CN108803555A - A kind of inferior health online recognition and diagnostic method based on performance monitoring data - Google Patents
A kind of inferior health online recognition and diagnostic method based on performance monitoring data Download PDFInfo
- Publication number
- CN108803555A CN108803555A CN201810231114.5A CN201810231114A CN108803555A CN 108803555 A CN108803555 A CN 108803555A CN 201810231114 A CN201810231114 A CN 201810231114A CN 108803555 A CN108803555 A CN 108803555A
- Authority
- CN
- China
- Prior art keywords
- data
- state
- class
- health
- training sample
- 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
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0256—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults injecting test signals and analyzing monitored process response, e.g. injecting the test signal while interrupting the normal operation of the monitored system; superimposing the test signal onto a control signal during normal operation of the monitored system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Automation & Control Theory (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a kind of inferior health online recognition and diagnostic method based on performance monitoring data, belong to fault diagnosis technology field.It initially sets up the initial model of probabilistic neural network state classification and to calculate threshold denominators poor, on-line monitoring and diagnostic classification are carried out to monitoring device using "current" model, and further identify and extract sub-health state data, and be put into sub-health state data group;If inferior health data group to be identified reaches storage tolerance or known state occurs, suspend storage work, all elements in the group are subjected to K-means clusterings, classification results is obtained and empties the memory space of inferior health data group.Then the sub-health state data set after clustering is merged with training sample before, new disaggregated model is obtained in update to initial model;Finally repeat the above steps identification sub-health state, the on-call maintenance in failure state.The present invention makes timely and effectively measure according to equipment state, is lost caused by mitigating failure.
Description
Technical field
The invention belongs to fault diagnosis technology field, be related to a kind of inferior health online recognition based on performance monitoring data and
Diagnostic method.
Background technology
Traditional method for diagnosing faults includes normal and failure two states, and in this, as partitioning standards, structure diagnosis
Model.However equipment is in the process of running, is not efficient always, high-accuracy work, there is also sub-health states;Therefore
It can cannot be reflected and be set by sub-health state mistaken diagnosis for normal condition or malfunction by two-state model normal, based on failure
Standby time of day.
Currently, in engineer application, generally report to obtain the data of equipment failure state by FMECA, to identify
Equipment failure state;However, it is very difficult to obtain the sub-health state of equipment, Zhi Nengsui by the hardware configuration and working mechanism of equipment
The passage for the time gradually obtains the status data of equipment by monitoring device operating status and data of monitoring point situation of change,
And then its sub-health state is identified according to status data.
But this status data run according to equipment, and then the method for identifying, being diagnosed to be the sub-health state of equipment,
Offline mode is mainly used at present, i.e., sub-health state is diagnosed to be by the historical state data of analytical equipment, thus cannot be real
When diagnostic device working condition.
Invention content
The present invention is to solve the above-mentioned problems, it is proposed that a kind of inferior health online recognition based on performance monitoring data and examines
Disconnected method, by the status data of continuous monitoring device operational process, the constantly sub-health state of identification and diagnostic device, in turn
State classification model is constantly updated, to effectively identify the sub-health state occurred in equipment running process.
It is as follows:
Step 1: being directed to certain monitoring device, establishes the initial model of probabilistic neural network state classification and calculate thresholding mark
It is accurate poor;
First, the measured data of the equipment normal operating condition is obtained, and obtains equipment event by way of direct fault location
The measured data of barrier state;Then, selected part is normally respectively and the measured data under malfunction is as training sample, establishes
The initial model of probabilistic neural network state classification.
Training sample set is combined into
Wherein:kiIndicate the number of the i-th class training sample;M indicates total class number of training sample.Indicate that the i-th class is trained
Kth in sampleiA training sample value, the value are tieed up for p.
Meanwhile calculating separately the p dimensions of all data under the p dimension standard deviations and malfunction of all data under normal condition
Standard deviation, the corresponding standard deviation respectively tieed up, selects the maximum value of each dimension poor as threshold denominators, threshold denominators difference is two-by-two
P is tieed up.
