ASHRAE IJHVAC 11-2：国际供暖、通风与空调研究期刊

版权申诉

102 浏览量更新于2024-07-20 收藏 10.5MB PDF 举报

"ASHRAE IJHVAC 11-2.pdf" 是一份由ASHRAE（美国采暖、制冷与空调工程师学会）发布的专业期刊，全称为《国际供暖、通风、空调与制冷研究》。这份期刊由Reinhard Radermacher博士担任编辑，他是美国马里兰大学机械工程系环境能源工程中心的教授兼主任。期刊还包括多名杰出的副编辑，如James E. Braun、Alberto Cavallini、Qingyan (Yan) Chen、Arthur L. Dexter和Srinivas Garimella等，他们在各自的领域有深厚的学术背景和实践经验。 ASHRAE作为业内权威机构，其出版物通常涉及以下关键知识点： 1. **供暖系统**：ASHRAE标准和指南覆盖了各种供暖系统的规划、设计、安装和维护，包括传统的燃油、天然气供暖设备，以及新兴的可再生能源供暖技术，如地源热泵和太阳能供暖。 2. **通风技术**：通风在建筑物的室内空气质量中起着至关重要的作用。ASHRAE的标准规定了合理的通风率、新风量和排风策略，以确保室内空气的健康和舒适性。 3. **空调系统**：涵盖了从传统的冷却系统到现代的变频多联机系统，以及能效优化和节能技术，如热回收和热泵应用。 4. **制冷技术**：包括食品储存、冷链物流、数据中心冷却等领域的制冷系统设计，以及环保制冷剂的研究和应用。 5. **环境能源工程**：强调建筑能源效率、能源审计、建筑外壳性能、照明系统优化等方面的研究，旨在减少建筑对环境的影响。 6. **热力学与流体力学**：这些基础科学在 HVAC 设计中不可或缺，涉及到传热、流体流动和空气动力学的计算。 7. **热系统实验室**：如副编辑Srinivas Garimella所负责的高级热系统实验室，可能涉及实验研究，开发新的热管理技术和设备，以提高能源效率和系统性能。 8. **标准与法规**：ASHRAE制定和更新行业标准，影响建筑规范和政策，指导设计和施工实践，确保合规性。 9. **国际合作**：期刊中的研究涵盖全球范围，反映不同国家和地区对HVAC&R的独特需求和解决方案，推动国际间的学术交流和技术合作。这份期刊通过发表科研文章、案例研究和技术报告，促进 HVAC&R 领域的创新和进步，为工程师、研究人员、建筑师和政策制定者提供宝贵的参考资料。其内容可能包括理论分析、实验数据、模拟计算、现场测试结果以及对现有技术的评估和改进建议。

176

HVAC&R

RESEARCH

Peitsman and Bakker (1996) used two types of black-box models (artificial neural networks

[ANNs] and auto regressive with exogenous inputs [ARX’]) to detect faults in the system and at

the component level of a reciprocating chiller system. The inputs to the system models included

condenser supply water temperature, evaporator supply glycol temperature, instantaneous power

of the compressor, and flow rate of cooling water entering the condenser (for the ANN only).

The choice of the inputs was limited to those that are commonly available in the field. Using

these inputs with both the ANN and

ARX

models,

outputs were estimated. For the

ANN

models, inputs from the current and the previous time step and outputs from

two

previous time

steps were used.

Peitsman and Bakker (1996) compared diagnostic capabilities of two types of models-a

multiple inpuîíoutput

ARX

model and ANN models. They used a two-level approach in which

system-level models were used to detect “faulty” operation and component-level models were

used to diagnose the cause of the fault. They developed 14 system-level models and 16 compo-

nent-level models to detect and diagnose faults

a chiller; however, only one example (air in

the system) is described in their paper. ANN models appeared to have a slightly better perfor-

mance than the

ARX

models in detecting faults at both the system and the component levels.

