CN1782672A - Method and apparatus for improved fault detection in power generation equipment - Google Patents
Method and apparatus for improved fault detection in power generation equipment Download PDFInfo
- Publication number
- CN1782672A CN1782672A CN 200510109895 CN200510109895A CN1782672A CN 1782672 A CN1782672 A CN 1782672A CN 200510109895 CN200510109895 CN 200510109895 CN 200510109895 A CN200510109895 A CN 200510109895A CN 1782672 A CN1782672 A CN 1782672A
- Authority
- CN
- China
- Prior art keywords
- sensor
- value
- confidence
- estimated
- absolute difference
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
A method and apparatus for detecting faults in power plant equipment is discloses using sensor confidence and an improved method of identifying the normal operating range of the power generation equipment as measured by those sensors. A confidence is assigned to a sensor in proportion to the residue associated with that sensor. If the sensor has high residue, a small confidence is assigned to the sensor. If a sensor has a low residue, a high confidence is assigned to that sensor, and appropriate weighting of that sensor with other sensors is provided. A feature space trajectory (FST) method is used to model the normal operating range curve distribution of power generation equipment characteristics. Such an FST method is illustratively used in conjunction with a minimum spanning tree (MST) method to identify a plurality of nodes and to then connect those with line segments that approximate a curve.
Description
The application number that the application requires the U.S. to apply on August 25th, 2004 is 60/604374 right of priority, and its full content is incorporated by reference at this paper.
Background technology
Large-scale machine is such as generating set, from purchase, installation, all very expensive to safeguarding, working.Therefore, determine that whether this kind equipment works is very important in desired running parameter scope.It also is very important that the situation that indication equipment is worked outside these desired parameters detects, and wherein equipment is worked outside these desired parameters and can be caused equipment impaired.In order to detect this class situation, adopt sensor to come the testing parameter usually, such as the pressure of various parts, temperature or the like,, determine to have taken place fault so if concrete measured value has surpassed the predetermined threshold of a certain special parameter.Recently, go to improve the accuracy whether detection failure exists more at large about the method for fault measuring system.Technique known such as neural network, polynary state estimation method (MSET) and fuzzy logic, has been used for above-mentioned purpose.These class methods all are to adopt the historical data of being collected by a plurality of sensors, represent past operate as normal and fault state, produce a model and are used for the following data that monitoring equipment work produces.If following data and historical data model bias are too many, can give a warning so and determine to have produced fault.
Fault detection method of the prior art depends on historical data usually, produces the estimated value of the working value that observes, and wherein these working values are that expectation detects with specific sensor.Then, use sensor real work value, and compare with estimated value.Then, the difference between sensor remainder (residue) or estimated value and the observed reading is calculated,, determined to have produced fault so if this remainder is higher than desired threshold value.Yet in the previous sensor estimation technique of this class, the estimated value of particular sensor has been adopted fault sensor and the influence of the measured value that obtains usually.Particularly, typical prior estimate method depends on several sensors and measures the measured value of same characteristic (measuring the leaf temperature of a turbine as a plurality of sensors) comes from each sensor expectation value with generation estimated value.The measured value that this class derives from several sensors is called vector at this paper.Usually make the error minimize between estimated value and the initial value in these methods, and the trend with the diffusion of the deviation between the value of individual sensor in all sensors is arranged.As a result, if a sensor is out of order, can produce obvious errors in its measured value, this error will be distributed in the sensor that does not all have fault, therefore will reduce the accuracy of the comprehensive estimated value of each sensor.This error profile is called at this paper overflows effect.
In order to reduce this overflowing, various estimation algorithms have been adopted, such as the method for utilizing known gradient decreasing function to find the solution.The example of these class methods can be referring to " the Robust Statistics " of R.J.Huber, and Wiley-Interscience published in 1981.Yet these methods need be selected controlled variable, come control function with speed convergence how soon.Select this controlled variable to be actually very difficult.In addition, these class methods are tended to concentrate restrain at leisure to obtain one and are optimized estimated value, are impracticable in many job applications therefore.Other is attempted to reduce this trial of overflowing effect and also comprises homing method, returns (kernel regression) or polynary state estimation method (MSET) such as known kernel function.These class methods in November, 2000 Washington hold about the international theme meeting of nuclear power plant equipment, control and human-machine interface technology on detailed description is arranged in " Use of Kernel BasedTechniques for Sensor Validation " article of delivering by A.V.Gribok, J.W.Hines and R/E.Uhrig, this piece article is incorporated by reference at this paper.Yet these homing method calculated amount are very big, the recurrence the number networks that need equate with number of sensors.In addition, this regression model also is inaccurate when fault sensor is arranged.
