It is unlawful to copy any content of this article including text and graphics without the premission of the authors.
Steven Meyers, Evan Mills, and Allan Chen
Center for Building Science Applications Team, Lawrence Berkeley National Laboratory
Department of Civil Engineering, University of California at Berkeley
Accurate, appropriate information is essential to understanding the dynamics of energy use, control strategies, and occupant comfort in buildings. Performance data for a building are valuable at all stages of its life cycleÑdesign, commissioning, and operation. However, transforming raw data into useful information and , utilities, and energy service companies) to operate their buildings optimally is no small task. Even if a feedback system is in place, poorly organized and displayed data may not yield value to the building operator. This article presents ualization techniques, illustrated by seven case studies for HVAC and lighting in commercial buildings.
Data visualization complements the process of data acquisition. For our purposes in a way that permits the analysis of data at different timescales with the objectives of identifying or verifying the energy savings achieved by energy-efficient technology or identifying malfunctions in building equipment. Effective data visualization depends on designing graphic presentation formats that clearly reveal technical phenomena relevant to the performance of the building. Incorrect visualization can mask critical problems.
Today, engineering estimates of energy use and savings are the basis of most energy-efficient building design and retrofit projects. Unfortunately, in the field and their performance varies with time. Measured data can help verify energy savings, illuminate control and operational problems in a building, and test the accuracy of predictive methods. Such data are increasingly valuable in energy performance contracting, utility demand-side management program evaluation, and other activities where regulatory requirements or financial contracts require documenting actual energy performance and costs.
In recent years, data collection has tended towards increasing disaggregation into shorter timesteps and into specific end uses. With the availability of larger d development of data-acquisition technology, computer workstations, and graphics software has made possible gathering and visualizing detailed buildings energy benefits:
As with other measures designed to achieve and maintain energy savings, the costs of data visualization must be weighed against the benefits. There is no general rule; different users of data have differing needs with respect to detail and accuracy. Regardless of the desired level of precision, effective data visualization adds value to the process of optimizing building performance.
This article addresses key issues of building monitoring pertaining to data visualization. The variables that need to be measured and the necessary accuracy of these measurements depend on the analysis to be performed on the data. Therefore data. The required individual sensor accuracy can then be back-calculated to meet targeted objectives. Three options exist for collection:
Common displays such as time-series and scatter plots have been employed for decades (Liu, 1994). Recent advances in software for data visualization in the building sector have led to the development of a broad range of powerful visual diagnostic tools. Declining prices of powerful personal computers and workstations have made possible more data-intensive color displays that can add value to data. Various types of contour plots, psychometric charts, and 3-D surface plots help raphics may be animated.
Advanced data displays convey both qualitative and quantitative information. A common application of qualitative data analysis is to verify that mechanical and electrical systems are running properly and only when required. Such information energy, but may cause occupant discomfort and decrease equipment life. To effectively interpret quantitative results, detailed information such as nameplate data for the building's subsystems must be available. For example, if a sensor indicates that the air flow in a duct is 12,000 cfm, but the mechanical plans indicate the fan is rated at 8,500 cfm, something is malfunctioning or the wrong fan was installed. Also, original design parameters such as the expected coefficient of performance (COP) of a chiller at a given loading can be compared with the COP calculated from a plot of the monitored air/water flows and temperatures.
The type of visualization and the speed at which it is executed depend on the hardware and software available to the analyst. Currently available tools range from standard spreadsheet packages to high-end systems designed specifically for displaying monitored building performance data. Features useful for analyzing building data are described below. The day, date, and time should be easily correlated to each data value. Engineering units and specific formats (such as load aningful trends occur on time scales ranging from minutes to years. A visualization system should be capable of handling both extremes. Large buildings can have years, a process that generates large data sets. It is also convenient to have access to all the monitored data at once because the right questions to ask about a system's performance are not always clear at first. Preliminary data analysis leads to questions requiring information from additional data points. For example, to address complaints of occupant discomfort in a particular zone, an operator may display a time-series of the zone and supply-air temperatures. This may confirm that a problem exists, but without simultaneous access to data from valves and motors in the zone's air-handler, it is unlikely that the operator will
Maintaining a building involves a variety of players. Since advanced data visualization requires specific skills and hardware, the analyst may not be the person hod of feeding the analysis back to the operations staff must be developed. It is important to develop an implementation plan emphasizing specific responsibilities to communicate results and develop a response. All players must understand why this analysis is being performed, and institutional barriers can challenge reaching this consensus (Lovins, 1989). For example, building operators may feel uncomfortable because another team is intruding on their area of responsibility t equipment and monitoring/visualization systems from which the tenants benefit through reduced utility bills. Also, in some cases, when the building's facilities department reduces the energy load, the building's owners respond by reducing lude an incentive for building operators that produces benefits for all the parties involved in the building.
