FLUID-FILLED ELECTRICAL EQUIPMENT INTELLIGENT
ANALYSIS SYSTEM AM® METHOD
BACKGROUND OF THE INVENTION
The invention relates generally to fluid-filled electrical equipment. More
particularly, the invention relates to an apparatus and method for determining
operating status, diagnostic status, and prognostics of electrical equipment in real
time and to electrical equipment incorporating the apparatus.
Electrical equipment, particularly medium-voltage or high-voltage
electrical equipment, requires a high degree of electrical and thermal insulation
between components thereof. Accordingly, it is well known to encapsulate
components of electrical equipment, such as coils of a transformer, in a
containment vessel and to fill the containment vessel with a fluid. The fluid
facilitates dissipation of heat generated by the components and can be circulated
through a heat exchanger to efficiently lower the operating temperature of the
components. The fluid also serves as electrical insulation between components
or to supplement other forms of insulation disposed around the components, such
as cellulose paper or other insulating materials Any fluid having the desired
electrical and thermal properties can be used. Typically, electrical equipment
is filled with an oil, such as castor oil, mmeral oil, or vegitable oil, or a
synthetic "oil", such as chlorinated diphenyl, silicone, or sulfur hexaflouride.
Often electrical equipment is used in a mission critical environment in
which failure can be very expensive, or eves catastrophic, because of a loss of
electric power to critical systems. Also, failure of electrical equipment
ordinarily results in a great deal of damage to the equipment itself and
surrounding equipment thus requiring replacement of expensive equipment.
Further, such failure can cause injury to personnel due to electric shock, fire,
or explosion. Therefore, it is desirable to monitor the status of electrical
equipment to predict potential failure of the equipment through detection of
incipient faults and to take remedial action mrough repair, replacement, or
adjustment of operating conditions of the equipment. However, the performance
and behavior of fluid-filled electrical equipment inherently degrades over time.
Faults and incipient faults should be distinguished from normal and acceptable
degradation.
A known method of monitoring the status of fluid-filled electrical
equipment is to monitor various parameters of the fluid. For example, the
temperature of the fluid and the total combustible gas (TCG) in the fluid is
known to be indicative of the operating state of fluid-filled electrical equipment.
Therefore, monitoring these parameters of the fluid can provide an indication
of any incipient faults in the equipment. For example, it has been found that
carbon monoxide and carbon dioxide increase in concentration with thermal
aging and degradation of cellulosic insulation in electrical equipment. Hydrogen
and various hydrocarbons (and derivatives thereof such as acetylene and
ethylene) increase in concentration due to hot spots caused by circulating
currents and dielectric breakdown such as corona and arcing. Concentrations
of oxygen and nitrogen indicate the quality of the gas presring system
employed in large equipment, such as transformers. Accordingly Adisolved gas
analysise (DGA) has become a well accepted method of discerining incipient
faults in fluid-filled electric equipment.
In conventional DGA methods, an amount of fluid is removed from the
containment vessel of the equipment mrough a dram valve. The removed fuild
is then subjected to testing for dissolved gas in a lab or by equipment in the
field. This method of testing is referred to herein as Aoff-line DGA. Since the
gases are generated by various known faults, such as degradation of insulation
material or other portions of electric components in the equipment, turn-to-turn
shorts in coils, overloading, loose connections, or the like, various diagnostic
theories have been developed for correlating the quantities of various gases in
fluid with particular faults in electrical equipment in which me fluid is contained.
However, since conventional methods of off-line DGA require removal
of fluid from the electric equipment, these methods do not, 1) yield localized
position information relating to any fault in the equipment, 2) account for spatial
variations of gases in the equipment, and 3) provide real time data relating to
faults. If analysis is conducted off site, results may not be obtained for several
hours. Incipient faults may develop into failure of the equipment over such a
period of time.
MICROMONITORS, INCJ and SYPROTECJ have each developed a
gas sensor which resides in the drain valve, or other single locations, of a
transformer and overcomes some of the limitations of off-line DGA. However,
location data relating to a fault is not discernable with such a device because it
is located in one predefined position and does not provide any indication of the
position of the source of the gas, i.e., the fault. U.S. patent 4,654,806 discloses
an apparatus for monitoring transformers including sensors tor delecting oil
temperature, gas in oil, and cabinet temperature. Raw data from the sensors is
collected by a microcoputered and periodically downloaded to remote host
computer. The microcompvter can compare varkws inea*ured parameters with
predetermined thresholds and can activate alarms or other warnings if the
thresholds are exceeded. The remote host computer can control a cooling
system of die transformer based on the parameters oat are periodically
downloaded to die remote host computer. Similarly, U.S. patent 3,855,303
discloses a remotely monitored transformer in which data from sensors is
downloaded to a remote computer and compared to predetermined thresholds.
If the thresholds are exceeded, die transformer can be de-energized. U.S.
patent 4,654,806 discloses that the individual thresholds can be varied based on
other thresholds. However, the devices disclosed in U.S. patent 4,654,806 and
U.S. patent 3,855,503 fall short of providing comprehensive and cohesive
diagnostics in real time because they do not account for the complex
relationships between the various operating parameters of fluid-filled electrical
equipment or the normal degradation over time of fluid-filled electrical
equipment. The article entided "Monitoring die Health of Power Transformers"
discusses research at Massachusetts institute of Technology relating to model
based diagnostic systems.