The authors also note that it is critical to find a global minimum when using

ANN

models. If an

incorrect initial state is chosen, it may lead to a local minimum rather than the global minimum.

Bailey (1998) also used an ANN model to detect and diagnose faults in an air-cooled chiller

with a screw compressor. The detection and diagnosis were carried out in a single step. The

faults evaluated included refrigerant under- and overcharge, oil under- and overcharge, con-

denser fan loss (total failure), and condenser fouling. The measured data included superheat for

heat exchanger circuits

and

subcooling from circuits

and

power consumption, suction

pressure for circuits

and

discharge pressures for circuits 1 and

chilled water inlet and out-

let temperatures

from

the evaporator, and chiller capacity. Each heat exchanger circuit has its

own compressor. The

ANN

model was applied to normal and “faulty” test data collected

from

70-ton laboratory air-cooled chiller with screw compressor.

Sreedharan and Haves (2001) compared three chiller models for their ability to reproduce the

observed performance of a centrifugal chiller. Although the evaluation was meant to find the

most suitable model for chiller FDD, no FDD system was proposed or developed. Two models

were based on first principles (from Gordon and Ng [I9951 and a modified ASHRAE Primay

Toolkit

from Bourdouxhe et al. [1997]) and the third was an empirical model. While each model

has some distinct advantages and disadvantages, they concluded that the accuracies of all three

models were similar. Hydeman et al. (2002) reported that the three models compared by

Sreedharan and Haves (2001) were not accurate in predicting the power consumption of chillers

with variable condenser water flow and centrifugal chillers operating with variable-speed drives

at low loads. They reformulated the Gordon and Ng model and found that it performed better

than the three models described above.

Castro (2002) used a physical model developed

Rossi (1995) along with a k-nearest neigh-

bor classifier to detect faults and a rule base to diagnose five different faults (condenser and

evaporator fouling, liquid line restriction, and refrigerant under- and overcharge) in a reciprocat-

ing chiller. The

FDD implementation detected and diagnosed condenser fouling, refrigerant

undercharge at faults level of

20%

or greater, and evaporator fouling and liquid line restriction at

fault levels of

30%

or greater.

‘Refer to

Box

and

Jenkins

976)

for

details

ARX

type

models.

Provided by IHS under license with ASHRAE

Licensee=Naval Aviation Depot/5918326100

Not for Resale, 01/19/2008 02:28:15 MST

No reproduction or networking permitted without license from IHS

--`,`,`,``,,`,`,,,`,``,,,`,`,,``-`-`,,`,,`,`,,`---

VOLUME

11,

NUMBER

APRIL

2005

111

Air-Handling

Units

There are several studies relating to FDD methods for air-handling units (both the airside and

the waterside); some of these are summarized in this section (Norford and Little

1993;

Glass et

al. 1995; Yoshida et al. 1996; Haves et al. 1996; Lee et al. 1996a, 1996b; Lee et al. 1997; Peits-

man and Soethout 1997; Brmbley et al. 1998; Katipamula et al. 1999; House et al. 1999;

Ngo

and Dexter 1999; Yoshida and Kumar 1999; Seem et al. 1999; Karki and Karjalainen 1999;

Morisot and Marchio 1999; House et al. 2001; Dexter and

Ngo

2001; Kumar et al. 2001; Sals-

bury and Diamond 2001; Carling 2002; Norford et al. 2002; Wang and Chen 2002; Pakanen and

Sundquist 2

003).

Norford and Little

(1993)

classi@ faults in ventilating systems, consisting of fans, ducts,

dampers, heat exchangers, and controls. They then review

two

forms of steady-state parametric

models for the electric power used by supply fans and propose a third, that of correlating power

with a variable-speed drive control signal. The models are compared based on prediction accu-

racy, sensor requirements, and their ability to detect faults.