When a supply fan and its variable frequency drive (VFD) perform at less than the optimum level, the fan wastes energy and can damage the drive. Figure 1, often called a surface or carpet plot, correlates the electricity use of a building supply fan in kilowatts against the day and time of use month period from April through September, 1993. Regularly spaced breaks in the data represent weekends. The fan ventilates a 22,000 ft2 (2000 m2) office building in Northern California, an area with a hot, dry summertime climate. The plot shows a nighttime offset of 1.3 kW for a period of 4 months beginning in April 1993. The forest of spikes in the central part of the plot portray the fan's daytime usage pattern. These fluctuations show the fan power profile as the VSD varies the fan's speed.
At a glance, this plot tells the building manager that the EMCS is not shutting down the fan's VFD at night as it should prior to 6/1/93. The 1.3 kW of energy represents an unnecessary cost of perhaps a thousand dollars a year. From the perspective of a building manager, this is a small cost and is passed on to the building's tenants. What cannot be overlooked is the effect of this situation on a piece of capital equipmen- the fan's motor. If a motor is not designed to run at low frequencies, it can overheat and burnout if driven too slowly by a VFD. The cost of a replacement motor is a "first cost" incurred by building maintenance nants, identifying this malfunction has produced a non-energy benefit to the building's owner.
Another interesting feature of this plot are the plateaus occurring on June 17 and 18. In this case, plateaus represent the intervention of a contractor who was hey could just as well have indicated a critical system malfunction.
To ensure that the offset is a real problem, and not the result of a miscalibrated sensor, the building's energy manager can examine other types of data if available. The measurement of airflow over time would show higher than expected nighttime values. Another measurement is the static pressure in the building's supply air duct, for which well-established procedures are available in ASHRAE standards, and which would be elevated during nighttime hours of operation.
During the commissioning of a new building or the retrofit of an existing one, contractors or operations personnel sometimes install measures that fail to save energy as planned because the measure was implemented incorrectly or because the measure.
The electricity usage of a bank of T-8 lamps between September 25, 1993 and March 29, 1994 is depicted in Figure 2 in a California office building. The lights, outfitted with occupancy sensors, illuminate 10,000 ft2 (1000 m2) of the 22,000 ft2 (2200 m2) building (Mills 1994a). The plot's dark ostly off. The dark vertical band on the left shows the off-hours of lighting energy use during the interval from midnight to about 6 a.m.
The plot reveals two problems with the lighting system's energy performance. The ring the late evening. The lights are on more than one would expect at this time ights in the absence of activity in the area. When building managers saw this first anomaly, they investigated and discovered that a security guard making his rounds until midnight each evening was tripping the occupancy sensors, reducing the expected energy savings. On November 11, new switches were installed which automatically turn off the lights but require manual turn on. Energy use went from 68 to 62 kWh per day, yielding an additional 9% energy savings.
The data show another problem relating to the dimmable ballasts that were installed as a retrofit energy conservation measure. On January 10, an electrician accidentally reset the ballasts to their maximum settings (from 6 to 7.5 kW). During the following two months--a period in which lighting data visualization was not carried out--, the lighting energy use increase of 18% to 73 kWh per day went unnoticed. The increase is visible on the diagram as the block of red horizontal stripes in Figure 2. When data for this period were retrospectively analyzed, the problem was identified, and on March 15, the dimmers were reset correctly. Data visualization played a critical role in detecting the problem. The human eye would not detect the small changes in illumination that have such a large effect on power consumption.
