Transcript
Fault Detection and Diagnosis Methods in HVAC Building Autom
ation S S
y
stem
toj an P
s using I ndustr ial S
er š in,
Bor is T
ov
olutions
or nik
Laboratory for process automation U
niv
ersity of M
aribor, S
F
acul ty of E
metanov
a 1
7
l ectrical ,
2
0
0
0
E
M
ng
ineering
aribor,
S
l ov
and
C
omputer S
cience
enia
Abstract Fault Detection and Diagnosis systems offers enhanced availability and reduced risk of safety haz ards w
hen comp
line FDD an ap
onent failure and other unex
p rop riate method an O
data for FDD in industrial ap bottom- up ap p
ap
ected events occur in a controlled p
lications and solution,
p roaches to diagnostic reasoning of w
p roaches w rop
p
p
n- line data are req
ere suggested.
osed and S
olution 2
S
olution
1
using
uired.
I t is q
using O
P
C
is suggested.
hole system w
q ualitative
using reference subsystem w
ere p
data
rop
lant.
For O
uite difficult to get O
ere rep
T
op
- dow
resented and tw
from “ similar”
n-
n- line n and o new
subsystems
w
as
osed.
1. Introduction
Modern plants are large scale, highly complex, and operate with a large number of variables. Processes are becoming more heavily instrumented, resulting in large q uantities of data becoming available for use in detecting and diagnosing faults. T req uirements, stability D
and
of
which
the
reliability.
T
here
etection and I solation ( F
he design of such systems req uires trade- off between several competing
most
D
I )
important
are
many
are:
accuracy,
approaches
available today. T
to
F
resolution,
ault
D
robustness,
etection
and
Pa a t eg
ra q
u
t i on a
of
l i t a
t i v
q
u
a
e k
n
n
t i t a
ow
t i v
l ed
e a g
n
d
S
y m
( R g m
e
p
asur e
m
e
n
p
t om
esi d en
u
st a -
n
t t ern
recog
roa
es
ch
t i st i ca
eu
ra
l
l
n
n
( F
D
diagnostic
D
)
or
F
ault
m
et h
od
et w
ork
s
S
l s)
t i on
y m
m
p
to
m
p
t om
esi d
ev sy
.
s
( R
ts
igure 1
i t i on
s
a
era
p
-
e
Process
sensitivity,
iagnosis
heir general structure consists of the three maj or stages:
symptom generation, symptom evaluation and fault diagnosis as is shown on F
I n
D
a
l u
a
u
a
s
R
ea
l s)
( B
t i on
s
F e
v
sy
aluate m
p
to
m
u
son
i n
z
ool ea z
y ,
T
g n
B
M
)
d s
faults Q m
u
a
n
t i v
el - b
od
el - b
a
sed
ob
serv
m
od
el - b
a
sed
p
ri t y
a
l - b
a
sed
m
p
a
a
ra
p
od
n
sed
e a
m
si g
a
t i t a
et h
m
p
roa
ch
es:
et er est i m
t i on
ru
er sp
od
a
Q
a
ce a
p
p
roa
T
he F
D
D
D
D
sa
igure 1
:
G
u
a
l i t a
l
m cy
rel a t i v
t i v
e a et h
ch
p
sh u
roa
od
eck
t i on
e si m
p
l a
eneral structure of F
es:
s i n
i p
ch
g
s
Process R
D
n- line data are req uired.
ery
D
one of the commercial
and one of the special data acq uisition boards for measurements. S
an appropriate method an O
ecov
t i on
method is in laboratory environment relative easy to implement using
mathematical tools ( i.e. Matlab) line F
ca
u
l i t a
sed
si st en
q
F
a
a
con ch
s
l e- b
u
o, for O
n-
2. Industrial solutions It is quite difficult to get On-line data for FDD in industrial applications, because there safety is on first place. A
T
hat m
plant m
eans also use of industrial solutions, none hom
ay easily hav
has to be v alidated and docum and ev
ery S
C
A
DA
ented, that m
ade products, standards and docum L
C
s and S
eans the connection betw
has to be v alidated, docum
proj ect database. and redocum
e-m
e thousands of I/ O points connected to P
ented, and tagged.