Step 2:On-line monitoring and diagnostic classification are carried out to monitoring device using "current" model, using classification results into one
Step identification and extraction sub-health state data, and be put into sub-health state data group;
"current" model is initially initial model;Diagnostic classification result is normal or failure probability value;
Step 2.1, using "current" model on-line monitoring equipment operating status, acquire p in real time and tie up status data as one group,
And calculate the Euclidean distance in current p dimension status datas and training sample between each element.
The current p acquired in real time ties up status data:
Euclidean distance:E=(d-xij)T(d-xij);
xijIndicate j-th of training sample value in the i-th class training sample;J=1,2 ..., ki, i=1,2 ..., m;
Step 2.2, with the neuron of Euclidean distance combination Gauss type function activation pattern layer radial basis function;
Gauss type function activates formula as follows:
σiIndicate the corresponding maximum value of each dimension in the standard deviation of the i-th class training sample.Pij(d) j-th of training sample is indicated
The neuron of this equivalence corresponds to the output of the i-th class training sample;
Step 2.3 acquires the probability that current p dimensions status data belongs to known class in probabilistic neural network summation layer:
fi,ki(d) indicate that current p dimensions status data d belongs to the probability of known class i;
Step 2.4 judges whether inferior health data group is empty, if so, entering step 2.5;Otherwise, 2.6 are entered step;
Inferior health data group initial value is sky.
Step 2.5 ties up the probability value that status data belongs to known class according to current p, judges whether that probability value is big
Belong to the average value of all kinds of probability sums in last moment p dimension status data, if so, entering step 2.7;Otherwise, it enters step
2.8;
Step 2.6 ties up the probability value that status data belongs to known class according to current p, judges whether that probability value is big
The mathematical expectation of probability of status data is tieed up in the p of the correct identification state of last moment energy;If so, entering step 2.7;Otherwise, enter
Step 2.8.
Step 2.7, current p dimension status datas can be correctly validated, that maximum one kind of probability value is that current p ties up state
The recognition result state of data.
Step 2.8, current p dimension status datas are inferior health data to be identified, are put into inferior health data group to be identified
In.
Step 3: judging whether inferior health data group to be identified reaches storage tolerance or whether known state occur;
If it is, entering step four;Otherwise, return to step two continue acquisition p dimensions status data in real time using "current" model and are divided
Analysis.
Known state includes normal condition and malfunction when initial, so detecting that malfunction just stops;
After model modification, it is known that state includes normal condition, sub-health state and malfunction, is detected wherein arbitrary
A kind of state should all stop.
Sub-health state data group to be identified:
In formula, n indicates the inferior health data number extracted, shnIndicate n-th group sub-health state data;
Step 4: all elements in the group are carried out K-means clusters point by the storage work of pause inferior health data group
Analysis, obtains classification results and empties the memory space of inferior health data group.
It is as follows:
All elements in inferior health data group are classified as one kind by step 4.1, initialization cluster number of clusters class_k=1,
It is poor to calculate class internal standard;
Step 4.2 judges whether each class internal standard difference is respectively less than poor equal to threshold denominators, if it is, such is obtained
Sub-health state data belong to a kind of, label is done to such data set and is terminated.Otherwise, 4.3 are entered step;
Step 4.3, class internal standard difference are poor more than threshold denominators, then the sub-health state data obtained are not belonging to one kind;
Step 4.4, will cluster number of clusters class_k from increasing 1, using K-means methods by the element in inferior health data group
The classification being polymerized to adds one, and it is poor to calculate separately class internal standard of all categories;
Step 4.5, return to step 4.2 obtain until the class internal standard difference of all categories is respectively less than poor equal to threshold denominators
Cluster division result;
Cluster division result is:C={ C1,C2,…,Ck};
Step 4.6 divides inferior health data set according to cluster division result;
Wherein:ti'Indicate the number of the i-th ' class inferior health data;Class_k indicates total class number of inferior health data set.
Step 5: the sub-health state data set after clustering is merged with training sample before, update is to initially
New disaggregated model is obtained in model;
Newly the training sample data of disaggregated model are:X'=x ∪ sh;
Then, probabilistic neural network disaggregated model is re-created using new training data, to the update of implementation model.
Step 6: two continuous iteration renolation disaggregated model of return to step, by the running state data of monitoring device,
Online recognition goes out sub-health state, and the on-call maintenance when monitoring device failure state.
