METHOD AND SYSTEM FOR DIAGNOSING COMPRESSORS
BACKGROUND
TECHNICAL FIELD
Embodiments of the subject matter disclosed herein generally relate to
methods and systems and, more particularly, to mechanisms and
techniques for diagnosing a machine in general, and a compressor in
particular.
DISCUSSION OF THE BACKGROUND
Today there are a large number of machines (industrial machines as, for
example, compressors) installed at various facilities and used to process
oil and gas. Such machines may experience symptoms that are
indicative of a fault or a failure mode. Due to the technical
complexities of these machines, the user of the machine might not have
the capability to address these symptoms. Thus, the manufacturer of
the machines, which has the technical capability to determine the
matters affecting the machine enters into maintenance and diagnostic
agreements with the user for ensuring that the machines are monitored
and maintained in an adequate state of operation. For this reason, the
manufacturer of the machines may have plural sensors installed at the
location of the user for monitoring the "health" of the machines. The
same manufacturer may have plural contracts with multiple clients.
Prognostics and Health Management (PHM) is an emerging technology
to support the efficient execution of contractual service agreements
(CSA) for assets such as locomotives, medical scanners, aircraft
engines, turbines, and compressors. One goal of PHM is to maintain
these assets' operational performance over time, improving their
utilization while minimizing their maintenance cost. PHM can be used
as a product differentiator, to reduce the cost of the original equipment
manufacturer service during the assets' warranty period, or to more
efficiently provide service under a CSA.
Figure 1 shows a traditional PHM system 10. According to this figure,
after performing the traditional preparation tasks, such as sensor
validation in a sensor validation unit 12 and input data pre-processing
in a processing unit 14, the PHM system 10 performs anomaly detection
and identification in unit 16, diagnostic analysis in unit 18, prognostic
analysis in unit 20, fault accommodation in unit 22, and logistics
decisions in unit 24. These actions are known by those skilled in the art
and for this reason their detailed description is omitted herein.
The anomaly detection unit leverages unsupervised learning techniques,
such as clustering. Its goal is to extract the underlying structural
information from the data, define normal structures and identify
departures from such normal structures. The diagnostics unit leverages
supervised learning techniques, such as classification. Its goal is to
extract potential signatures from the data, which could be used to
recognize different faults/failure modes s. The prognostics unit
produces estimates of Remaining Useful Life (RUL). Its goal is to
maintain and forecast the asset health index. Originally, this index
reflects the expected deterioration under normal operating conditions.
Later the index is modified by the occurrence of an anomaly/failure,
reflecting faster RUL reductions.
The above discussed functions are interpretations of the machine's
state. These interpretations lead to an on-platform control action and
an off-platform logistics, repair and planning action. The on-platform
control actions are usually focused on maintaining performance or
safety margins, and are performed in real-time. The off-platform
maintenance/repair actions cover more complex offline decisions. They
require a decision support system (DSS) performing multi -objective
optimizations, exploring frontiers of corrective actions, and combining
them with preference aggregations to generate the best decision
tradeoffs.
However, the traditional algorithms for calculating the relevance of a
determined diagnostic relative to the existent symptoms of the
compressor are not always accurate and sometimes they are ambiguous.
Accordingly, it would be desirable to provide systems and methods that
avoid these problems and drawbacks.
SUMMARY
According to an exemplary embodiment, there is a method for
diagnosing a compressor. The method includes generating a feature
vector of the compressor, the feature vector of the compressor including
components describing states of various parts of the compressor;
determining, based on fuzzy constraints, an aggregated anomaly vector
corresponding to the feature vector; defining rules for a preset list of
possible faults/failure modes of the compressor; calculating a
corroborating measure between the aggregated anomaly vector and the
rules; and identifying a faults/failure modes of the compressor based on
a result of the corroborating measure.
According to another exemplary embodiment, there is a system for
diagnosing a compressor. The system includes an interface configured
to receive measurement data about the compressor; and a processor
configured to receive the measurement data. The processor is
configured to generate a feature vector of the compressor based on the
measurement data, the feature vector of the compressor including
components describing states of various parts of the compressor,
determine, based on fuzzy constraints, an aggregated anomaly vector
corresponding to the feature vector, retrieve rules for a preset list of
possible faults/failure modes of the compressor, calculate a
corroborating measure between the aggregated anomaly vector and the
rules, and identify a faults/failure modes of the compressor based on a
result of the corroborating measure.
