Diagnostics
By Jessica Granderson
Applications
Marrying bottom-up and top-down performance analytics
Best practice in building energy and operational diagnostics begins with setting efficiency, demand, and energy consumption targets at the whole-building, system, and end-use levels. Performance is tracked relative to those targets on a monthly and annual basis. Continuous fault and energy anomaly detection is applied to hourly or sub-hourly data – again at the whole-building, system, and end-use levels, and also for specific components or pieces of equipment. The outputs of these continuous diagnostics are used in daily and weekly operational assessments to maintain persistent operational efficiency.
In a top-down approach, whole-building energy use is analyzed, followed by ‘drill-down’ investigations into systems and end uses, to gain more granular insights. In a bottom-up approach, the focus is on the constituent systems and end-uses. The most effective campus and building diagnostic strategies integrate both top-down and bottom-up approaches into a holistic data-driven energy management process.
Setting targets and tracking performance
Building targets might be set relative to existing portfolio energy use (for example, twenty percent below current consumption), or relative to national benchmarks (See Action-oriented Benchmarking). System and end-use targets can be set using energy metrics, or specific efficiency metrics such as watts per cubic feet per minute for fans (Figure 1). Submetering is implemented to disaggregate whole-building level energy use into more granular subtotals of system, and end-use, or even component level energy consumption. In best practice applications, performance with respect to targets is reviewed approximately monthly, using a 12-month rolling window for metrics that are expressed as annual totals or peaks.
Figure 1: Schematic illustration of energy, demand, and efficiency metrics used to establish performance targets.
Continuous diagnostics
Complementing performance tracking, diagnostics are applied to hourly or sub-hourly energy data streams, and operational trend logs from building automation systems (BAS). The results are viewed daily. These diagnostics are referred to in many ways; commonly used terms include fault detection, energy anomaly detection, exception reporting, monitoring based commissioning, and ongoing commissioning [see Commissioning section].
Today’s commercially available analytical software tools typically offer some combination of the following three diagnostic approaches.
Rule-based diagnostics are often specific to certain equipment or systems, and their associated control logic. They focus on operations as opposed to energy consumption, and are not generally applied at the aggregated end-use level. BAS trend logs are the primary source of data for rule-based diagnostics. Figure 2 shows an example of the graphical output from a rule-based automated fault detection diagnostic module.
Empirical energy and demand baseline models (for more info refer to Granderson et al. 2013) can be applied at the whole-building or submetered level to perform energy anomaly detection. These approaches focus explicitly on identifying periods of abnormal or excessive energy use, relative to typical historic usage patterns, while accounting for factors such as weather and time of day.
Like-equipment benchmarking is form of peer-to-peer, or “cross-sectional” benchmarking. Applied at the system, end-use, or component level, parameters such as energy use or run time hours are compared. These approaches are most effective in environments such as campuses, where there are sufficient numbers of like equipment to conduct the required comparisons.
Figure 2: Automated fault detection in a rule-based commercial diagnostic tool (Granderson et al. 2013). The y-axis shows economizer damper position as a percent of fully open, and the x-axis shows outside air temperature. Note: Points labeled "B" appear light green on some displays.
For each approach, a key technical issue is to determine the thresholds at which a fault is triggered. For example, in a rule-based system that stipulates that one parameter should be “greater” than another, it is necessary to determine how much greater, given uncertainties in sensor measurements. In baseline model approaches, and like-equipment comparisons, the same types of thresholds must be determined. To avoid false positives, thresholds are set more loosely, whereas to avoid false negatives, thresholds are set more tightly.
Continuous measurement and verification
‘Cumulative sums’ (CUSUM's) are a best practice diagnostic to quantify energy and cost savings, and therefore the effectiveness of data-centered energy management practices. They are also used to track and maintain savings persistence by identifying periods of abnormal energy consumption. CUSUMS plots provide a powerful at-a-glance view into: (a) whether savings or energy use is increasing, decreasing, or stable, and (b) the total accumulated value of energy or cost savings, over time (Figure 3).
Figure 3: Use of cumulative sum diagnostic plot to continuously track savings and to identify energy waste, due to a component fault (Granderson et al. 2013).
Bleeding Edge (emerging technologies and practices)
Linking diagnostics to controls
A limited number of commercial tools deliver automated control optimization based on continuous analytics and two-way integration with building automation systems (BAS). BuildingIQ’s Predictive Energy Optimization, Enerliance’s LOBOS, and Optimum Energy’s OptiCx are a few examples. These tools are still ‘emerging’, and their costs and benefits have not been widely demonstrated; however optimal control and the implementation of continuous corrective action-based on diagnostics, represent the future of advanced buildings. Model-predictive controls will becoming increasingly critical as the industry moves to more dynamic, grid-integrated buildings [See section on Demand Response & Electric Vehicles].
Virtual sensing
Virtual sensing refers to the use of available data to estimate the value of variables that are too difficult or expensive to sense directly. Today’s building monitoring systems commonly use calculated virtual sensors, for example adding or subtracting one electric meter from another or calculating thermal energy based on temperatures and flows. More sophisticated approaches such as CO2 or IT network traffic traffic might be used to derive information about occupancy. Ongoing LBNL research is showing that network traffic data can in some cases be used as an effective proxy for building occupancy, can improve the accuracy of baseline load prediction models.
