A novel metric for determining the constraining effect of resources in manufacturing via simulation

Abstract

This paper provides a novel method for determining the constraining effect of resources in a manufacturing system using discrete event simulation. Traditionally manufacturing systems are constrained by one or more bottlenecks. Eliminating or mitigating the bottleneck will speed up the system throughput. However, bottlenecking resources generally only refer to machines, and primarily focus on flow-shops not job-shops. One important resource we believe that is often overlooked is workers and their associated skills, and we propose that a particular skill could be flagged as a bottleneck resource. We define new metrics known as resource constraint metrics (RCM) for measuring the constraining effect of a resource on the entire manufacturing system. These metrics are flexible and differentiate between the constraining effects of machines and their requested skills. The metrics can also deal with complex workflows with alternative routing, alternative resources, calendars (a necessary consideration when dealing with workers), worker performance, and multiple modes of operation of machines (e.g. run, setup, and maintenance). The use of RCMs in simulation aids in real-world decision-making, by determining which resource should be focussed on and improved to reduce the overall system feeling constrained. This will have the effect of increasing throughput or at least providing the capacity for increased throughput.

Keywords:

• manufacturing
• enterprise
• resource
• machine
• skill
• starvation

Notes

Notes

1. Excess processing time that is incurred due to worker inefficiencies. See first paragraph on next page for a more detailed explanation.

2. EMRs will never be in queue for a machine. When a machine fails, an EMR is generated and instantly replaces the order that was on the machine at the time of failure.

3. This is a prediction within the simulation at run-time however, and may actually turn out to be wrong. For example, if the selected machine fails unexpectedly, the alternative machine may have been a better choice.

4. We use sample standard deviation, with Bessel's correction for measuring the spread of the results, which we store in a separate RDB. We do not use confidence intervals, since confidence intervals assume that the outputted results follow the normal model distribution which may not be the case.