The advantage of the invention is that:
1), a kind of inferior health online recognition and diagnostic method based on performance monitoring data, compensate for traditional fault diagnosis
The deficiency of the easy mistaken diagnosis of method, can reflect the time of day of equipment.
2), a kind of inferior health online recognition and diagnostic method based on performance monitoring data, can be transported by monitoring device
Status data when row, online recognition go out the sub-health state of equipment appearance, have in time so as to be made according to equipment state
The measure of effect mitigates the loss caused by failure.
3), a kind of inferior health online recognition and diagnostic method based on performance monitoring data are equipment sub-health state
Inline diagnosis, identification provide a set of standardization, feasible new method.
Description of the drawings
Fig. 1 is the present invention a kind of inferior health online recognition and diagnostic method flow chart based on performance monitoring data;
Fig. 2 is the model structure for the probabilistic neural network state classification that the present invention establishes;
Fig. 3 is present invention identification and extraction sub-health state data and the method flow diagram of storage;
Fig. 4 is the method flow diagram that the present invention carries out inferior health data K-means clusterings;
Fig. 5 is the initial training sample distribution figure extracted in the embodiment of the present invention;
Fig. 6 is monitoring of equipment data set and the sub-health state data set distribution not being correctly validated in the embodiment of the present invention
Figure;
Fig. 7 is sub-health state data clusters result to be identified in the embodiment of the present invention;
Fig. 8 is board state data set in the embodiment of the present invention;
Fig. 9 is training data distribution map in the embodiment of the present invention (by the end of failure state).
Specific implementation mode
Below in conjunction with drawings and examples, the present invention is described in further detail.
The present invention establishes state classification model using probabilistic neural network first, carries out on-line monitoring and diagnostic classification, and
Identification and extraction sub-health state data;Then, the method based on K-means clusters, carries out the cluster of sub-health state data
Analysis;Finally, the training data for updating state classification model, to obtain new disaggregated model.Iteration executes, constantly improve point
Class model continues to execute, to identify in equipment running process if equipment failure state after carrying out breakdown maintenance
All sub-health states occurred.
As shown in Figure 1, being as follows:
Step 1: being directed to certain monitoring device, establishes the initial model of probabilistic neural network state classification and calculate thresholding mark
It is accurate poor;
First, the measured data of the equipment normal operating condition is obtained, and obtains equipment event by way of direct fault location
The measured data of barrier state;Then, selected part is normally respectively and the measured data under malfunction is as training sample, establishes
The initial model of probabilistic neural network state classification, as shown in Figure 2.
Training sample set is combined into
Wherein:kiIndicate the number of the i-th class training sample;M indicates total class number of training sample.Indicate that the i-th class is trained
Kth in sampleiA training sample value, the value are tieed up for p.
Meanwhile calculating separately the p dimensions of all data under the p dimension standard deviations and malfunction of all data under normal condition
Standard deviation, the corresponding standard deviation respectively tieed up, selects the maximum value of each dimension poor as threshold denominators, threshold denominators difference is two-by-two
P is tieed up.
Step 2:On-line monitoring and diagnostic classification are carried out to monitoring device using "current" model, using classification results into one
Step identification and extraction sub-health state data, and be put into sub-health state data group;
"current" model is initially initial model;Diagnostic classification result is normal or failure probability value;
As shown in figure 3, being as follows:
Step 2.1, using "current" model on-line monitoring equipment operating status, acquire p in real time and tie up status data as one group,
And calculate the Euclidean distance in current p dimension status datas and training sample between each element.
The current p acquired in real time ties up status data:
Euclidean distance:E=(d-xij)T(d-xij);
xijIndicate j-th of training sample value in the i-th class training sample;J=1,2 ..., ki, i=1,2 ..., m;
Step 2.2, with the neuron of Euclidean distance combination Gauss type function activation pattern layer radial basis function;
Gauss type function activates formula as follows:
σiIndicate the corresponding maximum value of each dimension in the standard deviation of the i-th class training sample.Pij(d) j-th of training sample is indicated
The neuron of this equivalence corresponds to the output of the i-th class training sample;
Step 2.3 acquires the probability that current p dimensions status data belongs to known class in probabilistic neural network summation layer:
fi,ki(d) indicate that current p dimensions status data d belongs to the probability of known class i;
Step 2.4 judges whether inferior health data group is empty, if so, entering step 2.5;Otherwise, 2.6 are entered step;
Inferior health data group initial value is sky.