Another problem visible in the post-January-10 interval is a second rise in consumption during the night hours. This appears to have been caused by the installation of incandescent lamps in a number of previously unlamped sockets in the building's lobby where no occupancy sensors were present. Another feature here is the slight offset in the first five bands of morning data (upper left) between the < 1 kW and 1 to 5 kW zones. This offset was caused by the switch from daylight saving time to standard time in early November.
Motors that drive fans, pumps and compressors are often candidates for variable frequency drives as energy conservation measures. A common application is for fan systems where maximum air-flow is not always necessary to meet the load of the
VFDs often perform sub-optimally. In a recent commissioning survey, VFD performance problems defeated any of their benefits as an energy conservation measure in asons such as improper adjustment of the controlling parameters, and malfunctioning pressure sensors that controlled the drive.
Malfunctioning drives often go undetected because they usually fail in the full- speed mode and thus do not affect catastrophically the daily performance of the building. Simple displays of monitored data from VFDs allow building operators to determine quickly if a drive is performing to specifications. Figure 3 plots the energy (kW) consumed by a high-efficiency 2 HP motor driving a fan on an indirect evaporative cooling system in 10,000 ft2 (1000 m2) 2) of conditioned space in an office building in Northern California. The regularly spaced breaks in the data correspond to weekends when the building is not occupied and the mechanical systems are not running.
The plateaus in the Figure 3 show that the fan is running at constant speed most of the time. If the VFD were functioning correctly, we would expect to see spikes or rolling hills in this plot such as those observed in the supply-fan case study (Figure 1) . Displaying these data as a histogram also reveals in detail how much time the motor spent at a given frequency interval. If the drive were properly adjusted, the histogram would show events distributed over its full operating range (i.e. between 0 ogram in Figure 3b shows a disproportionate population of events occurring at roughly 1.8 kW when the fan was on full. Although histograms provide an additional level of detail, the 3-D surface plot in Figure 3a contains more information and is more intuitive to most operators.
Using hand calculations to quantify the energy savings from a malfunctioning VFD conditions at a particular moment. Displaying the data as a surface plot or a histogram as in Figure 3 is therefore useful in estimating the patterns of the drive's operating range.
Direct power measurements of the type used in these case studies are not always available. Most EMCSs offer an alternative and can record the signal sent to the an be used to determine a drive's status. However, if parameters are set incorrectly in the drive, the actual speed of the drive may differ from the EMCS's signal. Simultaneous recordings of the signal to the VFD, the power of the motor, and the sensor controlling the VFD (often a pressure sensor) allow a building operator to best evaluate the performance of a drive.
Faulty sensors are common in buildings and have significant implications for energy efficiency. Consider the data collected from an outside air temperature sensor during the entire year of 1992 in San Ramon, California. Most EMCSs use such a sensor to control the chiller lock-out or determine the effectiveness of economizers and evaporative coolers. At a first glance, a common time-series plot of the outside air temperature (Figure 4a ) suggests reasonable behavior of the variable. However, a careful observer would notice the curious plateaus indicated on the Figure. Visualizing the frequency at which a given temperature interval occurred as a histogram (Figure 4b) reveals an unsuspected problem with the sensor. A smooth Gaussian-type distribution around the city's mean temperature would be characteristic. The data displayed on the histogram suggest the implausible circumstance that a temperature of 64 F occurred more than twice as often as a temperature of 63 F and 65 F. This sensor failed by sticking every 9 F.
We have seen other problems with data from outside air temperature sensors. Sensors misplaced in the hot air stream of the condenser exhaust record erroneous values above the actual outside temperatures. In such a case, the problem would be visualized by plotting daily profiles of outside air temperature with another e to the roof of a building or not properly shielded, false high temperature readings may allow the EMCS to turn on energy-intensive mechanical cooling before it is really necessary. Visualizing daily, un-averaged temperature profiles such as the time-series in Figure 4a can help detect this error.
Economizers in HVAC systems bring in outside air to cool a building when the air ssors that cool the building mechanically. However, economizers are plagued by failed dampers or improper control settings and do not always perform optimally. The economizer in this case study is part of a 7.5 ton package HVAC unit on a 7 climate. Although these data were analyzed retroactively and no channels could be established for operator feedback, this case study presents three data displays that could provide building operators with progressively richer information to identify and diagnose economizer problems.