Known processes and apparatus do not provide accurate, real-time
diagnosis of incipient faults in, and prognosis of, fluid-filled electrical equipment
because the complex relationship between various operational parameters of
fluid-filled electrical equipment is not addressed fully by the prior art. For
example, a temperature rise outside of a normal range may be due to a
temporary increase in load and not to an incipient fault. Otiier parameters are
related in more complex ways mat are not addressed by the prior art. Also, the
devices discussed above do not account for dynamic change over time in
transformer behavior.
U.S. patent 5,845,272 discloses a system for isotating failures in a
locomotive or a process having a plurality of equipment. The system uses
outputs from various sensors as inputs in a knowledge base including causal
networks. However, U.S. patent 5,845,272 is not directed to diagnostics of
fluild filled electrical equipment and thus does not
relationships between parameters of fluid filled electrical equipment and the
dynamic change in behavior of fluid-filled electrical equipment over time.
In summary, known processes and apparatus do not take into account
analytical models of fluid-filled electrical equipment operation including thermal,
fluid flow, electric field, pressure-volume, chemical, failure mode, root failure
cause and gas-in-oil models, all of wakh are related in a complex manner and
change over time. Therefore, known methods and apparatus do not accurately
identify and predict failure modes and assess the life cycle of fluid-filled
electrical equipment.
SUMMARY OF THE INVENTION
The invention relates to a diagnostic apparatus and method for electrical
equipment. A first aspect of the invention is an intelligent analysis apparatus for
fluid-filled electrical equipment of the type having components surrounded by
fluid. The apparatus comprises electrical equipment having a containment vessel
configured to contain a fluid and at least one electrical component disposed in
the containment vessel, plural sensors configured to output signals indicative of
plural operating parameters of the electrical equipment, and a diagnostic device
coupled to the sensors and having a processor for determining operating
characteristics of the electrical equipment based on at least one analytical model
of the electrical equipment and the signals outputted by the sensors by applying
values of parameters calculated by the at least one analytical model and values
of parameters as indicated by the signals of the sensors in a causal network.
A second aspect of the invention is a method for inteligent analysis of
fluid-filled electrical equipment of the type having components surrounded by
fluid. The method comprises the steps of sensing plural operating parameters of
electrical equipment having a containment vessel confingured to contain a fluid
and at least one electrical component disposed in me containment vessel,
generating signals indicative of the plural operating parameters of the electrical
equipment sensed in the sensing step, and determining operating characteristics
of die electrical equipment based on at least one analytical model of the electrical
equipment and the signals generated in the generating step by applying values
of parameters calculated by the at least one analytical model and values of
parameters as indicated by the signals generated in the generating step in a
causal network.«
BRIEF DESCRIPTION OF THE DRAWING
The present invention can be more fully understood upon reading the
following detailed description of a preferred embodiment in conjunction with the
accompanying drawing in which:
Fig. 1 is a schematic illustration of a preferred embodiment of the
invention;
Fig. 2 is a flowchart of a diagnostic determming routine of the preferred
embodiment;
Fig. 3 is a graphical representation of possible failure modes of the
preferred embodiment; and
Fig. 4 is a graphical representation of a causal network for the failure
modes of Fig. 3.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
An embodiment of the invention uses analytical models of operation of
fluid-filled electrical equipment in combination with causal networks for the
purpose of determining operational status, diagnostics, and prognostics of fluid-
filled electrical equipment. The analytical models can include moddel of thermal
characteristics, electric and magnetic fields, temperature-pressure-volume,
failure modes, root failure cause, gas in oil, and chemical composition. The
analytical models are adjusted over tune to account for behavioral changes in the
fluid-filled electrical equipment. A belief network is used to dynamically adjust
the parameters of the causal network.
Fig. 1 illustrates a preferred embodiment of the invention. Diagnostic
system 10 comprises electrical equipment 20, an electrical transformer in the
preferred embodiment, and diagnostic device 30. Electrical equipment 20
comprises electrical components 22, including a core and coils/windings of the
transformer, and containment vessel 24 surrounding components 22.
Containment vessel 24 is adapted to contain fluid F, such as oil, for cooting and
insulating components 22. Fluid F can be pumped through containment vessel
24 and a radiator (not illustrated) by pump 26 disposed in or near containment
vessel 24. The radiator serves as a heat exchanger to cool fluid F and to thereby
conduct heat away from components 22 and can include any knows form of
pipes, conduits, heat exchanging surfaces, cooling elements, pumps, fans, or the
like. Cooling can be accomplished through thermal convection, thermal
conduction, molecular convection of fluid F, or in any other manner.
A plurality sensors 28a-28f are operatively coupled to electrical
equipment 20 in an appropriate manner. Sensors 28a-28f can be of any
appropriate type for sensing a desired parameter and generating, i.e. outputing,
a signal indicative of the value of the sensed parameter. In the preferred
embodiment, sensor 28a is a voltmeter, ammeter, or the like to measure the
electrical load on electrical equipment 20 and is coupled to load terminate 29 of
electrical equipment 20, sensor 28b is a temperature sensor disposed in fluid F
inside containment vessel 24, sensor 28c is a pressure sensor disposed in fluid
F inside containment vessel 24, sensor 28d is a molecular hydrogen sensor
disposed in fluid F inside containment vessel 24, sensor 28e is a fluid circulation
sensor, and sensor 28f is a fluid level sensor.
