Multi-Step Activities in EZStrobe — Part 1

Multi-Step Activities

Certain repetitive work, such as loading a flatbed truck with five prefabricated beams using a crane, can also be viewed as one multi-step activity that consists of n identical steps. A multi-step activity thus offers a choice between modeling all the work as (a) one long activity, or (b) as n consecutive shorter activities. For example, loading a flatbed can be modeled as one activity that represents loading all five beams, or as five consecutive activities each of which represents loading one beam.

The most straightforward way to represent a multi-step activity in an EZStrobe ACD is to model it as just one Activity and give it a duration that represents the time it takes to perform all n steps.

This approach is possible if all the resources required by a multi-step activity remain committed to the activity during all n steps. The commitment of resources to multi-step activities is illustrated in the EZStrobe network below (click it to see in detail).

EZStrobe Multi-Step Identical Network

There are two multi-step activities in this model: One for loading five beams at the fabrication site (LoadTrk) and another for unloading them at the job site (UnloadTrk). Notice that when combi activity LoadTrk starts, it captures the Loader (for example, a crane) and does not release it until it is done. This means that the crane (Loader) starts loading a flatbed only when five beams are available and it never stops to do something else until all five beams are loaded.

There are at least three possible approaches for specifying the durations LoadTm and UnloadTm for the multi-step activities LoadTrk and UnloadTrk in the above network.

A. The distribution of the time required to complete all n steps is available.

Suppose that the time it takes to load all five beams is known to follow the distribution ScaledBeta[5,10,8,8] (minutes). In this case, the duration of LoadTrk can be specified by an EZStrobe Result called LoadTm which is defined with the formula 'ScaledBeta[5,10,8,8]'.

Notice that it is often convenient to model the duration of an activity, such as LoadTrk, indirectly using an EZStrobe Result. An EZStrobe Result is like a dynamic Excel formula that is reevaluated every time a new activity instance needs to sample its own duration. Thus, every instance of LoadTrk will have a different duration.

Important: The duration of an Activity should never be defined by an EZStrobe Parameter. An EZStrobe Parameter is just a constant whose value could be initialized by a formula (if so desired). EZStrobe will evaluate the formula for a Parameter only once, at the start of the simulation, and will store this result as the fixed value of the Parameter from then on. Hence, using an EZStrobe Parameter is a mistake because it makes all instances of an activity have the exact same duration and zero variability.

B. Each step's duration distribution is available and the number of steps that make up the multi-step activity is relatively small.

For example, the time it takes to unload one beam follows the distribution Uniform[1,2] (minutes) and it takes unloading five beams to complete the activity.

In this case, the duration of the entire activity is most easily specified by an EZStrobe Result that is defined as the explicit sum of n samples from the distribution of the time for each step.

For example, the duration of activity UnloadTrk can be specified by the Result UnloadTm defined by the formula:


If the number of steps is large, then it becomes cumbersome to define the formula for the Result, even with cut-and-paste, although EZStrobe will actually allow the formula for a Result to be up to 64k characters in length.

Important: Note that the sum of 5 independent samples from Uniform[1,2] follows an Irwin-Hall distribution with parameter 5, that is scaled and translated to the range [5,10]. This distribution can be approximated well by a Beta distribution (with both shape parameters set to 8), or a Normal distribution (with mean 7.5 and variance 5/12, see below). These three distributions are very close to each other and they are all bell-shaped.

A common mistake is to define UnloadTm using the formula 5*Uniform[1,2], thinking that it is the same as above sum. That is not so. This last formula just multiplies one sample from Uniform[1,2] by 5. Clearly, the resulting duration follows a uniform distribution from 5 to 10. This distribution is flat and not bell-shaped, and although its mean is 7.5, its variance is 25/12 (much greater than 5/12).

C. Each step's duration distribution is available and the Central Limit Theorem applies.

When the number of steps in a multi-step activity is sufficiently large, it may be possible to invoke the Central Limit Theorem and approximate the total duration of a multi-step activity by a Normally distributed variate. One well-known necessary condition for the Central Limit Theorem is that the number of random variables being summed up is large enough. It is interesting to note that having just five uniformly-distributed steps is indeed enough for the sum of their durations to follow approximately a Normal distribution.

The duration X of each step with distribution Uniform[1,2] has mean = 1.5 and variance 1/12, hence the sum, T, of 5 such IID variates, T=X1+X2+X3+X4+X5, has mean 7.5 and variance 5/12. Thus, the duration of activity UnloadTm can be specified approximately by a Result defined by the formula:


This discussion continues with Multi-Step Activities in EZStrobe - 2, Non Identical Consecutive Steps which describes how to model multi-step activities when

    • The durations of the required steps are not Independent and Identically Distributed (IID), or
    • The number of steps required is variable, or
    • The multi-step activity can be interrupted between steps.

The EZStrobe model for the above example can be obtained here. It has a seed and uses multiple random number streams so that it yields the exact same output as other models that are supposed to be identical.

Stroboscope, EZStrobe, ProbSched, Vitascope and Vitascope++ are based upon work supported by the National Science Foundation under Grants No. 9733267, No. 0113890, and No. 0732560. Any opinions, findings, and conclusions or recommendations are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.