Using the three proposed models, four different types of faults associated with fan systems are

detected:

(i)

failure to maintain supply air temperature, (2) failure to maintain supply air pres-

sure setpoint,

(3)

increased pressure drop, and

(4)

malfunction of fan motor coupling to fan and

fan controls. Although the paper by Norford and Little (1993) lacks details on how the faults

were evaluated, error analysis and associated model fits were discussed. The results indicate that

ail three models were able to identi@ at least three of the four faults. The diagnosis of the faults

inferred afîer the fault

detected.

Glass et al.

995) use a qualitative model-based approach to detect faults in

air-handling

unit. The method uses outdoor, return, and supply air temperatures and control signals for the

cooling coil, heating coil, and the damper system. Although Glass et al.

995) mention that the

diagnosis is inferred from the fault conditions, no clear explanation or examples are provided.

Detection starts by analyzing the measured variables to veri@ whether steady-state conditions

exist. Then, the controller values are converted to qualitative signal data and, using a model for

expected values and measured temperature data, qualitative signals are estimated. Faults are

detected based on discrepancies between measured qualitative controller outputs and corre-

sponding model predictions based on the temperature measurements. Examples of qualitative

states for the damper signal include ?maximum position,? ?minimum position,? ?closed,? and

?in between.? When the quantitative value of the damper signal approaches the maximum value,

the corresponding qualitative value of ?maximum? is assigned to the measured controller out-

put. The results of testing the method on a laboratory

AHU

were mixed because the method

requires steady-state conditions to be achieved before fault detection is undertaken. Fault detec-

tion sensitivity and ability to deal with false alarms are not discussed.

Yoshida et al. (1996) use ARX and the extended Kalman filter approach to detect abrupt

faults with simulated test data for an

AHU.

Although the fault diagnosis approach is clearly

described, the authors note that diagnosis is not feasible with the ARX method but that the Kal-

man filter approach could be used for diagnosis. Fault detection sensitivity and ability to deal

with false alarms are not discussed.

Haves

al. (1996) use a combination of two models to detect coil fouling and valve leakage

in the cooling coil of an

AHU.

The methodology was tested with data produced by the HVAC-

SIM+

simulation tool (Clark 1985). A radial bias function

(RBF)

models the local behavior of

the AHU and is updated using a recursive gradient-based estimator. The data generated by exer-

cising the RJ3F over the operating range of the system are used in the estimation of the parame-

ters for the physical model

(UA

and percent leakage) using a direct search method. Detection is

accomplished by comparing estimated parameters to fault-free parameters.

Provided by IHS under license with ASHRAE

Licensee=Naval Aviation Depot/5918326100

Not for Resale, 01/19/2008 02:28:15 MST

No reproduction or networking permitted without license from IHS

--`,`,`,``,,`,`,,,`,``,,,`,`,,``-`-`,,`,,`,`,,`---

178

HVAC&R

RESEARCH

Lee et al. (1996a) used two methods to detect eight different faults (mostly abrupt faults) in a

laboratory test

AHLJ.

The first method uses discrepancies between measured and expected vari-

ables (residuals) to detect the presence of a fault. The expected values are estimated at nominal

operating conditions. The second method compares parameters estimated using autoregressive

moving average with exogenous input

(ARMX)

and

ARX

models with the normal (or expected)

parameters to detect faults. The faults evaluated included complete failure of the supply and

return fans, complete failure of the chilled-water circulation pump, stuck cooling-coil valve,

complete failure of temperature sensors, complete failure of the static pressure sensor, and fail-

ure of the supply and return air fan flow stations. Because each of the eight faults has a unique

signature, no separate diagnosis is necessary.

Lee et al. (1996b) used an

ANN

to detect the same faults described previously (Lee et al.