According to still another exemplary embodiment, there is a computer
readable medium including computer executable instructions, where the
instructions, when executed, implement a method for diagnosing a
compressor. The method include generating a feature vector of the
compressor, the feature vector of the compressor including components
describing states of various parts of the compressor; determining, based
on fuzzy constraints, an aggregated anomaly vector corresponding to
the feature vector; defining rules for a preset list of possible
faults/failure modes of the compressor; calculating a corroborating
measure between the aggregated anomaly vector and the rules; and
identifying a faults/failure modes of the compressor based on a result of
the corroborating measure.
BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings, which are incorporated in and constitute a
part of the specification, illustrate one or more embodiments and,
together with the description, explain these embodiments. In the
drawings:
Figure 1 is a schematic diagram of a conventional compressor
diagnostic system;
Figure 2 is a schematic diagram of a compressor diagnostic system
according to an exemplary embodiment;
Figures 3 and 4 are graphs showing fuzzy thresholds according to
exemplary embodiments;
Figure 5 is a flowchart illustrating a process for detecting
faults/failure modes of a compressor according to an exemplary
embodiment;
Figure 6 is a flowchart illustrating a method for ranking various
faults/failure modes of a compressor according to an exemplar
embodiment; and
Figure 7 is an exemplary system that can implement a method for
diagnosing a compressor.
DETAILED DESCRIPTION
The following description of the exemplary embodiments refers to the
accompanying drawings. The same reference numbers in different
drawings identify the same or similar elements. The following detailed
description does not limit the invention. Instead, the scope of the
invention is defined by the appended claims. The following
embodiments are discussed, for simplicity, with regard to the
terminology and structure of centrifugal compressors. However, the
embodiments to be discussed next are not limited to these compressors,
but may be applied to other compressors or complex structures that may
experience various that need to be identified.
Reference throughout the specification to "one embodiment" or "an
embodiment" means that a particular feature, structure, or characteristic
described in connection with an embodiment is included in at least one
embodiment of the subject matter disclosed. Thus, the appearance of
the phrases "in one embodiment" or "in an embodiment" in various
places throughout the specification is not necessarily referring to the
same embodiment. Further, the particular features, structures or
characteristics may be combined in any suitable manner in one or more
embodiments.
According to an exemplary embodiment, a system is configured to
receive measurements from at least one compressor, generate a feature
vector of the at least one compressor, the feature vector of the
compressor including components describing states of various parts of
the compressor, determine, based on fuzzy constraints, an aggregated
anomaly vector corresponding to the feature vector, define rules for a
preset list of possible diagnostics of the compressor, calculate a
corroborating measure between the aggregated anomaly vector and the
rules, and identify a fault/failure mode of the compressor based on a
result of the corroborating measure.
Such a system 30 is shown in Figure 2 and may be configured to
monitor plural compressors. In one exemplary embodiment, the system
30 is a centralized system connected to tens if not hundredths of
compressors via a data acquisition unit 32. The compressors may
belong to various customers and the centralized system provides
maintenance and technical support to these customers. The schematic
of such a system is discussed later with regard to Figure 7. The
collected data is stored in a database 34 and the collected data may be
related to various parameters of the compressors (temperatures,
pressures, speed, gas composition, etc.).
A job scheduler 36 is configured to receive as input the collected data
and to listen to a timer or start an application by timer. The job
scheduler 36 may be configured to check sequentially and at given time
intervals data received from the plurality of compressors and to
determine whether an anomaly identification process should be started.
The job scheduler 36 is configured to communicate with a data preprocess
unit 38. The data pre-process unit 38 is configured to receive
the collected data from database 34 and to perform various operations
on that collected data. For example, the pre-process unit 38 may
retrieve reference values for the collected data, may perform feature
extractions based on predetermined algorithms, may calculate various
parameters (features) of the compressor based on the collected data, etc.
The calculations performed by the pre-process unit 38 may be based on
a thermodynamic model of the compressor. The pre-process unit 38
may be receiving data from a machine configuration unit 40, which
includes machine-specific parameters, default values of those
parameters, normal values of those parameters, acceptable boundaries
for those parameters, etc.
The derived features determined by the pre-process unit 38 are provided
to an anomaly notification unit 42, which is configured to calculate an
anomaly matrix and an anomaly vector as discussed later. The anomaly
notification unit 42 is also connected to the machine configuration unit
40 and can retrieve desired data from this unit. The output from the
anomaly notification unit 42 is provided to a case manager unit 44. The
case manager unit 44 is configured, among other things, to open a new
case for a compressor that exhibits an anomaly.
