1、 In terna tio na1 Journa 1 of Heating, Ven tila ting, Air-conditioning and Refrigerating Research Editor Reinhard Radermacher, Ph.D., Professor and Director, Center for Environmental Energy Engineering, Department of Mechanical Engineering, University of Maryland, College Park, USA Associate Editors
2、 James E. Braun, Ph.D., P.E., Professor, Ray W. Hemck Laboratories, Alberto Cavallini, Ph.D., Professor, Dipartmento di Fisicia Tecnica, University of Padova, Italy Qingyan (Yan) Chen, Ph.D., Professor of Mechanical Engineering, School of Mechanical Engineering, Purdue University, West Lafayette, In
3、diana, USA Arthur L. Dexter, D.Phil., C.Eng., Professor of Engineering Science, Department of Engineering Science, University of Oxford, United Kingdom Srinivas Garimella, Ph.D., Associate Professor and Director, Advanced Thermal Systems Laboratory, Department of Mechanical Engineering, Iowa State U
4、niversity, Ames, Iowa, USA Leon R. Glicksman, Ph.D., Professor, Departments of Architecture and Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, USA Anthony M. Jacobi, Ph.D., Professor and Co-Director ACRC, Department of Mechanical and Industrial Engineering, University of I
5、llinois, Urbana-Champaign, USA Bjarne W. Olesen, Ph.D., Professor, International Centre for Indoor Environment and Energy Technical University of Denmark Nils Koppels All, Lyngby, Denmark Jeffrey D. Spitler, Ph.D., P.E., Professor, School of Mechanical and Aerospace Engineering, Oklahoma State Unive
6、rsity, Stillwater, Oklahoma, USA School of Mechanical Engineering, Purdue University, West Lafayette, Indiana, USA Editorial Assistant Lori Puente, CEEE Officeh4echanical Engineering, University of Maryland (30 1-405-5439) Policy Committee Special Publications Staff Daryl Boyce, Chair, Member ASHRAE
7、 P. Ole Fanger, Fellow/Life Member ASHRAE Curtis O. Pedersen, Fellow ASHRAE Reinhard Radermacher, Member ASHRAE Jeff Littleton, Associate Member ASHRAE W. Stephen Comstock, Associate Member ASHRAE Mildred Geshwiler, Editor Erin S. Howard, Associate Editor Christina Helms, Associate Editor Michshell
8、Phillips, Secretary W. Stephen Comstock Publisher Conditioning Engineers, inc., i791 Tullie Circle, Atlanta, Georgia 30329. All rights reserved. Periodicals postage paid at Atlanta, Georgia, and additional mailing offices. HVAC nor may any part of this book be reproduced, stored in a retrieval syste
9、m, . or transmitted in any form or by any means-electronic, photocopying, recording, or other-without permission in writing hm ASHRAE. Abstracts-Abstracted and indexed by ASHRAE Abstract Center; Ei (Engineering information, Inc.) Ei Compendex and Engineering Index; IS1 (Institute for Scientific Info
10、rmation) Web Science and Research Alert; and BSRIA (Building Services Research accepted July 20, 2004 Part II of this article will be published in Volume II, Number 2, April 2005. Poor similar process descriptions have been provided by Issermann VOLUME 1 1, NUMBER 1, JANUARY 2005 5 (1984) and Rossi
11、and Braun (1997). The first step is to monitor the physical system or device and detect any abnormal conditions (problems). This step is generally referred to as fault detec- tion. When an abnormal condition is detected, fault diagnosis is used to evaluate the fault and determine its causes. These t
12、wo steps constitute the FDD process. Following diagnosis, fault evaluation assesses the size and significance of the impact on system performance (in terms of energy use, cost, availability, or effects on other performance indicators). Based on the fault evaluation, a decision is then made on how to
13、 respond to the fault (e.g., by taking a corrective action or possibly even no action). Together these four steps enable condition-based mainte- nance, which is referred to as an automated FDD system in this paper. In most cases, detection of faults is relatively easier than diagnosing the cause of
14、the fault or evaluating the impacts aris- ing from the fault. FDD itself is frequently described as consisting of three key processes: fault detection, fault isolation, and fault identification. The first, fault detection, is the process of determining that some fault has occurred in the system. The
15、 second involves isolating the specific fault that occurred, including determining the kind of fault, the location of the fault, and the time of detec- tion. The third process, fault identification, includes determining the size and time-variant behavior of a fault. Together, fault isolation and fau
16、lt identification are commonly termedfault diagnosis. Review of the literature reveals a wide array of approaches used to detect and diag- nose faults. The sequencing of the detection and diagnosis varies. In some cases, the detection system runs continuously, while the diagnostic system is triggere
17、d only upon the detection of a fault. In other applications, the detection and diagnostic systems run in parallel, and in some instances, the detection and diagnostics are performed in a single step. Review of literature also indicates that most research and development in this field focuses on meth