1996a). The

ANN

was trained using the normal data and data that represented each of the eight

faults. Inputs to the

ANN

were values for seven normalized residuals, and the outputs were nine

values that constitute patterns that represent the normal mode and the eight fault modes. Instead

of generating the training data with faults, idealized training patterns were specified by consider-

ing the dominant symptoms of each fault. For example, supply fan failure implies that the sup-

ply fan speed is zero, the supply air pressure is zero, the supply fan control signal is maximum,

and the difference between the flow rates in the supply and return ducts is zero. Using similar

reasoning, a pattern

dominant training residuals was generated for each fault (see Table

4).

dominant symptom residual is assigned a value of +1 if the residual is positive and -1 if the

residual is negative; all other residuals are assigned a value of

The

ANN

was trained using the

pattern shown in Table

Normalized residuals were calculated for faults that were artificially

generated in the laboratory

AHU.

The normalized residuals vector at each time step was then

used with the trained

ANN

to identify the fault. Although the

ANN

was successful in detecting

the faults from laboratory data, it

not clear how successful this method would be in general

because the faults generated in the laboratory setting were severe and without noise.

Lee et al. (1997) extended the previous work described in Lee et al. (1996b). In the 1997 anal-

ysis, Lee et al. (1997) used two

ANN

models to detect and diagnose faults. The

AHü

is decom-

posed into various subsystems such as the pressure control subsystem, the flow-control

subsystem, the cooling-coil subsystem, and the mixing-damper subsystem. The first

ANN

model is trained to identify the subsystem in which a fault occurs, while the second

ANN

model

is trained to diagnose the specific cause of a fault at the subsystem level.

approach similar to

the one used in Lee et al. 1996b is used to train both

ANN

models. Lee et al. (1997) note that this

two-stage approach simplifies generalization by replacing a single

ANN

that encompasses all

considered faults with a number of less complex

ANNs,

each one dealing with a subset of the

residuals and symptoms. Although 11 faults are identified for detection and diagnosis, fault

detection and diagnosis are presented for only one fault in the paper.

Peitsman and Soethout

997) used several different

ARX

models to predict the performance

of an

AHü

and compared the predictions to measured values to detect faults. The training data

for the

ARX

models were generated using

HVACSIM+.

The

AHü

is modeled at two levels. The

first level is the system level, where the complete AHü is modeled with one

ARX

model. The

second level

is the component level, where the

AHU

is subdivided into several subsystems such

as the return fan, the mixing box, and the cooling coil. Each component is modeled with a sepa-

rate

ARX

model. The first level

ARX

model is used to detect a problem and the second level

models are used to diagnose the problem. Most abrupt faults were correctly identified and diag-

nosed, while slowly evolving faults were not detected. There is a potential for a conflict between

the two levels with this approach; for example, the top-level

ARX

model could detect a fault

with the

AHU,

while the second-level

ARX

models do not indicate any faults. Furthermore,

there is a potential for multiple diagnoses at the second level. Peitsman and Soethout (1997)

Provided by IHS under license with ASHRAE

Licensee=Naval Aviation Depot/5918326100

Not for Resale, 01/19/2008 02:28:15 MST

No reproduction or networking permitted without license from IHS

--`,`,`,``,,`,`,,,`,``,,,`,`,,``-`-`,,`,,`,`,,`---

180

HVAC&R RESEARCH

Ngo and Dexter (1999) developed a semi-qualitative analysis of measured data using generic

fuzzy reference models to diagnose faults with the cooling coil of an AHU. The method uses

sets of training data with and without faults to develop generic fuzzy reference models for diag-

nosing faults in a cooling coil. The faults include leaky valve, waterside fouling, valve stuck

closed, valve stuck midway, and valve stuck open. The fuzzy reference models describe in qual-

itative terms the steady-state behavior of a particular class

equipment with

faults present

and when each of the faults has occurred. Measured data are used to identie a partial fuzzy

model that describes the steady-state behavior of the equipment at a particular operating point.

The partial fuzzy model is then compared

each

the reference models using a fuzzy match-

ing scheme to determine the degree of similarity between the partial model and the reference

models.

Ngo

and Dexter (1999) provide a detailed description of fault detection sensitivity and

false alarm rates.