In the preferred embodiment, sensors 28b-28f are in contact with fluid
F. However, the invention requires only that the sensors 28b-28f be capable of
measuring parameters of fluid F. Accordingly, the sensors can be in a contact
or non contact relationship with fluid F depending on the type of sensors used,
as discussed in greater detail below. For example, sensors 28b-28f can be
positioned remotely from fluid F and can have sensing elements disposed in fluid
F. Alternatively, sensors 28a-28f can be entirely remote from fluid F and can
monitor parameters in fluid F from a distance, such as through optical means or
the like. Sensors 28a-28f can be disposed at any location and can sense
parameters ol electrical equipment 20 at any location as dictated by the type,
size, and shape of electrical equipment 20, the diagnostics and prognostics to be
evaluated, and any other details of the practical application. For example, it
may be desirable to sense values of winding temperature, hot spot temperature,
core, temperature, on load tap changer (OLTC) temperature, ambient
temperature, gas space pressure, fluid level, moisture in fluid, fluid dielectric
strength, acoustic partial discharge, sound pressure in the equipment, ambient
sound pressure, various gases in the fluid, fluid flow, fan/pump speeds and
currents, load currents, line voltage, and vibration. All of these parameters can
be sensed with known sensors. Further, plural sensors can be used to measure
the same parameter simultaneously at more man one location. Of course, there
can be any number of sensors depending oa the parameters to be measured and
me measurement locations desired.
Sensors 28a-28f can be fixedly disposed at the desired position on or in
electrical equipment 20 or can be removably disposed at desired locations by
being selectively inserted mrough sensor ports or other opeaiags formed through
walls of containment vessel 24 or other portions of electrical equipment 20.
Sensors 28a-28f can be of any appropriate type. For example, each sensor 28b-
28f can be one or more of metal insulator semiconductor diode sensors, fiber
optic probes, acoustic or optical waveguides, bimetal sensors, thin film sensors,
or any other appropriate sensor or transducer for measuring the parameters
noted herein. Sensor 28a can be any known type of electrical load sensing
device such as a resistive or inductive device. If sensors 28a-28f are electric or
electronic in nature and disposed inside a high EM field region of electrical
equipment 20, proper electrical shielding can be provided. Optical or other
types of sensors need not be electrically shielded regardless of location.
Sensors 28a-28f generate data or signals indicative of the operating parameters
of electrical equipment 20 sensed thereby.
Diagnostic device 30 is coupled to electrical equipment 20 for
determining various operating characteristics of electrical equipment 20 and
comprises processor 40, input/output interface (I/O) 50, user interface 60, data
bus 70 and power supply 72. Sensors 28a-28f and pump 26 are
communicatively coupled to I/O 50 through an appropriate conducting
mechanism. For example, if sensors 28a-28f are electronic or produce
electronic signals, electric conductors can extend from sensors 28a-28f to I/O
50. The conductors can include any appropriate terminal strip, connector, or the
like, for connection to I/O 50. Coupling, i.e. conducting signals, between
sensors 28a-28f and I/O 50 can be accomplished by wires, fiber optic strands,
radio frequency devices, infrared devices, or in any other known manner. Pump
26 can be coupled to I/O 50 in a similar manner. Diagnostic device 30 can
communicate with seams 28a-28f and pump 26 over a remote or local
communication link, such as a phone line, an RS232 serial link, universal serial
bus (USB) link, radio frequency link, infrared link, or the like. Power supply
72 is illustrated as a separate element. However, the power supply can be
integral to one more of the other components.
I/O 50 includes plural signal conditioning circuits 52a-52g which can be
of any type appropriate for conditioning nie signals or data output from sensors

28a-28f and pump 26. For example, signal conditioning circuits 52a-S2g can
include circuitry for smoothing, sampling, current limiting, choking, amplifying,
attenuating, or other, functions in a known manner. Note that pump 26 can
include a feedback capability to provide a signal or data representative of the
operating status thereof, such as one or more of a tachometer feedback, vibration
feedback, or load feedback. Similarly, each of signal conditioning circuits 52a-
52g are capable of conditioning output signals to be sent to sensors 28a-28f and
pump 26. Such signals can relate to adjustment of thresholds, linearization
parameters, sensitivity adjustments, speed adjustments (in the case of pump 26),
and the like as will be described further below. Signal conditioning circuits
52a-52g are illustrated as separate elements. However, one or more of the
signal conditioning circuits can be integral to the sensor, the processor, or other
components.
I/O SO also includes a multiple channel digital to analog and analog to
digital converter (D/A) 54 which provides an interface between the signals of
sensors 28a-28f and pump 26, which are analog in the preferred embodiment,
and processor 40, which is digital in the preferred embodiment. Of course, if
sensors 28a-28f, pump 26 and processor 40 are all analog or all digital, D/A 54
can be omitted. D/ A 54 is coupled to processor 40 through data bus 70 for two
way communication Processor 40 includes central processing unit (CPU) 42,
and memory device 44. CPU 42 excates a control program stored in memory
device 44. Memory device 44 can include a standard magnetic memory device,
such as a hard disk, for storing the control program and other data and also
includes a Awork space, such a random access memory (RAM) for CPU 42
to store data temporarily.
