grey box modeling of a low pressure electric boiler for domestic hot water s...
Ian_McLeod_Masters_Portfolio_Rough
1. IAN MCLEOD
PORTFOLIO
MASTER OF SCIENCE - MECHANICAL ENGINEERING
School of Engineering for Matter, Transport, and Energy
May 2016
2. EXECUTIVE SUMMARY
A requirement for obtaining a masters of science in mechanical engineering at Arizona State
University is a coursework portfolio. The following portfolio exemplifies what I consider to be
the two most outstanding projects performed in my graduate studies. The two papers included in
this portfolio are:
(1) the report submitted to Dr. Patrick Phelan for the project entitled “Calculating Loss
Coefficient of a Two-Cover Flat Plate Solar Collector” in the course of Solar Thermal
Engineering (MAE 585), and
(2) the report submitted to Dr. Ronald Calhoun for the project entitled “Statistical Methods of
Minimizing Risk for Small-Scale Wind Energy Investments” in the course of Wind Energy
(MAE 598).
The first paper included in this report, “Calculating Loss Coefficient of a Two-Cover Flat
Plate Solar Collector”, is notably multi-faceted. It involved an experimental set-up requiring
precise material manufacturing and assembly, presentation of complex theoretical equations in a
digestible format, and multiple packets of software coding in order to collect and analyze data.
Since the project drew upon so many skills learned throughout the mechanical engineering
curriculum, it effectively demonstrates the engineering acumen I have developed as a student.
Myself and two other graduate students performed this project as a group. Each of us worked
together during all stages of the project development. This ensured that each of us had the
opportunity to practice the variety of skills aforementioned, in addition to ensuring equal work
distribution. To summarize the project, a flat, two-cover solar panel was built that enabled
temperature measurements over time at varying key locations on the solar panel. My contribution
to the project can be explained in three stages: construction, modeling, and computation.
Constructing the solar panel necessitated use of several tools in the ASU machine shop. I
operated the drilling machine with a hollow cylindrical attachment to drill large holes, and a
circular saw to cut 2x4’s for structural support. I was also put in charge of coming up with a
blueprint in SolidWorks since I consider computer-aided design to be my specialty. This
blueprint can be viewed in figure 4 of the report. Lastly, I helped dremel cutouts in the PVC pipe
to allow lateral water flow between glass plates of the solar panel. The application of generous
quantities of sealant concluded the construction phase as our panel was securely watertight.
Validation of our group’s method was dependent upon the data we collected being in
agreement with theory. This goal required development of a list of equations that would enable
energy loss to be tabulated from temperature readings. I had to conduct a fair amount of research
in order to help flesh out the right equations for use with our model. The equations of thermal
resistance for use in the thermal network diagram shown in figure 2 were found in our course
book. I took these equations and applied the necessary simplifications to allow numerical
estimation based on the data we were going to collect. My colleague’s researched the equations
for convective heat transfer and loss coefficient dependency on both thermal resistances and heat
transfer coefficients. I also researched the available solar irradiance at the exact time, day and
latitude that the experiment was conducted. Together we built a derivation that led to an
estimation of the overall loss coefficient of the entire collector, meaning the total energy per unit
area lost by the collector for the time duration of the experiment.
Finally, thermocouple temperature data had to be digested in Matlab before being applied in
our model. I contributed to this process by developing a LabView code that recorded
3. temperatures from thermocouples placed on the solar collector and output them in a text file. I
then co-authored Matlab code that ultimately graphed overall loss coefficient based off of the
thermocouple temperature readings. Our experimental results were compared to theoretical
results published by Sekhar et al. as discovered by my teammate. Our numerical model proved to
be in agreement with their more lengthy experiment.
The second report included in this portfolio was what I consider to be my most creative
project out of my graduate courses. Professor Calhoun allowed full student autonomy in the
scope of the project so long as it was relative to wind energy in some way. I am especially
inspired by renewable energy technologies as the future power sources for the world. I found the
project to be both fascinating and practical as a means for any individual to calculate the amount
of energy able to be produced through wind power in a given region without the use of expensive
equipment. Additionally, this project was special because it employed statistical analyses in the
core of the project, which is a subject that is not heavily stressed in the mechanical engineering
curriculum. In this way, I consider this project to be one in which I “stepped outside the box” in
terms of creativity and development, and the results were encouraging.
