Systematic procedures to address these problems are beginning to emerge. For example, research at Texas A&M University has found that in almost all older buildings, and even in many new buildings, the use of the building is quite different from the original plan (Claridge, et al., 1994). These researchers use a monitoring process of "continuous commissioning" to tune building systems for optimal comfort and peak efficiency based on current operational requirements. Their methods have saved an average of over 20 percent of the total energy cost (over 30 percent of the heating and cooling cost) in over 80 buildings (Claridge, et al., 1994). While these researchers have demonstrated success in bringing in experts to "fix" building systems, few tools are available to the on-site engineer to conduct such improvements.
A related problem is that EMCS?s are becoming more complex over time and are difficult for the average operator to understand (Hyvärinen and Kärki, 1996). Furthermore, most EMCS?s do not include energy monitoring in their scope. Building operators have only the monthly utility bill to help track how much energy is used. One study that supplied building operators with energy use data found that after a few months of strong enthusiasm, building operators lost interest in standard energy use plots provided by the utility research project (Behrens & Belfer, 1996). Building operators need assistance in sifting through the large volume of data available with new monitoring technologies. Current EMCS?s have limited capabilities in collecting, archiving, and displaying important building performance data. We pose the question of whether higher-quality data are needed to perform important analysis. New techniques are needed to assist operators in extracting relevant information from the underlying data. Thus, some automation of diagnostics is needed to set off alarms that can tell an operator when the diagnostic system has identified a performance problem or deviation from normal operation.
This research also addresses the variations in the professional activities of building operators, including their knowledge and technical sophistication. The vast majority of research in the fault detection, commissioning, and diagnostics area has focused on using expert engineers as the analysts, rather than working directly with building operations staff. Controls companies report that building operators are not interested in analyzing large, complex building performance data sets. This project has specifically addressed the question: is there a market for high-quality, high-resolution, archived building performance data if the tools required to simplify the analysis are also provided? Finally, this project addressed the problem of understanding how property management companies identify and analyze new technologies. Are these processes similar from company to company?
1. Maintain, enhance and finalize the IMDS specification. The initial IMDS specification was developed prior to the selection of the building and was used to present the technology concept to the prospective demonstration site partners. During the past year we have made some changes to the software.
2. Evaluate IMDS performance. To evaluate the energy savings and other non-energy benefits of IMDS use in technical terms. Our aim was to reduce total energy use and energy cost by 15 percent without sacrificing any other building services or performance issues. Energy use in most commercial buildings is dominated by electricity use; energy costs are dominated by electricity costs.
3. Develop and demonstrate techniques to automate fault detection and diagnosis. Although the project has focused on supporting and evaluating the usefulness of the manual, human-based diagnostic tool (IMDS), we have explored two approaches for automating fault detection and diagnosis. One is a steady-state chiller model. The other is evolutionary programming for self -learning systems.
4. Evaluate decision making and technology adoption processes in the commercial buildings sector. The objective of this aspect of the project is to provide a description of the innovation adoption process and a road map for the adoption of practical, business applications of new energy technology in the commercial building industry, focusing on the role of third-party property managers.
5. Evaluate the costs and economic potential of IMDS. In this report we ask the simple question, "What is the economic value of the data?" While there is no simple answer, we examine this question in detail, from a variety of perspectives.
Table 1-1. Project spending by task
|
|
|
|
|
|
Revise work statement, task deliverables, and budget |
|
|
|
Prepare quarterly progress reports |
|
|
|
Maintain and enhance IMDS |
|
|
|
Functional specification |
|
|
|
Prototype review |
|
|
|
Test automated diagnostics |
|
|
|
Technology transfer |
|
|
|
Prepare interim report |
|
|
|
Prepare final report |
|
|
|
Final meeting |
|
|
Total
|
|
N/A = Not applicable
The IMDS is oriented toward deploying the basic infrastructure for an advanced information system. This demonstration will allow the controls industry to examine the value of such systems that greatly exceed today's current EMCS technology. Such a system is the starting point for more advanced, automated diagnostics, such as those based on fuzzy logic or neural networks. The system is a distributed data collection and analysis system. The primary elements of the system are:
The IMDS development in Phase One included creating nine standard plots available for viewing of key performance data. The operations staff was trained to interpret these plots. The IMDS also offers a series of more sophisticated browsing and statistical analysis tools. These more sophisticated tools will likely be of greater use to the remote researchers. Researchers in several locations will have access to the data, plus the identical analysis software, allowing them to analyze the building performance and test the automated diagnostic systems. The PC server will offer a subset of the real-time analysis graphics from the demonstration site to the public over the World Wide Web. The purpose of these graphs is to demonstrate the technology to interested organizations and potential service providers such as energy service companies, utilities, and control companies.