Step 2.5 ties up the probability value that status data belongs to known class according to current p, judges whether that probability value is big
Belong to the average value of all kinds of probability sums in last moment p dimension status data, if so, entering step 2.7;Otherwise, it enters step
2.8;
Step 2.6 ties up the probability value that status data belongs to known class according to current p, judges whether that probability value is big
The mathematical expectation of probability of status data is tieed up in the p of the correct identification state of last moment energy;If so, entering step 2.7;Otherwise, enter
Step 2.8.
Step 2.7, current p dimension status datas can be correctly validated, that maximum one kind of probability value is that current p ties up state
The recognition result state of data.
Step 2.8, current p dimension status datas are inferior health data to be identified, are put into inferior health data group to be identified
In.
Step 3: judging whether inferior health data group to be identified reaches storage tolerance or whether known state occur;
If it is, entering step four;Otherwise, return to step two continue acquisition p dimensions status data in real time using "current" model and are divided
Analysis.
Known state includes normal condition and malfunction when initial, so detecting that malfunction just stops;
After model modification, it is known that state includes normal condition, sub-health state and malfunction, is detected wherein arbitrary
A kind of state should all stop.
Sub-health state data group to be identified:
In formula, n indicates the inferior health data number extracted, shnIndicate n-th group sub-health state data;
Step 4: all elements in the group are carried out K-means clusters point by the storage work of pause inferior health data group
Analysis, obtains classification results and empties the memory space of inferior health data group.
As shown in figure 4, being as follows:
All elements in inferior health data group are classified as one kind by step 4.1, initialization cluster number of clusters class_k=1,
It is poor to calculate class internal standard;
Step 4.2 judges whether each class internal standard difference is respectively less than poor equal to threshold denominators, if it is, such is obtained
Sub-health state data belong to a kind of, label is done to such data set and is terminated.Otherwise, 4.3 are entered step;
Step 4.3, class internal standard difference are poor more than threshold denominators, then the sub-health state data obtained are not belonging to one kind;
Step 4.4, will cluster number of clusters class_k from increasing 1, using K-means methods by the element in inferior health data group
The classification being polymerized to adds one, and it is poor to calculate separately class internal standard of all categories;
Step 4.5, return to step 4.2 obtain until the class internal standard difference of all categories is respectively less than poor equal to threshold denominators
Cluster division result;
Cluster division result is:C={ C1,C2,…,Ck};
Step 4.6 divides inferior health data set according to cluster division result;
Wherein:ti'Indicate the number of the i-th ' class inferior health data;Class_k indicates total class number of inferior health data set.
Step 5: the sub-health state data set after clustering is merged with training sample before, update is to initially
New disaggregated model is obtained in model;
Newly the training sample data of disaggregated model are:X'=x ∪ sh;
Then, probabilistic neural network disaggregated model is re-created using new training data, to the update of implementation model.
Step 6: two continuous iteration renolation disaggregated model of return to step, by the running state data of monitoring device,
Online recognition goes out sub-health state, and the on-call maintenance when monitoring device failure state.
Embodiment:
The present embodiment selects direct-current power supply converting circuit, including three parts:Respectively 18V power-supplying circuits, 18V are arrived
12V power-switching circuits and 12V to 5V power-switching circuits.
The direct-current power supply converting circuit is provided with 3 monitoring points, respectively 18V electric power output voltages VOUT, 12V power supplys altogether
Output voltage S+12V, 5V electric power output voltage S+5V.The voltage output of this 3 monitoring points is monitored respectively, data are used in combination
Card collection voltages data, in order to carry out the health status evaluation of the circuit.
This 3 monitoring point voltages of VOUT, S+12V, S+5V reflect the critical function of circuit, and this 3 monitorings
The voltage output of point is independent from each other, so the voltage monitoring value of this 3 monitoring points is selected to evaluate DC power supply conversion electricity
The health status on road;Remember VOUT, S+12V, the monitoring points S+5V are respectively T1, T2, T3。
The present embodiment chooses normal and each 1200 groups of fault state data, and initial model is built in this, as training sample.
