Learn how to use the implemented probability distributions in INOSIM to add stochastic failures or process parameters to your model, and how to set up your custom distribution.
Probability distributions in simulation
One advantage of simulations are the always reproducible and therefore comparable results. That allows us to compare different scenarios, identify and solve bottlenecks, and last but not least, it simplifies the debugging of models. But there are situations in which this reproducibility is not desired. For example, if failure times of equipment should be investigated, certain input parameters of a model are subject to uncertainties, or, if a specific operation should only be executed in a specific fraction of times, randomness must be brought into the simulation.
Using the implemented probability distributions
With INOSIM, this is either possible by using already implemented probability distributions or a random number generator. Using the implemented distributions comes with the advantage that all generated random values are actually pseudorandom. This means they are randomly generated but at the same time reproducible as long as the seed value in the settings of the simulation or the sequence of random numbers is not changed (see Fig. 1). This again makes simulation runs comparable and simplifies the debugging.
Figure 1: Property window of the project (left) and the experiment (right). Here, the seed values and the sequence of random numbers can be changed.
There are already ten different probability distribution types available in INOSIM. They can either be used directly, for example in the property window of units for failures or maintenance (see Fig. 3) or via VBA. Via VBA, either the distribution itself can be assigned to a property (e.g., the FailureDuration property of a unit) or a sample value of the distribution to an arbitrary variable in INOSIM (see examples below).
Figure 2: Failures tab (left) and Maintenance tab (right) of the property window of a unit. Here the implemented probability distributions can be chosen to define the intervals and the duration of failures and maintenance.
An example of how an assignment via VBA could look like is shown below. Here, a specific operation should only be executed in 20 % of the times. To achieve this, an alternative branch with a link condition is used and the shown control is assigned to the link condition. In this case a uniform distribution is used.
Sub LinkCondition((cond As OrderLinkCondition) Dim distr As New Uniform Dim value As Double distr.Start = 0 'Lower bound distr.Stop = 1 'Higher bound value = distr.Sample 'Uniformly distributed value is saved in variable if value <= 0.2 Then cond.value = True 'Operation will be executed Else cond.value = False 'Operation won't be executed End If End Sub
Another example of using a distribution in your simulation is given below. A normal distribution is used to define the duration and an exponential distribution to define the interval of failures of a specific unit in the initialization phase of the simulation. Here, instead of a sample value of the chosen distribution the distribution itself is assigned to the FailureDuration or the FailureInterval property.
Private Sub Simulation_Init() Dim nor As New Normal nor.Mu = 2 * 3600 '[h] nor.Sigma = 2 * 3600 * 0.1 '[h] Set Units("Stirred-Tank Reactor").FailureDuration = nor Dim expo As New Exponential expo.Beta = 24 * 3600 '[h] Set Units("Stirred-Tank Reactor").FailureInterval = expo End Sub
Definition of custom distributions
If the given distribution types are not suitable for your problem, it is possible to create any other custom distribution types out of the given ones. Custom distributions are often used to create discretely or triangularly distributed values. A discrete distribution can easily be generated out of the uniform distribution implemented in INOSIM as shown below. The Round function returns the uniformly generated variable rounded to an integer.
Function discrete_distribution() As Double Dim distr As New Uniform distr.Start = 0 'Lower bound distr.Stop = 10 'Higher bound discrete_distribution = distr.Sample 'Uniformly distributed value is saved in variable discrete_distribution = Round(discrete_distribution) 'Uniformly distributed value is rounded to an integer End Function
The triangular distribution is often used as a simplification of the normal distribution if the sample data is not sufficient to determine a significant mean or standard deviation. It also shows the typical behavior of a higher probability of occurrence the closer the randomly generated value is to the expected value. On the other hand there is no standard deviation needed to describe the distribution (see Fig. 1). One way to create a triangular out of a uniform distribution in INOSIM is the function triangular_distribution shown below. It needs the low and high bound as well as the most probable value as inputs and returns a triangularly distributed value of the type Double.
Figure 3: Triangular distribution with the lower bound a, the higher bound b and the mode c
Function triangular_distribution(d_Min As Double,d_Mode As Double, d_Max As Double) As Double Dim dRand As Double Dim dc_a As Double Dim db_a As Double Dim dc_b As Double If d_Mode <= d_Min Or d_Max <= d_Mode Then Console.Error "Mode is not between lower and higher bound." Exit Function End If dc_a = d_Max - d_Min db_a = d_Mode - d_Min dc_b = d_Max - d_Mode Dim dre As New Uniform 'Uniformly distributed value between 0 and 1 is generated dre.Start = 0 dre.Stop = 1 dRand = dre.Sample If dRand < db_a / dc_a Then 'Uniformly distributed value is converted to triangulary distributed value triangular_distribution = d_Min + Sqr(dRand * db_a * dc_a) Else triangular_distribution = d_Max - Sqr((1 - dRand) * dc_a * dc_b) End If '[see Vose: Risk Analysis, 2nd ed., p. 128] End Function
The function can be called in your simulation as shown below. Here an uncertainty is assigned to the mass flow through a specific unit. This time the triangular distribution is used to describe the uncertainty.
Sub call_triangular_distribution Dim mass_flow As Double Dim Mean_flow As Double Dim min_flow As Double Dim max_flow As Double Mean_flow = 100 'Expected value of the triangular distribution min_flow = 80 'Lower bound max_flow = 120 'Higher bound mass_flow = triangular_distribution(min_flow,Mean_flow,max_flow) End Sub
A third example of how a custom distribution can be implemented in INOSIM is shown here with the beta distribution. The beta distribution is a family of continuous probability distributions that can take different shapes depending on the parametrization (see Fig. 2). A beta distributed value can easily be generated out of two gamma distributed values (G. S. Fishman. Monte Carlo – Concepts, Algorithms, and Applications. Springer, New York, 1. Edition, 1996). The gamma distribution is already implemented in INOSIM and therefore a beta distribution can be created with the function Beta_distribution shown below. It needs the two shape parameters α and β as inputs and can be called in the same way as the function for the triangular distribution.
Figure 4: Beta distribution with different shape parameters α and β
Function Beta_distribution(alpha As Double, beta As Double) As Double Dim X As Double Dim Y As Double Dim gam1 As New Gamma 'First gamma distributed value is generated gam1.Alpha = alpha gam1.Beta = 1 X = gam1.Sample Dim gam2 As New Gamma 'Second gamma distributed value is generated gam2.Alpha = beta gam2.Beta = 1 Y = gam2.Sample Beta_distribution = X/(X+Y) 'Gamma distributed values are converted to beta distributed value End Function