A key cause of variability in manufacturing systems can be the breakdown of equipment. The science of modelling breakdowns is, to my mind, underdeveloped. This is recognised in seminal texts by simulation luminaries such as Averill Law. It also matters. Too often in models historical patterns are used with little analysis of underlying causes of variability. Historical patterns are good but can offer problems of limited data, non-stationary data, and data which has been superseded by change.

It seems obvious to say it but the right approach is to model at the correct level of detail. Where necessary an experienced WITNESS modeller will use the multiple breakdown options on a machine to define many different causes of interruption. This more accurately represents patterns of breakdown than a single set of breakdown and repair time distributions. The distributions used are vital too. It is too easy to choose ‘random’ failure distributions, such as the Exponential, whereas in reality there are many different effects that build up a complex curve. These include effects such as ‘infant mortality’ where a badly machined component may fail on the first few cycles, or ‘wear out’ where a component assumes higher failure rates after a set period of use. Maintenance patterns only complicate the picture.

All of this warrants careful thought and application.