(normal condition label is 1 to initial training data, and malfunction label is 9) distribution situation, as shown in Figure 5:
Normal and malfunction measured data standard deviation is determined respectively, and the higher value of each dimension in standard deviation is made
It is poor for the threshold denominators of device status data;Standard deviation calculation formula is:
Normalcy is poor:0.0150,0.0137,0.0074
Malfunction standard deviation:0.0095,0.0050,0.0026
Threshold denominators difference is:0.0150,0.0137,0.0074
On-line monitoring and diagnostic classification are carried out to monitoring device using "current" model, the results showed that:Monitoring device runs shape
State, in the 2007th group of data, which belongs to the mathematical expectation of probability that normal and failure probability is respectively less than one group of data, shows
Circuit board starts unknown state occur at this time.The storage tolerance of circuit board selected by present case is 2000 groups of data, i.e., goes out certainly
Existing unknown state starts to add up 2000 groups of status datas (i.e. the 2007th to 4006 group data), extracts inferior health number to be identified
According to collection, clustering is carried out.Monitoring of equipment data set and the sub-health state data set distribution situation not being correctly validated, it is such as attached
Shown in Fig. 6.
The 2000 groups of inferior health data to be identified extracted, standard deviation are:0.0085,0.0092,0.0211
Since the standard deviation is poor more than threshold denominators, then show that extracting data is not belonging to one kind, carries out K-means clusters
Analysis.
As shown in fig. 7, K-means cluster results:
Cluster number of clusters:Class_k=2;
Mean vector:u1=| 17.9950,11.6779,4.5838 |, u2=| 17.9913,11.6708,4.4727 |
Class internal standard is poor:s1=| 0.0086,0.0091,0.0042 |, s2=| 0.0054,0.0104,0.0047 |
Inferior health data set:{[3x1928double],[3x72double]}
By in sub-health state update to disaggregated model, new disaggregated model is obtained, new training data is as shown in table 1:
Table 1
Monitor power panel on-line, disaggregated model is improved in continuous renewal, until power panel failure state break-off, into
Row breakdown maintenance.Whole process is total to 10615 groups of gathered data, and data graphs are as shown in Figure 8.
Through the above steps, it can effectively extract, 3 class inferior healths of the identification circuit board in addition to normal and malfunction
State, to realize online recognition and the diagnosis of inferior health.
The training sample set obtained by the above process, data distribution are as shown in Figure 9:
[3x1200double], [3x1200double], [3x1928double], [3x72double],
[3x2000double]}
The inferior health data set identified:
{[3x1928double],[3x72double],[3x2000double]}
Board state is as shown in table 2:
Table 2
Claims (3)
1. a kind of inferior health online recognition and diagnostic method based on performance monitoring data, which is characterized in that be as follows:
Step 1: being directed to certain monitoring device, establishes the initial model of probabilistic neural network state classification and to calculate threshold denominators poor;
Step 2:On-line monitoring and diagnostic classification are carried out to monitoring device using "current" model, further known using classification results
Not and sub-health state data are extracted, and is put into sub-health state data group;
"current" model is initially initial model;Diagnostic classification result is normal or failure probability value;
Identification and the process for extracting sub-health state data are:
Step 2.1, using "current" model on-line monitoring equipment operating status, acquire p in real time and tie up status data as one group, and count
Calculate the Euclidean distance between each element in current p dimension status datas and training sample;
The current p acquired in real time ties up status data:
Euclidean distance:E=(d-xij)T(d-xij);
xijIndicate j-th of training sample value in the i-th class training sample;J=1,2 ..., ki, i=1,2 ..., m;
Step 2.2, with the neuron of Euclidean distance combination Gauss type function activation pattern layer radial basis function;
Gauss type function activates formula as follows:
σiIndicate the corresponding maximum value of each dimension in the standard deviation of the i-th class training sample;Pij(d) j-th of training sample etc. is indicated
The neuron of valence corresponds to the output of the i-th class training sample;
Step 2.3 acquires the probability that current p dimensions status data belongs to known class in probabilistic neural network summation layer:
Indicate that current p dimensions status data d belongs to the probability of known class i;
Step 2.4 judges whether inferior health data group is empty, if so, entering step 2.5;Otherwise, 2.6 are entered step;
Inferior health data group initial value is sky;
Step 2.5 ties up the probability value that status data belongs to known class according to current p, and it is upper to judge whether that probability value is more than
One moment p dimension status data belongs to the average value of all kinds of probability sums, if so, entering step 2.7;Otherwise, 2.8 are entered step;
Step 2.6 ties up the probability value that status data belongs to known class according to current p, and it is upper to judge whether that probability value is more than
The mathematical expectation of probability of the p dimension status datas of the correct identification state of one moment energy;If so, entering step 2.7;Otherwise, it enters step
2.8;
Step 2.7, current p dimension status datas can be correctly validated, that maximum one kind of probability value is that current p ties up status data
Recognition result state;
Step 2.8, current p dimension status datas are inferior health data to be identified, are put into inferior health data group to be identified.