Summary of the invention
The invention provides a kind of method and apparatus that utilizes fault in sensor degree of confidence (sensor confidence) checkout equipment, a kind of improving one's methods of generating set normal range of operation of determining when utilizing these sensors to measure also is provided.
Particularly, according to an embodiment of the invention, give the degree of confidence of sensor assignment proportional with the remainder of this sensor.If sensor has high remainder, distribute to the very little degree of confidence of this sensor so.If sensor has very low remainder, distribute to the very high degree of confidence of this sensor so, and give this sensor and the suitable weighting of other sensor settings.Subsequently with this degree of confidence produce to the power generation equipment characteristics observed reading through the finishing estimated value.
According to another embodiment of the present invention, adopt feature space track (FST) method to simulate the normal operating range curve distribution of power generation equipment characteristics.Specifically, be this FST method and minimal span tree (MST) method to be combined to make be used for determining a plurality of points, then these points are coupled together with the line segment that is roughly curve.In case this curve roughly forms, the method for detecting sensor degree of confidence as previously discussed, can be used for determining and improving the sensor estimated value.
For a person skilled in the art, by with reference to following detailed description and accompanying drawing, can know and understand these and other advantage of the present invention.
Description of drawings
Fig. 1 shows sensor vector and estimates synoptic diagram and the observed reading of utilizing monitoring system to collect, and these estimator/values are how to compare with the turbogenerator normal range of operation of giving an example;
Fig. 2 shows the schematic chart of sensor confidence value function for example, and wherein this function is used for giving the sensor assignment degree of confidence according to the principle of the invention;
Fig. 3 shows the method according to one embodiment of the present invention, the estimated value X that it utilizes the sensor confidence value function shown in Fig. 2 to be improved;
Fig. 4 shows illustrative features space tracking (FST) simulation of the normal range of operation of a certain characteristic of generating set;
Fig. 5 shows a kind of method in accordance with the principles of the present invention, and simulation improves to the FST among Fig. 4 for it;
Fig. 6 shows and has adopted the feature space track of the training data of curve representation among Fig. 4, wherein shows three centroid position V ' for example
1, V '
2, V '
3, and utilize minimal span tree (MST) method to couple together;
Fig. 7 shows the synoptic diagram of computing machine, and this computing machine is suitable for the calculating sensor confidence value and/or carries out and calculate and definite method relevant with FST/MST, and this method is used for determining the normal range of operation of generating set.
Embodiment
Fig. 1 shows sensor vector and estimates synoptic diagram and the observed reading of utilizing monitoring system to collect, and how these estimation/values to compare with normal range of operation, for example, and the temperature of the parts in the power generation turbine.Particularly, with reference to 1, two sensor of figure, this paper specified sensor 1 and sensor 2 are parts of monitoring system.For example sensor 1 is the fault sensor, and sensor 2 does not have fault.These sensors for example are to be provided for monitoring aforementioned turbine rotor characteristic, and schematically for instance, operating characteristic is the blade path temperature of blade in the turbine.Those skilled in the art will appreciate that as seeing that from Fig. 1 a plurality of temperature measured values such as the measured value that is obtained by sensor 1 and sensor 2, can be represented with vector 102 and 103 respectively.Specifically, the measured value of vertically representing respectively with the vector 102 and 103 of horizontal direction for example is blade path temperature measured value x
1(recording) and x by sensor 1
2(recording) by sensor 2.Therefore, the simple one dimension zone that measured value forms two-dimensional diagram rather than adopts the temperature measured value that is drawn by single-sensor, wherein this two-dimensional diagram is the temperature measured value x that obtains from sensor 1 and sensor 2 respectively
1And x
2Function.Therefore, each a some vertical component of expression and a horizontal component among Fig. 1, wherein this vertical component is the one or more measured values that recorded by sensor 1, horizontal component is the one or more measured values that recorded by sensor 2.
Normal range of operation 101 is to represent for example to be the curve of the normal range of operation of generating set, and determining by known method, is to utilize the sensor that is located on the generating set desired position to collect the historical data relevant with this equipment work in this known method.These data or partial data are used for estimating, and adopt known statistical model method to describe out the normal range of operation of this equipment.In the equipment work process, if the normal range of operation that the obvious diasporometer of a certain measured value is calculated can determine to have occurred fault so.To further discuss hereinafter the definite of generating set normal operating range curve 101.
