Вы находитесь на странице: 1из 5

Journal of Manufacturing Systems 28 (2009) 41–45

Contents lists available at ScienceDirect

Journal of Manufacturing Systems


journal homepage: www.elsevier.com/locate/jmansys

Technical note

Factors affecting the implementation rates of energy and productivity


recommendations in small and medium sized companies
Farouq Alhourani a,∗ , Umesh Saxena b,1
a
Information and Technology Management Department, John Sykes College of Business, University of Tampa, 401 W. Kennedy Blvd, Box O, Tampa, FL 33606, United States
b
Department of Industrial and Manufacturing Engineering, College of Engineering and Applied Science, P.O. Box 784, University of Wisconsin-Milwaukee, Milwaukee, WI 53201,
United States

article info abstract


Article history: The Industrial Assessment Center (IAC) at the University of Wisconsin-Milwaukee has worked with
Received 25 October 2008 more than 500 manufacturing companies to identify energy and productivity saving opportunities for
Received in revised form small and medium sized companies. The most common saving recommendations are summarized. The
22 February 2009
implementation rates of the recommendations were only about 40%. The logistic regression analysis
Accepted 14 April 2009
Available online 9 May 2009
is used to investigate the factors that affect the implementation rates of the recommendations in
147 manufacturing plants from seven different types of industries. Six possible factors affecting the
implementation rates of the recommendations are investigated; payback period, sales, number of
employees, plant area, annual working hours, and the industry type. It is determined that each type of
industry deals differently with implementing the recommendations. Furthermore, the recommendation
type (energy or productivity) and payback period are the main factors that affect the implementation
rates. It was also found that the companies which run on full capacity or work more hours are less likely
to implement the recommendations.
Published by Elsevier Ltd on behalf of The Society of Manufacturing Engineers.

1. Introduction principles involved in energy management. Smaller plants are


less likely to be able to afford a full- or part-time energy
The manufacturing sector in the United States accounts for or productivity manager on site, or conduct a comprehensive
more than 42% of the total energy consumption. The small energy and productivity audit to identify and implement saving
and medium sized manufacturing plants (those with fewer opportunities.
than 500 employees) represent more than 98% of over 375,000 The US Department of Energy (DOE) has operated an Industrial
establishments in the United States [1]. It was established that Assessment Center (IAC) program since 1976 to provide informa-
there is a high potential to save energy, improve productivity, and tion to the smaller firms and give them access to the benefits of
reduce waste in the manufacturing industry sector which has not energy and productivity audits. The program comprises a network
yet been taken by firms. It is estimated that if the industry exploited of 26 university-based centers that provide energy assessments for
fully the practical conservation energy potential, in three to four small and medium sized manufacturing plants at no cost to compa-
years a savings rate equivalent to 6 million barrels of oil per day nies. Beginning in 1994, the program was expanded to offer assess-
could be achieved [2]. ments that address productivity assessments in addition to energy
Energy cost in small and medium sized manufacturing plants use.
is usually a small portion of the total production cost. Thus, A study by the DOE summarizes the IAC program and the
energy cost is receiving relatively little attention from the financial results through 1994 [4]. It reports that the recommendations
point of view [3]. The lack of initiative, information availability implemented by the participating companies have resulted in $
and expertise are some of the barriers in introducing energy 517 million in savings, at a federal program cost of $ 27 million.
management in small and medium sized manufacturing plants. Olszewski, Leach and McElhaney [5] showed that in some cases,
Another problem is the lack of knowledge about the underlying the non-energy benefits identified (e.g., reduced scrap, increased
productivity, and maintenance reduction) have been greater
than the energy savings [5]. Chejne et al. [6] found that about

Corresponding author. Tel.: +1 813 253 6221 x 3505; fax: +1 813 258 7408. 50% of all savings could be carried out without investment
E-mail addresses: falhourani@ut.edu (F. Alhourani), uksaxena@uwm.edu when the recommended actions included adequate operation of
(U. Saxena). processes and equipment, satisfactory maintenance practices, and
1 Tel.: +1 414 229 4052; fax: +1 414 229 6958. implementation of new technologies.
0278-6125/$ – see front matter. Published by Elsevier Ltd on behalf of The Society of Manufacturing Engineers.
doi:10.1016/j.jmsy.2009.04.001
42 F. Alhourani, U. Saxena / Journal of Manufacturing Systems 28 (2009) 41–45