Step 3: judging whether inferior health data group to be identified reaches storage tolerance or whether known state occur;If
It is then to enter step four;Otherwise, return to step two continue acquisition p dimensions status data in real time using "current" model and are analyzed;
Known state includes normal condition and malfunction when initial, so detecting that malfunction just stops;
After model modification, it is known that state includes normal condition, sub-health state and malfunction, detects any of which
State should all stop;
Sub-health state data group to be identified:
In formula, n indicates the inferior health data number extracted, shnIndicate n-th group sub-health state data;
Step 4: all elements in the group are carried out K-means clusterings, obtained by the storage work of pause inferior health data group
To classification results and empty the memory space of inferior health data group;
Step 5: the sub-health state data set after clustering is merged with training sample before, initial model is arrived in update
In obtain new disaggregated model;
Newly the training sample data of disaggregated model are:X'=x ∪ sh;
Then, probabilistic neural network disaggregated model is re-created using new training data, to the update of implementation model;
Step 6: two continuous iteration renolation disaggregated model of return to step, by the running state data of monitoring device, online
Identify sub-health state, and the on-call maintenance when monitoring device failure state.
2. a kind of inferior health online recognition and diagnostic method based on performance monitoring data as described in claim 1, feature
It is, the step one is specially:
First, the measured data of the equipment normal operating condition is obtained, and obtains equipment fault shape by way of direct fault location
The measured data of state;
Then, selected part is normally respectively and the measured data under malfunction is as training sample, establishes probabilistic neural network
The initial model of state classification;
Training sample set is combined into
Wherein:kiIndicate the number of the i-th class training sample;M indicates total class number of training sample;Indicate the i-th class training sample
Middle kthiA training sample value, the value are tieed up for p;
Meanwhile calculating separately the p dimension standards of all data under the p dimension standard deviations and malfunction of all data under normal condition
Difference, the corresponding standard deviation respectively tieed up, selects the maximum value of each dimension poor as threshold denominators two-by-two, and threshold denominators difference is tieed up for p.
