Periodic Pull Strategy - itemtype specific

Hello all!

I have a question regarding the pull strategy in the Sink of my model. I want to execute two tasks:

1) The sink should pull the exact quantity of items according to their itemtype (see columns in the global table "Demand"). The itemtypes are located in their specific queues.
2) The pulling should happen periodically (Period length, e.g. 20 time units). The periods are the rows in the global table "Demand".

The Pull Strategy in the Sink offers me the option "Using Global Lookup Table". This is certainly the right option that brings me closer to my task. In order to go through the periods and execute row per row I tried a for loop:

int period;
for(period; period<=gettablerows("Demand"); period++)
{
result = gettablenum("Demand",period,getitemtype(item));
}
return result;

Unfortunately this loop doesnt work.
More precisely asked: How can I pull the exact quantity of items from the respective queue (itemtype specific) PER ROW? Isn't it something like: gettablenum("Demand", x , getitemtype(item));?

Additionally the periodic pulling inflates the complexity of my issue. I would have said that the above loop needs to have a separate (higher ranked) loop too, that dictates the model to perform the Demand loop periodically.

But I dont know if this is the "golden way" to do these two tasks. I am aware of the fact that ths periodic demand issue can be solved by a User Event too. But wouldnt it be simpler to do all the tasks in one place (Pull Strategy of the Sink)?

Thank you very much for your help! I appreciate any suggestions and coding hints!

All the best!
Simon

Hi simon,

1)The easiest thing is to keep a combiner infront of the sink and pull the itemtype that has variable batch size.Here is a model Ralf has built few years ago

2)If you want to sent items periodically then give the combiner a processtime of 20 length units or store this parameter as a label on the item so it knows how long the "processing" takes.

Simon,

there are two independent things you want to do:

1. Periodic pulls
The pull strategy evaluation is event triggered, not time based, therefore you have to do something beside the pure pull strategy to get the timimng right. You could use a user event to openinput of the sink periodically causing the pull strategy to fire. I would use the same event to read the itemtype and number of items to be pulled from your table and set the numbers on labels on the sink.

2. Pull right number and itemtype
It is not enough just to return the number of items. This is what the pull requirement does: It is fired for every single upstream flow item that is available for transfer. If the pull requirement is met for this item, it is transferred. Which means you have to check two conditions in your pull requirement:
- Does the item have the correct itemtype to be pulled according to label 1?
- Do I still have to pull more items of this type according to label 2?
You will need an additional counter or just have to count down label 2 to zero.

After you have pulled all items you need, you have to close the input again to make sure, no additional items are pulled.

Is this enough of help to get you up to speed?

Good luck
Ralf
FlexSim

Thank you Sagar and Ralf for your replies!

The idea with teh combiner is ok. But i am not sure if it increases complexity too much.

1. Periodic pulls: Alright, I got the idea. It is basically what was done in the Single-Item case. Sending a message with a label value via UserEvent to the sink. The sink has several item-type specific labels that are filled with the information from the message. The OnEntry trigger counts down the entering items and closes input when the specific label gets to zero. The pull strategy is fired periodically...
2. Pull right number and item type: what do you mean with label 1 and label 2? I got the idea with the additonal counter that counts down to zero. What do you mean with label 1? Is it an additional label that indicates the itemtype?? i am confused....

sorry for the obstacles and many thanks for the help!

Simon,

I thought of reading the itemtype and no of items out of a global table as you wanted to originally. Then set label 1 to the itemtype and label 2 to the no. of items to be pulled.

You should check in the pull requirement, if the no. of items is already reached instead of OnEntry.

Good Luck

Ralf
FlexSim

OK, got it I guess!

I am currently working on the User Event suggested by you. I formulated:

/**DEMAND PROCESS*/
treenode current = ownerobject(c);
if(time() == 0)
{period = 0;}
period ++;
if(period > gettablerows("Demand"))
{
       stop();
       msg("UserEventDemandDay", "No further Demand!");
}
int P1 = gettablenum("Demand", period, 1);
int P2 = gettablenum("Demand", period, 2);
int P3 = gettablenum("Demand", period, 3);
int P4 = gettablenum("Demand", period, 4);
int P5 = gettablenum("Demand", period, 5);
int P6 = gettablenum("Demand", period, 6);
int P7 = gettablenum("Demand", period, 7);
senddelayedmessage(SinkDemand, 0, model(), P1);
senddelayedmessage(SinkDemand, 0, model(), P2);
senddelayedmessage(SinkDemand, 0, model(), P3);
senddelayedmessage(SinkDemand, 0, model(), P4);
senddelayedmessage(SinkDemand, 0, model(), P5);
senddelayedmessage(SinkDemand, 0, model(), P6);
senddelayedmessage(SinkDemand, 0, model(), P7);

