logistic_control

Logistic control is a means of achieving lean production, by making workpieces flow more smoothly within a factory. In a smoothly-running production system, there are fewer requirements for safety stocks, which can release capital normally tied up in inventory.

It is important not to reduce inventory too far, since the risk of disruption to the manufacturing system will increase. it is possible to become too lean! When a system becomes too lean, new costs are introduced, in the form of penalties, costs to expedite late orders and less tangible things like the loss of reputation. Fortunately, our models can be employed to find out just how far a production control strategy can be pushed.

There are a variety of different production control strategies, each with strengths and weaknesses. Exactly which is best for a particular business unit – and with what parameters – can only be determined by experimentation.

Let’s begin by considering a simplified, fictional manufacturing system where a product must undergo processes on three machine tools:

The model of our fictional manufacturing system needs some more information, in order to be fully defined. For example, how long does each process take? If all the operations have the same duration, that’s great... we have a flowline, and it won’t require much logistic control, as long as we can keep it supplied with raw materials. In reality, life tends to be more complicated. Let’s imagine our processes are of different duration:

Above, you see the simulation as it would look when the elapsed time was zero. At the start of the simulation run, all the workpieces are ‘heaped’ into the system, so they go straight to the queue for process one. One workpiece moves on and starts to be machined; the others remain in the queue.

Nothing changes for twenty minutes. Then...

That’s what our fictional manufacturing system would look like, represented within Arena. Each of the processes is now represented by a ‘seize, delay, release’ module. It’ll attempt to get the resources it needs (machine tools, operators and maybe fixtures, etc.), for the durations we have specified. Note the bars above each process; these will fill up to show parts queuing to begin the process.

At the moment, it has no logistic control at all. This is sometimes referred to as a ‘heap and hope’ approach to logistic control. The theory here is that if we start every job as soon as possible, some workpieces must get through the system, so we can start making deliveries.

Here’s what happens...

Thus, we need the simulation to be driven by a schedule of deliveries, and this is our first real system of logistic control. Since it takes time to produce finished goods (in a real system it might be many days or even weeks), we don’t use the schedule of deliveries directly, to launch the products into the factory. Instead, we adjust the schedule to reflect the time a part will normally spend inside the factory. For example, if a product was due to be shipped on August 19th, and we know that it will normally take four days to make that product, we’ll start making the product no later than August 15th. This offset is called the standard lead time and this system of logistic control is called Material Requirements Planning (MRP).

One workpiece has now finished process one, and moves on to process two. Because the machine for process two is standing idle, there is no need for the part to enter the queue. In fact, because process one takes longer, you should never see a part in the queue for process two under normal circumstances. Note that another workpiece has begun process one, as well.

The simulation continues running in this fashion, until all the workpieces have been processed. If you watched this simulation running, here’s what the simulation would show when three hours had elapsed...

Three workpieces have now been completed, and have left the simulation. It is now clear that process three is the bottleneck; the queue there is building steadily. You could ‘heap’ as many workpieces into this manufacturing system as you wished; it won’t make them come out any faster.

The example presented here is a very simple simulation; you could perform the same experiment by hand, using tokens such as coins, and moving them between workcentres drawn on a sheet of paper.

Computer-based simulation really comes into its own when there is more complexity. What if there were several product variants, and each required a slightly different process sequence? What if process times were somewhat variable? What if the simulation was to be run for a year? What if resources are more constrained, and there are only two workers available, to serve the three machine tools? Having created a computer-based model, it is easy to explore alternative scenarios like these.

Note that most of the workpieces spend most of their time waiting (in queues), rather than being worked upon. Since inventory costs money, this could indicate an inefficient system. Can we do better? Of course we can! There are a great many systems of logistic control that might be applied, even to this little system.

The most obvious thing to do is to adjust the rate at which work flows into the manufacturing system. Instead of ‘heaping and hoping’, you would introduce a new workpiece every forty minutes, since this is the best rate at which finished goods can leave the system, due to the time required for process three. To be still more clever, we wouldn’t just make products as quickly as possible, but in response to customer demand, so as to be sure we are doing work that somebody is going to pay for.

Kanban is well-suited to a manufacturing system where there are few variants, and where volumes are reasonably constant. It is a control system rather than a planning one, holding the level of work-in-progress strictly in check, because there cannot be more parts in the system than there are kanbans, and the manufacturing system can be fine-tuned by adjusting the number of tokens in circulation. Because the system responds directly to customer orders as they arrive, it is important that processes are reliable and quality is consistent. Long machine set-up times should also be eliminated, where possible.

An interesting variant of this approach is Kanban Squares. Here, there are no tokens in circulation. Instead, the buffers themselves are monitored and production only takes place if there is a space in the buffer immediately downstream. The squares can actually be storage racks or painted areas on the factory floor. When constructing a simulation, the kanban squares approach is relatively easy to model, because there is no need to represent the movement of tokens, only workpieces.

Under MRP, manufacturing systems are dependent upon firm orders being placed in advance, or good forecasts. A measure of control can be achieved by varying the standard lead time, to balance on-time delivery against the cost of inventory. Note that MRP does not take manufacturing system capacity into account, so an unusually large order might not be completed in time.

That system of logistic control still involves ‘pushing’ the products into the manufacturing system, and waiting for them to come out. Alternatives exist where products are ‘pulled’ from the delivery end instead. One such system is Kanban. In this system, no process is carried out without a work instruction. Instructions take the form of tokens, and it is these that give the system its name. (Kanban is Japanese for ‘card’.) The tokens flow ‘upstream’ and the products flow ‘downstream’. When an order arrives, it is fulfilled with components from a buffer of finished goods. If there are insufficient finished goods available, process three draws some partially completed products from a buffer, upstream. If there are not enough of these available, process will be triggered, and so on.

There are many other systems of logistic control, including drumbeat systems (where all activities are forced to operate with a common cycle time), conwip systems (where the amount of work-in-progress at a group of machines is controlled by kanbans, rather than controlling indiivdual buffers), and period batch control systems (where operations are grouped, to have similar overall cycle times).

In reality, few real systems can be controlled by a ‘pure’ implementation of any one method of logistic control, and our case studies were no exception. The mixture of product types being made, in varying volumes, with processes that had widely different cycle times, meant that a hybrid system of logistic control is often a realistic choice. Ultimately, there is no ‘best’ solution; only the one that most closely matches the requirements of the customer, the product and the production system – including the people who must operate it.