Efficiency is a measure of relative performance, but relative to what? Defining energy efficiency requires a choice of a reference point against which to compare energy use. Energy efficiency measures can be developed through a variety of means, such as engineering and theoretical estimates of performance or through observing the range of actual levels of performance. The choice of method used to define efficiency depends on the need to define a reference point for energy efficiency. One of the challenges with using energy efficiency measures based on engineering or theoretical estimates is that they may be dismissed by the industry as being economically infeasible. Consequently, these case studies have focused on developing energy efficiency measures based on actual or observed operational performance rather than theoretical estimates of potential efficiency levels. Additionally, the EPA needed to identify a method that would be perceived by users as providing economically feasible performance targets.
The reference point for economic potential (observed practice) depends, in part, on the reason for measuring efficiency as well as the available information to create a reference. Generally, the
ceteris paribus principle (“all other things being equal or held constant”) is usually desired in creating the reference point. From a practical perspective, there is a hierarchy of measures and methods by which one can “hold constant” things that influence
energy use that are not part of
energy efficiency. The first is some measure of production activity, either production of the final product or, alternatively, a ubiquitous input into the production process. This is most commonly done by computing the ratio of energy use to activity, a measure of energy intensity. Energy intensity is a common metric that controls for changes in production and is commonly confused with energy efficiency, as in the statement “
the industry or plant’
s energy efficiency has improved based on the fact that the corresponding energy intensity has declined over time.” This type of statement brings us to the second way that one may approach the ceteris paribus principle for measuring efficiency, comparing energy intensity of a particular plant, firm, or industry to itself over time. This approach is a plant
2-specific
baseline comparison or
intraplant efficiency measure. The baseline approach has the advantage of controlling for some plant-specific conditions that do not change during the comparison period.
Intensity metric selection
Murray (
1996) discusses a variety of intensity metrics for efficiency measurement, including thermodynamic approaches. Intensity ratios provide a basic metric for measuring energy efficiency and performance compared to a baseline. To measure intensity, you need a measure of energy and something for the denominator. For the numerator, these case studies use total source energy, defined as the net Btu total of the fuels (Btu) and electricity (Kwh) with electricity converted to Btu based on the level of efficiency of the US grid for delivered energy, i.e., including generation and transmission losses. A net measure is needed for when energy is transferred offsite, most commonly in the form of steam or electricity.
The choice of the denominator is a major issue for measuring intensity. Ideally, the denominator should capture some measure of production. Freeman et al. (
1997) show that industry level trends in energy intensity based on value, both total and value added, can differ dramatically from those based on physical quantities. As Freeman et al. have observed, there are many challenges with creating efficiency measures based on price indexes, cost, and other value measurements.
3
Given issues with linking energy use with price indexes, these studies have focused on using metrics based on physical quantities. For physical production to be meaningful, it needs to be at a high level of industry specificity. For example, the “Dairy” industry produces many products that cannot be aggregated, but “Fluid Milk” can. Therefore, within industries, it is necessary to differentiate between specific types for plants and manufacturing operations.
Similarly, measures of building energy use commonly use physical size (ft
2) as the main denominator for energy intensity, but for most industrial facilities, this is not appropriate.
4 While commonly used for commercial buildings where energy use is primarily tied to plug loads, lighting and heating, ventilation and air-conditioning (HVAC) systems, energy intensity based on size (sq. ft) does not correspond well with manufacturing process energy uses.
Multifactor efficiency measures
While energy intensity ratios are commonly used for intraplant level baseline comparisons in an industrial energy management setting, their value for developing interplant comparisons may be limited. For interplant comparison, there are multiple factors that must be considered. To make an interplant comparison for energy management, one wants to know which plant is performing better, so the company can learn from the top performers and focus efforts on the bottom performers. The comparison needs to be “fair”; i.e., an “apples-to-apples” comparison. To do so, we must “normalize” or “remove” the differences between the plants that influence energy use, but we do not consider being part of the “energy efficiency” of the plant. This raises the question of what should be included, or “normalized for,” in the efficiency measure and what should not. Other research has tried to address this question. May et al. (
2015) lays out a proposal to develop multiple energy key performance indicators. Tanaka (
2008) develops a framework to address measures of energy efficiency performance, ranging from simple intensity to more inclusive approaches as are proposed here.