Diagnostic device 30 can also include user interface 60 comprising
display 62 and input device 64. Input device 64 can be any type of keyboard,
mouse, switch or switches, or any other device for allowing the user to input
settings, parameters, instructions, or the lite to processor 30. Display 62 can
be any type of display for indicating operating status, such as an LCD or CRT
display, a pilot lamp or series of pilot lamps, an audible alarm, or the like.
Power supply 72 provides power to other elements of diagnostic device 30 and
can be any type of known power supply, such as a battery, a fuel cell, or a
rectifier for providing DC power from an AC input. Diagnostic device 30 can
be a microprocessor based device, such as a personal computer or programmable
logic controller, a hardwired logic device, or any other device for accomplishing
the requisite processing disclosed below.
Processor 40 contains a preprogrammed control program stored in
memory device 44 for determining characteristics, such as diagnostics,
prognostics, performance characteristics, and life assessment of electrical
equipment 20 in the manner described below. Specifically, the control program
includes various analytical models of behavior of electrical equipment 20, a
causal network, and a belief network. Data bus 70 can utilize any appropriate
type of hardware and/or software protocols for transmitting and receiving data
or signals. For example, data bus 70 can be a standard ISA bus, DCI bus, GPIB
bus, or the fike. Data can be tramitted to a received from a remote or local
host computer to provide further diagnostics, prognostics, and control and to
coordinate diagnostics and operation of a plurality of fluid-filled electrical
equipment.
In operation, containment vessel 24 is fully or partially filled with fluid
F, such as oil. In this slate, sensors 28b, 28c, 28d, 28e and 28f are in contact
with, or otherwise can sense parameters in, fluid F. In the preferred
embodiment sensor 28b senses the temperature of fluid F, sensor 28c senses the
pressure in fluid F, sensor 28d senses the content of molecular hydrogen in fluid
F, sensor 28e senses circulation of fluid F, and sensor 28f senses the level of
fluid F. Other sensed parameters can include, but are not limited to the content
of various gases (such as acetylene, carbon, monoxide, and ethylene), winding
temperature, hot spot temperature, core temperature, on-load top changer
(OLTC) temperature, ambient temperature, gas space pressure, fluid level,
moisture in fluid, fluid dielectric strength, acoustic partial discharge, sound
pressure, ambient sound pressure, gas content, fluid flow, pump speed, and
vibration. Of course, any parameter which is helpful in determining the
operational status and/or is considered in an analytical model of electrical
equipment 20 can be sensed.
Fig. 2 is a flowchart of a diagnostic determining routine in accordance
with the preferred embodiment. The routine can be in the form of software
stored in memory device 44 and written in any appropriate language or code mat
can be read by CPU 42. For example, the software routine can be written in
Basic, C+ + , or the like. Output signals, i.e. sensor data, from D/A 54 are
representative of parameters of equipment 20 and are input to processor 40 over
bus 70. The sensor data is first subjected to validation step 100 to determine
if the corresponding sensors are working properly. For example, validation step
100 can include a step of comparing the senior data to predetermined minimum
and maximum thesholds s mat correspond to (although not necessarily
denrable) values of of parameters sensed by sensors 2th-28f. For example if
the sensor data corresponding to sensor 28b (a temperature sensor) indicates a
temperature higher or lower than a possible oil temperature, for example lower
than an ambient teaapcrature or much higher than the temperature rating of
components 22, it can be assumed that tne sensor 28a is not working properly.
Also, validation step 100 can include a step of checking for impossible
fluctuations in the value indicated by the sensor data which are indicative of an
intermittent problem in sensor 28a. Validation step 100 is conducted in a similar
manner for each of sensors 28a-28f. If one of tfae sensors 28a-28f is indicated
as not functioning properly in validation step 100, an appropriate error message
is displayed on display 62 or on a remote display or otherwise logged or
communicated to an operator or a remote computer or the like. Data from a
defective sensor can be ignored until the sensor is repaired or replaced.
Alternatively, the parameter measured by the defective sensor can be calculated
by one of the models in the manner described below.
The routine then advances to calculating step 110 in which the various
parameters are calculated based on other sensed parameters in accordance wira
models developed for the particular parameter in equipment 20. For example
a hydrogen model is an algoridim uiat calculates the theoretical value of
molecular hydrogen (H^ in fluid F of equipment 20 based on equipment
configuration information, i.e the size, relative dimensions, components, type
of fluid, etc. of equipment 20. The preferred embodiment includes hydrogen
model 112, temperature model 114, and pressure model 116. Any of various
known models can be used for each parameter. For example the AMIT
Hydrogen Model® developed at the Massachusetts Institute of Technology can
be used as hydrogen model 112. The MIT Hydrogen Model uses Ae following
Any of various known pressure models can be used. The models are
configured (i.e., the constants, and coefficients are calculated) in accordance
wim the rwtirular physical characteristics of
example, me rated load, the average conductor temperature rise over top oil, the
top oil rise, the load loss ratio, the cooling characteristics, the loss, the thermal
capacity, the weight of components 22, the weigth of containment vessel 24, and
the fluid capacity of electrical equipment 20 can be considered in a known
manner to develop the appropriate models.