See Fig. 1 once more,, can obtain the measured value of operating characteristic (such as temperature) subsequently by sensor 1 and sensor 2 in case determined normal range of operation 101.Perhaps in other words vector x 107 expressions, are the values of real work blade path temperature by the position of the perfect estimation value of the temperature value of sensor 1 and sensor 2 acquisitions.Yet, suppose that once more sensor 1 is a fault sensor, thereby its measured value is inaccurate.As shown in Figure 1, in this case, these are observation vector y110 by sensor 1 and 2 measured values that obtain.This point can see that vectorial y110 vertically is offset a certain amount of from perfect estimation value x107 position, and this is the error that is caused by fault sensor 1.This vertical shift is called the sensor remainder of sensor 2 in this article, and directly cause by the fault in the sensor 1.
In the fault monitoring system formerly,, will make any error minimize in the measured value so usually as possible in case measure observed reading such as sensor vector y.This effort generally includes observation vector is mapped on the point nearest in normal range of operation, with this closest approach as actual measured value.With reference to figure 1, according to this method, by determining to arrive on the normal range of operation 101 closest approach of vectorial y110, the error minimize that vectorial y110 is represented with respect to the equipment normal range of operation.This closest approach is used
Expression, it is the point of distance vector y110 bee-line 104 on normal range of operation 101.The point
Depart from perfect estimation value 107 distances 105 and 106 along the vertical and horizontal directions respectively.Those skilled in the art can observe, and the original vectorial y110 that observes is not with respect to the skew of x107 along continuous straight runs, point
Offset distance 106.Distance 106 is called the overflow error of sensor 2 in this article, and as previously discussed, this overflow error is the error that is caused in the operate as normal measurement value sensor by fault sensor 1.In this situation, the vectorial y110 that effort will observe is mapped on the normal range of operation, can directly cause overflowing, and this overflows is because the mistake that fault sensor 1 causes.
According to principle of the present invention, this overflow problem can roughly be resolved.Specifically, according to an embodiment of the invention, give the sensor assignment degree of confidence, this degree of confidence is directly proportional with the relevant remainder of this sensor.If this sensor has very high remainder, distribute to the very little degree of confidence of this sensor so.If sensor has very low remainder, then distribute to the very high degree of confidence of this sensor, and offer this sensor and other sensor proper weight.Specifically, give the degree of confidence of i sensor definition, w
i, be:
w
i=g (d
i) equation 1
W wherein
iBe the degree of confidence of i sensor, d
iBe the observation sensor value of i sensor and the normalized absolute difference between the estimated sensor value.Concerning a specific sensor, along with the value and the increase of the difference between the estimated value of this sensor, the remainder of this sensor also increases.Specifically, d
iAs give a definition:
Wherein, emphasize once more,
It is the sensor vector estimated value of utilizing in all sensors combination that the conventional statistics model estimates;
Be to utilize this model, the sensor vector estimated value in i sensor; y
iIt is the observation sensor vector of i sensor; Y is the observation sensor vector that records from all sensors combination.Normalized absolute difference is used for reducing the different sensors remainder by normalized influence.Fig. 2 shows the schematic chart of confidence value function 201, and wherein this function is used for the method for the description according to the present invention to the sensor assignment degree of confidence.Referring to Fig. 2, it will be understood by those skilled in the art that the degree of confidence g (d that distributes to sensor along Z-axis 202 once more
i) be one along with d
iIncrease and from 1 to 0 function that reduces is represented with transverse axis 203.Specifically, Fig. 2 shows the schematic confidence value function g (d) with following equation definition:
G (d)=exp (γ d
2) equation 3
Wherein, d such as front are defined, and γ is selected convergency value, wherein γ＜0.The chart that goes out as shown in Figure 2, γ are to select according to the mode of g (1)=0.001.
By adopting such confidence value function, can obtain through upgrade, sensor vector estimated value more accurately
Specifically, the sensor vector estimated value of this improvement of i sensor
Can calculate by following formula:
Wherein, the variable in the equation 4 as previously mentioned.This point as can see from Figure 1, the new that upgrades is close to perfect estimation value x significantly in the horizontal direction, is therefore overflowed effect and can be reduced greatly by what fault sensor 1 caused.