Physical model-based diagnostics
Best practice diagnostics used today are based on measured data from sensors and meters. Bleeding-edge applications such as that shown in Figure 4, are beginning to demonstrate hybrid empirical/physical simulation based approaches. Whereas purely data-driven approaches permit analysis of current vs. historic performance, and representation of engineering-based operational rules, diagnostics that incorporate physics-based simulation models enable representation of first principles and design intent. (Bonvini et al. 2014; Bailey et al. 2011).
Figure 4. Design of a model-based diagnostic application demonstrated in partnership with the Department of Defense. Estimation-based diagnostic algorithms are used with physics-based Modelica models and measured equipment data, to identify operational faults; faults are output to maintenance and operations staff through a graphical user interface.
Economics
Over a decade of case study research has shown the value and rapid payback associated with whole-building and system level diagnostics. Site energy savings from 10% to over 20% have been reported in the literature, with paybacks often ranging from one to three years (Granderson 2011; Henderson 2013; Motegi 2003; Smothers 2002).
The cost of end-use submetering can be minimized by employing ‘design for meterability’ in new buildings. This approach makes an effort to design electrical distribution systems in which end use loads are disaggregated, and therefore measurable with fewer meter points. In most existing buildings, end-use loads are mixed within electrical panels that serve areas of the building, as opposed to specific end-uses.
Some of the highest costs in implementing advanced diagnostics are associated with BAS trend log integration. Tremendous time and cost may be required to (a) commission BAS sensor data to ensure accuracy and quality, (b) map trend log names to the associated variables required for diagnostic algorithms, and (c) integrate data from multiple vendors’ BAS for use in central diagnostic tools. To minimize these costs, procedures that standardize data protocols, naming conventions, and BAS vendors across buildings and campuses are critical.
If cash flow and capital availability are constrained, it may be practical to implement top-down, phased metering and diagnostics, beginning with whole-building diagnostics and modest submetering, ad exiting BAS-based diagnostics. After savings begin to accrue, more granular submetering can be implemented to enable persistence and deeper savings.
Other considerations
The most significant co-benefits of a comprehensive diagnostics strategy include improved occupant comfort and indoor environmental quality, the ability to ensure persistence of savings over time, and the availability of data that to streamline reporting practices. For example, enterprises may use the data from their diagnostics platforms to support ESCO contracting or to calculate and report Scope 1 and 2 greenhouse gas emissions associated with building operations. In addition, the metering, communications and software infrastructure used for energy and operational diagnostics may be leveraged and extended to support monitoring of all ‘WAGES’ resources (water, air, gas, electricity, and steam).
Institutional requirements & capacity
User expertise and training are important considerations in the successful deployment of diagnostics. While some diagnostics are fully automated, others require a ‘human-in-the-loop’ to view and interpret data. Given the many possible ways to ‘slice and dice’ data, it is critical to establish a comprehensive plan to meet organizational energy performance goals, including which diagnostics will be employed and how frequently, and what data is required.
The industry is beginning to more tightly couple diagnostics with maintenance, for example by integrating analytical tool outputs with work order systems. This is usually achieved through work stream process standardization (although customized system/platform integration is possible). Although it has long been acknowledged that ongoing diagnostics can extend equipment life and improve maintenance costs, condition-based maintenance and prognostic solutions are not yet as common in building applications as they are in other engineered systems.
References
Bailey, T, O'Neill, Z, Shashanka, M, Bhattacharya, P, Haves, P, Pang, X. Automated continuous commissioning of commercial buildings. Lawrence Berkeley National Laboratory, September 2011. LBNL Report No. 5734-E.
Bonvini, M, Piette, MA, Wetter, M, Granderson, J, Sohn, M. FDD Bridging the gap between simulation and the real world: An application to FDD. Proceedings of the 2014 ACEEE Summer Study on Energy Efficiency in Buildings, Pacific Grove, California.
Granderson, J, Piette MA, Ghatikar, G. 2011. Building Energy Information Systems: User Case Studies. Energy Efficiency 4(1): 17-30.
Granderson, J, Piette, MA, Rosenblum, B, Hu, L, et al. Energy information handbook: Applications for energy-efficient building operations. CreateSpace, 2013, ISBN 1480178276 / 9781480178274; LBNL Report No. LBNL 5272E.
Henderson, P, Waltner, M. Real-time Energy Management A Case Study of Three Large Commercial Buildings in Washington, D.C. Natural Resources Defense Council, October 2013. Report Number CS-13-07-A.
Motegi, N, Piette, MA, Kinney, S, Dewey, J. 2003. Case Studies of Energy Information Systems and Related Technology: Operational Practices, Costs, and Benefits. Report prepared for the California Energy Commission, Public Interest Energy Research, HPCBS # E5P2.2T1e. LBNL Report No. 53406. Proceedings of the International Conference for Enhanced Building Operations.
Smothers, Frederic J., and Kristopher L. Kinney. 2002. Benefits of enhanced data quality and visualization in a control system retrofit." Proceedings of ACEEE Summer Study on Energy Efficiency in Buildings, Pacific Grove, California.