Efficient Process Optimization

A computer-based system for modelling and optimizing processes is presented. The Design Structure Matrix (DSM) process representation was used to model the processes due to its compact, generic and easily quantifiable nature. The system is capable of calculating a number of process performance metrics that are focussed towards determining the degree of iteration and concurrency within the process, however, the system is easily extendible to include other process performance measurements. The paper describes the use of a Genetic Algorithm (GA) to optimise the sequence of activities with the focus of reducing the amount of iteration by reducing the number of feedback loops and hence reducing the number of initial guesses that are needed in order to undertake highly dependent tasks. Previous investigations have attempted to define a generic structure for combinatorial optimisation using GAs [Todd, D. (1997). Multiple Criteria Genetic Algorithms in Engineering Design and Operation, Ph.D. Thesis, Engineering Design Centre, University of Newcastle upon Tyne, UK.], however this paper demonstrates that the structure of the GA is intrinsically tied to the domain. The focus of this paper was an investigation to determine the most efficient and timely structure for the GA with respect to process optimisation. Additional criteria are included within the system and it is has been demonstrated that the structure is applicable for these criteria. It is therefore assumed that if the criteria are dependent upon the matrix representation, in particular, the sequence of the activities and dependencies, then the GA structure will remain applicable. This assumption was demonstrated to be correct when the DSM and GA were used with the same GA structure to optimise component modularity using different optimisation criteria [Whitfield R.I., Smith J.S. and Duffy A.H.B. (2002). Identifying Component Modules, Seventh International Conference on Artificial Intelligence in Design AID’02, Cambridge, UK, 15–17 July 2002.]. The results indicated that the independent position based crossover and shift mutation operators with 60 and 20% probabilities respectively was the most successful structure for the GA. A relationship between the number of activities and the number of evaluations was determined and may be used to eliminate unnecessary computation in future investigations.