Fig. 3 shows the method according to one embodiment of the present invention, and equation above wherein utilizing and the sensor confidence value function among Fig. 2 have obtained the estimated value x through improving.Especially, with reference to figure 3, in step 301, for calculating observation sensor values y and sensor estimated value
Between normalized absolute difference d
i, observation sensor vector y is input in the equation 2.Next, in step 302, this d as calculated
iValue utilizes the confidence value Function Mapping shown in Fig. 2 to particular confidence level value w
i=g (d
i) on, just as discussed above like that.In case determined w
iValue, in step 303, observation sensor vector y
iAnd initial estimate
Be input in the equation 4 to obtain value as previously discussed
In case for each sensor has calculated this value, in step 304, to new
Calculate, this value is the estimated value of observation sensor vector improvement, and wherein this observation sensor vector is to have considered to reduce the degree of confidence of distributing to sensor 1 and by refine.Then, in step 305, the value that this is new
Be input to subsequently in the statistical model to determine value new, through upgrading
In step 306, need determine whether new
Calculate well to previous
Between distance less than desired threshold value.If less than, so in step 307, current value
Then be used as the optimum value of perfect estimation value x.If opposite, in step 305, new
Good with previous calculating
Between distance greater than desired threshold value, program will be got back to step 301 and with this new value so
Calculate the d of renewal according to equation 2
i, new
Value.According to the program of proceeding described above up to current value
To preceding value
Between distance less than desired threshold value.According to this method, because therefore the confidence value of having given sensor assignment can significantly reduce fault sensor overflowing normal working sensor measured value.
In order to ensure above-described sensor degree of confidence method is accurately, must guarantee normal range of operation, as the curve among Fig. 1 101, identification be accurate identification.The present invention has been found that in many examples, and working equipment such as generating set, has the sensor of one or more groups of height associations, such as the blade path temperature sensor in the aforementioned turbine.These sensors are known as highly related, be because these sensors are to be positioned on the known location physically with respect to another sensor, and the measured value by another sensor can relatively accurate ground, the following measured value of a sensor is predicted on error ground.Those skilled in the art can find, because this incidence relation, the distribution of the sensor of any a pair of height association on two-dimensional space is all as individual curve.Because this incidence relation, can suppose that the distribution of the sensor vector be made up of the measured value of these sensors also is a curve.In order to form the normal operating range curve of curve and then acquisition generating set according to historical measurement value sensor, known use principle linearity curve and same known variation method have been adopted.Especially, this method comprises the curve of determining to pass in the sensor vector space training data center.Yet this method usually is ill-considered, because they can not converge desired curve rightly, especially when curve is complicated shape.As a result, the specific curves of expression generating set normal range of operation is difficult to determine sometimes.
Therefore, the inventor has realized that: the feature space method of loci can be used for simulating the normal operating range curve distribution of power generation equipment characteristics.These characteristic space method in other field also by common general knowledge, for example image recognition, therefore, only described here to making everybody understand the degree of the principle of the invention.Usually, these FST methods a plurality of nodes are discerned and these nodes are connected with the line segment of curve of approximation aspect helpful.Fig. 4 has provided a kind of FST simulation that is used for for example, and we will further inquire into it hereinafter, and in this example, simulation comprises three line segment: v
1v
2, v
2v
3And v
3v
4As mentioned above, reuse test input vector y and add up the value of obtaining
In this example,
Be to make the estimated value that produces minor increment 401 between y and the line segment.As discussed earlier, in case determined value
So also just determined the sensor degree of confidence, the sensor estimated value of improvement also can iteration form.
In order to calculate FST, the FST among Fig. 4 400 for example, typically, must know among Fig. 4 such as v
1, v
2, v
3And v
4Node, and will know and the order of these nodes these nodes could be connected successively like this, a curve formed.Yet in the present example, node in the training data and their order all are ignorant, therefore must obtain these information from training sensor vector data set.Therefore, the inventor has realized that: k mean cluster (k mean clustering) can be used for determining to be present in the sensor vector in the training data, and then obtains being used for the node of FST.In this k mean cluster, by the distance between data point in the specified data group (for example sensor training data) and the centroid position and according to each point and the minor increment between barycenter data point is divided into groups, can determine a plurality of centroid positions in one group of data.This cluster is known, is not described further here.