Recommendation %
Tonn and Martin [7] conducted a follow-up survey of 42 20
companies; the survey concluded that the program positively 15
influenced the companies’ attitudes about energy efficiency and
10
productivity improvement decision making. They also noticed a
5
significant positive change in the firms’ energy efficiency decision
0
making within a relatively short period of time after receiving

ts

ity

s
le

so
er

or

er
lin
AV
gh

tiv
i
one of the three IAC benefits; a direct energy assessment, the

Bo

ov

ot

es

th
du
Li

uc
H
M

O
ec

pr

he

od
employment of a student alumnus of the IAC Program, or the use

om
tR

Sc

Pr
C
ea
of energy efficiency information from an IAC website.

H
Recommendation Type
Beyene [8] studied the energy consumption data for 300 man-
ufacturing plants from different Standard Industrial Classification Fig. 1. Percentage of each type of recommendation.
(SIC) codes in Southern California. Beyene found that unless an en-
ergy recommendation represents large saving, industry reaction to 3. Common energy and productivity recommendations
recommendations of payback in more than two years is cautiously
enthusiastic or casual. Beyene [8] also studied the relationship be- A database is developed for 147 companies for which recom-
tween the SIC and energy use profile. He found that the indus- mendation reports were submitted. The total number of recom-
try classification standards do not reflect the energy consumption mendations submitted was 1189. A summary of the most common
and offer little hint to draw pertinent conclusions about the energy submitted energy and productivity recommendations is as follows:
consumption profile for any group of manufacturing plants defined
by the SIC classification codes. This is because energy intensity is
3.1. Energy recommendations
more a function of the process, not the industry type.
The IAC at the University of Wisconsin-Milwaukee has worked
• Boilers: The boiler’s efficiency can be improved by simply
with more than 500 manufacturing companies in Wisconsin and
adjusting the boiler air–fuel ratio manually or with an
Illinois states. Data from 147 companies from 7 different SIC codes
automatic oxygen trim system.
are used to summarize the most common energy and productiv-
• Heat recovery: Significant amount of heat is lost through the
ity recommendations submitted to companies. This study also ex-
boiler stack. This heat can be recovered and used by installing
amines the factors that influence a firm’s decision to implement a
an economizer in the stack. Also, waste heat from hot flue gases
recommendation. The factors of interest are payback period, sales,
can be used to preheat combustion air.
number of employees, plant area, number of working hours per
• Motors: For motors with large horsepower and variable loads,
year, type of the recommendation, and the industry type based on
it is energy efficient to install variable frequency drives.
the SIC code. Logistic regression is used to develop a model that
Also, ordinary V-belts in the motors should be replaced by
estimates the probability of implementing a recommendation.
synchronous (energy efficient) belts.
• Compressors: As it is easier to compress cold air than hot air,
2. The assessment process using a ductwork to bring cooler air from outside to the intake
of the compressor will reduce the energy use. Furthermore, in
To qualify for an IAC assessment, a manufacturing plant must many plants, up to 15% of compressed air is wasted due to
meet the following criteria: leaks in the system. Repairing these leaks also lowers the energy
consumption. Also, reduce the pressure of compressed air to the
• Within Standard Industrial Codes (SIC) 2000–3900. minimum required.
• Within 150 miles of the host campus. • Lighting: Many plants have HID lighting (Metal Halide and
• Gross annual sales below $ 100 million. High Pressure Sodium). Replacement with energy efficient (T8)
• Fewer than 500 employees at the plant site. lamps can save energy. Also, office lighting can be changed to T5
• Annual energy bills more than $ 100,000 and less than $ 2 with electronic ballasts. Furthermore, occupancy sensors can be
million. installed in the low traffic areas.
• No professional in-house staff to perform the assessment. • Heat, Air Ventilation and Cooling (HAVC): In many plants, doors
are not well protected. Installation of an air curtain (or vinyl
Each assessment team conducts an extensive data gathering
strips) can reduce the heat loss through the doors during hot
preassessment and then carries out a one-day site visit to an
seasons. Also, timers or thermostats can be installed to control
industrial plant. Following the site visit, the team prepares a report
the temperature.
for the plant that includes information about several assessment
• Scheduling: Rescheduling plant operations or reducing the load
recommendations. These recommendations are a set of cost saving
can help in avoiding peak times during the day. Also, a shift can
opportunities in energy and productivity. Each recommendation
be made from daytime to nighttime operations.
included in the report contains sufficient technical and cost details
including the anticipated savings, implementation cost, and simple Figs. 1 and 2 show the percentage and the implementation rate
payback period. The average annual savings is about $ 45,000. of each type of energy recommendation respectively. The figures
After six months to one year of submission of a technical show that the compressor, lights, and HAVC are the most common
report outlining the recommendations, each company is contacted recommendations.
to find out which recommendations were implemented or will
be implemented and which recommendations are not. The 3.2. Productivity recommendations
results of this interview, along with the data on the plant and
the original recommendations, are maintained in an available • Automation: Automation can improve productivity by reducing
database. It is found that the average implementation rate of manufacturing costs through better control of production.
the recommendations is only about 40%. Thus, the goal is to Parts are loaded, fed, and unloaded on machines more
increase the implementation rates by investigating the factors that efficiently, machines are used more effectively, and production
influence it. is organized more efficiently.
F. Alhourani, U. Saxena / Journal of Manufacturing Systems 28 (2009) 41–45 43