A diagnostic reasoning unit 46 is connected to the case manager unit 44
and provides the case manager unit 44 with a ranked list of possible
diagnostics (fault/failure modes) for the exhibited symptoms of the
compressor. The details for generating the ranked list of possible
diagnostics are discussed later. In case that more data is necessary, the
case manager unit 44 may request the necessary data from a data
request unit 48 which is capable to communicate with compressors 50.
Having the ranked list, the case manager 44 may store the data in a
database 52, or may present the data, for example, via a dedicated user
interface 54, to a user 56. Alternatively or in addition, the case
manager 44 may present the results of the analysis to a customer 58 via
a communication unit 60. Thus, the customer 58 has the opportunity to
provide feedback to the case manager 44.
According to an exemplary embodiment, more details are provided
about the anomaly notification unit 42. Suppose that plural sensors
(not shown) are distributed on the compressors and these sensors
measure various parameters of the compressors, as for example,
temperatures, pressures, speeds at various locations of the compressor,
etc. A vector including this data is referred herein to feature vector F.
Vector F may have k components (corresponding to the measured k
parameters) and each component has a time stamp, i.e., each component
is associated with a time at which the corresponding parameter is
measured. This vector may include data received from the job
scheduler 36 as processed by the data preprocess unit 38.
Using dedicated mathematical notations for vectors and their
components, the feature vector is given by F(t) = 1(t),...,ò k(t)]· Each
component of vector F represents a dynamic feature, as for example, a
measured value, a difference between the measured value and a
reference value, a percentage of a threshold, a trend of the measured
parameter, etc. Components 1 to k of the feature vector F are measured
and/or derived at sampling points in time. The measurements of these
components are repeated at a given time interval and these
measured/calculated components may be stored in a database for further
processing.
According to an exemplary embodiment, the components are filtered to
maintain only those components that are not in a transient mode, i.e., a
steady state system is desired to be analyzed. Then, one or more
components are compared to corresponding fuzzy thresholds that
include two values, an attention value and a risk value. The attention
value makes aware the operator that the respective parameter should be
observed as the compressor may start to behave in an undesirable
fashion. The risk value indicates that the compressor may be in danger
of failing and measures needs to be taken to correct the parameters that
are past this threshold.
It is noted that, based on the inventors knowledge, fuzzy thresholds
have not been used previously in the field of diagnosing compressors.
For this reason, fuzzy thresholds are now discussed in more details.
For each component fi(t) of vector F a lower threshold LT (fi) and/or an
upper threshold UTi(fi) can be defined to map the domain variable f
into the interval [0, 1]. For example, the lower threshold may
beLTi (ò i) :ò i ®[0,1] and the upper threshold may be UTi (ò i) :ò i ®[0,1]
Figure 3 illustrates the lower fuzzy threshold and Figure 4 illustrates
the upper fuzzy threshold. It is noted that Figures 3 and 4 present
specific examples of fuzzy thresholds. However, the disclosed
exemplary embodiments are compatible with other fuzzy thresholds and
the specific fuzzy thresholds shown in Figures 3 and 4 are for
illustrative purposes only.
Figures 3 shows the attention value Ci and the risk value a while Figure
4 shows the attention value i and the risk value Ci . Figure 3 shows that
a change of the parameter fi of the compressor past value zero triggers
the attention value Ci to be exceeded while a change of the parameter
towards value one triggers the risk value a, to be exceeded. A similar
explanation is valid for Figure 4. As the monitored parameters f have
either an upper limit or a lower limit, the appropriate upper fuzzy
threshold or the lower fuzzy threshold need to be used.
Further, it is assumed that Ti(fi) defines a generic fuzzy threshold of
feature f , with the understanding that Ti(fi) is to be replaced by either
LTi(fi) or UTi(fi) depending on a direction of the criticality. Thresholds
Ti(fi) are considered to be constraints that need to be satisfied to
maintain a normal operation state of the compressor. A difference
between fuzzy constraints and traditional constraints is reflected in the
fact that the traditional constraints are routinely step functions while
the fuzzy constraints are represented by non-step functions.
The feature vector F may be considered to be a k x N matrix M, where k
is the number of features and N is the number of samples of
components f taken over a temporal window. A temporal window is a
predetermined amount of time over which measurements of components
f are taken. For simplicity, consider k=20 parameters fi of the
compressor being measured every minute over a two hours temporal
window. Other numbers may be used based on the compressor and
needs. This data, when assembled as matrix M, has (i,j) elements and
can be represented as shown in Table 1.
Table 1
By applying the thresholds Ti(fi) in vector format to the columns of
matrix M, the matrix shown in Table 2 is obtained.
