QDOBA Burrito Line Simulation Input Modeling Report
Xxxx xxxx
Xxxx xxxx
Xxxx xxxx
Xxxx xxxx
Fall ####
IEE 475 Pavlic
Introduction to System
Qdoba Mexican grill located in the Memorial Union at ASU’s Tempe campus is known
for having one of the longest wait times of any restaurants on campus. Starting as early as 10 in
the morning the 3 switch back queue can be filled with as many as 30 people waiting to receive
burritos, tacos, or burrito bowls. Qdoba believes that as ASU Tempe continues to expand this
queue times will only increase and as they do the number of customers that will begin to
choose other venues will steadily increase. In order to prevent from this increasingly large
opportunity cost Qdoba has requested suggestions on different ways to improve the burrito
line efficiency in order to match this increasing demand.
This report will create a proposal for input models to be used in a simulation. These
input models will be constructed using input analysis of data that has been collected over the
previous week. These input models will allow our team to test multiple factors and alternative
methods that could be implemented within the line to improve process lead time.
Input Modeling Study
Pay Pay Burrito Prep Or
der
30
Person
Inputs:
There are four inputs that we analyzed in order to model this system. The first will frequency of arrivals
per 5 minutes. This data can be used to extrapolate interarrival times of our system. For the actual
assembly line we will be analyzing the time that each of the three process steps taken: Ordering time,
Burrito build time, and checkout time.
Decision Variables:
The objective of the study will be to improve the process lead time of purchasing a burrito. Below are
the decision variables that have been identified that when adjusted may affect the results:
1. Number of burrito builders working in the prep line
a. This number can be between 1-4 based on the available space in the system.
2. Number of cashier stations open
a. This can be either one or two as Qdoba is not willing to purchase additional cash
registers.
3. Number of ordering stations
a. This can be either a 1 or 2 as Qdoba has expressed they may be willing to order an
additional one ordering grill if the study proves it will improve timing
4. Number of burritos allowed in the system at one time
a. Currently there is no limit to the number of orders allowed to be waiting in the system
Qdoba’s current strategy is to have as many orders as possible being processed in the
system. However, we will experiment with limiting the number of burritos that are
allowed to be processed in the system.
5. Assembly line vs One piece flow
a. Currently the prep line is operated as an assembly line where there considerable non
value add time the team will experiment with the potential of turning the process into a
single piece flow.
State Variables
In our simulation we plan to model customers as the entity. Relevant state variables include
order status which will be measured as: In queue, ordered, BIP (burrito in production), paying, paid.
Outputs and Potential performance measures:
There are several metrics that have been identified in order to determine if a simulation will be
determined to be a plausible solution to improve process lead time. The first metric is, of course,
average process lead time per customer. The simulation will also track human metrics such as maximum
number of customers waiting as well as employee utilization. Employee utilization will be measured by
employees overall as well as by station in order to ensure process leveling. Finally the system will
measure factors like maximum number of customers in line.
Data Results
Methods of Data Collection:
Data was collected in two 2.5 hour periods on Thursday October 31st and Friday November 1st.
From 11:30 AM to 2:00 PM. To collect frequency of arrival times the number of customers that arrived
were counted at the 5 minute mark every 2 hours. The process step times for the burrito prep process
times were calculated using a running clock strategy. The times at which each orders activity changed
states were noted and then activity times were calculated using the difference between each time
period. All collected data can be found in the linked spreadsheet (link removed). The arrival rate will help
to calculate interarrival times and will be used to control the create block. The process times such as
ordering, preparing, and paying will model the process block time. We will be able to test different
parameters in the system by altering available resources of the process blocks.
Data Summaries:
Frequency of Arrivals per 5 minutes
Selected Distribution Normal Distribution
Distribution Parameters Mean: 10.1 People per 5 minutes
Standard Deviation: 2.73
Justification: For the arrival rate for 5 minutes a normal distribution was chosen. This is an acceptable
distribution for this process because counting the arrivals per 5 minutes is essentially a counting
process and normal is a good distribution to model a counting process. We find that the distribution
has an average of 10.1 people per 5 minutes. Which means the interarrival times of our system could
be modeled using an exponential distribution with a parameter value of 29.7 seconds or 1 customer
every 29.7 seconds. This distribution did not have the lowest MSE however it was very close and the
p-value of the chi square test could not disprove the data.
Ordering Times
Selected Distribution Empirical
Distribution Parameters CONT or DISC (0.000, 2.500, 0.196, 5.500, 0.304,
8.500, 0.348, 11.500, 0.543, 14.500, 0.652,7.500,
0.783, 20.500, 0.870, 23.500,0.913, 26.500,0.957,
29.500, 0.957, 31.500)
Justification: The initial distribution histogram for the ordering times was very complex and non
confirmative. Which made it very difficult to model with a distribution. When discussing the reason
for this data our group came up with a possible reason for the results. Our method for data collection
modeled ordering time as the time it takes for the persons in order to reach the prep station. If the
order was a burrito bowl this process is extremely quick, burrito requires pressing one tortilla on the
press and tacos require pressing three smaller tortillas on the press. Each of these takes a very
different amounts of time and there is a lot of stochasticity to measure. Our first idea was to use a
bimodal distribution and separate the data at the gap that is visible. However, this left us with too few
data points on each side to have strong statistical proof that these distributions we were selecting
actually matched the process. Therefore we selected an empirical model for the time being. Going
forward we would like to take more specific data and find out the percentage of customers that chose
each type of order and we can use a decision block to transfer data to pass entities to each of the
possible distributions.
Prepared times
Selected Distribution Normal Distribution
Distribution Parameters Mean: 141
Standard Deviation: 27.4
Justification: Building a burrito is at its core an additive process because you are adding ingredients to
the burrito one at a time and every ingredient takes about the same amount of time to add to the
burrito. In addition, after watching the burrito line for a number of hours we saw that most people
order extremely similar number of items. Therefore, it would make sense that the process time would
appear to be normally distributed around and average time with outliers for people who like very little
or alot on their burritos. While the mean squared error of this test was very low, we can see that the
p-value for the chi squared test was only .08 which is abnormally low however still high enough to be
unable to disprove the distribution the KS test statistic was also greater than .15 so it is also unable to
disprove the distribution.
Checkout times
Selected Distribution Weibull Distribution
Distribution Parameters Shape: 33.9
Scale: 1.9
Justification: A weibull distribution was determined to be the distribution for the checkout times since
it seems to be the most reasonable distribution for it. Also, the p-value for the Chi Square test is larger
than alpha = 0.05 which leads in failing to reject the hypothesis that is saying it is a Weibull
distribution. As a result, a weibull distribution was chosen.
Research Questions:
Below are the research questions identified during the experiment some are questions identified by the
voice of the customer and some were identified from examining the system.
1. Is it possible and by how much are we able to reduce the process lead time by altering the
available decision variables?
2. Does hiring extra employees or buying additional tortilla presses improve the overall process
lead time?
3. Is the assembly line structure the best way to run the system or would single piece flow improve
the process lead time?