18、ods for FDD itself rather than on decision processes and tools. Approaches to FDD range from methods based on physical and analytical models to those driven by performance data and using artificial intelli- gence or statistical techniques. Little has been published on the fault evaluation and decisi
19、on stages of the overall process in which FDD is used. Prognostics Prognostics focus on predicting the condition of an engineered system or equipment at times in the future. As with FDD, prognostics are used along with evaluations of impacts to make operation and maintenance decisions. Use of progno
20、stics enables transition from maintenance based on current conditions of engineered systems and equipment (condition-based mainte- nance) to predictive maintenance. Predictive maintenance is based on anticipated future condi- tions of the equipment, its remaining time before failure (or time before
21、reaching an unacceptable level of performance), the rate of degradation, and the nature of the failure if it were to occur. Prognosis generally uses some figure of ment (FOM) to quanti the “degree of fault.” Pre- dicting the future state and its acceptability requires three factors: (i) a measure of
22、 the systems current FOM, (2) a model of the progression of faults, and (3) the value of the FOM at which the system fails (i.e., a fault occurs) or reaches an unacceptably poor level of Performance (Greitzer and Pawlowski 2002). The current value of the FOM at any time can be determined from sensor
23、 measurements and FDD methods. The lowest value of the FOM that is acceptable is based on judgment or a mapping from FOM to failures. The model for progression of faults can be based on a theory of fault progression, on empirical data, or a combination of both. The key difference between FDD and pro
24、gnostics is the need to model fault progression. Fault progression is very application-specific, and much of the work in this field is documented in the literature of the var- ious application areas. This aspect of prognostics will not be reviewed further in this paper, which will instead focus on F
25、DD, which, in addition to providing a basis itself for condi- 6 HVAC Katipamula et al. 1999, 2003). Other DOE-funded activities include the work of Salsbury and Diamond (2001), who developed a simplified physical-model-based FDD for air-handling units, and the investigation by Sreedharan and Haves (
26、2001), who compared three chiller models for their ability to reproduce the observed performance of a centrifugal chiller for FDD application. Also in the mid to late 1990s, a newly formed ASHRAE Technical Committee on Smart Building Systems became active in FDD research, sponsoring several research
27、 projects (Norford et al. 2002; Cornstock et al. 1999,2001; Reddy and Andersen 2002; Reddy et al. 2003). Also around the same time, the California Energy Commission through the public interest energy research (PIER) program funded projects that built on some of the previ- ous FDD work in the HVAC Ka
28、tipamula et al. 2002). FDD meth- ods for application during initial building commissioning might differ from those applied later in the buildings lifetime. At start-up, no historical data are available, whereas later in the life- time, data from earlier operation can be used in FDD. Selection of met
29、hods must consider these differences; however, automated functional testing is likely to involve short-term data collec- tion, whether performed during initial building commissioning or during routine operation later in the buildings lifetime, and, therefore, the same methods can be used regardless
30、of when the functional tests are performed. Such a short time period is generally required for functional test- ing that it obviates the possibility that the system undergoing testing may change (e.g., perfor- mance deteriorate) during the test itself. Besides use in functional testing, FDD methods
31、could be used to test for the proper installation of equipment without requiring visual inspection. Use of labor might be minimized by only performing visual inspections to confirm installation prob- lems after they have been detected automatically. Methods for automatically detecting and diag- nosi
32、ng physical configurations have not yet been developed. During building operation, FDD tools could detect and diagnose performance degradation and faults, many of which go undetected for weeks or months in most commercial buildings. Many building performance problems are compensated with automatic c
33、ompensation by con- 8 HVAC Haves et al. 1996; Salsbury and Diamond 200 1 ; Norford et al. 2002; Castro 2002). VOLUME 11, NUMBER 1, JANUARY 2005 11 Strengths of Quantitative Models. Strengths of FDD based on quantitative models include: O Models are based on sound physical or engineering principles.
