Using probability to indicate machine/operator availbility

Hi all,

I remember reading some where in this forum about ways to set the machine or operator availability by probability. Could any kind souls lead me to that thread? I.e. say 90% that the resource is available.

Thanks

David

Hello everybody. I bring this thread up again because I have a similiar question.

Thank you in advance!
Jens

I think there are two ways.

1. Timetables and setting shifttimes (if you know the exact down time)
2. Breaks (Tab) and setting a Distribution for MTBF/MTTR

You even can use a mix of it.

Careful. A probability does not define frequency/behavior. As an example lets say 2 machines are 90% available:

Machine1 is down for 1 minute every 10 minutes.

Machine2 is down for 10 minutes every 100 minutes.

Note that Machine1 breaks down far more frequently compared to Machine2. But when Machine2 fails, it is down for a much longer period of time. In reality there is a big difference between these two machines, yet they are both 90% available.

In simulation we attempt to model the behavior of parts of a system and thus use the approaches that Sebastian posted.

I'm not saying that we can't make an operator take a job based on a probability in Flexsim. But before you do, you might want to ask yourself if you are trying to cut corners that you shouldn't be. It's easy to think this way when lacking good data.

Yes exactly. But how would you approch this problem, if you don't know the real planned downtimes so far.
I just know that we plan our production cells (including automation and the real process) with a TEEP of 75% (and this is still very optimistic)

Would you brake it down to hours (say: one shift has 8 hours, so i could include a 2 hrs downtime with time tables in each shift) or include this percentage somehow (I don't really know how, so far).

Thank you in advance!

I wouldn´t do it by time tables any way. It´s to static. And with breakdowns you can set extra states, while your object is down. This may help at scoring results later.
Probably you even want to add shift times anytime, because if failures are important, breaks could be relevant too!?
Normaly I had a look what kind of failure could happen and how long they would take. But if you can´t find out, setting the distribution for every hour sounds well and should be easy handling later.

Using the hourly distribution also seems to be the most appropriate approach in my eyes.

My question now is: How to realize this in the best way?

Would you just use fixed expressions for
MTBF (3600*75% = 2700)
and MTTR (3600*15% = 900)

Note: The simulation uses seconds as the basic time unit.

// It works, but as you said .. somehow this is a really botched way ... I try to get real planning data asap =) This solution doesn't make me really happy.

Would use the exponential distribution.
MTBF: exponential(0,2700,1)
MTTR: exponential(0,900,0)
Does what you want. You even could split it down, if you want.
Don´t forget to set down the First Failure Time!
Only for interessts, do you want to use this even for TE´s?

Could you be so kind and give a small explanation why you would use exponential distribution? Because it makes the simulation less static?

I hope I can get more background data for the TE's.
I already know the break times of the shift model and maybe over additional information like work plans and determined control tasks I might be able to calculate realistic breakdown times or availabilities.

I normaly use distributions to make the models less static.
Advantage is, that sometimes situations were found, that nobody thought about. This way I try to analyse weak points in my model or even in planings of projects. And only with distribution a number of Replications make sence. So I use it as often as possible in my models.

Exponential distribution is the best one for this kind of task I think, and it´s easy to understand. But you even can use the "normal" distribution.
If you want to set any values by percentage it´s better to use "bernoulli".

I remember there had been good explanations about using different distributions in the user manuals!?

TEEP (Total Effective Equipment Performance) is a high-level (aggregate) measure of performance. Like OEE (Overall Equipment Effectiveness), it is good for validating the performance of your simulation model. However they are both model outputs, and not inputs.

TEEP is measured using calendar time, and includes all forms of downtime and loss of productivity:
-scheduled down (off-shifts, breaks/lunch)
-maintenance
-starved
-blocked
-yield
-machine failure (breakdown) and repair
-etc.

Note that only the first 2 in the list are "planned", and so Time Tables can be used. Your model will handle starved and blocked states naturally. Yield can be modeled using the SendToPort under the Flow tab. Finally, machine failures are simulated using the MTBF MTTR functionality.