% Implemented 50 60
40
50
30

% Implemented
20 40
10
30
0
r

ity
20

ts
ile

so
er

or

lin
AV

tiv
gh
Bo

ov

ot

es

du

uc
H
Li
M
ec

pr

he

od
10
om
tR

Sc

Pr
C
ea
H

0
Recommendation Type 26 27 30 33 34 35 36
SIC
Fig. 2. Implementation rate for each type of recommendation.
Fig. 3. Implementation rate in each industry.

Table 1
Some important characteristics of the companies studied. 100.00 SIC 26
80.00 SIC 26 SIC 33
Characteristics Mean Std. Deviation
SIC 34 SIC 26
SIC 35 SIC 36
Annual sales ($ Million) 25.7 19.9 60.00 SIC 30
No. of employees 179.3 124.6
40.00
Plant area (1000 ft2 ) 144 120.3
Annual working hours 5512 1649.7 20.00
Energy use (MMBTU) 50,945 93,738
Energy saving (MMBTU) 2610.1 3107 0.00

ry

ity
ts
ile

or

or

in
AV
Productivity saving ($) 43,639 96,998

ve

tiv
gh
Bo

l
ot

du
o

es

uc
H
Li
M
ec

he
pr

od
tR

Sc
om

Pr
ea

C
• Work cell and assembly line balancing: Many manufacturing

H
plants suffer a great deal from the problem of bottlenecks
Fig. 4. Highest implementation rates for each recommendation type.
in work cells and assembly lines. This imbalance can create
huge idle times that result in high production cost and longer 50000
throughput times. Thus, all people, cells, and departments need 45000
to be balanced with each other. This can be done by taking some 40000
work away from a busy station and giving it to someone who 35000
does not have enough work or by redesigning the bottleneck
Savings ($)