Once the various values for each parameter have been calculated by
processor 40 in accordance with the models in step 1 It), the calculated values of
each parameter are compared to the measured values, i.e. the sensor data, of the
corresponding parameter in anomaly detection step 120. If the measured values
are within a prescribed tolerance or range of the calculated value, no anomaly
for that parameter is detected and no alarm is sounded. On the other hand, if the
measured value of a particular parameter is not within the prescribed-tolerance
or range, an alarm can be sounded on display 62, a separate alarm device, a
remote display or the like or otherwise logged or communicated, thus providing
a preliminary status indication.
m step 130, the differences between the measured and calculated
parameter values are applied as indicators of a causal network. The causal
network is part of the routine and thus can be stored in memory device 44. The
causal network can be developed in advance in the manner described below.
Each causal network has a cause and effect relationship between a plurality of
nodes, wherein some of the nodes represent root causes-associated wife failures
causal networks has a prior probability indicating me likehood of the particular
failure.
Each of the nodes in the causal network also has conditional probabability
node to its failure mode, i.e., the cause and effect relationships between failures
and observable symptoms for electrical equipment 20. Thus, in order to develop
the causal network there has to be an understanding of how each component in

fluid-filled electrical equipment operates and of the observable symptoms of each
failure mode. Some of the possible failure modes mat electrical equipment 20
may be subject to are failure of pump 26 (including failure of me motor and
damage to the fan blade), leakage of containment vessel 24, failure of
component 22, failure of insulation on component 22, an overload condition,
dielectric breakdown of fluid F, and a radiator leak.
After all of the possible failure modes have been identified for electrical
equipment 20, the causal network for electrical equipment 20 is developed. Fig.
3 illustrates the above-identified failure modes for electrical equipment 20.
These failure modes are intended to be exemplary and the list is not all-
inclusive. Each of the failure modes is designated as a failure mode node, or a
cause, and is represented as a box with rounded corners. Each cause has some
higher level effect on electrical equipment 20. It is also possible mat several
causes may have the same effect. At some point, an effect manifests itself such
that it can be measured or observed. When the state of a single observable
symptom or the state of several observable symptoms is unique to a single cause,
then it will be possible to unambiguously identify the problem.
Fig. 4 illustrates an example of tine cause and effect reiatiooslups for each
of the failnre modes identified in Fig. 3, i.e., a causal network for electrical
equipment 20. Tbe pbiase Acausal
algorithm, or the like indicating potential failure, and their probability
relationship to various manifestations. The cause and effect relationship between
each of the nodes (failure modes and observable manifestations) is shown with
an arrow pointing in4he direction of the causality, in Ftg. 4, the faikire modes
of pump motor failure and pump Made failure are each shown to have an effect
that is characterized by the observable manifestation of low fluid circulation.
The low circulation node is coupled to an indicator node, low circulation, mat
indicates whether the fluid circulation is low as measured by sensor 28e. The
indicator node is a node that is always an effect that directly represents the value
of a measured parameter, a calculated value of the parameter, or the difference
between the measured parameter and the calculated value thereof and is
represented by a circle.
The pump blade failure node, and the pump motor failure node are each
shown to have an effect that is characterized with low fluid circulation through
containment vessel 24, as indicated by the data from sensor 28e as compared to
values calculated by pressure model 116 (in step 110 above). The containment
vessel leak node is shown to have an effect that the fluid level will be low as
indicated by fluid level sensor 28f in containment vessel 24. In addition, the
pump motor failure mode is coupled to an indicator node that correspoads to the
feedback, e.g. a tachometer, from pump 26. At a higher level, the effects of low
fluid circulation and low fluid level have an effect on electrical equipment 20
that is characterized by inadequate cooling capacity because the fluid is not
properly circulated through a radiator. This effect is coupled to an indicator that
checks if the fluid temperature is above normal, as indicated by the difference
between the temperature value calculated by temperature model 114 (in step 110
described above)
For each failure mode in the causal aetwork,
probability indicating the likelihood of a failure is assigned. The coadiuoaal
probabilities are factors assigned to each failure mode mat indicates the relative
probability that the cause is present. Fig. 4 shows an example of
probabilities assigned to each of the failure modes
The conditional probabilities are listed as decimal numbers under the
corresponding failure mode node. Note that in cases where a component has
multiple failure modes, the conditional probability of failure due to each failure
mode is required. Also, note that the conditional probability magnitudes of
failure mode nodes grouped together dictate the relative likelihood that a
particular failure mode is the problem. For example according to the causal
network of Fig. 4, if mere was low circulation of fluid F, then it would be
predicted that a pump motor failure (conditional probability of 0.1) was ten
times more likely to be the cause of low fluid circulation than a pump blade
failure (conditional probability of .001).
After a conditional probability has been assigned to each of the failure
modes, an edge probability estimating the strength of the relationship between
the failure mode and a next level manifestation is assigned for each relationship.
The edge probabilities, are listed in accordance with the line connecting the
failure mode nodes with the manifestation nodes and represent the probability
mat the manifestations will exist, given that the failure mode is already known
to exist. If all of the failure modes are independent, and if all of the failure
modes do not exist, then the manifestation does not exist. In the preferred
embodiment, a single parameter between 0 and 1 is used as an edge probability
to represent the strength of the relationship (1 being indicative of a direct
relationship or a Aone-to-one@ relationship) however any indication representing
the relationship between the failure mode node and the maniestation node can
be used.