The first basic principle proposed is to normalize for things “beyond the control of the energy manager.” There are obvious things, like weather, that meet this criterion. However, should one also consider long-lived capital investment in a particular energy using process? A business is unlikely to relocate to a favorable climate for energy reasons, but they may also be unlikely to replace a large piece of capital solely for reasons of energy efficiency. This first principle is viewed as necessary, but not sufficient condition to be part of the normalization.
Consider the example of location and the influence of weather (climate) on energy. The location of the plant is an example of the second principle, a “
market-driven business decision.” The location decision may be to best serve a local market, have access to raw materials at reasonable prices, etc. It is unlikely to consider the energy use related to the climate location, unless those cost are very large. Like the location of the plant, there are other basic categories of
market-driven business decision that should be included in the normalization if they have significant impact on energy use. These should include
5the following:
Materials selection can influence energy use most commonly in the “make vs buy decision.” If a plant manufactures an intermediate input then it is using energy that would have been used elsewhere. To compare two plants that make juice, one from concentrate and one directly from fruit, it is necessary to account for the material choice. The final product may be virtually indistinguishable, but the plant-level energy use will differ. This “make or buy” business decision to be vertically integrated up the supply chain is market driven, has impacts on energy, and should be in the normalization. This issue is a common problem of defining the “boundary” for a benchmark, i.e., what plants to include only fully integrated plants or only one with identical inputs. The approach in these case studies is to include input choice in the analysis and allow the statistical models to normalize for those decisions. There are also issues of material quality that may imply more or less energy. Some materials have lower energy intensity but may impact product quality or be in short supply. In this case, the normalization should be to the average industry practices regarding input quality. Sand or cullet in glass making is an example. In this case, the appropriate comparison for normalization is to what the industry average rate of cullet use is.
Plant size and utilization rates can have impacts on energy use. Large plants may have advantages in terms of energy use due to economies of scale. The size of the plant will also be a market-driven decision regarding investment, market size/share, etc. A “small” plant cannot become “big” to reap the energy advantage if the company does not have the market share to sell higher product volumes. It is important to conduct the energy benchmark based on size, to the extent that size is a large impact on energy use, and there are different-sized plants in the industry that reflect differing market conditions. During economic downturn, energy intensity may rise for many reasons, one of which is the inherent levels of “fixed” vs “variable” energy use. While it is important for energy management to minimize “fixed” energy, the efficiency measure needs to account for this. Not all plants will experience the same market conditions at the same time and the normalization of utilization on energy due to fixed energy use should depend on industry practice. If industry practice on fixed energy management improves over time, then the efficiency measure should reflect this and normalize “less” for utilization.
The most ubiquitous component of normalization is product mix. Even within an industry like cement, the product is not entirely homogeneous. Some products that are in demand in the market may be more energy intensive. Masonry cement is more energy intensive than ASTM 1; milk in small cardboard cartons are more intensive (per gallon) than milk in gallon plastic jugs; orange juice is more intensive than apple juice; complex cast steel components are more intensive than simple cast steel item. In some cases, there is no way to substitute one energy-intensive product for another less-intensive product. In other cases, market preferences may change toward “green” or “low carbon” products over time, but companies have limited influence over these consumer preferences but react to those market demands. Milk in smaller containers (typically cardboard) is a different product than milk in gallon jugs from a consumers’ perspective. A company will not try to convince the consumer to drink the less energy-intensive apple juice instead of orange juice, but the efficiency measure for juice manufacturing can allow an “apples-to-apples” comparison for juice plants. The efficiency measure can then be used to inform the energy efficiency of the product that consumers demand.
Some aspects of manufacturing may appear to satisfy the two principles above, particularly when one considers differences in underlying processes. For example, a small-scale, wet-process (energy intensive) cement kiln installed 30 years ago may be (1) “beyond the control of the energy manager” now and (2) “market driven,” at least at the time the decision was made. Similar examples exist for the choice automotive painting processes and paint booth design. In these cases of process choice, the 3rd, “same product,” principle needs to be applied.