Figure 1-1. Components of the Information and Monitoring Diagnostics System
Four types of measurements are taken by the IMDS: temperature (including wet-bulb), electric power, flow speed, and pressure. The installed system consists of 57 physical and 28 calculated points for a total of 90 points sampled at one-minute intervals. The sensors include high-grade thermistors, electric power transducers, magnetic flow meters, and aspirated psychrometers. Table 1-2 summarizes the monitoring scope. Further details of the sensors and sensor accuracy were presented in Piette, et al., 1998.
Table 1-2. Systems and sensors in the IMDS
| System to be evaluated |
|
|
| Whole building | Electric power |
|
| Two chillers | Differential pressure (water)
Water temperatures Flow rates (water) Electric power |
8 5 2 |
| Four pumps | Differential pressure (water)
Electric power |
4 |
| One cooling tower | Drybulb temperature
Wetbulb temperature Water temperatures Electric power |
2 6 2 |
| One air handler | Drybulb temperatures
Electric power Static pressure |
2 4 |
| Local micro-climate | Drybulb temperature
Wetbulb temperature |
1 |
| Miscellaneous (lights & plug) | Electric power |
|
|
Total
|
|
The IMDS is designed to be permanently installed and continuously active. This is necessary because buildings continuously change. For example, some problems recur, such as those from modifications to schedules to handle special events. These modifications often lead to equipment being left on when not needed. The diagnostic system is designed to operate in parallel with an existing EMCS, rather than expanding or modifying the EMCS. The IMDS is therefore not constrained by EMCS data collection capabilities, which can be problematic with 50 points of one-minute data. This technology may, however, be incorporated in future EMCS?s. The EMCS at 160 Sansome focuses on scheduling and controlling building HVAC systems including air temperatures and flows and monitoring zone conditions. By contrast, the IMDS measures energy, weather and water-side variables (temperatures, pressures, and flows).
Figure 1-2. A model-based fault detection scheme
The left hand diagram in Figure 1-2 shows the general principle and the right hand diagram shows the application to a heating coil, in which the inputs to the model are the inlet temperatures and the control signal and the output of the model is the outlet air temperature.
Reference models used for fault detection can be either qualitative or quantitative. One common form of qualitative model is a set of IF/THEN rules, which would be implemented in a knowledge-based system, together with an inferencing engine. There are two main types of quantitative models: first-principles and empirical. Two common methods of implementing empirical models are polynomial curve fits and neural networks. In each case, the model is only as good as the data used to train it; training data that cover the operating range of interest are required, since empirical models generally have poor extrapolation properties.
If the reference model is a steady-state model, i.e. it does not treat dynamic response, it is necessary to use only steady-state data, both when training the model and making comparisons between measurements and model predictions. Various forms of steady-state detectors have been employed in automated diagnostics; a summary is given in Hyvärinen and Kärki, 1996. In the terminology of the reference, the detector employed here is based on the geometrically-weighted average of the functional variation. This average is calculated recursively and is given by:
where yk is the value of signal at sample-time k
and a is the forgetting factor (0 <
a
< 1). The signal is deemed to be in steady state when 
is
less than a selected threshold, t.
Figure 1-3. Ideal types of innovators found in any typical population. The estimated percentage of the population in each category of innovation is shown (Rogers, et al., 1983).
It is rare that any substantive innovation enters a population without first passing through innovators and early adopters. Frequently, there is a break, between the adoption by early adopters and adoption by the early majority, where products and services die (Moore, 1999). Innovators and early adopters will look at more technologies than the wider population and will discard some innovations. The first and critical step in developing an understanding of the adoption of any technology is to understand who the innovators are. Innovators act as gatekeepers for technologies within their own companies. These innovators are easily identified by others in their industry even though they are not formally appointed (Allen & Nochur, 1992; Tornatsky, et al., 1983). The adoption processes described in this report are separated into two main categories: routine innovation and radical innovation.
Routine innovation - Routine innovations are generally innovations to existing products. This innovation can be large (new chiller systems) or small (hand-held tools). This method of adoption is frequently used and the introduction of these types of innovations does not change the way that the company does business.