The project was entitled “Statistical Methods of Minimizing Risk for Small-Scale Wind
Energy Investments”, and was also a group project. However, I only worked with one other
person, and I ended up producing the vast majority of the project alone. I solely authored the
Matlab code that produced the discriminant analyses and respective quadratic classifiers, the t-
test for comparing summer and winter data, and error estimation. The procedure for analyzing all
data was also my construction. My partner was tasked with helping research wind energy
empirical data in Cold Bay, Alaska, helping write up the report, and also conducting the multiple
linear regression involved that happened to fail based upon the variables that he used.
Nevertheless, the results of the project exemplified how simple statistical measures could reveal
a great deal regarding wind energy generation for the small-time financier. I may even recreate
this project in the near future if I am able to settle down in an area where wind energy could be
used for home power generation, and thus this report has become more than just an assignment
in my eyes.
On the next page starts the report “Calculating Loss Coefficient of a Two-Cover Flat Plate
Solar Collector” as given to Professor Phelan in the fall semester of 2015. Following that,
“Statistical Methods of Minimizing Risk for Small-Scale Wind Energy Investments” is included,
as given to Professor Calhoun in fall of 2014.
11. Statistical Methods of Minimizing Risk for
Small- Scale Wind Energy Investments
By Erik Misiak and Ian McLeod
Professor Calhoun
Dec. 2nd, 2014
Executive Summary
The purpose of this project is to explore ways in which the typical individual
could sample local data in order to better understand the available wind power in
their region. The project is focused on using past data to extrapolate on potential
returns in energy and money. This is intended as an alternative to expensive
meteorological software. The project purpose is accomplished using statistical
analyses in Matlab, such as discriminant analyses, multiple linear regressions, t-
tests for comparison, and other methods of estimating probabilities. An estimation
of potential power generation over a six-month period in Cold Bay, Alaska, is
achieved. An explanation of the business impact of operating small-scale, residential
turbines is also provided. It is recommended that the reader employ the methods in
this paper toward implementing their own turbine, and thus contributing to the
global shift toward renewable energy.
12. McLeod and Misiak 2
Introduction
A. Problem Statement
In the modern energy market, alternative energies such as solar and wind
power are becoming more and more competitive against fossil fuels. However,
reusable energy is not currently as reliable as fossil fuels for human needs. Of the
many issues brought about, it can be difficult to select a site to farm wind because of
inaccuracies in wind power estimation techniques. Advances in technology such as
lidar equipment and meteorological software have significantly reduced the risk
involved in commercial wind farm investments. However, these technologies are
still so rare and expensive that the average citizen cannot afford them. This limits
wind farming potential. The purpose of this project is to demonstrate how the
average individual might take published data and employ statistical methods
towards understanding the nature of wind speeds in their area. A theoretical utility
for this knowledge could be setting up a personal wind farm on a residential
property, or perhaps even designing a small-scale wind farm. The ability to use
statistical estimation to predict areas that can produce the most energy is a vital
resource for implementing personal wind turbines. Wind power is reliant on
sustained wind speeds.
The business aspects pushing alternative energies can also be complex and
multidimensional. State and federal agendas often conflict, and it appears that many
people are uneducated about the energy resources they have available. It is the
objective of this project to present the reader with simple energy policy knowledge
in addition to statistical techniques regarding wind energy. Understanding these
principles can empower the average citizen to defy the unpredictability in weather
and feel comfortable operating their own turbine.
B. Background Information
As global dependency on coal continues to threaten the environment,
alternative energy is on the rise. Many states have instituted a percentage quota of
renewable energy to reach within the next few decades. Wind energy has become a
leader in the pursuit of clean energy sources. New ideas are being explored every
day in terms of capturing wind power, however a significant portion of the business
aspect of energy relies on prediction. As any meteorologist can attest, it is very
difficult to predict wind speeds since they are a byproduct of global changes in
weather and chaotic variables. However, statistics can be employed in order to
determine locations where there is a high probability of sustained wind speeds. To
demonstrate this process, the selected location for this report was Cold Bay, Alaska
since it is a relatively unexplored area in terms of wind energy. Cold Bay rigorously
documents statistics in relation to wind energy in an online database as well. While
sustained wind speeds may be lower in Alaska than in the contiguous United States,
it is interesting to examine the possibility of lucrative power generation from a
conditional probability standpoint. Such results can be a testament to the true
versatility of wind energy.
