Stochastic simulation studies of dispatching rules for production scheduling in the capital goods industry

1. Stochastic simulation studies of dispatching rules for production scheduling in the capital goods industry Chris Hicks, Business School Fouzi Hossen, Mechanical & Systems Engineering

2. Introduction • Dispatching rules are used to choose which part to process next when there is a queue. • Typical rules are: first come first served, earliest due date first, shortest operation first etc. • Most research has focused upon simple situations such as job shops • Previous research has ignored uncertainties and other operational factors.

3. Objectives • To investigate the significance of operational parameters (minimum set-up, machining and transfer times and the data update period) in companies that produce complex products in low volume, with Beta distributed processing times and infinite capacity. • Find the configuration which gives the best performance under infinite capacity conditions. • To investigate the relative significance of these factors and dispatching rules with finite capacity. • To find the best/worst dispatching rules at component and product levels. • To find the importance of dispatching rules relative to the operational parameters.

4. Capital Goods Companies • Produce complex products with many levels of assembly in low volume. • Produce different product families e.g. main product, spares and subcontract products. These families are subject to different time scales, competitive criteria etc. • There is a lot of contention for resources. • There are many sources of uncertainty: process times, machine breakdown, delivery of materials, engineering changes etc.

5. Typical Product

6. Case Study • Based upon an 18 months schedule obtained from a collaborating company. • 56 products from 3 product families. • 3,360 components with 5,539 operations processed on 36 resources. • Main products had up to 8 levels of assembly. Other product families involved the manufacture of components or single level assemblies. • Measure performance in terms of mean tardiness.

7. Screening Experiment

8. Regression equations Only statistically significant factors included. Predictive models identify ‘optimum’ performance and impact of the factors

9. Results • Company significantly behind schedule at the start of the simulated period. • Not possible to meet due dates, even with infinite capacity. • All coefficients were positive indicating that the best results would be obtained with the low levels of the factors. • The Beta distribution was statistically significant in all cases, but the impact was small. • Beta distribution used was most important for the main products, which had many levels of assembly.

10. Finite capacity experimental design

11. Results • The Beta distribution did not change the relative significance of the factors • The best rule was different at component and product level and varied by product family • Minimum transfer time was the most significant factor, followed by the data update period.

12. Conclusions Infinite capacity experiments • Minimum transfer time most important factor, followed by data update period. • The Beta distribution used was statistically significant, but the difference in mean tardiness was small. Finite capacity experiments • Relative performance of dispatching rules not affected by the Beta distribution used. • The ‘best’ rule varied by product family and was different for components and products.