If an additional data acquisition system
C
A
een ev
E
v
DA
system
av oiding this procedure is »
nothing to change«
connection P
L
C
to the S
industrial system
C
A
s ( Figure 2
DA
system
is OP
C
, w
is added that m
en data for S
c
e
s
C
A
DA
s
s
e
r v
e
M
odern
system m
V
A
B
buildings
s
( E
M
C
S
S
OPC
uilding
are
) ,
A
being
often
utom
G
CA c
D
l i e
A n
a FDD system
ow
does
instrum
V
A
C
w
y
ith
industrial
this
ented w
ent.
equipm
happen?
ith the m
A
L
T
second ex
rend data from
effectiv ov
e w
erlook
um
of
are
num
sensor
a
num
T
N
prov
iv
M
em
U
porarily allev
C
2
M
S
M
C
S
lack
t l a
b
CA
r f a
c
M S
D
A
a i m
t l a u
b
l i n
N
ber
of
ex
S
hav
e
C
energy
seem
m
ingly
anagem lim
ance ex
ent
itless
onetheless, building heating, v
planations.
ation
is
a
significant
pectations env
onitor system
ay buildings are m
First, plem
H
V
A
C
for
entilating and air-
equipm
ent
is
A
third ex
erw
to
assessing
helm
the
operation
how
ev
the
ent status.
an, and this is not a cost-
planation is that building operators m plem
er, there is little argum
ay
ented.
anual ov
A
errides
ay lead to unintended and undetected operating problem
ist;
isory
of
s building operators because
ay not fully understand the control strategies im
, but m
typically
ent basic local-loop and superv
barrier
hen analyz ed by a hum
operation.
ent that there is v ast room
s in for
onitored.
hat can be done to im
prov
e the perform
the tools necessary to
)
detect that problems (often referred to as faults) exist, and
)
assist building
operators in diag
control
isioned at design, so
of understanding of sophisticated control strategies leads to m
planations ex
and
capabilities
ation into a clear and coherent picture of equipm
are useful, but only w
iate a problem
k
e
sophisticated C
ollegiate Dictionary defines diagnosis as it is used in this contex
oday' s E 1
S
ndoubtedly other ex
ent in the w
his is difficult
s.
s of a failure because they m
en this set of barriers, w
ew
C
planation is that lack
ay tem
the future. im
today’ s E
ptom
t e
ber of sensors sufficient to im inform
a
planation is that the data that is collected ov
ay to continuously m
sym
related ex that m
G
E
ent routinely fails to satisfy perform
here
inim
ack
T
s
increasingly
there is little effort to consolidate the inform T
ste m
solutions.
could be useful in such a system
control strategies. equipm
)
he possibility for
t
etting On-line data using OP
ation S
designed
using
:
onitoring and controlling the conditions in buildings.
conditioning ( H
H
C
eans going back
T
OPC r
Figure 2
in H
C
hich can be also used for FDD-On-line connection in
i n
D
L
industrial solutions for
OPC
D
ery P
) .
M
. F
entation.
ery data point
es part of the
needs.
One of the new
OPC Pr o
v
ery single I/ O point, ev
for FDD system
using already tak
E
ery line of code becom
enting things, v alidate connections etc and first of all to stop the process.
because the source code of the protocol is usually not av ailable.
4
s.
nosing
the problems that arise.
ance of H t as follow
V
A s:
C
equipm
ent?
W
ebster' s
Having the capability to quickly diagnose operational problems in HVAC equipment means that equipment w
ill operate as intended a higher percentage of
HVAC equipment are listed below
F
improved occupant comf ort and health
improved energy ef f iciency
longer equipment lif e
reduced maintenance costs
reduced unscheduled equipment dow
igure
3
provides
a
buildings and show top- dow
n
representation
s tw
approach,
w
uses
of
ome of
the benef its of
properly operating
n time
the
perf ormance
hierarchical
structure
of
HVAC
measures
f rom
er- level causes of
higher
levels
of
T
systems
and
subsystems
he f irst approach,
the
in
termed the
building/ system/ controller
degradations to those higher level measures.
F
or
hole building energy use is one high level measure that provides usef ul inf ormation about the
perf ormance of
a building.
I f
building energy use ex
signif icant,
top- dow
probable ex
planations f or the ex
T
S
o approaches commonly used f or diagnostic reasoning.
hierarchy to reason about possible low instance,
the total run time.
:
n reasoning w
he second approach,
ceeds its ex
ould be used to navigate dow
pected value by an amount considered to be n through the hierarchy and isolate the most
cess energy use.
termed the bottom- up approach,
uses perf ormance measures at low
er levels of
the
hierarchy to isolate problems and then propagates that problem up through the hierarchy to determine its impact on building perf ormance. problem w
I f
the impact w
ould be given a high priority.
nothing at this time. isolate f aults at low
P
I f
ere considered to be large or potentially large,
the impact is considered to be small,
correcting the
the decision may be to do
erf ormance measures at intermediate levels can be used in a top- dow
er levels,
and also in a bottom- up approach to determine the impact of
igure 3
op- dow
n approach to the f ault at the
building level.
F
I n general,
there is a f ew
:
T
data f or usef ully model –
getting additional data f rom rest of I n f igure 4 of
is show
n and bottom- up approaches to diagnostic reasoning.
the system.
T
observed
subsystem
and
qualitative
comparative subsystems is used, equipped w
D
D
in HVAC systems.