Table 2
The result in Table 2 is a k x N matrix E, which indicates a degree to
which each element of matrix M satisfies a corresponding normality
constrain, where the constraints are the fuzzy constraints discussed
above. Each element E(i, j ) of matrix E is given by
E(i,j) =[Ti (M (i,j ))]=[Ti {f i (j ))] , which has values in the interval [0, 1].
A departure from zero of an element E(i, j ) indicates a potential
departure from normality for the corresponding parameter f . For
example, matrix E in Table 2 shows that a value for parameter fi has
reached the attention value at sampling point j and it reached the risk
value at sampling point N.
However, due to the potential high number of sampling points, it may
be desired to aggregate the numbers shown in matrix E. Thus, the
degree of anomaly for each feature may be aggregated over a time
window that is appropriate for the sampling rate and a time constant
related to the feature. In one application, the window may have length
N. More generally, the aggregation may be accomplished using a
moving window over the N columns of matrix M.
An aggregation may need to be performed over the elements included in
the window. One example of an aggregation may be an exponential
moving average, which gives more relevance to recent values. Other
known aggregation function may be used, as for example, a generalized
weighted means. The result of such aggregation is a vector A, whose i
element indicates an overall degree of anomaly of feature fi, i.e., A(fi) .
For example, the maximum value of the anomaly for a set of measuring
times may be chosen as shown in Table 3 . In another application, an
average of various anomalies measured over plural measuring times
may be considered as A(fi) . In the exemplary embodiment shown in
Table 3, after the aggregation, it is assumed that there is an anomaly in
feature fi and a partial anomaly in feature fi but no anomaly in feature
fk. This information may be integrated with other sources of anomaly
detection (for example, information encoded as error messages).
While such error messages are treated as Boolean data (i.e., anomalies
with strength 1), there is full visibility over the evolution of the
anomaly strengths through matrix E.
Table 3
Schematically, the process of applying the constraints to the features fi
and aggregating the anomalies is shown in Table 4 .
Table 4
According to an exemplary embodiment, the diagnostic unit 46 may
store or may request data regarding the compressor from an external
storage unit.
Table 5
The diagnostic process may create a (partial) order over the potential
fault/failure modes by matching the degree of anomaly of each feature,
represented by the vector A, with the anomalies expected when a
particular fault/failure mode is present. This domain knowledge is
captured in the matrix shown in Table 5, in which the first column
represents possible fault/failure modes and the remaining columns
represent the corresponding expected degree of strength of the anomaly
for the corresponding fault/failure mode. In this context, the vector of
anomaly strengths A is considered as a vector of symptoms which is
driving the diagnostics system. In other words, the columns in Table 5
are populated based on the experience and expectancies of the operator
of the compressor with entries H, L, , ic, or nil which stand for High,
Medium, Low, Indirect Consequences and no influence. These values
may be input manually by the manufacturer and/or the operator of the
compressor and these values correspond to the expected anomalies and
their strength when certain faults/failure modes are present in the
compressor. It is noted that a given fault/failure mode may determine
more than one part of the compressor to exhibit anomalies. While
Table 5 includes data specific to a centrifugal compressor, other tables
may include data relevant to another machines to be diagnosed.
As an example, consider that the symptoms noted in the first row of
Table 5 represent the various components fi, and the Fault i row is one
fault/failure mode of the compressor. For such a fault/failure mode it is
expected, according to this exemplary embodiment, that values for the
Symptom 1 and Symptom k are L. In other words, an algorithm
implemented on a microprocessor identifies the fault/failure mode Fault
i with low (L) confidence if the two parameters Symptom 1 and k are
outside of the threshold value and all other parameters are insignificant.
It is noted that this is an example and each compressor may have its
own parameters and characteristics monitored as desired.
The faults/failure modes noted in the first column of Table 5 may be
identified based on rules defined for each fault/failure mode. Entries H,
L, M, and ic may be interpreted as the expected strengths of the
symptoms for each corresponding fault/failure mode.
The matrix shown in Table 4 may be interpreted as a set of association
rules having elements Rule,., with
where Y is a fault/failure mode, for example, the elements listed in the
first column in Table 5, and Xi represents a plurality of symptoms, for
example, the entries in each row in Table 5.
The anomaly vector A is matched with each fault/failure mode in the
matrix and the it row in the matrix of Table 5 is interpreted as the
association rule Rule*, where Rule is given by
, the Left Hand Side (LHS) of association rule Rulei is given by:
and where each component Xi,j of Xi has one of five values H, M, L, ic
or nil with nil being zero. It is noted that vector Xi has k components
and k is equal to the number of columns in Table 5 and i is the number
of rows in Table 5.