First, I didn't indicated precisely the columns when i defined int P1...P2.
Instead I wrote: int P1 = gettablenum("Demand", period, getitemtype(item));
But it didnt work out. What is the problem with this reference to the column? There must be a way to get this more flexible in case I want to add some more product types, right?
When I do it the way I indicated it above the labels in the Sink (P1...P2) are increased by the sum of the row although I explicitely indicate the column (1,2,...7). How does this come?

Thank you!

Simon,
you do not have to create all calls in the code of the user event. Make the user event repeat periodically and you do only have to call once.
You can calculate the period from the time the event is called or just make it a global variable.
You cannot use "item" in user event codes, since it is not specified. In general "item" references the item that caused a function to fire (see OnEntry). For these cases it is declared and assigned in the top rows of the function.
I thought your table would only have two columns: One for the item type you want to pull and one for the number of items to be pulled of that type?

Best wishes
Ralf
FlexSim

Thanks for the reply!

No, my global table looks like the global table in the model above attached!

The columns are the Product (Item) types and the rows are the periods.
For each period I want to generate a specific demand (quantity) per item type. in other words: in each period each item type has a particular demand quantity.
This is the reason I call several times in the User Event...

Well,
in that case only call it once and loop through the columns of the table (getttablecols()). Create labels on the sink for each column for the number of items per type only. You can call labels not only by name, but also by rank: getlabelnum(current, 2); would refer to the second label on your current object.
This way you can easily be flexible on future changes like adding new itemtypes. You can even check, if the number of labels is matching the number of columns in the table (content(labels(current))) and add labels dynamically, if you need them (addlabel()).
Or you just add a large number of labels that you will never reach in itemtypes and never worry again

Here is my question to make it more complicated: Do you assume that you can fulfill demand at every periodic call or is there a chance, that demand can't be fulfilled for an itemtype? If so, what are you going to do?

Best wishes
Ralf
FlexSim

A solution attached.

Global variable "rememberdemand" set to 1 or 0

Approach:
buffers hold items if there is no unsatisfied demand for that type or rememberdemand is 0;
the sink message releases items.

hi all!

sorry for the late reply!

first of all, thank you very much for your helpful comments and suggestions!

Special thanks to Jason for his model! This really helps me alot!

After having studied your contribution I have some questions:

In many periods I encountered a very strange behavior for some queues (itemtypes).
1) In period 8 the value of the label "unsatisfieddemand" indicates 5 in Queue1 (item type 1) although the queue has several items to fullfill the demand.

2) In period 3 only 2 instead of 9 items of item type3 (queue 3) are released to the sink. Is this a "random system beaviour"??

3) In period 4 again the label "unsatisfiedddemand" remains positive (1) although items are available to satisfy demand. (queue 4)

4) The exact situation as in 3) for period number 8 (queue 7).

5) In period 6 only 2 items are released to the sink. The label "unsatisfieddemand" becomes 5 (the total demand in this period is 7) ALTHOUGH sufficient items remain in the queue to satisfy the total period demand (queue 6).

Queue 2 and 5 work perfectly! Is this somehow related to the demand structure? I tried it with a different demand pattern - same bad result.

Or maybe I don't get the logic of the model...I am not sure, I thought I did...

Maybe you can have a second look on it? I checked the coding in the queues. There are any differences at all that could explain the deviations.

Thanks again!

Best wishes!
Simon

Ok, some of my critics vanish in the case I use initial stock for the queues (this is a prerequiste of my model!).

Attached the jason's model modiefied with the initial stock (coded by me). Now queue 7 behaves strange. Again the label "unsatisfieddemand" takes on a positive value although sufficient items are in queue 7 to fullfill demand.

Is there a logical issue I don't get? Or am I right if I suppose a little bug in the model?

Thanks again for helping me out! I am vey thankful for this expert board here!

Sorry - I didn't make enough checks - please try this fix. You shouldn't need starting stock for it to work.

I only modified the message trigger (above) so you could copy that to your version with initial stock.

