While much of the work at this stage addresses the gathering of information that is to be built into the model, it is also important to decide what to leave out. Some factors do not need to be described to a full level of detail – indeed, modelling them may actually make the simulation less informative. For example, in the business unit under study there were a number of operations performed at shared resources; these are located outside the focused factory, and they work on a broad range of other parts. It would have been highly complex to model the arrival of all these parts at the shared resources. Instead, a statistical distribution describing the time each TEC had spent outside the focused factory does the same job.

 

The software used to create the model of the focused factory was Arena, from Rockwell. In this software, a simulation is constructed as a network of nodes that represent activities or events. The icons do not represent machine tools! The diagram below shows a simplified example of an Arena model:

The construction of a simulation is a lengthy process, and it is important that the work is approached in a realistic and methodical fashion. The sequence of activities shown on the right indicates good practice; note the amount of iteration that is necessary.

A lot of work must be done before the simulation software is used at all. First of all, what are the aims of the simulation exercise, and what is its scope? The simulation should be sufficiently complex to allow meaningful results to be obtained, but it might well be a mistake to attempt to simulate a large system in great detail.

Key tests are that the model will yield information that is useful, and that the information is sufficiently accurate. When sufficient confidence has been obtained, the parameters for a programme of experiments must be determined.

Once experiments have been performed, the results are analysed, and this might well lead to different parameters being used in another round of experiments, or changes to the model itself.

This was the process that was followed for the Volvo Aero focused factory where turbine exhaust casings (TECs) are made. We began by selecting an appropriate scope for the model. Other simulations might have been made, examining the sequence of operations performed at one machine tool, or examining the entire factory, but a trade-off must be made between model development effort and the kind of results that are sought. Our brief called for logistic control to be investigated at the business unit level, so suitable information was gathered:

A sequence of operations for each product type:

A list of all resources available within the business unit:

Other information that was required included typical production volumes, and the kinds of disruption to which the focused factory was prone. (It was found that the major problems were the late arrival of raw materials, varying demand patterns, and occasional problems with machine breakdowns or components failing an inspection.)

The process plan for each product was detailed in two documents; an “operations flow diagram” identifying the sequence of operations that were performed, and a table supplying additional information. Historical data had been employed so that timings could be expressed as a statistical distribution.

The model of the focused factory was much more complicated, but it was basically constructed in the same way, dragging modules into place and giving them parameters to identify the resources that each activity requires, and so on. The completed model contains something like 2,000 icons, although most are held within ‘sub-models’ to keep the main screen of the model as simple as possible. The screenshot reproduced below shows the finished model of the focused factory; it has been made to resemble the operations flow diagrams for the two main product types within the factory.

Completing a large model is an achievement, but much remains to be done. In order to meet the aims of the study, this must be tested to ensure that it is sufficiently accurate, and then it must be used in a programme of experiments. Even then, the results obtained might lead to further model development. These stages are described in the pages that follow.