For each fault/failure mode Y (row in Table 5), the diagnostic unit 46
may be configured to compute a measure of corroborating evidence
C(Yi, A) (the expected symptoms for Yi to match the anomaly vector
A), and a measure of refuting evidence R(Yi, A) (lack of match between
the expected symptoms for Yi and the anomaly vector A). There are
multiple ways of computing these two measures. For a better
understanding, exemplary functions for C and R are next discussed.
The anomaly vector, which is the input for the diagnostics process, is
defined as :
A=[A1,A2 ,...Ak ]Î[0,1],
where each value Ai has a value between zero and one.
To account for the partial match of the inputs with their corresponding
references, taking into account the High (H), Medium (M), or Low (L)
probability of observing that symptom, or the fact that it might be an
indirect consequence (ic), the LHS , of association rule Rulei may be
decomposed into:
X
where X " binary indicator of the value H :
etc.
Further, a set of weights W reflecting the importance of matching
inputs with the entries H, M, L, ic, nil of Table 5 is introduced as
follows:
W = { , W , W ,W
ic ,W
nil }
with W being, for example, {0.8, 0.5, 0.2,0.1, 0.0}. Based on this
example, the corroborating evidence measure C can be computed to
account for the amount of evidence expected for a fault/failure mode Yi
and the anomaly A. In other words, C may be the weighted, partial
match between the anomaly vector A and , the LHS of rule R /e, , as
given by:
where the inner product is implemented using the traditional
scalar product Note that A =[A1,A2,...AK] where A [0,1], while
To exemplify the calculation of a fault/failure mode Yi, the example
shown in Table 6 is discussed next. Assume that the anomaly vector A
has the values shown in the first row of Table 6. Also assume that the
rule (a row not shown in Table 5) is given by Xi as shown in the second
row of Table 6. The rule i in the second row is decomposed into
subrules that correspond to the H, M, and L values as shown in the
third, fourth, and fifth rows of Table 6. As no ic or nil values are
present for the fault/failure mode Yi, corresponding rules are not shown
in Table 6. While the values of anomaly vector A are calculated based
on (i) measured parameters of the compressor and (ii) predetermined
constraints as discussed above with regard to Tables 1-4, the rule Xi is
likely determined based on the experience of the operator of the
compressor and/or the manufacturer of the compressor. Using the
corroborating evidence measure C defined above and the weights noted
above, the value 82% is determined for the fault/failure mode Yi.
Table 6
Another example that has more symptoms is shown in Table 7 . Here
there are symptoms with a low value L besides the high H and medium
M values. The same weighting factors and the same corroborating
evidence measure C are used for Table 7 to arrive at a 50.7% match
between the fault/failure mode Y and the anomaly vector A.
Table 7
As also discussed above, it may be relevant to calculate the amount of
evidence expected to see for a fault/failure mode Y missing from
anomaly vector A. A complimentary metric to C may be used, such as
R(Yi, A) = 1 - C(Yi, A). However, this metric R does not provide any
additional information.
A better measure is to compute a discount factor based on the
probability of Y being the cause of the faults/failure modes without
observing the expected symptoms. For computational simplicity it is
assumed that all symptoms are independent from each other. This is
equivalent to taking the negation of probabilistic sum of the missing
value in each degree of anomaly (i.e., complement to one), each
weighted by its corresponding weight. The probabilistic sum is one of
many Triangular conorms (T-conorms) that can be used to evaluate the
union, and it is defined as: S(a,b) = a + b - a-b. Like all T-Conorms,
this function is associative. This means that when there are 3 or more
arguments, the function can be computed recursively, i.e., S(a,b,c) =
S(S(a,b), c).
For example, in the example presented in Table 6, the following
missing values are present: for high strength anomaly (WH =0.8) there
is 0 .1 in A2 and 0.2 in A3, and for low strength anomaly (WL =0.2) there
is 0.8 in A and 0.2 in A6. Based on these values, the measure of
refuting evidence R(Yi, A) may be calculated as:
S( S( S(0.8 · 0 .1, 0.8 · 0.2), 0.2 · 0.8), 0.2 · 0.5) = 0.415 when S(a,b) = a + b - a b.
If a different T-conorm is used, such as S(a,b) = Max(a, b), then the
above expression becomes:
S( S( S(0.8 · 0.1, 0.8 · 0.2), 0.2 · 0.8), 0.2 -0.5) = 0. 16.
By computing the measure of refuting evidence using the maximum
operator instead of the probabilistic sum, a less drastic view of the
missing information is taken.
There are many parameterized families of T-conorms known in the art.
Other T-conorms would be suitable to compute the refutation measure.