As an example, a leak in containment vessel 24 will result in a detected
effect of low level of fluid F 90% of the time, as indicated by an edge
probability of 0.9. The other 10% of the time, the teak will be too slow to affect
fluid level significantly. Hole that the conditional probability information is
derived from the edge probability information. This information can be
determined experimentally or mathematically.
The causal network described above, is applied in step 130 of Fig. 2 and
used to determine the operating status of electrical equipment 20. The indicators
of the causal network, i.e. the parameters measured by sensors 28a-28f and the
differences between these parameters and calculated values thereof yielded by
models 112, 114 and 116, are evaluated in accordance with the conditional
probability information and the edge probability information. Processor 40
continually recalculates the probabilities of the causal networks according to the
status of the mapped indicators, as indicated by sensors 28a-28f and the
appropriate models. In particular, the probabilities are recalculated in step 140
using a known belief network solution algorithm, such as a Bayesian Belief
Network, and fed back to the causal network of step 130. For example, if there
is low fluid circulation detected by sensor 28e and the tachometer of pump 26
indicates normal pump motor status, the probabilities are adjusted to increase the
likelihood that there is a pump blade failure. Processor 40 then evaluates the
recalculated probabilities in the causal network. In addition, processor 40 can
provide a list of the most likely causes for any abnormality, as well as a list of
corrective actions to be taken to correct or avoid failure. The probabilities are
recalculated by the belief network based on information learned from the causal
network using the previous probabilities. The recalculations can be based on
known inference techniques, influence techniques, or Bayes' theorem.
The routine illustrated in Fig. 2 is conducted is essentially a confnmfms
mariner in the pseferred embodiment. However, the process can be conducted
in a periodic manner automatically, or upon request by an operator. The various
constants and coefficients of the models are adjusted over time to compensate for
normal behavioral changes in electrical equipment 20 over time. The constants
and coefficients can be determined mathematically or experimentally in a known
manner. The output of the causal network can be processed in any maimer for
diagnostics, prognostic, or the like. For example, status reports relating to
operating characteristics of the electrical equipment can be generated, alarms can
be effected, or the operation of the equipment can be adjusted.
The invention can be applied to any fluid-filled electrical equipment.
Any desired parameters can be detected. Sensor data or signals can be
processed in any way to provide indication of incipient faults prognostics, life
assessment, maintenance schedules, fault root cause identification, or other
status of the electrical equipment based on experimental or mathematical models.
Additionally, the invention can provide performance characteristic evaluations
such as utililization factors, load scheduling, efficiency, energy loss, power
factor, harmonics, and on load tap changer performance.
The diagnostic device can be local, i.e., closely situated with respect to
the electrical equipment, or remote, i.e., located at a remote location with
respect to the electrical equipment. Histories of the values of the various
parameters as measured and as calculated, can be compiled to assist further in
fault determination. The various sensors can be polled at regular intervals and
the intervals can be increased at times of heavy load on the equipment or upon
indication of an abnormal state of the equipment.
The invention has been described through a preferred embodiment.
However, various modifications can be made without departing from the scope
of the invention as defined by the appended claims.

WE CLAIM
1. An intelligent analysis apparatus (10) for fluid -filled electrical
equipment (20) of the type having components (22) surrounded by fluid (F), said
apparatus comprising:
electrical equipment (20) having a containment vessel (24) configured to
contain a fluid (F) and at least one electrical component (22) disposed in said
containment vessel (24);
plural sensors (28a-e) configured to output signals indicative of plural
operating parameters of said electrical equipment (20); and
a diagnostic device (30) coupled to said sensors (28a-e), said diagnostic device
(30) having a processor (40) operative to determine operating characteristics of said
electrical equipment (20) based on at least one analytical model (112, 114, 116) of
said electrical equipment (20) and the signals outputted by said sensors (28a-e) by
applying values of parameters calculated by the at least one analytical model (112,
114, 116) and values of parameters as indicated by the signals of said sensors (28a-e)
in a causal network.
2. An apparatus as recited is claim 1, wherein said diagnostic device (30)
compares a parameter calculated by the at least one analytical model (112, 114, 116)
with a correspondin parameter and uses a result of me comprison as an
indicator in the causal network
3. An apparatus as recited in claim 2, wherein the probobitites of the
causal network are updated based on the probability of the indicators (130) obtained
from the analytical model (112,114,116) or sensed parameters.
4. An apparatus as recited in claim 3 where in variables
analytical model (112, 114, 116) are adjusted over time in correspondence to
acceptable behavioral changes of said electrical equipment (20) over time.
5. An apparatus as recited in claim 4, wherein said sensors (28a-e)
comprise a temperature sensor (28b) configured to output a signal indicative of fluid
(F) temperature inside said containment vessel (24), a gas sensor (28d) configured to
output a signal indicative of gas content of fluid (F) inside said containment vessel
(24), a load sensor (28a) configured to output a signal indicative of the electrical load
on said electrical equipment (20), and a pressure sensor 28(e) configured to output a
signal indicative of pressure in said containment vessel (24).
6. An apparatus as recited in claim 5, wherein said gas sensor (28d) is
configured to output a signal indicative of a hydrogen content of fluid (F) inside said
containment vessel (24).
7. An apparatus as recited in claim 6, wherein the at least one analytical
model (112, 114, 116) comprises a temperature model (114) and a hydrogen model
(112).
