|
|||||||||||||||||||
|
|||||||||||||||||||
     |
|||||||||||||||||||
|
|||||||||||||||||||
|
Building simulations can be easy but what can make it hard and often gets overlooked till mid way through a simulation build, is managing your data. Do you have data to input? If not how do you collect the data you need? Do you have too much data? How do you summarize the data to get it into your simulation? Data woes can mean your simulation project will fail, after all "garbage in, garbage out". Data, What Data? Having too little or no data is the most common problem you will face and the problem that most people think is the hardest to overcome. After all if you have no data to input what can the simulation possibly tell you about your system? The most obvious thing to do is to collect the data you need. You can do this in a variety of ways: surveys, interrogate work flow management systems or get yourself down to the factory floor with your stop watch. Data collection is difficult and error prone. You might decide to collect data on a really bad, or really good day thus biasing the results. The way you collect the data has to be consistent across all repeated measures. This includes trying to keep the environment the same. What if the factory is really warm on the second day you collect data this would probably slow down all the workers and could skew your data so should probably be disregarded. By far the biggest issue is that data collection is a social process, that is it involves people, both as subjects and as data recorders. When people are being observed they will often behave differently, usually better. You can try to observe them covertly but this isn't always possible. Depending on how you collect the data you may influence it. For example, if you sit with a stop watch and time stages in the process you might not start your watch at exactly the time the step starts. Also you have to make decisions about whether to follow one part through the entire system or just record hundreds of parts going through one step then do the same at the next. Each approach could be right or wrong in different circumstances. If there's more than 1 person collecting the data how do you make sure you don't duplicate collection. If each person collects data for one section how do you match the data together, you need some method to identify which observations relate to the same parts. These are just a few of the many, many issues with data collection. Of course you could create or invest in a software system to collect your data. This would get rid of most of these issues, but the amount of time it could take to get a system up and running could be very lengthy, the need for your simulation could even have disappeared by then. There is an alternative. You could identify "experts" in the process. This could be the person who works on the shop floor or it could be the manager; it will vary. You're looking for the person who knows the process inside out. Through discussions with experts you can often get all the data you require. Although, you do have to be careful how you conduct the conversions/interviews as how you phrase the question. The relationship you establish could effect the answers you get. If you can't find any experts or don't feel confident with the data they have provided you can still complete your simulation project. Assuming you have managed to collect enough detail on the process flow through your system (and the decision rules used) you can use your simulation to find the limits of the system. Test the maximum number of parts the system can process per day or how many staff are required if calls take 15 or 30 minutes to process. This kind of information can still provide lots of valuable insights into your system. It may also generate enough interest for you to get support to then go on and collect the necessary data. Data, Data Everywhere It may be hard to believe, but often having too much data can be just as big a problem as having too little. To begin with you have to plough through streams of data to get to the handful of entries that are of relevance to you. The data will probably not have been collected with your purpose in mind so you'll have to interpret or try to amend the data to fit your purpose. By doing so you may invalidate the data. There is also the issue of interpretation. Was the data collector's definition of a result the same as yours? For example, some people may define the cycle time of a machine as including the breakdown times; others may not. Such assumptions could be crucial to the accuracy of your simulation. As the section above suggested collecting data is hard. You'll have to find some way to make an assessment of how well the data collection exercise was conducted and hence whether you trust the data you've been given. This may be difficult to do. If there is a wealth of data available to you, even if it does not quite fit your needs there will be pressure on you to use it. Especially if you are an external consultant you may have no choice but to use it. The solution (as with having no data) is to find an expert in the process or the data who can help you interpret it Summarizing and Inputting the Data Let's say that you have overcome all the issues of data collection and now have just the right amount of data, you still have to find a way to summarize this data and input it into your simulation. If you have lots of data points to summarize it's probably worthwhile investing in distribution fitting software. Such software will take your raw data from spreadsheets, text files, or manual input and convert that data into the appropriate distribution. Generally they'll provide relative comparisons between distribution types, and an absolute measure of each distribution's acceptability. Stat::Fit for SIMUL8 is ideal because it will let you import the results directly into SIMUL8. If you're data doesn't neatly fit any particular distribution you may want to consider feeding SIMUL8 the actual data, say the exact times work items arrived at, using Visual Logic. See the Learning Zone for an example of how to do this. Alternatively you may want to use SIMUL8's probability profiles. These distributions can be configured entirely under your control to generate any shape of distribution desired. They look like a histogram but you can drag the blue bars up or down to change the probability (percentage chance) of each sample value occurring.
If you didn't manage to collect any data or are working with your expert with "best guesses" then as a general rule use an Exponential distribution in a Work Entry Point and a Normal or Average distribution in a Work Center. More Reading I have briefly dealt here with just a few data woes and how to input your data to SIMUL8. If you'd like to read more on this topic here are some references you might find interesting. Robinson S, Simulation
The Practice of Model Development and Use, Chpt 7, John
Wiley & Sons 2004. Our website is always being updated with new support resources. Here are details on some of the downloads that were added this month. New SIMUL8 Service Pack Version 11 Build 644 New Stat::Fit for SIMUL8 Service Pack Version 2 Build 8 Which Breakdown Quickly Add Resources Online Questionnaire for Simulation Users Thomas Pohl, a Ph.D. student at Sheffield Hallam University in the UK is conducting research into the flexibility of simulation models and ways to improve that flexibility. He is also working on a comparison of the most widely used simulation programs. In order to obtain more substantive data from simulation software users he has created an online user survey. If you could spare a minute or two to fill in his questionnaire it would really help him out. The results of this survey available at some point in the not too distant future. If you have any questions, please feel free to contact him at Thomas.Pohl@shu.ac.uk Last month we launched Master Classes. Master Classes give you a chance to get access to our most experienced consultants. You learn lots of non-documented hints and tips and get a chance to make suggestions on how we can improve SIMUL8. Master Classes cover different technical support topics each month and include a Question and Answer session. Master Classes are free and are conducted over the internet via WEBEX Conferencing software. The next scheduled Master Class is on 23rd July at 10am (EST) and will be presented by Mark Elder our CEO who has over 20 years experience in simulation modeling. The topic will be 'Writing Fast Simulations'. See our support website for more details and to sign up. If you have any questions generated from, or comments about this
newsletter then email Newsletter@SIMUL8.com
|
|
|||||||||||||||||
