Simulationof…
System: collections of parts organized for a purpose
Natural, Designed physical, Designed abstract (math model), Human (family)
Purpose of simulation: understanding, controlling, changing, managing systems
Advantages of simulation: time compression, cost, understanding, real system does not exist
Disadvantages: Can be expensive (relative to benefit), time, data hungry, skill, OVERCONFIDENCE
Types of simulation: Throwaway, ongoing use, regular use, generic and reusable
Simulate when you have a complex enough system that deterministic models do not model system and when cost of simulation is covered by possible outcomes.
Conceptual model: description of the simulation model that is to be developed describing the objectives, inputs, outputs, content, assumptions, simplifications (black boxes, rules, and data), data flows, business rules, interactions/interfaces with other process, resource needs and costs. The model should be:
Valid: sufficiently accurate for purpose at hand (‘good’ data is available: good:
when model is executed, it mimics real system)
Credible: clients perceive the model as one that is sufficiently accurate for purpose at hand
Utilitarian: both modeler and client believe output of model will be adequate to base decisions on
Feasible: both modeler and client believe can create computer model
Rules: routes, processing times, schedules, allocation of resources, queue order and length, check offs, quality inspection, …..
Methods of modeling variability, both quant and qual (description of process and actions of players used to develop conceptual model):
Data types/nature
Trace: (assume independence) queue time, cycle time, down time, what happened over time in the processes
Empirical distributions: distribution of trace
Statistical distributions: fit of trace data to distribution, then use rand()
For this class use a software to fit data, or visual of histogram, pp138-152 tells how to determine type of distribution using stat fit in ProModel
Problem with normal distribution (negative values), might have to use a triangular distribution.
Bootstrapping: selecting with replacement of trace data at random
Data to gather
TPT
Takt time (assuming market demand and production rate are same)
WIP cost queue and in process
Resource Utilization
Queue times
Resource cost
Arrival rate and interarrival times
(distributions)
Step cycle times and distributions (parameters)
Set up
Mean time to breakdown
Duration of breakdown
Maintenance schedule
Vacation
Sick leave
Coding, testing, DOCUMENTING (20 to 30% of time should be donated to documentation, it saves time in the long run)
Separate data from model (do not hard coed data in an equation, reference data)
Three types of documentation:
Model documentation: the model, assumptions and simplifications, input data including interpretation and sources, results format,
Project documentation: meeting minutes, project specifications (cost, time, quality, scope), verification and validation, scenarios executed and results, final report, project review
User documentation: why you did what so you can do a better job next time
Terminating: business closes or lunch hour is over
NonTerminating: capacity, just let it run (steady state) (warm up period = initial transient and provides initialization bias)
Transient output: stochastic
Determining warm up period: min is achieving steady state
Run length, 10 times warm up period
Number of replications, enough for reliable CI
We are going to use TOC and Lean principles to determine how to improve a model and hypotheses testing of mean differences to determine if we succeeded in improving the model.
Implementation: a concise exploration if this subject will be needed in your project
what barriers do you see that will impede implementation of the new system
resource (time, money, availability of skilled employees, equipment)
people issues (buy in or not)
how would you avoid those impediments
Verification: conceptual model is translated into simulation model and the simulation model works as designed
Validation: accurate model
Impossible to verify completely, do best can in time you have
Check against real world (statistics, means testing)
Check against simple model
Does it make sense????