Yoshida and Kumar (1 999) evaluated

two

model-based methods to identi@ abrupt (sudden)

faults in an AHü. They report that both

ARX

and adaptive forgetting through multiple models

(AFMM) seem promising for use in on-line fault detection of

AHUs.

They report that ARX

models require only

minimal knowledge of the system, and the potential limitation of the tech-

nique is that it requires long periods to stabilize its parameters. On the other hand, Yoshida and

Kumar (1999) report that the

AFMM

method requires long moving averages to suppress false

alarms. When this is done, faults of lesser magnitude cannot be easily detected. Implementation

details are not provided, and only one example of fault detection

provided.

Morisot and Marchio (1999) use an ANN-based approach to detect degradation of perfor-

mance of a cooling coil in an

AHU. The ANN network includes an input layer (six inputs), a

hidden layer (four nodes), and an output layer

(two

outputs). The inputs include entering air

temperature and humidity, entering and leaving water temperatures, fan-control signal, and

cooling-coil-valve-control signal. The outputs are the leaving air temperature and humidity. The

authors highlight the difficulties

using ANNs with real measured data, which include a need

for an exhaustive training data set and the inability of the ANNs to extrapolate values outside the

range of the training data. The proposed alternative is to use a simulation model to generate the

training data for the

ANN.

Using this alternative approach, the authors test the ability of the

ANN

to detect two faults (air-side fouling and a sensor fault).

Dexter and Ngo

(2001) outline a multi-step fuzzy model-based

FDD

approach to detecting

and diagnosing faults with

AHUS.

This approach involves classiQing measured data with fuzzy

rules and comparing them to a set of fuzzy reference models for normal and faulty operations.

The fuzzy reference models for a specific system are developed from data that are generated

from simulations. Each rule is assigned a rule-confidence in the range from zero to one, where

zero indicates no confidence and one indicates complete confidence in the rule correctly describ-

ing the behavior. Rule-confidence values are estimated from the data. The authors state that this

method prevents false alarms because it accounts for major sources of uncertainty. The

multi-step approach

shown to be capable of detecting and isolating faults in a cooling coil

(leaking valves and fouling).

Kumar et al.

(2001)

propose a method based on an auto regressive exogenous model and a

recursive parameter estimation algorithm

to detect faults with AHUs. They conclude that

changes in parameter estimates from real data cannot be directly used to detect faults; instead a

statistical analysis of the frequency response of the model parameters is needed to detect faults.

Salsbury and Diamond

(2001)

develop a simplified physical model-based approach to both

control and detect faults in

AHUs.

Results from a field test

a single AHü demonstrate the

fault detection capabilities but also highlight some of the practical implementation difficulties

including selection

model parameters, reliability of sensor signals, and difficulty in establish-

ing a baseline of “correct” operation of the

AHU.

Provided by IHS under license with ASHRAE

Licensee=Naval Aviation Depot/5918326100

Not for Resale, 01/19/2008 02:28:15 MST

No reproduction or networking permitted without license from IHS

--`,`,`,``,,`,`,,,`,``,,,`,`,,``-`-`,,`,,`,`,,`---

剩余178页未读，继续阅读

制冷技术咨询与服务

粉丝: 4069
资源: 2万+

ASHRAE IJHVAC 11-2：国际供暖、通风与空调研究期刊

ASHRAE IJHVAC 11-4: 美国采暖与制冷工程研究

ASHRAE IJHVAC 14-1：HVAC&R Research期刊文章

ASHRAE IJHVAC 8-4: 美国采暖与制冷工程研究期刊

ASHRAE IJHVAC 11-1.pdf

ASHRAE IJHVAC 11-4.pdf

ASHRAE IJHVAC 11-3.pdf

ASHRAE IJHVAC 7-2.pdf

ASHRAE IJHVAC 2-4.pdf

ASHRAE IJHVAC 9-2.pdf

ASHRAE IJHVAC 13-2.pdf

最新资源