Solving the buffer allocation problem using simulation-based optimisation

This dissertation solves the BAP, using SBO. Chapter 1 describes the role buffers play within a production line. Buffer is allocated to decouple stations and reduce the effect station failure and differing process times have on the complete line throughput. Adding buffer to the line increases the working capital required to run the line, thus the optimal number of buffer to add to a line is an important design question. The problem statement is introduced. The research design is discussed in this Chapter. A DES is specifically designed to simulate a production line for the BAP. An inner and outer loop is used to generate various buffer configurations. The ATS for the inner loop is modified by exploiting theory of constraints. The ideology states that the overall performance of a line can be improved by ensuring the bottleneck is never starved nor blocked. The proposed TOCT uses the block or starved ratio of stations to generate neighbours, reducing the number of neighbours that need to be evaluated per iteration. Small-sized, medium and large-seized lines with various process time, failure rate and repair time is used to scrutinise the proposed method, detailed computation results are shown. In Chapter 2 a literature study is done on approaches for solving the BAP. Three main objectives are considered when solving the BAP. This dissertation focuses on objective two, a prescribed throughput must be achieved with the minimum total buffer size. Various search methods, meta-heuristics and complete enumeration is discussed as an option for the evaluative method. The works of Demir et al. [9, 11] show that TS is an effective approach as a generative method for the BAP. Analytical methods and simulation is discussed as evaluative methods. The decomposition method is most common in literature. The disadvantage is that the decomposition method has limitations on the line topology it can solve as well as distribution types for the machine parameters. Alternatively, simulation or meta-models have been employed to represent more complex lines. Simulation has a more accurate representation of the production line than a meta-model but takes longer to compute. To overcome this DES can be used to create a specific program. In Chapter 3 a specific solution approach for the BAP is developed. The use of DES to design a simulation in Java specifically for the BAP problem addresses the computational burden of generic simulation models. Two event types are created for the model, departure and failure events. The works of Demir et al. [9, 11] establish a solid foundation for the use of tabu search in a two-loop manner to solve the BAP. In Chapter 4 the proposed method for creating and simulating serial production lines using the program designed in Java is comparable to a commercial program and is faster. Initial tests using DES with the ATS inner loop proved that SBO is capable of solving the BAP. However, long execution time makes it unusable for medium and large-sized problems. Two alterations are considered for the ATS and a new inner tabu loop is proposed based on theory of constraints, TOCT as well as a list that saves previously tested buffer scenarios and their throughput so that same scenarios are not re-evaluated with the simulation model. The proposed method is 18 times faster than ATS for small-sized problems and 5.5 times for medium-sized problems. The proposed method is used to solve the BAP for objective two, showing that SBO is an effective model. Finally, in Chapter 5, the solution approach scalability is tested on an actual, complex automotive production line.Dissertation (MEng (Industrial Engineering))--University of Pretoria, 2020.Industrial and Systems EngineeringMEng (Industrial Engineering)Unrestricte