Performance Evaluation and Buffer Sizing

Industrial factory planners who are responsible for building up sufficient production capacity in an economical way are confronted with a number of design factors that have an effect on the throughput of a planned production system subject to stochastic influences (breakdowns, random processing times). In addition to technical considerations, such as the definition of the processes and the specification of the required production and material handling resources, there are several organizational issues that have to be decided upon.

In the following, we deal with the question of how the negative consequences caused by random phenomena such as breakdowns or the variability of the processing times can be predicted and how the resulting loss of throughput can be regained through the introduction of buffer spaces between the stations.

Many flow production systems comprise special-purpose machines that repetitively perform a certain number of tasks on a single product type. In this case, the processing times are deterministic. Randomness arises from breakdown and repair processes.

There are several situations where stochastic processing times must be taken into account. First, if the task is repetitively performed by a human operator, then processing times will usually be random, as a certain amount of variability is inherent in human nature. Empirical studies show that task durations of human operators will have a coefficient of variation that is considerably less than one, which would be the case for an exponential distribution. Secondly, if a number of flexible automatic machines or robots assigned to a station are able to process a mixture of product variants in any sequence, then — from the point of view of an external observer — the processing times of the variants can be considered as random.

In industrial practice there is only limited use of mathematical planning models and software. In most cases only simulation is applied to evaluate a design specified by the planner based on his personal experience. To develop and use a simulation model in order to produce statistically significant results takes a lot of time. For example, a detailed simulation of the design of a car body shop may last about five to ten minutes on a fast PC. A planner who is trying to find the optimum system configuration will have to perform multiple simulation runs. Consequently, many flow lines found in industrial practice are improvable.

Buffer optimization provides answers to the following questions:

1. How many buffers are required in the flow production systems in order to achieve a target production rate?   This is called the Primal Problem.
2 How should these buffers be distributed among the stations?   This is called the Dual Problem.

Both questions are tightly related. In any case, buffer optimization requires the evaluation of a given system configuration, including the buffer sizes. This is evaluation done by performance evaluation methods.

Software for performance evaluation and buffer optimization of flow lines:
POM Flowline Optimizer A software system for the design of flow lines with limited buffers. Read more