Making products to serve the market is the raison de entre for manufacturing companies. If the product is, for all practical purposes, identical to another, then the plant using the more efficient process should be the reference point. Some examples may appear on the surface to satisfy this principle. Steel “mini-mill” plants are touted as being more energy efficient, using scrap and electricity to make steel. The products they produce, re-bar, etc. are not the same as sheet steel from more energy-intensive, integrated steel mills. In fact, the emergence of the mini-mill drove traditional steel out of those product lines to specialize in higher quality steel products (Boyd and Karlson
1993). The product demanded in the market and the process to make it became linked in the competitive markets; the efficiency measure should treat these two types of mills separately, not because of process, but product.
When making intraplant comparisons, it necessary to consider a variety of factors that do not neatly fit under the denominator of an energy intensity ratio. All plants may make a common product and other differences can significantly affect energy intensity. The difficulty with estimating an industry-level interplant efficiency measure is controlling for interplant difference other than production volume. While the things that differ between plants are numerous, there is a common thread across the case studies that the primary difference that have the most impact on energy fall into the following categories.
-
Product mix
-
Process input choices (i.e., “make or buy” upstream integration)
-
Size—physical or productive capacity and utilization rates
-
Climate (and other location-specific factors)
The choice of factors to include in the analysis depends upon the nature of the production process, the configuration of the industry (e.g., is upstream integration common or rare?), the availability of data to represent these factors, and the outcome of the statistical testing. In order to address these types of factors, these studies use a multivariate approach to normalization where multiple effects are simultaneously considered (Boyd and Tunnessen
2007). The next sections discussed the four basic categories of effects that are commonly considered. There is further elaboration on the way this is implemented in the section on the specific case studies.
Product mix
Not all plants produce exactly the same product. In fact, many plants produce multiple products. The diversity between plants gives rise to a mix of derived demands for specific processes and energy services. To the extent that the final product is the result of a series of energy using steps, the energy use of the plant will depend on the level and mix of products produced. Rather than specifying each process step individually, the approach used here is to identify those products that use significantly more (or less) energy and then use statistical analysis to estimate the different energy requirements of these products.
One approach to controlling for product mix is to segment the industry into cohorts based on product categories. This works best when there is no overlap between plants that produce the various basic products and there are sufficient numbers of plants to conduct the statistical comparison between those resulting groups. This means each subgroup is effectively treated as a separate industry for evaluation purposes. A good example is the glass industry where containers and flat glass are distinct industry segments.
When such natural subsectors do not exist and multiple products are produced within a plant, additional approaches are needed. The statistical approach is well suited to testing if a particular grouping of products is appropriate for benchmarking differences in energy. When industries produce a mix of products that differs across plants, then the product mix (share of activity) of distinct products is needed. This approach was first used in wet corn mills (Boyd
2008) and was later applied to other sectors.
In the absence of meaningful data on discrete product classes, an alternative is a continuous measure of product differentiation. Price is often taken as a measure of quality difference. To the extent that such quality differences arise for additional energy using processes, then value of shipments may be an appropriate proxy for product mix. Differences in value may not involve higher energy use, as in luxury cars or specialty beers, but may be the case in creating different types of glass bottles or more complex cast metal products. Given available data, the link between energy and value (price) can be treated as largely an empirical issue, but preferably with some underlying hypothesis about the industry in the case study. Value of shipments might be used instead of a physical production variable or in conjunction with physical outputs. In the latter case, the ratio of value to physical product is price and becomes an implicit variable in the analysis. Other measures of energy-related product differences are industry specific; as in the case of vehicle size in automobile assembly.
Size
Size and associated capacity utilization rates may directly impact energy use. Size may impact specific engineering and managerial advantages to energy use. If there is a substantial “fixed” level of energy use in the short run, the utilization rates may have a non-linear impact on energy intensity. In order to include size (and utilization) as a normalizing factor a meaningful measure of size or capacity is needed. It may be measured on an input basis, output basis, or physical size. In some cases, there may be advantages to larger scale of production, i.e., economies of scale. If it is the case that a larger production capacity or larger physical plant size has less than proportionate requirements for energy consumption, then there are economies of scale with respect to energy use. For example, in the cement industry, the scale is quite important. The larger size of the kiln (rather than several smaller kilns) has advantages in terms of energy use.