Figure 5a shows a common profile of the mixed air, supply air, and outside air temperatures, as well as the compressor's power during pressor status) with this profile suggests that the economizer is not functioning correctly because the compressor is running even when the outside air is cooler than the supply air and latent cooling loads are not significant.
Although scrolling through daily profiles provides detailed information about the economizer, it does not provide a picture of the economizer's overall performance. An XY-plot of the difference between the return air temperature and the outside air temperature (Tret-Toutside) versus the difference between the return air temperature and the mixed air temperature (Tret-Tmixed) summarizes three months of data relating to the economizer (Figure 5b ). This Figure displays data collected between May 29 through September 9, 1993 at 15-minute intervals. The dotted line represents an outside air fraction (OAF) of 1 where 100% outside air supplies the building. The data in this display suggest three possible operating regions:
Although the density of data in Figure 5b suggests that the economizer is often stuck at an OAF of 18%, some information is lost when s and dates cannot be correlated with particular data points. The conditions and chronology corresponding to each of the three regions cannot be known. Since regions of the data display become saturated with data points, the relative time ending this display into three dimensions by creating an XY-plot for each day ( Figure 5c ) shows that there are only about five days that deviate from region 1 described above.
While visualizing a 3D display as in Figure 5c , a sectional plane can be passed or even animated through any day. Data from this single day can be displayed simultaneously as an inset XY plot (Figure 5d ). Animating this 2D XY-plot through time adds an additional r the hours when the building is not occupied. When the sectional plan was passed through the entire 3D display, the XY inset clearly showed collinear data points falling around 18% OAF for all except five days. These five days primarily had mysterious patterns when data points fell in the lower right quadrant (described above as region 3) as shown in Figure 5d . The user can view the dates and times of these anomalies and then display the anomaly differently, for example, in a detailed time-series plot. The analyst can display other relevant parameters such as compressor power, fan power, or relative humidity and see the exact conditions under which the anomaly occurred. In this case, this analysis showed all of the mysterious days corresponded to times when the supply fan was not operating and heat from either the burner, sunlight, or compressor heated the mixed air. This discovery indicated that the economizer he correct displays quickly allowed analysts to answer a variety of diagnostic questions as they surfaced.
Quantifying the energy savings from malfunctioning economizers is difficult because the economizer contributes to a varying fraction of the cooling load. In some climates, latent loads must be considered as well. Comparing the monitored and mate the energy lost due to the malfunctioning economizer in this case, we took the difference between calculated the total measured energy used by the compressor and measured energy used by the compressor only when the outside temperature was lower than the supply air temperature. During the six-month period from June rily to cool the building.
Although 11% of the compressor's total energy could have been saved by correctly compressors prematurely ages a valuable piece of capital equipment. To estimate these effects, we compared the total hours that the compressors were running and wer than the supply air temperature. During the six-month period, 28% of the multi-stage compressor's total runtime was unnecessary from an environmental control standpoint. However, from a maintenance standpoint, periodic operation may be beneficial. Although not the case in this particular example, this type of analysis can also identify problems that could compromise indoor air quality.
The mechanical systems in large modern buildings can contain hundreds of components. Not surprisingly, these complicated systems do not always perform optimally onents helps building operators in an area of diagnostics commonly referred to as relational checks. We define relational checks as comparing two or more systems to ensure that both are operating as designed. For example, an air handler should not be simultaneously heating and cooling the supply air (assuming latent conditioning is not required). Although these relational checks may appear obvious, they are common malfunctions in buildings because the building systems can often meet the required load (albeit inefficiently) in spite of them.
A first step in performing relational checks is reviewing the mechanical drawings or control diagrams of the building to determine how the systems were intended ould be that the chilled-water pump not be running if the compressors are not running1. However, the daily profile in Figure 6 shows t hat on May 21, 1993, the compressor never turned on, but the pump was running almost all day at its maximum 0.6 kW. Scrolling through months of data, we noticed n this pump, unnecessary operation prematurely wears the pump and, in this case, the VFD that drives it. An error in the EMCS program controlling the chiller caused this malfunction. The building's mechanical contractors were notified and the software problem was corrected.