Thank you Jason! It works perfectly!!!

Hi all!

After I have implemented Jason's idea for the demand pattern, I extended the model by a second stage. The logic remains the same for both stages. Once a predefined ReorderPoint is undershot, a order with a predefined ReorderQuantity is given to the upstream supply chain partner. In my model the problem is now that the single object "SourceRaw" should execute this logic. I defined several labels for the specific itemtypes in the Source. Once the source receives a message from I1-7, the label should take on the value of the message parameter and release the quantity defined. OnExit the labels get reduced successively and as they become "0" the output gets closed.

The centerport connection between SourceRaw and the queues I1-7 is important for the messaging between the objects. Is there a mistake in my definition for the for the messaging partner (OnExit Trigger of I1-7)?

treenode SourceRaw = centerobject(current, 2);

I defined the treenode SourceRaw in all the queues (I1-7) this way. Or maybe the problem is located elsewhere?

Why am I don't using separate sources that are releasing me the particular itemtypes?
The idea is to give EACH item a specific numberID independently of its itemtype. I modelled this in the Single-Product case in the OnExit Trigger of the Source:

//Set ItemID
int UnitNumber;
inc(label(current, "UnitNumber"), 1);	
setlabelnum(item,"ItemID", getlabelnum(current, "UnitNumber"));
//Write ItemID into Global Table
settablenum("FlowTime", getlabelnum(item, "ItemID"), 1, getlabelnum(item, "ItemID"));
settablesize("FlowTime",gettablerows("FlowTime")+1,gettablecols("FlowTime"));

With several sources this might be not possible?! Is there a way to do this with one?



What is the whole idea of my project? The following paragraph gives you a clue how I would proceed:

Several scenarios (batchsize combinations on the different stages) are investigated and I want to track the flow time of each item through particular areas in the model and write it (flow time) in a global table according to its ItemID. This is easily done - setting labels on the items themselves, etc.
I already implemented this for the Single-Product case. I assume for the Multi-Product case it is similar. After each scenario I have a huge global table with thousands of ItemIDs and the respective flow times. I do an Excel export (csv files) and calculated manually the average flow time per item for each scenario. This is a doable but rather cumbersome way to reach my final goal. Maybe there is a suggestion from your side that to make this whole procedure less time consuming? I state the whole problem here because I think it is related to my specific question above. If I do not need the ItemIDs I could implement several sources...


Again thank you very much for your patience and help! I hope I made all points clear - if not, please ask me!

All the best!
Simon

Hello Simon,

I don't understand what are you looking for. You are investigating the batch size in different stages of your model. What is your goal:
batch size to setup time,
batch size to spending time,
change interval at the processors to stay time,
queuing capacity to batch size,
...?

Jörg

Hi Jörg,

the aim is to depict the average flow time per item as a function of the different lot sizes on the two manufactruing stages.

In order to achieve this, I programmed my model (Single-Product case) as I described it in my post above. For different batch size combinations (=scenarios) I tracked the flow time (waiting time plus setup plus processing time) for EACH item through the system. After each scenario Flexsim generated an Excel file that contained the flow time for each item which has passed through the system and I manually calculated the average flowtime per item (sum of all flow times divided by the number of items). Is it more understandable now?

Thank you!

Additional thoughts:

I asked myself if it was easier to use the "Performance measures" in the Experimenter tool in order to achieve the same as I described above. As a performance measure you can extract the "average staytime" from an individual object or a group of objects ("Statistic by...").

The question is: How is the "average staytime" defined? In this definition for example:

value = getnodenum(stats_staytimeavg(current));

Is "stats_staytimeavg" the average time (waiting+operating) of an ITEM in an object during the simulation run time?

Please write me, if you are not clear with my question. Although it is a rather simple task I want to achieve, it is quite difficult to explain it properly.

Are you looking for the time starting on creation until the item leaves the model.
The value is:

double value=time()-getcreationtime(item);

in the On Entry Trigger of the sink.

Edit:
If your start time isn't the creation of the item, you can add a label to the item, which you set to the value of the function time(), when the event function (trigger) is called. You build in the the sink the difference of the actual time to the saved label value.

Is "stats_staytimeavg" the average time (waiting+operating) of an ITEM in an object during the simulation run time?

This is the average staytime of ALL itemS of the Object.


Jörg

Attached you find a model with the average staytime of three products differing itemtypes.
Jörg