13. McLeod and Misiak 3
C. Design Considerations
It is important to take caution when exploring statistics since false
interpretations can proliferate. As the law of large numbers states, the larger the
sample size the closer the statistics of the observations approach population
parameters. In this project, yearly data consisting of hundreds of data points was
examined. The data was taken from Alaska Energy Authority, which has an extensive
inventory of statistics. As with any statistical method, error must be accounted for ,
and the student’s took caution not to extrapolate results beyond reasonable
applications.
Procedure and Results
The project organization later expands in complexity, but first relevant
observations were noted from simple scatter plots. The following plot exemplifies
the average frequency of wind versus direction for an entire year in Cold Bay.
Higher frequencies are observed around the 150° and 300° directions.
The same quantity of data points were used to plot average wind speeds
versus direction for the same year in Cold Bay. One can observe a similar spike in
wind speed at the 150° mark as in the previous frequency plot. High frequency wind
with high velocity is good for power generation, so it is important to identify which
direction these conditions occur most often.
14. McLeod and Misiak 4
Discriminant analyses were used for this purpose. Discriminant analyses can
be used to help identify associations between pairs of variables. In order to prepare
for a discriminant analysis, variables must be selected that are somehow related. In
this case, average wind speed per direction was plotted against frequency per
direction. Both variables are related in that they were recorded by direction. This
yielded a plot of average wind speeds per frequency.
The purpose of this set up was to identify directions in Cold Bay where high
frequency of wind and high wind speeds occur simultaneously. Naturally, this
15. McLeod and Misiak 5
combination is ideal for wind energy production. Normally, sustained wind speeds
of 12 mph are typical constraints for wind farm investment. However, as observed
in the previous plot, most wind speeds average below 9 mph in Cold Bay. Thus, it
may not be economical to operate a wind turbine throughout the entire year. The
goal of subsequent discriminant analyses is to identify directions and times of year
where the probability of sustained wind speed occurring above 9 mph is high. This
is similar to a conditional probability analysis. In other words, it is desirable to
identify directions where the probability of wind speed occurring above 9 mph
exists given also that a high frequency of wind exists in that direction. Quadratic
discriminant classifiers were employed towards toward this end as well.
The following plot groups the data by four quadrants of direction. It is clear
that the highest frequencies with simultaneous high wind speeds occur most often
in the second quadrant. This implicates that an individual would have the best odds
of generating quality power in Cold Bay by facing their turbine between 90° and
180°. The plot also shows that the lowest frequencies in combination with low wind
speeds occur in the third quadrant, between 180° and 270°. The quadratic
classifiers display decision boundaries. These decision boundaries account for the
flexibility in approximation, and identify regions of the plot that belong to the same
category. An individual could use these boundaries as probabilistic guidelines for
sustained wind speed and frequency. There is confidence that the classifiers are
reasonably accurate because the resubstitution losses of the classifiers in both plots
were as low as the cross-validated losses.
16. McLeod and Misiak 6
qerror =0.4468
cverror = 0.4514
The following plot displays the same quadratic discriminant analysis,
however the data is grouped by season instead of direction. It is evident that the
boundaries in this analysis are much less clear-cut as with the directional grouping.
However, the trend generally demonstrates that higher wind speeds and
frequencies are observed in the fall and winter seasons rather than the spring and
summer seasons.
qerror = 0.6319
cverror =0.6435
Combining the results of the two discriminant analyses concludes that
operating a turbine between the months of October and March facing in a direction
between 90° and 180° yields the highest probability of generating the most power.