8. An apparatus as recited in claim 4, wherein said processor (40) is
disposed outside of said containment vessel (24) and said sensors (28a-e) are disposed
inside said containment vessel (24), said apparatus further comprising .means for
conducting the signals from said sensors (28a-e) to said processor.
9. An apparatus as recited in claim 4, said diagnostic device (30) further
comprising a user interface module (60) having a display (62) for displaying an
indication of the operating characteristics of said electrical equipment (20) and an
input device (64) for permitting a user to input at least one of data aad commands to
said diagnostic device (30).
10. An apparatus as recited hr claim 5, wherein said gas sensor (28d) is
configured to output a signal indicative of at least one of hydrogen, carbon monoxide,
carbon dioxide, oxygen, nitrogen, hydrocarbons, and hydrocarbon derivatives.
11. Aa apparatus as recited in claim 4, wherein said processor (40)
comprises a computer having a CPU (42) and a memory device (44).
12. An apparatus as recited in claim 4, wherein said electrical equipment
(20) comprises a transformer.
13. An intelligent analysis apparatus (10) for fluid -filled electrical
transformers, said apparatus comprising:
a transformer (20) having a containment vessel (24) configured to contain a
fluid (F) and a core and coil (22) disposed in said containment vessel (24);
plural sensors (28a-e) configured to output signals indicative of plural
operating parameters of said transformer (20); and
a diagnostic device (30) coupled to said sensors (28a-e), said diagnostic device
(30) having a processor (40)operative to determine operating characteristics of said
transformer (20) based on at least one analytical model (112, 114, 116) of said
transformer (20) and the signals outputted by said sensors (28a-e) by applying values
of parameters calculated by the at least one analytical model (112, 114, 116) and
values of parameters as indicated by the signals of said sensors (28a-e) in a causal
network.
14. An apparatus as recited in claim 13, wherein said diagnostic device
(30) compares a parameter calculated by the at least one analytical model (112, 114,
116) with a corresponding measured parameter and uses a result of the comparison as
an indicator in the causal network.
15. An apparatus as recited in claim 14, wherein the probabitities of the
causal network are adjusted based on the probability of the indicators (130) obtained
from the analytical model (112,114,116) or sensed parameters.
16. An apparatus as recited in claim IS, wherein of the at least one
analytical model (112, 114, 116) is adjusted over time in correspondence to
acceptable behavioral cnanges of said transformed (20) over time.
17. An apparatus as recited in claim 16, wherein said sensors (28a-e)
comprise a temperature sensor (28b) configured to output a signal indicative of fluid
(F) temperature inside said containment vessel (24), a gas sensor (28d) configured to
output a signal indicative of gas content of fluid (F) inside said containment vessel
(24), a load sensor (28a) configured to output a signal indicative of the electrical load
on said transformer (20) and a pressure sensor 28(e) configured to output a signal
indicative of pressure in said containment vessel (24).
18. An apparatus as recited in claim 17, wherein said gas sensor (28d) is
configured to output a signal indicative of a hydrogen content of fluid (F) inside said
containment vessel (24).
19. An apparatus as recited in claim 18, wherein the at least one analytical
model (112, 114, 116) comprises a temperature model (114) and a hydrogen model
(112).
20. An apparatus as recited in claim 16, wherein said processor (40)is
disposed outside of said containment vessel (24) and said sensors (28a-e) are disposed
inside said containment vessel (24), said system further comprising means for
conducting signals from said sensors (28a-e) to said processor.
21. An apparatus as recited in claim 16, said diagnostic device (30) further
comprising a user interface module (60) having a display (62) for displaying an
indication of the operating characteristics of said transformer (20) and an input device
(64) for permitting a user to input at least one of data and commands to said
diagnostic device (30).
22. An apparatus as recited in claim 17, wherein said gas sensor (28d) is
coafigured to output a signal indicative of at least one of hydrogen, carbon monoxide,
carbon dioxide, oxygen, nitrogen, hydrocarbons, and hydrocarbon derivatives.
23. An apparatus as recited in claim 16, wherein said processor (40)
comprises computer having CPU (42) and a memory device (44).
24. An intelligent analysis aparatus (10) for fluid-filled electrical
equipment (20) of the type having components (22) surrounded by fluid (F), said
apparatus comprising:
electrical equipment (20) having a containment vessel (24) configured to
contain a fluid (F) and at least one electrical component (22) disposed in said
containment vessel (24);
sensing means (28a-e) for sensing plural operating parameters of said
electrical equipment (20) and for outputting signals indicative of the plural operating
parameters of said electrical equipment (20); and
diagnostic means (30) for determining operating characteristics of said
electrical equipment (20) based on at least one analytical model (112, 114, 116) of
said electrical equipment (20) and the signals outputted by said sensing means (28a-e)
by applying values of parameters calculated by the at least one analytical model (112,
114, 116) and values of parameters as indicated by the signals of said sensing means
(28a-e) in a causal network.
25. An apparatus as recited in claim 24 wherein said diagnostic moans (30)
comprises means for comparing a parameter calculated by the at least one analytical
model (112, 114, 116) with a corresponding measured parameter and means for using
a result of the comparison as an indicator in the causal network.
26. An apparatus as recited in claim 25, wherein said diagnostic means
(30) comprises means for adjusting probabilities of the causal network based on the
probability of the indicators (130) obtained from the analytical models (112,114,116)
or sensed parameters.