Radical innovation - This type of innovation is sufficiently unique to the prospective user that he or she must first understand how it fits into his or her existing business. While it may share key features with other technologies, it must be sufficiently different that its introduction changes the way the company or building operates (Dewar and Dutton, 1986).
The adoption process includes several stages:
Knowledge. The prospective user becomes aware of the innovation.
Persuasion. Users follow up on their original information to determine if the innovation suits their needs.
Decision. The user decides to adopt or reject the innovation.
Implementation. The innovation is put into place and the adopter gains first-hand experience with the technology.
Confirmation/denial. The now-experienced user determines whether the innovation is suitable for their purposes and determines a course of action.
While the stages of the adoption process are artificial constructs developed by the researcher to categorize events in what is really a continuous process, they are standard among industry users and with minor variations are commonly used in innovation research (Rogers, 1983).
The research activities in Phase Three to address the five objectives are described below.
Maintain, enhance, and finalize the IMDS specification - All of the data were examined for completeness on a regular basis. A series of problems were identified and corrected. The final specification is available on the web at http://poet.lbl.gov/tour/.
Evaluate the IMDS performance - The operations staff was interviewed every two weeks to determine how they had been using the IMDS. These discussions included evaluating the performance of various building components and system. The IMDS was installed in May 1998. We used data from May and June as a baseline to evaluate the performance of the building. This evaluation was done to ensure that we had at least a few months of observations prior to fully training the on-site staff in the IMDS operation. The building staff began using the IMDS on a daily basis in late summer 1998. We have collected all of the data and examined them for completeness. A detailed analysis of the multi-year energy use data was conducted to examine whether there have been any energy savings between July 1998-June 1999 and the previous year.
Develop and demonstrate techniques to automate fault detection and diagnosis - This project has included chiller modeling for model-based fault detection and development of evolutionary programming techniques.
Model-based fault detection for chillers - The general approach for using a model-based fault detector for a particular HVAC subsystem is as follows:
1. Select the model. Select the type of model to be used as a reference model (quantitative versus qualitative, first principles versus empirical, steady state versus dynamic).
2. Tune the steady-state detector. If the model is a steady-state model, set up a steady-state detector to indicate when the comparisons between the model and the real system are valid. This involves tuning the parameters of the detector so that it removes as many unsteady points as possible while not removing an undue number of steady points.
3. Configure the model. If the model is an empirical model, obtain training data from the real system when it is known (or assumed) to be operating correctly. These training data are measurements of the inputs and output(s) of the model taken from the real system. Ideally they should cover the operating range of interest. The best way to obtain these training data is from a systematic functional test procedure, of the sort used in commissioning. This ensures that data covering the whole operating range are collected over a short period of time, before the system has had a chance to degrade or be modified, and allows the data to be compared easily with the manufacturer?s performance data. For empirical models, the configuration process involves fitting the model to the training data, usually using some form of least-squares procedure. The standard deviation of the residuals provides a general indication of the goodness of fit and hence of the adequacy of the model structure. A more appropriate test is to examine the points with the largest residuals and determine if there is any objective reason why they should have been excluded a priori. If not, these residuals determine the threshold for the fault detector, since they result from what are deemed to be valid measurements from a correctly operating system.
4. Implement, monitor. Use the model to monitor performance and test for faults. This monitoring can either be done in real time, with the model running on-line and making comparisons sample by sample, or it can be done off-line at suitable intervals, e.g., every day.
Evolutionary programming research on automated diagnostics -During Phase One, which took place during the mid-1990?s, the research team developed the IMDS concept based on interviews with innovative building operators. The IMDS design would include accurate, high frequency data collection with a top-down design. The top-down design would provide performance data for the whole-building, major systems (e.g., cooling plant) and sub-systems (such as chillers and cooling towers). The IMDS would compare the measured data with benchmark data from simulations, rules of thumb, or models of correct performance. and operators could be. One of the concepts of the IMDS is that rather than trying to detect specific faults, the system was specified to monitor performance in areas of the building that were performing poorly. The intent was to tell building engineers: "There is something wrong with subsystem X," rather than, "The front bearing on pump 6 is failing." The IMDS was designed with the concept that once building operators are made aware of a problem in a specific subsystem, they are reasonably good at tracking it down. The principal advantage of using human-based diagnostics is that one can immediately generate useful detectors for potential efficiency increases. By partnering with actual building operators, we can progress in an orderly fashion toward increased specificity without running the risk of missing "big-picture" issues.