30000
station to make it more efficient. 25000
• Redesign workplace: Reorganizing the workplace and the 20000
development of proper tools and fixtures can make the station 15000
more efficient and productive and may reduce the number of 10000
operators required to do the job. 5000
• Trash: Installation of a compactor for trash can reduce waste 0
disposal costs. 0 10000 20000 30000 40000 50000 60000
• Evaporators: Installation of an evaporator that will evaporate Implementation Cost ($)
water from non-hazardous liquid waste will reduce the volume
Fig. 5. Total potential annual saving vs. total implementation cost.
and cost of disposal of the coolant used.
• Coolant: Coolant life is increased significantly by filtration.
Portable coolant filtering systems will reduce the cost of implementation rates for HAVC recommendations are the highest
disposal of the coolant used. in SIC 33 industries.
It is found that 64% of the implemented recommendations
have a payback period of 1 year or less. It is also noted that,
4. Data analysis
in general, those recommendations with high implementation
costs also have higher potential savings, but at an ever-declining
A database for 147 companies for which reports were submitted
rate, a manifestation of the law of diminishing marginal returns.
is developed. This includes 21 companies from 7 different SIC
However, as Fig. 5 shows, there are many exceptions.
codes: 26 (paper and allied products), 27 (printing, publishing, and
Let Y be a dichotomous random variable for which the value
allied industries), 30 (rubber and miscellaneous plastic products),
1 represents the implementation of a recommendation and 0 rep-
33 (primary metal industries), 34 (fabricated metal products), 35
resents no implementation. The probability that a recommenda-
(industrial and commercial machinery and computer equipment),
tion in this population is implemented can be estimated by the
and 36 (electronic and other electrical equipment and components
sample proportion (p̂). Thus, the probability for a recommendation
except computers). A complete implementation report is available
can be estimated by:
for each company. Few main characteristics of these companies are
shown in Table 1. 481
The total number of the recommendations (events) submitted p̂ = = 0.4045. (1)
1189
for the 147 companies is 1189 recommendations of which 481
were implemented. Fig. 3 shows the implementation rate in Overall, 40.45% of these recommendations were implemented.
each type of industry. The figure shows that SIC 35 and SIC 26
industries have the highest implementation rates. Fig. 4 shows the 5. Probability model development
highest implementation rate of each type of recommendation in
the industries studied. The implementation rate for compressors The logistic regression statistical technique is used to accurately
and heat recovery recommendations are the highest in SIC 26 determine the significant factors that affect the implementation
industries (about 85% and 70% respectively). Furthermore, the rates of the recommendations. The determined significant factors
44 F. Alhourani, U. Saxena / Journal of Manufacturing Systems 28 (2009) 41–45

with a certain level of significance are used to develop a A statistical significance level of 0.10 for p-values of potential
probability model to estimate the probability of implementing a factors is used for entry or retention in the equation. The stepwise
recommendation. procedure starts with one variable and then adds in the next
most significant variable. After each step, the remaining potential
5.1. Logistic regression models variables are tested for their ability to explain the remaining model
error. The variable that explains most of the remaining model error
Logistic regression is a statistical technique that uses one or is the next to be added. Using the stepwise regression technique
more independent explanatory variables (factors) to predict the helps one to avoid any multicollinearity problem in the regression
probability of a categorical response. Logistic regression is well model between the independent variables.
suited for modeling the probability or likelihood of the occurrence The best logistic regression equation determined from the data
or non-occurrence of some event [9]. The response or dependent in this study that could be used for estimating the probability of a
variable in logistic regression is the logarithm of the odds ratio recommendation to be implemented is given in Box I where:
p/(1 − p), where p is the probability of a data value being in one
p is the probability of implementing a recommendation,
of the categories [10]. The logit transformation converts a variable
exp is approximately 2.7182,
constrained between zero and one into a continuous variable that
X1 Dummy variable to represent SIC code 26 industry,
is linear with respect to the vector of the explanatory variable, X :
X2 Dummy variable to represent SIC code 27 industry,
Y = Log[p/(1 − p)] = B0 + B1 X1 + · · · Bk Xk (2) X3 Dummy variable to represent SIC code 35 industry,
X4 Dummy variable to represent SIC code 36 industry,
where X5 Recommendation type (1: if energy recommendation, 2:
Y is the response variable; [p/(1 − p)] is the odds ratio; if productivity recommendation),
B0 is the intercept; X6 Recommendation’s payback period, and
B1 is the slope coefficient for the first explanatory variable X1 ; and X7 Number of the working hours in thousands of hours per
Bk is the slope coefficient for the kth explanatory variable Xk . year.