The selection of the most appropriate T-conorm is driven by the
conservative or liberal position that is desired to be taken in grading the
impact of the missing information in the overall diagnostic process.
According to an exemplary embodiment, the processor in which the
diagnostic method is implemented may be configured to offer the
operator of the processor which function to choose for the refutation
measure. As one skilled in the art would recognize, the process
described above may be implemented in hardware, software or a
combination thereof.
According to an exemplary embodiment, the C and R measures may be
combined and an overall result may be presented as the interval formed
by [C(Yi, A), max(C(Yi A), 1 - R(Yi, A))] or by aggregation of these
two measures C and R in which 1 - R(Yi, A) is used to discount
C(Yi, A).
For example, with regard to the example shown in Table 6, C(Yi, A) =
0.82, R(Yi, A) = 0.415 (when using S(a, b) = a + b - a-b), and 1 -
R(Yi, A) = 0.585. The interval noted above is actually a point [0.82,
0.82] and the discounted score is C(Yi, A)·(1 - R(Yi, A)) = 0.48.
Using the interval or the discounted factor, it is possible to rank the
potential fault/failure modes to form an ordered list based on which it is
possible to suggest correcting actions for the most likely fault/failure
modes.
The process of detecting the fault/failure mode may be represented as
shown in Figure 5. In step 500, the processor on which the process is
implemented receives measurement data. The measurement data is
indicative of various parameters of the compressor. A feature vector F
is generated in step 502. The feature vector F may have k components.
The number of components may be changed by the operator of the
compressor. Some of the components are the measured data and some
other components are calculated based on the measured data. In step
504, which may be performed prior to steps 500 and 502, fuzzy
thresholds as discussed above are defined for each components of the
feature vector F. Depending on the situation, the fuzzy thresholds may
account for an increase or decrease of the component. In one
application, a combination of fuzzy thresholds and step thresholds may
be used for the components of the feature vector F.
Based on the fuzzy thresholds and the feature vector F, an anomaly
vector A is calculated in step 506. The procedure for calculating the
anomaly vector may differ from application to application as already
discussed above. A set of rules is retrieved in step 508 from a database.
The set of rules is previously stored in the database or may be changed
live by the operator. The set of rules maps a set of faults/failure modes
to corresponding symptoms. For example, as shown in Table 5, the
first column identifies plural faults/failure modes and for each
fault/failure mode a set of symptoms (listed in the corresponding row)
are expected. The considered symptoms are listed in the first row of
Table 5 and their expectancy for a given fault/failure mode in row "n"
are listed in row "n". As seen in Table 5, not all the symptoms are
present for a given fault/failure mode, some have a high expectancy,
some have a low expectancy and some have a presence but not directly
related to the fault/failure mode. For these reasons, the expectancy to
see a symptom for a given fault/failure mode is quantified as high,
medium, low, null, and indirectly related. It is noted that according to
an exemplary embodiment, the set of rules are manually entered by a
technician who is familiar with the operation and failures of the
compressor.
Based on the rules and the anomaly vector A, a corroborating evidence
measure C is calculated in step 510. This measure provides a reliance
about an identified fault/failure mode to correspond to detected
symptoms. The measure may be expressed as a percentage, with 100%
being a fully reliable result. However, according to optional step 512, a
refuting evidence measure R may also be calculated and this measure
indicates the lack of reliability of the identified fault/failure mode.
According to another optional step 514, the two measures C and R may
be combined to produce in step 16 a more reliable ranking of the
possible fault/failure mode corresponding to a set of detected symptoms.
According to an exemplary embodiment illustrated in Figure 6, there is
a method for diagnosing a compressor. The method includes a step 600
of generating a feature vector of the compressor, the feature vector of
the compressor including components describing states of various
components of the compressor; a step 602 of determining, based on
fuzzy constraints, an aggregated anomaly vector corresponding to the
feature vector; a step 604 of defining rules for a preset list of possible
diagnostics of the compressor; a step 606 of calculating a corroborating
measure between the aggregated anomaly vector and the rules; and a
step 608 of identifying a diagnostic of the compressor based on a result
of the corroborating measure.
Optional steps may include calculating a measure of refuting evidence
R that quantifies an amount of evidence that is missing for a
fault/failure mode, where the measure of refuting evidence R may be
based on a triangular conorm function; or combining the corroborating
measure C with the measure of refuting evidence R to rank each
fault/failure mode of the compressor.