Consistent with the overall project objectives, the UCSD team had the specific objective of interfacing an analysis prototype with the real data and so provide a means by which the building operators could investigate the data, have access to benchmark data sets and readily available plots. A demonstration system, implemented in Matlab, was developed that has a rudimentary capability to learn from real data. (Matlab is a mathematics and data analysis program commonly used in academic institutions. The idea was that the operators would make some specification such as, "On such and such a date and time, X was fixed." and the automated system would learn by example and create a test to determine whether data presented to the system indicated that X was fixed or faulty. This would be done for problems the operators consider important and frequent enough to require automatic detection.
In the development of the evolutionary programming techniques andthe decision to use graphical approaches to extract information from the IMDS data, a method is needed for generating filters which automatically recognize specified patterns in the data and alert the operators when such detection has occurred. The patterns were to be specified in advance (as opposed to the self-learning capability described below). These filters were of three types:
1. Range checks (potentially multidimensional)
2. Two-dimensional plots in which the axes were arbitrary functions of the raw data and regions to be detected were of arbitrary shape and not necessarily connected.
3. Three-dimensional plots in which the axes were arbitrary functions of the raw data and the regions to be detected were of arbitrary shape and not necessarily connected.
The UCSD diagnostics team explored these capabilities. A method was developed for automatically creating a robust test for a building condition that was important to operators and that could be generated from a statement from the operator such as, "X was fixed on -/-/- at ---- hours and - minutes." Such an on-line capability would also be able to obey operator commands, such as, "Let me know if the following condition recurs".
Evaluate decision making and technology adoption processes in the commercial buildings sector - The evaluation of decision making and technology adoption processes used Participant Action Research (PAR) techniques (Whytte, et al., 1991). In PAR, the participants are informed about the research goals and objectives, and are allowed access to the research results. We have agreed to allow the industry participants to slightly conceal descriptions of their adoption processes to ensure full cooperation and obtain the most accurate representation of the processes. Results are provided that accurately represent the process while protecting the identity of the participants. The results are generalized and are not unique to any participant unless noted otherwise. All participants have agreed to be identified by company name for the purposes of this report. The subjects in this study are third-party property managers who specialize in operating commercial buildings they do not own. Third-party property manager?s core business is the management, operation, and leasing of buildings. The companies who have participated in this research are:
· Jones Lang LaSalle
· Kennedy Wilson
· Cushman Wakefield (Northwest area)
· Cushman Wakefield (Southwest area)
· Pacific Properties Limited
The project tested the following three hypotheses:
1. Experienced innovative managers of commercial buildings have developed a way of identifying and analyzing new technologies. These processes may be generally well known to innovators in their industry, but no scientific research has been conducted to formally document the process.
2. The processes are sufficiently similar from company to company that there is a general adoption model that can be developed.
3. The adoption process for radical and routine innovations are sufficiently different to warrant separate discussion and model development.
The industry participants were selected by their peers in the Building Owners and Managers Association (BOMA). A general description of an innovator was provided to the BOMA local offices and they were asked to provide a list of ten companies in their local area that were innovators. A dialogue was opened with industry innovators in Northern and Southern California. The purpose of the dialogue has been to use the knowledge and experience of the industry participants to develop a technology that was useful and would desired by practitioners. Potential users have guided the selection of the IMDS technology from the earliest days. The technology was developed to overshoot commercial applications - that is, it was not intended to provide an incremental advance in the technology of building operation. In 1992-1993, researchers anticipated the development and adoption of Internet technologies, the continued integration of control devices in new products and the expected lower cost of instruments, software and hardware. In 1992-1993 the research team expected the monitoring system to evolve toward a system similar to the present IMDS. The IMDS technology was unavailable when this project began. It was not expected or required to be cost-effective as it is the subject of a research project.
Participants were made aware that the technology was experimental, costly and would not provide a traditional industry payback in the experimental stage. No promises of energy savings, operational savings or other benefits were made. The participants had to agree to participate without any guarantee of benefits.
Evaluate the costs and economic potential of IMDS - The economic evaluation included examining the costs to procure and install the IMDS. These costs were compared with operational savings. Statewide savings potential, assuming significant market penetration of the IMDS, was estimated.
Section 2. Conclusions and recommendations
Section 3. Discussion
Section 4. References
Section 5. Appendices