In case determined node, then must determine the order of node and they are linked in sequence according to this.In the embodiment, adopt minimal span tree (MST) algorithm to finish this task in accordance with the principles of the present invention.As known, the MST algorithm helps to connect a plurality of points according to the shortest mode of summation (span) that connects length.This result plots tree-shaped curve map usually.Yet in this example, desired tree will be simulated the normal range of operation of generating set.Because like this, the inventor thinks, applies certain restriction by the function of giving the MST rule, and the node that previously described FST method is produced couples together so probably, and normal operating range curve is carried out modelling accurately.Especially, illustrated among Fig. 5: the method according to one embodiment of the present invention is improved the FST among Fig. 4.Specifically, in step 501, initial k the node that will connect determined k=3 in this example.The initial barycenter number of determining is corresponding in this number and the training data, and this number also is necessary minimum counting of curve of simulation (two points can only be linked to be a straight line that connects these two points).Next in step 502, use the k mean cluster and determine three centroid positions.With reference to figure 6, adopt training data by curve representation among Fig. 4, schematically determined three centroid position v '
1, v '
2And v '
3Then,,, these k nodes are used the MST algorithm so that they are linked in sequence, in Fig. 6, represent with line 601 and 602 as foregoing in step 503.Yet, can form curve in order to ensure these nodes, in step 504, determine whether two end points belong to (promptly being only a node to be connected with another node) on one side.If belong to a limit, in step 505, determine all to remain node (promptly two mid-side node between) and whether belong to two limits (only that is to say two other nodes are connected them). Step 504 and 505 effect are to guarantee that these nodes can form curve model rather than other shape, such as having the single separately number of section.If in step 505, determined that each intermediate node belongs to two limits, so in step 506, whether the angle θ that further determines to form between the adjacent edge is greater than predetermined angular, such as 30 degree.This is to prevent that the training data border that MST forms has jagged edge.If in step 506, determined so in step 507, k to be revised as k=k+1 for being, k=4 under this situation, program turns back to step 502.In this case, when k=4, schematic FST will be improved to and have line segment v among Fig. 4
1v
2, v
2v
3And v
3v
4To one skilled in the art obviously, many more in accordance with the node number that limits above, then the estimated value of the normal range of operation 101 of Fig. 1 will be accurate more.Therefore, top program is along with increase the carrying out repeatedly of k value, and a step in step 504,505 or 506 is defined as not.In this case, in step 508, the result of MST program is as the final ecbatic of being determined the normal range of operation of generating set by training data.
It will be appreciated by those skilled in the art that, utilize the monitoring system of the FST/MST method of sensor confidence value and/or detection generating set normal range of operation, as described above, can carry out on programmable computing machine, this computing machine is suitable for the computer program step and calculates confidence value function and/or FST/MST.With reference to figure 7, this monitoring system 700 can be carried out on any suitable computing machine, and this computing machine is suitable for receiving, store and transmit the data such as aforementioned telephone directory information.Especially, for example schematically monitoring system 700 has processor 702 (perhaps a plurality of processor), whole courses of work of its Control and Monitor System 700.This work is limited and is carried out by processor 702 by the computer program instructions that is stored in the storer 703.Storer 703 can be the computer-readable medium of any kind, and is electric, magnetic or optical medium without limits.Yet figure 7 illustrates a memory cell 703, should be understood that memory cell 703 can comprise a plurality of memory cells, the sort memory unit comprises the storer of any kind.Monitoring system 700 also comprises schematic modulator-demodular unit 701 and network interface 704.Monitoring system 700 also comprises storage medium, for example is used to store data and according to the computer program that is suitable for the principle of the invention described above such as computer hard disc driver 705.At last, monitoring system 700 also comprises one or more input/output devices, represents with terminal 706 among Fig. 7, is used for allowing carrying out man-machine conversation with for example technician or data base administrator.Those skilled in the art it will be understood that address monitoring system 700 in fact only is schematically, and the various hardware and software components with same advantage also are suitable for use in accordance with the principles of the present invention in the computing machine.
Should be appreciated that the detailed explanation in front is schematically and illustrative, rather than restrictive, the invention disclosed herein scope is not to determine from above-mentioned detailed description, but in the limit that allows according to Patent Law, determines during accessory rights requires.Should be appreciated that the embodiment that illustrates and describe only is that the principle of the invention is described, the various adjustment of those skilled in the art in not departing from the scope of the invention and spiritual scope all are fine.Those skilled in the art can will carry out the combination of various alternate manners in not departing from the principle of the invention and spiritual scope.
Claims (33)
1, a kind of method is used for improveing the operational data measured value of the device with monitoring system, and described monitoring system comprises a plurality of sensors, and described method comprises:
Reception is from the operational data unit of a sensor in described a plurality of sensors, the measured value of the described device observation condition of described operational data unit's expression;
The confidence value of described sensor is defined as the function of described operational data unit;
The estimated value of described measured value is defined as the function of described confidence value.
2, according to the process of claim 1 wherein that described device comprises generating set.