According to an exemplary embodiment, a system including a processor
may be configured to receive values of the components of the feature
vector over a predetermined time window, apply the fuzzy constraints
to the components of the feature vector to determine corresponding
instant anomalies at certain times during the predetermined time
window, and aggregate the corresponding instant anomalies at the
certain times to generate the aggregated anomaly vector for the entire
predetermined time window and for all the components of the feature
vector.
According to another exemplary embodiment, the processor is further
configured to divide each rule into subrules, calculate a scalar product
between each subrule and the aggregated anomaly vector, and
determine a percentage value that is indicative of a fault/failure mode
experienced by the compressor based on the calculated scalar products
which form the corroborating measure.
Additionally or alternatively, the processor is further configured to use
a corroborating measure C between each rule and the aggregated
anomaly vector, where C is defined as
with A being the aggregated anomaly vector, Y representing a fault/failure
mode, Xi representing rule "i", Wh, wm, wi, and W C being weighting
factors, and Xi
H
, Ci , Xi
L and Xi
i being the subrules of rule Xi.
According to another exemplary embodiment, the processor is further
configured to calculate a measure of refuting evidence R that quantifies
an amount of evidence that is missing for a fault/failure mode, where
the measure of refuting evidence R is based on a triangular conorm
function. The processor may be further configured to combine the
corroborating measure C with the measure of refuting evidence R to
rank each fault/failure mode of the compressor.
According to an exemplary embodiment, a computer readable medium
includes computer executable instructions, where the instructions, when
executed, implement a method for diagnosing a compressor. The
method includes generating a feature vector of the compressor, the
feature vector of the compressor including components describing
states of various parts of the compressor; determining, based on fuzzy
constraints, an aggregated anomaly vector corresponding to the feature
vector; defining rules for a preset list of possible faults/failure modes of
the compressor; calculating a corroborating measure between the
aggregated anomaly vector and the rules; and identifying a
faults/failure mode of the compressor based on a result of the
corroborating measure.
For purposes of illustration and not of limitation, an example of a
representative system capable of carrying out operations in accordance
with the exemplary embodiments is illustrated in Figure 7. The
processor discussed above for implementing the diagnostic process may
be part of the system. Hardware, firmware, software or a combination
thereof may be used to perform the various steps and operations
described herein.
The exemplary system 700 suitable for performing the activities
described in the exemplary embodiments may include server 701 . Such
a server 701 may include a central processor (CPU) 702 coupled to a
random access memory (RAM) 704 and to a read-only memory (ROM)
706. The ROM 706 may also be other types of storage media to store
programs, such as programmable ROM (PROM), erasable PROM
(EPROM), etc. The processor 702 may communicate with other
internal and external components through input/output (I/O) circuitry
708 and bussing 710, to provide control signals and the like. For
example, the system 700 may communicate with plural compressors for
monitoring any symptoms that might appear. The plural compressors
may be distributed over a large geographical area while the system 700
may be a centralized system. The processor 702 carries out a variety of
functions as is known in the art, as dictated by software and/or
firmware instructions.
The server 701 may also include one or more data storage devices,
including hard and floppy disk drives 712, CD-ROM drives 714, and
other hardware capable of reading and/or storing information such as
DVD, etc. In one embodiment, software for carrying out the above
discussed steps may be stored and distributed on a CD-ROM 716,
diskette 718 or other form of media capable of portably storing
information. These storage media may be inserted into, and read by,
devices such as the CD-ROM drive 714, the disk drive 712, etc. The
server 701 may be coupled to a display 720, which may be any type of
known display or presentation screen, such as LCD displays, plasma
display, cathode ray tubes (CRT), etc. A user input interface 722 is
provided, including one or more user interface mechanisms such as a
mouse, keyboard, microphone, touch pad, touch screen, voicerecognition
system, etc. The operator of the system 700 may input any
information via the interface 722, as for example, modifying the
number of components of the feature vector F.
The server 701 may be coupled to other computing devices, such as the
landline and/or wireless terminals and associated watcher applications,
via a network. The server may be part of a larger network
configuration as in a global area network (GAN) such as the Internet
728, which allows ultimate connection to the various landline and/or
mobile client/watcher devices.
The disclosed exemplary embodiments provide a system, a method and
a computer program product for determining a fault/failure mode of a
compressor. It should be understood that this description is not
intended to limit the invention. On the contrary, the exemplary
embodiments are intended to cover alternatives, modifications and
equivalents, which are included in the spirit and scope of the invention
as defined by the appended claims. Further, in the detailed description
of the exemplary embodiments, numerous specific details are set forth
in order to provide a comprehensive understanding of the claimed
invention. However, one skilled in the art would understand that
various embodiments may be practiced without such specific details.