34、They provide the most accurate estimators of output when they are well formulated. o Detailed models based on first principles can model both normal and “faulty” operation; therefore, “faulty” operation can be easily distinguished from normal operation. o The transients in a dynamic system can only
35、be modeled with detailed physical models. Weakness of Quantitative Models. Weaknesses of FDD based on quantitative models include: o They can be complex and computationally intensive. The effort required to develop a model is significant. O These models generally require many inputs to describe the
36、system, some for which values may not be readily available. Extensive user input creates opportunities for poor judgment or input errors that can have Sig- nificant impacts on results. Suitability of Quantitative Models. FDD based on detailed physical models is unlikely to emerge as the method of ch
37、oice in the near future because of the weaknesses listed above, but simplified physical models will continue to make inroads into FDD applications. Qualitative Model-Based Methods Fault detection and diagnostics based on qualitative modeling techniques represent another broad category that is based
38、on a priori knowledge of the system. Unlike quantitative modeling techniques in which knowledge of the system is expressed in terms of quantitative mathematical relationships, qualitative models use qualitative relationships or knowledge bases to draw con- clusions regarding the state of a system an
39、d its components (e.g., whether operations are “faulty” or “normal”). Some qualitative models are obtained by deriving knowledge statements from process history data (such as for expert systems where human experience with a process is used to derive rules governing proper and faulty operation). Qual
40、itative model-based methods can be further subdivided into rule-based and qualitative physics-based models (see Figure 2). Both these qualitative modeling techniques employ causal knowledge of the process or system to diagnose faults. Other qualitative causal models include bond graphs (Ghiaus 1999)
41、 and case-based reasoning (Dexter and Pakenen 200 1). Qualitative models can also be based on abstraction hierarchies based on decomposition Venkatasubramanian 2003b), which is the ability to draw inferences about the behavior of the overall system solely from the laws governing the behavior of its
42、subsystems. More details on abstraction hierarchies and their application to problems in the process industry can be found in Venkatasubramanian (2003b). Commonly used measurement techniques provide quantitative output (e.g., temperatures, pressures, and humidities). Some qualitative methods accept
43、quantitative inputs directly, but others require qualitative inputs (e.g., linguistic statements) so that preprocessing is required before quantitative data are used. Fuzzy logic provides a mechanism for converting such quanti- tative data to qualitative information. A commonly used example is the c
44、onversion of air tem- perature measurements to qualitative categories of hot, warm, comfortable, cool, and cold. The boundaries between these qualitative categories are “fuzzy” and fuzzy logic provides a mecha- nism for conversion. The outputs of “fuzzification” can then be used as inputs to qualita
45、tive diagnostic processes. Rule-Based Systems The rule-based modeling techniques use a priori knowledge to derive a set of if-then-else niles and an inference mechanism that searches through the rule-space to draw conclusions. 12 HVAC Katipamula et al. 1999), although to our knowledge, none has been
46、 successfully commercialized or has achieved widespread acceptance. A detailed description of use of expert systems in process industries is provided by Venkatasubramanian (2003c), and Patel and Kam- rani (1996) provide a table of expert systems for diagnosis and maintenance. The strengths of expert
47、 systems are ease of development, transparent reasoning, ability to rea- son even under uncertainty, and the ability to provide explanations for the conclusions reached (Venkatasubramanian 2003b). The main weaknesses are that they are very specific to a system, can miserably fail beyond the boundari
48、es of the knowledge incorporated in them, and are diff- cult to update or change. A second category of methods under rule-based approaches uses rules derived from first prin- ciples (Brambley et al. 1998; Katipamula et al. 1999; House et al. 2001). As part of its mission in commercial buildings rese
49、arch and development, the U.S. Department of Energy (DOE) in collaboration with industry developed a tool that automates detection and diagnosis of problems associated with outdoor-air ventilation and economizer operation. The tool, known as the out- door-air economizer (OAE) diagnostician, monitors the performance of AHUs and detects prob- lems with outside-air control and economizer operation, using sensors that are commonly installed for control purposes (Brambley et al. 1998; Katipamula et al. 1999). The tool diagnoses the operating conditi