3, according to the process of claim 1 wherein that the step of described definite confidence value also comprises the following expression formula of calculating:
Wherein
It is the sensor vector estimated value of utilizing all sensors that statistical model combines;
It is the sensor vector estimated value of i sensor; d
iBe the operational data unit of measured value of i sensor and the normalized absolute difference between estimated value; y
iIt is the observation sensor vector at i sensor place.
4, according to the process of claim 1 wherein that the step of estimated value of described definite improvement comprises following calculation expression:
Wherein, w
iIt is the degree of confidence of i sensor;
It is the estimated value after the described measured value improvement;
It is the sensor vector estimated value of i sensor; y
iIt is the observation sensor vector at i sensor place;
5, according to the process of claim 1 wherein that the step of described definite confidence value comprises:
Calculating observation sensor values and estimated sensor value
Between normalized absolute difference;
Described normalized absolute difference is mapped to a confidence value.
6, according to the method for claim 5, wherein said mapping step comprises described normalized absolute difference is mapped on the confidence value function that described function limits with following expression formula:
g(d)＝exp(γd
2)
Wherein, d is normalized absolute difference; γ is the funneling factor of selecting.
7, according to the method for claim 6, wherein γ is that mode according to g (1)=0.001 chooses.
8,, comprise that also the improvement estimated value of utilizing described measured value replaces described operational data unit to represent described observation condition according to the method for claim 1.
9, according to the method for claim 5, wherein calculating observation sensor values and estimated sensor value
Between normalized absolute difference, comprising:
Determine a plurality of nodes in one group of training data, described node is represented described normal range of operation; With
By other node of each node and at least one in described a plurality of nodes being connected the curve that estimates the described device normal range of operation of expression, above-mentioned connection is to carry out according to the minimized mode of the summation of this connection length.
10, according to the method for claim 9, wherein said estimated sensor value
It is the vector that is positioned on the described curve.
11, a kind of device is used for improveing the operational data measured value of the device with monitoring system, and described monitoring system comprises a plurality of sensors, and described device comprises:
Be used for receiving device, the measured value of the described device observation condition of described operational data unit's expression from the operational data unit of a sensor of described a plurality of sensors;
The confidence value of described sensor is defined as the device of the function of described operational data unit;
Described measured value estimated value is defined as the device of the function of described confidence value.
12, according to the device of claim 11, wherein said equipment comprises generating set.
13, according to the device of claim 11, the device of wherein said definite confidence value also comprises the device that calculates following expression formula:
Wherein
It is the sensor vector estimated value of utilizing in all sensors that statistical model combines;
It is the sensor vector estimated value of i sensor; d
iBe the operational data unit of i sensor and the normalized absolute difference between the described estimated value; And y
iIt is the observation sensor vector at i sensor place.
14, according to the device of claim 11, the device of the estimated value of wherein said definite improvement comprises the device that calculates following formula:
Wherein, w
iIt is the degree of confidence of i sensor;
It is the estimated value after the described measured value improvement;
It is the sensor vector estimated value of i sensor; With yi be the observation sensor vector at i sensor place.
15, according to the device of claim 11, the device of wherein said definite degree of confidence comprises:
Be used for calculating observation sensor values and estimated sensor value
Between the device of normalized absolute difference;
Be used for described normalized absolute difference is mapped to device on the confidence value.
16, according to the device of claim 15, the wherein said device that is used to shine upon comprises described normalized absolute difference is mapped to device on the confidence value function that described function limits with following expression formula:
g(d)＝exp(γd
2)
Wherein, d is normalized absolute difference; γ is the funneling factor of selecting.
17, according to the device of claim 16, wherein γ is that mode according to g (1)=0.001 chooses.
18, according to the device of claim 15, wherein said calculating observation sensor values and estimated sensor value
Between the device of normalized absolute difference comprise:
Be used for the device of a plurality of nodes of definite one group of training data, described node is represented described normal range of operation; With
Be used for being connected the device that estimates the described device normal operating range curve of expression by other node of each node and at least one with described a plurality of nodes, above-mentioned connection is to carry out according to the minimized mode of the summation of this connection length.
19, according to the device of claim 18, wherein said estimated sensor value
It is the vector that is positioned on the described curve.
20, a kind of method, be used for improveing the operational data measured value of device with monitoring system, described monitoring system is suitable for first observed reading that at least one is relevant with described device operating characteristic and the normal range of operation of this characteristic compares, and described method comprises:
Determine a plurality of nodes in one group of training data, described node is represented the normal range of operation of described device;
According to the minimized mode of the summation of described connection length, other node of each node and at least one in described a plurality of nodes is connected.