27. An apparatus as retained in claim 26, wherein said diagonesitic means
(30) comprises means for adjusting variables of the as least one analytical model (112,
114, 116) over time in correspondence to acceptable behavioral changes of said
electrical equipment (20) over time.
28. An apparatus as recited in claim 27, wherein said sensing means (28a-
e) comprises temperature sensing means (28b) for outputting a signal indicative of
fluid (F) temperature inside said containment vessel (24), gas sensing moans (28d)
for outputting a signal indicative of gas content of fluid (F) inside said containment
vessel (24), load sensing means (28a) for outputting a signal indicative of the
electrical load on said electrical equipment (20) and pressure sensing means (28c) for
outputting a signal indicative of pressure in said containment vessel (24).
29. An apparatus as recited in claim 28, wherein said gas sensing means
(28d) comprises means for outputting a signal indicative of a hydrogen content of
fluid (F) inside said containment vessel (24).
30. An apparatus as recited in claim 29, wherein the at least one analytical
model (112, 114, 116) comprises a temperature model (114) and a hydrogen model
(112).
31. An apparatus as recited in claim 27, further comprising means for
conducting signals from said sensor means to said processor.
32. An apparatus as recited in claim 27, said diagnostic means (30) further
comprising user interface means (60) for displaying an indication of the operating
characteristics of the electrical equipment (20) and input means for permitting a user
to input at least one of data and commands to said diagnostic means (30).
33. An apparatus as recited in claim 28, wherein said gas sensing means
(28d) comprises means for outputting a signal indicative of an amount of at least one
of hydrogen, carbon monoxide, carbon dioxide, oxygen, nitrogen, hydrocarbons, and
hydrocarbon derivatives
34. An apparatus as recked in claim 27, wherein said diagnostic means
(30) comprises a computer having a CPU (42) and a memory device (44).
35. An apparatus as recited in claim 27, wherein said electrical equipment
(20) comprises a transformer (20).
36. A method for intelligent analysis of fluid-filled electrical equipment
(20) of the type having components (22) surrounded by fluid (F), said method
comprising the steps of:
sensing plural operating parameters of electrical equipment (20) having a
containment vessel (24) configured to contain a fluid (F) and at least one electrical
component (22) disposed in said containment vessel (24);
generating signals indicative of the plural operating parameters of the
electrical equipment (20) sensed in said sensing step; and
determining operating characteristics of the electrical equipment (20) based on
at least one analytical model (112, 114, 116) of me electrical equipment (20) and the
signals generated in said generating step by applying values of parameters calculated
by the at least one analytical model (112, 114, 116) and values of parameters as
indicated by the signals generated in said generating step in a causal network.
37. A method as recited in claim 36 wherein said determining step
comprises the steps of comparing a parameter calculated by the at least one analytical
model (112, 114, 116) with a corresponding measured parameter as indicated by
signals in said generating step and using a result of said comparing step as an
indicator in the causal network.
38. A method as recited in claim 37, wherein said determining step
comprises the step of adjusting the probabilities of the causal network based on the
probability of the indicators (130) obtained from the analytical model (112,114, 116)
or sensed parameters.
39. A method as recited in claim 38, wherein said determining step
comprises the step of adjusting variables of the at least one analytical al moid (112,
114, 116) over time in correspondence to acceptable behavioral changes of the
electrical equipment (20) over time.
40. A method as recited in claim 39, wherein said something step comprises
me steps of sensing temperansre inside the containment vessel (24), seasing gas
content of fluid (F) inside the containment vessel (24), sensing the electrical load on
the electrical equipment (20) and sensing pressure in the containment vessel (24).
41. A method as recited in claim 40, wherein said step of sensing gas
content comprises sensing a hydrogen content of fluid (F) inside the containment
vessel (24).
42. A method as recited in claim 41, wherein the at least one analytical
model (112, 114, 116) used in said determining step comprises a temperature model
(114) and a hydrogen model (112).
43. A method as recited in claim 39, further comprising the steps of
displaying an indication of the operating characteristics of the electrical equipment
(20) and inputting at least one of data and commands.
44. A method as recited in claim 40, wherein said step of sensing gas content
comprises sensing at least one of hydrogen, carbon monoxide, carbon dioxide,
oxygen, nitrogen, hydrocarbons, and hydrocarbon derivatives.
45. An apparatus as recited in claim 7, wherein the at least one analytical
model (112,114,116) further comprises a pressure model (116).
46. An apparatus as recited in claim 19, wherein the at least one analytical
model (112,114,116) comprises a pressure model (116).
47. An apparatus as recited in claim 30, wherein the at least one analytical
model (112,114,116) comprises a pressure model (116).
48. A me&od as recited in claim 42, wherein the at least one analytical
mode! (112,114,116) comprises a pressure model (116).

An intelligent analysis apparatus (10) and method for fluid (F) filled electrical
equipment (20) includes sensors (28a-e) for measuring various parameters of the
electrical equipment (20). Analytical model (112, 114, 116)s calculate parameters
based on measurements of other parameters. The measured and calculated parameters
are compared and the result of the comparison is used as an indicator in a causal
network. The probabilities of the causal network are recalculated by a belief network.
The analytical model (112, 114, 116)s are adjusted over time to account for
acceptable changes in behavior of the equipment over time. The output of the causal
network can be used for diagnostic and prognostic indication.