Slope coefficients for logistic regression models are fit to the The p-values for these factors are: X1 (0.084), X2 (0.089),
categorical response variable by using the method of maximum X3 (0.066), X4 (0.039), X5 (0.00), X6 (0.001), and X7 (0.000).
likelihood, which optimizes the probability that the observed The above results show that the payback period, the recom-
data will be estimated from the set of slope coefficients [11]. mendation type, and the number of working hours per year are
To calculate the predicted value of the response variable (a the most important factors that affect the implementation of the
probability), the logistic transformation (the inverse of the logit recommendations. A one year increase in the payback period low-
transformation) is used. The resulting equation has the form: ers the log odds of implementation by 0.174. This indicates that
the companies mainly invest in the recommendations that have a
(exp(B0 + B1 X1 + ..Bk Xk )) lower payback period. Also, if the recommendation is an energy
p= . (3)
(1 + (exp(B0 + B1 X1 + · · · Bk Xk ))) recommendation, it increases the log odds of implementation by
0.611, which indicates that the companies are interested in im-
The expression on the right, called a logistic function, cannot
plementing energy savings recommendations more than the pro-
yield a value that is either negative or greater than 1 [12]. Slope
ductivity recommendations. Furthermore, each 1000 h increase in
coefficients for logistic regression models are fit to the categorical
the annual working hours in the company lowers the log odds of
response variable by using the method of maximum likelihood,
implementation by 0.137. This is mainly because companies that
which optimizes the probability that the observed data will be
work more hours are busier in their production and have lower pri-
estimated from the set of slope coefficients [11]. The implied
ority to implement recommendations. The differences in log odds
calculations have all been computerized and many statistical
for the variables – number of employees and amount of sales – are
packages can generate multiple logistic analyses.
not likely to differ from zero in the population.
The response or logistic regression is used to examine the effect
The analysis also shows how each type of industry deals
of variables on implementation rates of the recommendations as
with saving opportunities. The SIC 34 industry has no dummy
the dependent variables 0 or 1 (0 means unimplemented and 1 variable and is represented by giving all the other dummy variables
means implemented). that represent the other industry types ‘‘0’’ values. The dummy
variables of the SIC codes 30 and 33 were not significant and did
5.2. Developed logistic regression model not enter the model. This means that these types of industries
are also similar to the SIC 34 industry; while in the SIC 26
All potential factors: payback period, sales, number of employ- industry the log odds of implementation increases by 0.324.
ees, number of working hours per year, and the recommendation This means that SIC 26 industries are more likely to imple-
type (energy or productivity) are evaluated by using the logistic re- ment the recommendations than SIC 34 industries. The SIC 27
gression stepwise selection procedure to help determine the best and 35 industries are also more likely to implement the recomm-
possible logistic regression equation. In addition to these factors, endation than SIC 34; while the SIC 36 industries are less likely
the type of industry based on the SIC codes 26, 27, 30, 33, 34, 35, to implement the recommendations than SIC 34 industries as
and 36 is also evaluated. Six dummy variables are used to repre- the log odds of implementation decreases by 0.395. Therefore,
sent the seven types of industries used in the analysis. Each type it can be concluded that the SIC 35 industry (industrial and
of industry is represented by one dummy variable with a value of commercial machinery and computer equipment) has the highest
‘‘1’’ while the other dummy variables have a value of ‘‘0’’. The SIC probability of implementation and the SIC 36 (electronic and other
34 industry has no dummy variable and is represented by giving a electrical equipment and components except computers) has the
value of ‘‘0’’ to all the dummy variables. Furthermore, an additional least probability of implementation.
dummy variable is used to represent the type of the recommenda- Goodness-of-fit of the potential logistic regression is evaluated
tion. This dummy variable has a value of ‘‘1’’ when the recommen- with the Hosmer and Lemeshow [9] goodness-of-fit test, which
dation is an energy recommendation and a value of ‘‘0’’ when the compares the observed to the predicted implementation values.
recommendation is a productivity recommendation. The classification table in the SPSS software output in Table 2 helps
F. Alhourani, U. Saxena / Journal of Manufacturing Systems 28 (2009) 41–45 45

exp(−0.004 + 0.324X1 + 0.321X2 + 0.346X3 − 0.395X4 + 0.611X5 − 0.174X6 − 0.137X7 )


p=
1 + exp(−0.004 + 0.324X1 + 0.321X2 + 0.346X3 − 0.395X4 + 0.611X5 − 0.174X6 − 0.137X7 )

Box I.