There are three ways that process inputs are important for benchmarking. The first is that inputs such as materials, labor, or production hours may be good proxy measures of overall production activity when measures of production output are not available or have specific shortcomings.
6 The second is in the identification for upstream (vertical) integration, i.e., whether a plant makes an intermediate product or purchases some preprocessed input. This is an important “boundary” issue for the energy footprint of a plant, even when two plants produce identical outputs. The third way is a variation of the second, relating to material “quality.” If there are alternative input choices that differ qualitatively and also with respect to energy use, then input quality measures can be introduced into the benchmark.
The first way process inputs can be helpful in developing a statistical benchmark of energy use is that inputs such as materials, labor, or production hours may be good proxy measures of overall production activity when measures of production output are not available or have specific shortcomings. If a physical measure of output is not readily available and pricing makes the value of shipments a questionable measure of production, then physical inputs can be a useful proxy. For some industries, the basic material input is so ubiquitous that it makes sense to view energy use per unit of basic input rather than (diverse) outputs. Process inputs may also be useful in measuring utilization, either directly or indirectly. Corn refining is a good example of this approach. The industry uses a ubiquitous input, corn. In some industries, physical production data may not be available but material flows are and can be used instead. For example, sand, lime, soda ash, and cullet (scrap glass) are the primary inputs to glass manufacturing.
The second way that process inputs are important for interplant benchmarking is when vertical integration is common in a sector but varies in degree from plant to plant. Industries are categorized by the products they produce, but some sectors may face a “make-or-buy” decision in the way they organize production. A plant may purchase an intermediate product or produce it at the plant as part of a vertically integrated plant. For example, an auto assembly plant may stamp body panels or ship them in from a separate facility. The energy use of these two facilities is not directly comparable. The interplant benchmarking approach must account for those “make-or-buy” decisions in the specific plant configurations. Examples range from food processing, where plant may make juice from concentrate or fresh fruit or paper mills which may purchase market pulp or recycled fibers.
The third way that process inputs are important for interplant benchmarking is when differences in material quality may also be related to energy use. For example, if the materials mix to produce a product directly impacts energy uses, then the statistical model can apply different weights to the material mix in the same manner that it does with product mix. In other words, product/process level differences in energy use can be inferred from the volume and types of materials used in production. To be considered in the statistical normalization, they must be measured on a consistent plant-level basis across the industry. For cement plants, the hardness and moisture content of the limestone is hypothesized to influence energy use, but no consistent data is available for this, leaving it the subject of future analysis if data can be collected.
One ubiquitous input is labor. Labor may be helpful in capturing the quasi-fixed nature of energy if there are production slow-downs or non-production periods of operation, but when both labor and energy are being used. In this way, labor captures a plant activity level that is related to energy use, even when product output is not being generated. As an empirical issue, the statistical significance, or lack thereof, of labor in the analysis can capture this potentially industry-specific phenomenon.
Climate
There are many things under the control of a plant or energy manager, but one they cannot control is “the weather.” In most manufacturing plants, HVAC contributes to energy demand and weather determines how much is required to maintain comfort. Since the approach used here is annual, seasonal variation does not enter into the analysis, but differences due to the location of a plant and annual variation from the location norm will play a role. The approach that has been taken for all sectors under study is to include heating and cooling degree days (HDD and CDD) into the analysis to determine how much of these location-driven differences in “weather” impact energy use.
In principle, all plants have some part of energy use that is HVAC related, but when the HVAC component of energy use is small relative to total plant consumption, the statistical approach may not be able to measure the effect accurately enough to meet tests for reliability. For some sectors, weather is a factor in the process, like auto assembly. It is a factor because of paint booths and the climate control technology needed for those systems. Pharmaceutical manufacturing, where “clean room” production environment is common, is another good example. The climate impact in this sector is only applicable to the “finish and fill” production stage. The more energy-intensive chemical preparation stage is not sensitive to climate. Even in industries where the HVAC component is not an obvious or large part of energy use, there may be production process-related effects that analysis needs to test for. For example, processes that use chillers may be sensitive to CDD (summer) loading. Process heat furnaces may be sensitive to cold outside air so HDD (winter) effects might be included in the model.