F
D
D
can be applied if
n a possible solution to acquiring additional qualitative data f rom other subsystems.
the data driven methods can be used f or observing.
f rom
based F
hose additional data can be either qualitative or quantitative.
ith additional elements.
w
data
I n reasoning there is a combination of
f rom
other
systems.
I n
here a ref erence subsystem is built.
f igure
5
a
O
ne
quantitative data
combination
of
tw
o
A ref erence subsystem has to be
Q
subsystem 1
u
al it at iv
( d
at a d
E
W C
M U C
P
L
subsystem 2
P
N
e d
r iv
at a
en M
f a
et h
od
s) :
b
ser v
S
U
Q
r o
ba
p
r o
ba
f a
S
u p
p
ul t:
f
1
bi l i ty f o
r
f 1
r
f 2
r
f 3
M
-
L
p
a
S
r ti a -
n
l
o
l ea
n
st sq
i ter a
ti v
ua
r es
e P
L
Reasoning:
S
B
ant it at iv
ar am
e d
et er
u
z
st at e est im anal y
t ic
z
B
f
2
y
M
f a
at a:
est im
ul t:
bi l i ty f o
ool ean
T
ed
ba
A
F
O
r o
A
I P
subsystem 3
p
ul t:
f
3
bi l i ty f o
at ion
at ion
al
r ed
u
nd
anc
y
f a
subsystem n p
r o
ba
ul t:
f
m
bi l i ty f o
r
f
m
Figure 4: FDD can be applied assuring additional (qualitative) data from similar subsystems
A subsystem 1
dditio
f o
r
nal
eq
r ef er enc
s
ub
s
s
ens
o
p
r o
es
c
y
s
tem
uip
m
ent
f a
e
r s
,
ac
s
o
r s
tuato
r s
s
eas o
o
ning
o
ar am s
tate es anal y
eter
tic
tim al
ba
p
r o
ba
:
z
B
1 r
f 1
es
tim
atio
atio
ul t:
f
r
f 2
r
f 3
2
bi l i ty f o
y
M
f a
Quantitative data: p
r o
f
l ean
uz T
er ved
p
ul t:
bi l i ty f o
f a
R
F
b
ba
,
B
O
r o
:
subsystem 2
subsystem 3
p
ul t:
f
3
bi l i ty f o
n
n
r edundanc
y
f a
subsystem n p
r o
ba
ul t:
f
m
bi l i ty f o
r
f
m
Figure 3: FDD can be applied assuring additional (qualitative) data from reference subsystem
5. Conclusion
T
h e difference betw
een real H
typically instrumented w
ith
th
V
A
C
applications and laboratory test system is th at H
V
A
C
equipment is
e minimum number of sensors sufficient to implement basic local- loop and
supervisory control strategies.
A
single FDD system can be used for many pieces of equipment,
and th H T
V
A
ereby allow C
industry.
op- dow
approach S
T
h
e use of more ex
pensive sensors.
e integration of FDD meth
n and bottom- up approach es w
olution 2
FDD in H
s th
ere suggested.
S
A
C
h ich
improves th
ods into individual controllers w h
ole system w
using qualitative data from “ similar”
e cost- to- benefit ratio e FDD for th
e
ould appear to be nex
t step.
ere represented and tw
o new
subsystems w
as proposed and
as proposed.
subsystem is possible if getting additional data.
same as in industrial applications.
For acquiring data w
no
data
additional
w
h is is a logical initial deployment of th
es to diagnostic reasoning of w
olution 1
using reference subsystem w
V
T
T
h e problem h ow
to get O
n- line data is th e
as suggested an industrial solution using O
acquisition
board
is
P
C
w
h
ere
needed.
References
[1] Sourabh Dash and Venkat Venkatasubramanian: Challenges in the Industrial applications of fault diagnostic sy stems, [2
] J ames E
R
esearch,
[3 '9 [4 B [5
. v
B
ol.
] Isermann R 4
,
E
spoo,
] J ohn M
2 .
H
,
5
,
v an R
Verlag L
aboratory
no.
for Intelligent P
rocess Sy stems,
P
urdue U
2
utomated fault detection and diagnostics for the H ,
19 9
5
9
7
- 6
ouse and G
2
eorge E
esearch to P
0
0
0
ersity , C&
R
U
SA
,
19
industry ,
9
9
H
VA
C&
R
IF
A
C Sy mposium SA
F
E
P
R
O
CE
SS
12
ussel et al: Data driv
ondon,
niv VA
9
. : Integration of fault detection and diagnosis methods, pp.
uildings: from R ] E
L
raun: A
.
K
elly
ractice,
P
: A
n ov
acific E
erv
iew
nergy
of building diagniostics, Center,
San F
rancisco,
19
9
Diagnostics for Commercial 9
en methods for fault detection and diagnosis in chemical processes,
Springer-
