M/M/k queues with preemptive priorities for large problem instances (many classes and servers) using class aggregation and server reduction. We compared our approximations to exact and simulation results and found that our approach yields small-to-moderate approximation errors.

Scheduling vehicles in automated transportation systems

One of the major planning issues in large scale automated transportation systems is so-called empty vehicle management, the timely supply of vehicles to terminals in order to reduce cargo waiting times. Motivated by a Dutch pilot project on an underground cargo transportation system using Automated Guided Vehicles (AGVs), we developed several rules and algorithms for empty vehicle management, varying from trivial First-Come, First-Served (FCFS) via look-ahead rules to integral planning. For our application, we focus on attaining customer service levels in the presence of varying order priorities, taking into account resource capacities and the relation to other planning decisions, such as terminal management. We show how the various rules are embedded in a framework for logistics control of automated transportation networks. Using simulation, the planning options are evaluated on their performance in terms of customer service levels, AGV requirements and empty travel distances. Based on our experiments, we conclude that look-ahead rules have significant advantages above FCFS. A more advanced so-called serial scheduling method outperforms the look-ahead rules if the peak demand quickly moves amongst routes in the system.