In the interest of saving money and effort, a small-scale turbine operator in Cold Bay
could activate their turbine for these few months of the year and not have to worry
about creating a responsive system that constantly adjusts the turbine into the face
of the wind. For the spring and summer seasons, the wind energy density may not
be high enough and the operator may desire to switch to solar energy due to the
extended daylight that occurs in areas close to the north pole.
In order to decisively conclude that wind speed averages in the summer of
Cold Bay are significantly different than those in winter, a two-sample t-test can be
implemented. A two-sample t-test involves two hypotheses: a null hypothesis that
the sample means are not significantly different, and an alternative hypothesis that
17. McLeod and Misiak 7
the sample means are significantly different. The t-test uses a 95% confidence
interval that false acceptance or rejection of the null hypothesis does not occur.
Using the matlab function “ttest2”, comparing wind speed averages in the winter
versus the summer validated the null hypothesis. This means that the average wind
speeds in the winter were significantly different than those in the summer. A plot of
the day of the season versus average wind speed was created to compare the two
data sets to better visualize their differences.
A final statistical tool that can be used by the average individual is multiple
linear regression. Multiple linear regression is a tool that allows multiple
independent predictor variables to be tested for relationship to a categorical
dependent variable. In this project, the variables of time of sunrise (IV), time of
sunset (IV), maximum temperature (IV), and minimum temperature (IV) from
January, February, and March of 2006 were tested in order to determine if any of
these individual variables could be used to predict wind speed (DV) in Cold Bay. The
results are as follows:
Show results of b-matrix and interpret meaning.
b=
0.4024
0.0109
-0.0094
0.0010
0.0023
18. McLeod and Misiak 8
Show results of stats and interpret meaning.
R^2 =[0.0062]
Fstat =[0.1319]
Prob =[0.9703]
Variance =[1.443]
Construct a normal probability plot of the residuals and interpret this graph.
The normal plot of the residuals does not show any evidence of nonlinearity,
which concludes that the regression results are trustworthy. Each [b] coefficient
represents the magnitude of correlation between wind speed and each variable
listed above, accordingly. The first [b] coefficient is simply the intercept of the
equation, while the remaining four [b] coefficients represent the magnitude of the
relationship between each variable and wind speed. It is evident that each
independent variable had almost zero correlation with wind speed, as each [b]
coefficient was especially small. The stats results also indicated a low value of
coefficient of determination (𝑅2
), meaning that the data was not fitted very
effectively. The highest correlation involved minimum daily air temperature.
Maximum daily air temperature was negatively correlated to wind speed, meaning
that lower maximum temperature correlated to higher wind speed in Cold Bay. The
time of sunset was about twice as much of a predictor as the time of sunrise. Using
the minimum daily air temperature and time of sunset, a surface fit plot was created
to demonstrate their correlation with wind speed.
19. McLeod and Misiak 9
Although the results had little correlation, an individual could select different
independent variables in order to discover variables that actually predict wind
speed. Understanding how such variables correlate to wind speed would allow the
individual to operate their turbine with greater confidence. Due to time limitations,
the students were not able to identify such variables, however the method is still
useful to learn for an individual interested in estimating wind speeds.
Returning to the discriminant analyses, it is possible to roughly estimate the
amount of power an individual turbine could generate while operating throughout
fall and winter and facing only in directions between 90° and 180° in Cold Bay,
Alaska. A Tycon Power Systems turbine was selected for this analysis since it is
affordable and designed for residential use. The rotor swept area of this turbine is
0.665𝑚3
. The following plot displays the wind speeds during Fall and Winter
between 90° and 180°.
20. McLeod and Misiak 10
A power curve was generated for this data according to the following
formula:
𝑃𝑜𝑤𝑒𝑟 =
1
2
∗ 𝜌 ∗ 𝑉3
∗ 𝐴
The startup wind speed for a Tycon Power Systems turbine is 2.1 m/s, so all
values are included.
