Study On Optimal Sequence Of A Mixed-model Assembly Line With Stochastic Operation Times

With the trend of diversification of customer demand, increasingly more manufacturers are dopting the mixed-model assembly lines (MMAL) for production, and nowadays MMALs have been extensively used in the automobile and electric appliance industries. MMALs are flexible production systems, which allow manufacturers to produce a large variety of different models of a common base product on the same line. With this kind of line, massive customization and consequently rapid response to market changes can be achieved, while inventories are kept at a low level.The effective utilization of a mixed-model assembly line requires solving two problems in a sequential manner as follows: (1) assignment of tasks to workstations (i.e. the line balancing problem) and (2) sequencing of different models on the line (i.e. mixed model sequencing problem). This thesis is focused on the sequencing problem of a mixed-model assembly line with stochastic operation times. Two goals of optimization are considered—minimization of total expected overload time and minimization of total overload probability. It is assumed that an effective line balancing has been already achieved and overloads are handled by utility workers without conveyor stoppages.The paper first extends the model established in [21] to the case of a multi-station MMAL with stochastic operation times, and then derives the approximate algorithm for the calculation of the expected overload time, expected idle time, overload probability and idle probability. Afterwards, several optimization algorithms including simulated annealing (SA) and heuristics are proposed to deal with the two goals aforementioned. Numerical experiments show that the effectiveness and efficiency of the SA are not very satisfactory, and that the SA is sensitive to the annealing parameters. It is also observed that the generation method of neighboring solutions based on problem specific knowledge does not necessarily perform better than the traditional way of random generation. Intensive research is performed on the heuristic algorithms which are classified into two categories—the greedy algorithms and the alternate algorithms. For each kind of heuristic algorithm, massive experiments are carried out against different problem sizes, operation time distributions and coefficients of variation. The results illustrate that some of the heuristic algorithms are quite effective. Specifically, on the one hand, they obtain the optimal solutions most of the time when applied to small scale problems, and on the other hand, they derive heuristic values that are close to the lower bound with nearly constant relative positions when applied to large scale problems. Therefore, when applied to practical problems which are usually large scale, the quality of the heuristic solutions can be estimated by the relative positions of the heuristic values versus the lower and upper bounds. Besides, the differences in operation time distributions and coefficients of variation have little influence on the performance of all the heuristic algorithms, indicating that the heuristic algorithms are robust.In addition, a simple algorithm is suggested for further improvement of the heuristic solutions. Numerical evaluation demonstrates that the improvement algorithm is quite effective for the solutions by some of the heuristic algorithms. Hence when solving practical problems, we can obtain the initial solutions with the best few heuristic algorithms, then adjust them by the improvement procedure, and ultimately choose the best improved heuristic solution as the production sequence.All the numerical examples as well as the detailed analysis convince that all the heuristic algorithms for both optimization goals are robust, and, in particular some of them are very effective and of practical value.