As also will be appreciated by one skilled in the art, the exemplary
embodiments may be embodied in a wireless communication device, a
telecommunication network, as a method or in a computer program
product. Accordingly, the exemplary embodiments may take the form of
an entirely hardware embodiment or an embodiment combining
hardware and software aspects. Further, the exemplary embodiments
may take the form of a computer program product stored on a
computer-readable storage medium having computer-readable
instructions embodied in the medium. Any suitable computer readable
medium may be utilized including hard disks, CD-ROMs, digital
versatile disc (DVD), optical storage devices, or magnetic storage
devices such a floppy disk or magnetic tape. Other non-limiting
examples of computer readable media include flash-type memories or
other known memories.
Although the features and elements of the present exemplary
embodiments are described in the embodiments in particular
combinations, each feature or element can be used alone without the
other features and elements of the embodiments or in various
combinations with or without other features and elements disclosed
herein. The methods or flow charts provided in the present application
may be implemented in a computer program, software, or firmware
tangibly embodied in a computer-readable storage medium for
execution by a specifically programmed computer or processor.
This written description uses examples of the subject matter disclosed
to enable any person skilled in the art to practice the same, including
making and using any devices or systems and performing any
incorporated methods. The patentable scope of the subject matter is
defined by the claims, and may include other examples that occur to
those skilled in the art. Such other example are intended to be within
the scope of the claims.

CLAIMS
1. A method for diagnosing a compressor, the method comprising:
generating a feature vector of the compressor, the feature vector
of the compressor including components describing states of various
parts of the compressor;
determining, based on fuzzy constraints, an aggregated anomaly
vector corresponding to the feature vector;
defining rules for a preset list of possible faults/failure modes of
the compressor;
calculating a corroborating measure between the aggregated
anomaly vector and the rules; and
identifying a fault/failure mode of the compressor based on a
result of the corroborating measure.
2. The method of Claim 1, wherein the step of generating a feature
vector comprises:
measuring first plural parameters of the compressor; and
estimating second plural parameters of the compressor based on
the first plural parameters,
wherein the feature vector includes a part of the first plural
parameters and the second plural parameters.
3. The method of Claim 1, wherein the various parts of the
compressor include bearings and a rotor and the states include at least
one of pressure, temperature, amplitude, differential pressure, mass or
volumetric flow or rotor speed.
4. The method of Claim 1, further comprising:
defining the fuzzy constraints for deviations of the components
of the feature vector from reference values, wherein the fuzzy
constraints are thresholds that change continuously from a low value to
a high value.
5. The method of Claim 4, wherein at least one fuzzy constraint is
defined by an attention value and a risk value, the attention value
indicating that a corresponding parameter needs to be monitored as an
anomaly is likely to occur, and the risk value indicating that the
anomaly has occurred.
6. The method of Claim 1, wherein the step of determining an
aggregated anomaly vector comprises:
receiving values of the components of the feature vector over a
predetermined time window;
applying the fuzzy constraints to the components of the feature
vector to determine corresponding instant anomalies at certain times
during the predetermined time window; and
aggregating the corresponding instant anomalies at the certain
times to generate the aggregated anomaly vector for the entire
predetermined time window and for all the components of the feature
vector.
7. The method of Claim 1, wherein the calculating step comprises:
dividing each rule into subrules;
calculating a scalar product between each subrule and the
aggregated anomaly vector; and
determining a percentage value that is indicative of a
fault/failure mode experienced by the compressor based on the
calculated scalar products which form the corroborating measure.
8. The method of Claim 7, further comprising:
using a corroborating measure C between each rule and the
aggregated anomaly vector, where C is defined as
with A being the aggregated anomaly vector, Y, representing a
fault/failure mode, Xi representing rule "i", Wh, wm, wi, and WiC being
weighting factors, and Xi , Xi , Xi
L and Xi
i being the subrules of rule
Xi.
9. The method of Claim 1, wherein the identifying step comprises:
calculating a measure of refuting evidence R that quantifies an
amount of evidence that is missing for a fault/failure mode.
10. A system for diagnosing a compressor, the system comprising:
an interface configured to receive measurement data about the
compressor; and
a processor configured to receive the measurement data and to,
generate a feature vector of the compressor based on the
measurement data, the feature vector of the compressor including
components describing states of various parts of the compressor,
determine, based on fuzzy constraints, an aggregated
anomaly vector corresponding to the feature vector,
retrieve rules for a preset list of possible faults/failure
modes of the compressor,
calculate a corroborating measure between the aggregated
anomaly vector and the rules, and
identify a fault/failure mode of the compressor based on a
result of the corroborating measure.