21, according to the method for claim 20, wherein said determining step comprises utilizes the k means clustering algorithm to determine a plurality of barycenter in one group of training data.
22, according to the method for claim 21, wherein said determining step comprises determines the distance between each data element and at least one barycenter in the described training data.
23, according to the method for claim 22, wherein said determining step comprises by with each data element and be associated from the nearest barycenter of described data element and form a plurality of nodes.
24,, utilize the minimal span tree method that described node is connected according to the method for claim 23.
25, a kind of monitoring system that is used for the checkout equipment fault comprises:
First sensor in a plurality of sensors is used for the collection work data element, the measured value of the described device observes condition of described operational data unit's expression;
Utilize the device of the confidence value of the definite described sensor of described operational data unit;
The estimated value of described measured value is defined as the device of described confidence value function.
26, according to the monitoring system of claim 24, wherein said equipment comprises generating set.
27, according to the monitoring system of claim 24, the device of wherein said definite confidence value also comprises the device that calculates following expression formula:
Wherein
It is the sensor vector estimated value of utilizing all sensors that statistical model combines;
It is the sensor vector estimated value of i sensor; d
iBe the operational data unit of described i sensor and the normalized absolute difference between estimated value; y
iIt is the observation sensor vector of i sensor.
28, according to the monitoring system of claim 24, the device of the estimated value of wherein said definite improvement comprises the device that calculates following expression formula:
Wherein, w
iIt is the degree of confidence of i sensor;
It is the estimated value after the described measured value improvement;
It is the sensor vector estimated value of i sensor; y
iIt is the observation sensor vector at i sensor place.
29, according to the monitoring system of claim 24, the device of wherein said definite confidence value comprises:
Be used for calculating observation sensor values and estimated sensor value
Between the device of normalized absolute difference;
Be used for described normalized absolute difference is mapped to the device of a confidence value.
30, according to the monitoring system of claim 29, wherein said mapping device comprises the device that described normalized absolute difference is mapped to a confidence value function, and described function limits with following expression formula:
g(d)＝exp(γd
2)
Wherein, d is normalized absolute difference; γ is the funneling factor of selecting.
31, according to the monitoring system of claim 30, wherein γ is that mode according to g (1)=0.001 chooses.
32, according to the monitoring system of claim 29, wherein calculating observation sensor values and estimated sensor value
Between the device of normalized absolute difference, comprising:
Be used for the device of a plurality of nodes of definite one group of training data, described node is represented described normal range of operation;
Be used for being connected the device that estimates the described monitoring system normal operating range curve of expression by other node of each node and at least one with described a plurality of nodes, above-mentioned connection is to carry out according to the minimized mode of the summation of this connection length.
33, according to the monitoring system of claim 32, wherein said estimated sensor value
It is the vector that is positioned on the described curve.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US60437404P | 2004-08-25 | 2004-08-25 | |
| US60/604374 | 2004-08-25 | ||
| US11/202861 | 2005-08-12 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1782672A true CN1782672A (en) | 2006-06-07 |
| CN100478650C CN100478650C (en) | 2009-04-15 |
Family
ID=36773117
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2005101098953A Expired - Fee Related CN100478650C (en) | 2004-08-25 | 2005-08-25 | Method and apparatus for improved fault detection in power generation equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100478650C (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101307756B (en) * | 2008-07-08 | 2010-06-16 | 河北省电力研究院 | Thermoelectric generating set auxiliary machinery fault fast-reducing load feed water pump trip behavior feasibility determination method |
| CN102298978A (en) * | 2011-05-17 | 2011-12-28 | 哈尔滨工程大学 | MFM (multilevel flow model)-based indeterminate fault diagnosis method for nuclear power plant for ship |
| CN103140812A (en) * | 2010-10-11 | 2013-06-05 | 通用电气公司 | Systems, methods, and apparatus for signal processing- based fault detection, isolation and remediation |
| US9151786B2 (en) | 2010-10-11 | 2015-10-06 | General Electric Company | Systems, methods, and apparatus for detecting shifts in redundant sensor signals |