Hopefully now you can see the dangers with calculating MTBF as TEEPx3600s
1. You are assuming that the machine breaks on average every hour (you are assuming the frequency)
2. You are assuming that ALL the other items in the list above do NOT contribute to the TEEP value. For instance, if the machine in your model is ever starved or blocked you have broken this assumption.

A simulation model is always based on a set of assumptions, and you might even be happy with the above assumptions. However, if you don't have any failure data (MTBF/MTTR) you might want to consider leaving it out of your model. Then once the model is complete, you could do a simulation study to assess the sensitivity of the overall system performance to various given MTBF/MTTR behaviors. Then you will know if it is even worth the effort to collect the data... if it's even possible. At least then you will understand its impact on your model's objective, which you wouldn't get from using a bad guess.

Hey Kris,

I'm quite aware of the role of this figure. This is why I'm not really happy with my solution so far, nor with leaving it out completely.

Especially because I'm apporaching the problem from the "planning" site and am only assuming the planned TEEP with 75%.
The TEEP itself as a "naked figure" makes not so much sense in my eyes anyway. It's just useful with the background information about all the factors and happenings that influenced the resulting figure.

Still, we have the experience that because of several circumstances we only can operate our machines with a TEEP around 75%. So we want to include this figure in the simulation.

In my optinion the best way would be to create an algorithm that randomly generates breakdowns, but still the random function is influenced in such a way that we reach an average TEEP of 75%

Regarding the operator availability I received the information now, that the operator will have free time to move the material from A to B for a total of 31 minutes a day. The rest of the time he is busy doing machine maintanance, measurements, etc ...
How do I realize this in a way that makes sense regarding the points we were speaking about in your post and above?

Since I don't have a good solution for the "static problem" of my "randomized" functions yet, I would approach the problem with making the intervalls just smaller. so the worker has so many free seconds / each minute or so ...

Originally Posted by Jens Mühlheimer

I'm quite aware of the role of this figure.

Sorry Jens, I defined TEEP for other viewers from different industries. I didn't mean to be offensive.


Let me answer the question of this thread very clearly:
"How do I use a probability (percentage, ratio) to control machine/operator availability in a simulation model?"

Answer: You can't.

The net availability of a resource in a simulation model is governed by the sum of all factors that I listed in my previous post. In a simulation model I can only define periods when the resource is unable to do the activities that I included in the model. I can define planned periodic breaks, or unplanned failure periods based on statistical distributions.

If all I know about a resource, is that it is unavailable 20% of the time, then I only know what a simulation model is going to try to estimate for me. Say if this 20% represents the ratio of time lost while in repair due to failure, there are no math tricks that will tell me how frequently the resource fails from this one number. Without frequency and duration of the aforementioned periods, a simulation cannot mimic the behavior. There is no way of converting this single value (backwards) into a statistical distribution that encompasses a behavior. Because I don't have definition or even sample of the behavior itself.

Even if I know an average frequency(MTBF) and an average duration(MTTR), there is no way for me to convert these averages into statistical distributions without assuming that the machine behaves like another similar machine that you have data for and curve fit previously. Often machine failure data fits to Weibull or gamma distributions. But that's not a rule of thumb, it really depends on the machine characteristics, including the modes of failure.

@Jens
In regards to your operator, you need to investigate the frequency and duration of the other activities (maintenance, measurements, etc.) that keep him from doing the A-B transport job. As a short term solution, you can proceed and see if the jobs you included in the model take more than 31 minutes. If so, you know you are in trouble even before collecting the other data.

Good morning Kris,

don't get me wrong, I was not offended, but I was more endorsing your statement about the role of the TEEP.

As you said, the TEEP is such a "big" figure with all the background information that is taken into account, which makes it also almost impossible to extract applicable data for the simulation at the same time when you approach the problem "backwards".

I'm totally on your side that it makes not much sense without good and exact data about work plans and additional information about things like TPM, QM, first piece inspections and so on ...