Table 2 Further research on other potential factors that affect the imple-
SPSS classification table. mentation rates of the recommendations should be investigated.
Observed Predicted A survey of decision makers to shed some light on the manner
IMP Correct prediction percentage in which they treat saving recommendations may lead to higher
.00 1.00 implementation rates. Another factor of interest may be the lack
Step IMP .00 610 98 86.2 of understanding of the saving potential and the technical imple-
0 mentation of the recommendations by the decision makers in the
1.00 275 206 42.8 company. A second visit to the company for submitting a detailed
Overall 68.6
report with a presentation and discussion with the decision mak-
percentage
ers in the company about the proposed recommendations may also
increase the implementation rates.
in assessing the performance of the model by cross tabulating
the observed response categories with the predicted response References
categories. For each case, the predicted response is the category
treated as 1, if that category’s predicted probability is greater [1] Woodruff MG, Muller MR, Roop JM, Jones TW, Seely HE, Dowd J. Analysis of
than the specified cutoff of 50%. Cells on the diagonal are correct energy efficiency investment decisions by small and medium manufacturers.
predictions. Cells off the diagonal are incorrect predictions. The Technical report No. 4. Washington (DC): US Department of Energy;
overall correct prediction percentage for this model is 68.6%. 1996.
[2] Bevington EM. Payoffs from corporate energy conservation. Harvard Business
Review 1981;59:42–6.
6. Conclusions and recommendations for future research [3] Kannan R, Boie W. Energy management practices in SME-case study of a bakery
in Germany. Energy Conversion and Management 2003;44:945–59.
[4] Woodruff MG, Jones TW, Dowd J, Roop JM, Muller MR. Evidence from the
This paper summarizes the most common saving opportunities
industrial assessment program on energy investment decisions by small
submitted to 147 small and medium sized manufacturing plants and medium-sized manufacturers. In: Thirty second intersociety energy
from different industries. The average implementation rate for conversion engineering conference. 1997, pp. 2138–42.
147 companies for which reports were submitted to in the last [5] Olszewski M, Leach R, McElhaney K. DOE/OIT plant wide energy assessment
four years is about 40%. The potential factors investigated using experience summary. Energy Engineering 2002;99:56–71.
logistic regression are payback period, sales, number of employees, [6] Chejne F, Rojas JC, Botero S, Hill AFJ, Figueoa EA, Perez JC, Cardona N,
Ramos EM. Rational use of energy in the industrial sector served by the
number of working hours per year, and the recommendation medallion public utility, Colombia. Energy Engineering 2000;97:39–48.
type which is either an energy or productivity recommendation. [7] Tonn B, Martin M. Industrial energy efficiency decision making. Energy Policy
In addition to these factors, the type of industry based on the 2000;28:831–43.
SIC codes 26, 27, 30, 33, 34, 35, and 36 is also evaluated. The [8] Beyene A. Energy efficiency and industry classification. Energy Engineering
results show that each type of industry deals differently with 2005;102:59–80.
[9] Hosmer DW, Lemeshow S. Applied logistic regression. New York: John Wiley;
the recommendations. It is also found that the probability of
2000.
implementing a suggested recommendation to a company is [10] Helsel DR, Hirsch RM. Statistical methods in water resources. Amsterdam,
mainly dependent on the recommendation’s payback period as Netherlands: Elsevier; 1992.
it is found that 64% of the implemented recommendations have [11] Battaglin WA, Ulery RL, Winterstein T, Welborn T. Estimating the susceptibility
a payback period of 1 year or less. The companies are also of surface water in Texas to nonpoint-source contamination by use of logistic
regression modeling. Technical report, US Geological Survey Water-Resources
generally interested in implementing energy recommendations.
Investigations Technical Report No. 03-4205. Washington (DC): Department
Furthermore, if companies operate on their full capacity or work of Interior; 2003.
more hours, they do not generally give priority for implementing [12] M, Pagano K, Gauvreau. Principles of Biostatistics, Duxbury, Pacific Grove,
the recommendations. 2000.

Вам также может понравиться