| CN113790766A (en) * | 2021-11-15 | 2021-12-14 | 江苏未来智慧信息科技有限公司 | Intelligent detection system and method for power plant |
| WO2022007330A1 (en) * | 2020-07-08 | 2022-01-13 | 深圳技术大学 | Battery electric vehicle state monitoring method and system |
| CN116990621A (en) * | 2023-09-26 | 2023-11-03 | 国网吉林省电力有限公司通化供电公司 | Dynamic fault monitoring system for electric power safety tools |
-
2005
- 2005-08-25 CN CNB2005101098953A patent/CN100478650C/en not_active Expired - Fee Related
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101307756B (en) * | 2008-07-08 | 2010-06-16 | 河北省电力研究院 | Thermoelectric generating set auxiliary machinery fault fast-reducing load feed water pump trip behavior feasibility determination method |
| CN103140812A (en) * | 2010-10-11 | 2013-06-05 | 通用电气公司 | Systems, methods, and apparatus for signal processing- based fault detection, isolation and remediation |
| US9151786B2 (en) | 2010-10-11 | 2015-10-06 | General Electric Company | Systems, methods, and apparatus for detecting shifts in redundant sensor signals |
| CN103140812B (en) * | 2010-10-11 | 2016-08-24 | 通用电气公司 | For fault detect based on signal processing, isolate and correct system, method and apparatus |
| CN102298978A (en) * | 2011-05-17 | 2011-12-28 | 哈尔滨工程大学 | MFM (multilevel flow model)-based indeterminate fault diagnosis method for nuclear power plant for ship |
| WO2022007330A1 (en) * | 2020-07-08 | 2022-01-13 | 深圳技术大学 | Battery electric vehicle state monitoring method and system |
| CN113790766A (en) * | 2021-11-15 | 2021-12-14 | 江苏未来智慧信息科技有限公司 | Intelligent detection system and method for power plant |
| CN116990621A (en) * | 2023-09-26 | 2023-11-03 | 国网吉林省电力有限公司通化供电公司 | Dynamic fault monitoring system for electric power safety tools |
| CN116990621B (en) * | 2023-09-26 | 2024-01-19 | 国网吉林省电力有限公司通化供电公司 | Dynamic fault monitoring system for electric power safety tools |
Also Published As
| Publication number | Publication date |
|---|---|
| CN100478650C (en) | 2009-04-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7953577B2 (en) | Method and apparatus for improved fault detection in power generation equipment | |
| CN100478650C (en) | Method and apparatus for improved fault detection in power generation equipment | |
| CN106505557B (en) | Remote measurement error identification method and device | |
| CN109765490B (en) | Power battery fault detection method and system based on high-dimensional data diagnosis | |
| US9310345B2 (en) | Sensor system, computer, and machine | |
| CN105930900B (en) | The Forecasting Methodology and system of a kind of hybrid wind power generation | |
| CN105512011B (en) | A kind of electronics testability modeling appraisal procedure | |
| CN113408068A (en) | Random forest classification machine pump fault diagnosis method and device | |
| CN110645934A (en) | Online calibration method of displacement sensor | |
| CN108960334A (en) | A kind of multi-sensor data Weighted Fusion method | |
| CN109738014A (en) | The intelligent diagnosing method and system of city integrated piping lane equipment fault | |
| CN115392037A (en) | Equipment fault prediction method, device, equipment and storage medium | |
| CN115577587A (en) | Historical building health state monitoring method and system | |
| CN108574290B (en) | Method, device, terminal and readable storage medium for positioning oscillation source of forced oscillation | |
| CN111797545B (en) | Wind turbine generator yaw reduction coefficient calculation method based on measured data | |
| CN113237619A (en) | Fault early warning method, device, equipment and storage medium for variable-speed rotating machinery vibration | |
| CN117406027A (en) | Distribution network fault distance measurement method and system | |
| CN115685045B (en) | Online evaluation method for voltage transformer | |
| CN111738878A (en) | Bridge stress detection system | |
| CN116245233A (en) | Wind speed downscaling method of wind power plant based on super-resolution reconstruction technology | |
| CN109556861A (en) | A kind of bearing real-time fault diagnosis system of case-based reasioning | |
| CN113487141B (en) | Bearing state evaluation method based on multisource information clustering fusion | |
| KR20180115448A (en) | Apparatus and method for authenticating time-varying signal in online via kernel regression model | |
| CN110048402B (en) | Two-stage power grid parameter estimation method | |
| CN113418146A (en) | Leakage auxiliary positioning control method for water supply pipe network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| ASS | Succession or assignment of patent right |
Owner name: SIEMENS AKTIENGESELLSCHAFT Free format text: FORMER OWNER: SIEMENS CORPORATE RESEARCH, INC. Effective date: 20100727 |
|
| C41 | Transfer of patent application or patent right or utility model | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20100727 Address after: new jersey Patentee after: Siemens AG Address before: new jersey Patentee before: Siemens Corporate Research, Inc. |
|
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090415 Termination date: 20130825 |