Collecting the data necessary to perform relational checks is relatively easy. Most EMCSs monitor each component's status and can therefore indicate when it is running or the total hours of operation. However, such status indicators cannot quantify parameters such as speed, current, or energy consumption. Some relational checks can be performed with run-time data alone. For example, in the system described above, the pump and the compressors should be running almost the same amount of time. If one runs significantly more often than the other, there is likely to be a problem. Monitoring run-time is generally one of the least expensive measurements because only a single scalar value (time of operation) is recorded as opposed to an array of values necessary to express a time series. We have running by the increase in temperature!
Returning to Figure 6, we cannot ignore the 3.5 F error in the calibration of the supply and return water temperatures. Looking again e refrigerant loop is not moving and these sensors should read the same temperature. ASHRAE recommends that delta-temperature measurements in a water loop be accurate within ±0.5 F (ASHRAE Standard 114-1986). These results were presented to ffects from this miscalibration are subtle. If the sensors were connected to an EMCS, then the particular HVAC components controlled by the chilled water temperature could have been triggered before they were necessary. Modeling the actual and designed performance in a building simulation package would be the most reliable method of determining these energy effects. However, the sensors in this case study were installed to evaluate the energy consumption of this chiller, which was installed as a retrofit energy conservation measure. If we assume a full-load operating delta-temperature of 10ûF in the water loop, the error in calculating the COP of this chiller will be at least 35% 2 and therefore unreliable. Visualizing monitored data from these systems helped to identify this measurement problem.
Most EMCSs have alarms that can automatically alert the building operators in the event of relational malfunctions. However, in some cases it is difficult to program these alarms so that they do not trigger falsely. When inundated with false alarms from the control systems, an operator's first reaction is to disable them. Although the relational alarms are often helpful, a knowledgeable operator will diagnose more problems using an efficient interface to scroll through data manually. For example, this case study began by discussing relational checks and progressed to discussing sensor calibrations. As indicated by this progression, it is common to find problems other than those originally sought when diagnosing HVAC systems. Until more applications of fuzzy logic appear in the building controls industry, manually visualizing monitored data using efficient interfaces and data displays will be an effective method for HVAC diagnostics.
This HVAC case study illustrates the importance of choosing the appropriate timestep for data collection and visualization. As discussed in the Methods section rmed on the data. However, we can illustrate how some important features of monitored data can be missed if data are sampled infrequently. Figure 7a shows a profile of the supply and return water temperatures along with the power from the chilled water pump and the compressor sampled on June 29, 1993 displayed as hourly averages. However, Figure 7b shows this same data displayed as 15 minute averages. The 15 minute averages instantly reveal spikes throughout the day, indicating that the compressor is short-cycling. Even finer resolution may be necessary for some analysis. Short cycling prematurely wears out the motor, and the current spikes also consume unnecessary energy. Viewing the data hourly did not accentuate this malfunction. The plot suggests that the control strategy for this system needs to be tuned. Scrolling through relational HVAC data recorded at small time intervals quickly reveals problems to building operators that would likely otherwise be undetected.
The case studies presented above demonstrate the use of data visualization as a tool to facilitate the detection and quantification of performance problems. Incorporating data visualization with operation and maintenance requires that several issues be addressed.
There are many feasible approaches to the implementation of regular data visualization. Two possible approaches are presented here.
These two approaches could be combined with overall guidance being provided by a regional Center.
The discussion and case studies presented in this article indicate that data visualization can help achieve substantial improvements in energy management, equipment maintenance, and occupant comfort in commercial buildings. Data visualization alone cannot improve a building's performance and must be integrated to an O& M plan to be effective.
This work was funded by the Assistant Secretary for Energy Efficiency and Renewable Energy, Office of Building Technologies of the U.S. Department of Energy under Contract No. DE-AC03-76SF00098. The authors wish to thank Lee Eng Lock and Arthur Rosenfeld for motivating this research. We also appreciate many helpful conversations with Grant Brohard, Kristin Heinemeier, George Hernandez, and Mary Ann Piette.