Speed (m/s) Approx
Frequency
5.5 10
6 15
6.5 14
7 29
7.5 28
8 42
8.5 17
9 18
9.5 14
10 11
Total 198
21. McLeod and Misiak 11
𝑃𝑜𝑤𝑒𝑟 𝑃𝑟𝑜𝑑𝑢𝑐𝑒𝑑 = Σ𝑃𝐷𝐹𝑤𝑖𝑛𝑑𝑠𝑝𝑒𝑒𝑑 ∗ 𝑐𝑜𝑟𝑟𝑒𝑠𝑝𝑜𝑛𝑑𝑖𝑛𝑔 𝑝𝑜𝑤𝑒𝑟
𝑃𝑜𝑤𝑒𝑟 = (
10
198
∗ 70.55) + (
15
198
∗ 91.6) + (
14
198
∗ 116.5) + (
29
198
∗ 145.5)
+ (
28
198
∗ 178.9) + (
42
198
∗ 225.4) + (
17
198
∗ 260.4) + (
18
198
∗ 309.1)
+ (
14
198
∗ 363.6) + (
11
198
∗ 436.9)
𝑃𝑜𝑤𝑒𝑟 = 213.6 𝑊𝑎𝑡𝑡𝑠
𝑃𝑜𝑤𝑒𝑟 𝑃𝑟𝑜𝑑𝑢𝑐𝑒𝑑 𝑂𝑣𝑒𝑟 𝐹𝑎𝑙𝑙 & 𝑊𝑖𝑛𝑡𝑒𝑟 = 213.6 𝑊𝑎𝑡𝑡𝑠 ∗ 4335 𝐻𝑜𝑢𝑟𝑠
𝑃𝑜𝑤𝑒𝑟 𝑃𝑟𝑜𝑑𝑢𝑐𝑒𝑑 𝑂𝑣𝑒𝑟 𝐹𝑎𝑙𝑙 & 𝑊𝑖𝑛𝑡𝑒𝑟 = 0.926 𝑀𝑊 ∗ ℎ𝑟𝑠
Conclusion and Recommendations
The major take-away from this project is that wind energy has great
potential to spread into residential environments. The statistical methods involved
in this project demonstrate how an individual could optimize a personal wind
turbine. Provided that past wind data is available, employing discriminant analyses
can help the user to identify the best times of year and directions to operate their
turbine. Multiple linear regressions can be employed to identify the magnitude and
nature of correlations between independent variables and wind speed. The reader
should take time to identify independent variables that are better correlated than
those presented in this project if they intend to minimize risk with this method. T-
tests and data plotting further helps to verify significant differences in wind speeds.
This improves the researcher’s confidence in their results. It is recommended that
the reader reproduce the procedure of this project using data from their hometown
in order to calculate how much energy and money they can generate by purchasing
a personal turbine such as those offered by Tycon Power Systems. Further
recommendations are to stay up to date with federal and local government
incentives and tax breaks, as these are sure to shift each year.
References
Cold Bay Longterm Data. (2007, August 3). Retrieved December 2, 2014, from
http://www.akenergyauthority.org/PDF files/Wind Resource
Assessment/Cold-Bay_longterm-data.txt
Conditional Probability. (2012, July 31). Retrieved December 2, 2014, from
http://www.epa.gov/caddis/da_exploratory_4.html
Decision Boundaries. (2001, April 7). Retrieved December 2, 2014, from
http://www.cs.princeton.edu/courses/archive/fall08/cos436/Duda/PR_sim
p/bndrys.htm
22. McLeod and Misiak 12
Discriminant Analysis. (n.d.). Retrieved December 2, 2014, from
http://www.mathworks.com/help/stats/discriminant-
analysis.html?refresh=true#brah8i2
Horizontal Wind Turbine. (2010, December 14). Retrieved December 2, 2014, from
http://www.flyteccomputers.com/ext/Tycon Power/TPW-200-12.pdf
Joint, Marginal & Conditional Frequencies: Definitions, Differences & Examples.
(n.d.). Retrieved December 2, 2014, from http://education-
portal.com/academy/lesson/joint-marginal-conditional-frequencies-
definitions-differences-examples.html#transcript
Multiple Regression. (2009, September 19). Retrieved December 2, 2014, from
http://peoplelearn.homestead.com/MULTIVARIATE/Module11MultipRegres
s1.html
Wind Power: A Clean and Renewable Energy. (2014, June 17). Retrieved December
2, 2014, from http://www.renewableenergyst.org/wind.htm