3. a kind of inferior health online recognition and diagnostic method based on performance monitoring data as described in claim 1, feature
It is, the step four is as follows:
All elements in inferior health data group are classified as one kind, calculated by step 4.1, initialization cluster number of clusters class_k=1
Class internal standard is poor;
Step 4.2 judges whether each class internal standard difference is respectively less than that be equal to threshold denominators poor, if it is, such obtained Asia
State of health data belongs to a kind of, makees label to such data set and terminates;Otherwise, 4.3 are entered step;
Step 4.3, class internal standard difference are poor more than threshold denominators, then the sub-health state data obtained are not belonging to one kind;
Cluster number of clusters class_k is increased 1 by step 4.4 certainly, is polymerized to the element in inferior health data group using K-means methods
Classification add one, it is poor to calculate separately class internal standard of all categories;
Step 4.5, return to step 4.2 obtain cluster and draw until the class internal standard difference of all categories is respectively less than poor equal to threshold denominators
Divide result;
Cluster division result is:C={ C1,C2,…,Ck};
Step 4.6 divides inferior health data set according to cluster division result;
Wherein:ti' indicate the i-th ' class inferior health data number;Class_k indicates total class number of inferior health data set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810231114.5A CN108803555B (en) | 2018-03-20 | 2018-03-20 | Sub-health online identification and diagnosis method based on performance monitoring data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810231114.5A CN108803555B (en) | 2018-03-20 | 2018-03-20 | Sub-health online identification and diagnosis method based on performance monitoring data |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108803555A true CN108803555A (en) | 2018-11-13 |
CN108803555B CN108803555B (en) | 2020-08-21 |
Family
ID=64095208
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810231114.5A Active CN108803555B (en) | 2018-03-20 | 2018-03-20 | Sub-health online identification and diagnosis method based on performance monitoring data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108803555B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110070076A (en) * | 2019-05-08 | 2019-07-30 | 北京字节跳动网络技术有限公司 | Method and apparatus for choosing trained sample |
CN110133488A (en) * | 2019-04-09 | 2019-08-16 | 上海电力学院 | Switchgear health status evaluation method and device based on optimal number of degrees |
CN111956198A (en) * | 2020-10-21 | 2020-11-20 | 北京妙医佳健康科技集团有限公司 | Equipment state determination method and device |
CN117725538A (en) * | 2024-02-07 | 2024-03-19 | 河海大学 | Water gate deformation multi-measuring-point joint diagnosis system and method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103512751A (en) * | 2013-07-03 | 2014-01-15 | 辽宁大学 | Bearing health state identification method based on probabilistic neural network |
CN103927446A (en) * | 2014-04-18 | 2014-07-16 | 北京航空航天大学 | Product fault diagnosis method in consideration of sub-health status and single input single output constraints |
CN203879690U (en) * | 2014-05-20 | 2014-10-15 | 北京国电思达科技有限公司 | Fault real-time pre-warning and smart control platform for wind power plant |
CN105782071A (en) * | 2016-03-04 | 2016-07-20 | 长沙有色冶金设计研究院有限公司 | Water isolation pulp pump fault diagnosis method based on probabilistic neural network |
CN107153841A (en) * | 2017-04-24 | 2017-09-12 | 南京康尼机电股份有限公司 | A kind of inferior health Forecasting Methodology of urban rail transit vehicles door system |
-
2018
- 2018-03-20 CN CN201810231114.5A patent/CN108803555B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103512751A (en) * | 2013-07-03 | 2014-01-15 | 辽宁大学 | Bearing health state identification method based on probabilistic neural network |
CN103927446A (en) * | 2014-04-18 | 2014-07-16 | 北京航空航天大学 | Product fault diagnosis method in consideration of sub-health status and single input single output constraints |
CN203879690U (en) * | 2014-05-20 | 2014-10-15 | 北京国电思达科技有限公司 | Fault real-time pre-warning and smart control platform for wind power plant |
CN105782071A (en) * | 2016-03-04 | 2016-07-20 | 长沙有色冶金设计研究院有限公司 | Water isolation pulp pump fault diagnosis method based on probabilistic neural network |
CN107153841A (en) * | 2017-04-24 | 2017-09-12 | 南京康尼机电股份有限公司 | A kind of inferior health Forecasting Methodology of urban rail transit vehicles door system |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110133488A (en) * | 2019-04-09 | 2019-08-16 | 上海电力学院 | Switchgear health status evaluation method and device based on optimal number of degrees |
CN110133488B (en) * | 2019-04-09 | 2021-10-08 | 上海电力学院 | Switch cabinet health state evaluation method and device based on optimal grade number |
CN110070076A (en) * | 2019-05-08 | 2019-07-30 | 北京字节跳动网络技术有限公司 | Method and apparatus for choosing trained sample |
CN111956198A (en) * | 2020-10-21 | 2020-11-20 | 北京妙医佳健康科技集团有限公司 | Equipment state determination method and device |
CN111956198B (en) * | 2020-10-21 | 2021-01-15 | 北京妙医佳健康科技集团有限公司 | Equipment state determination method and device |
CN117725538A (en) * | 2024-02-07 | 2024-03-19 | 河海大学 | Water gate deformation multi-measuring-point joint diagnosis system and method |
CN117725538B (en) * | 2024-02-07 | 2024-04-16 | 河海大学 | Water gate deformation multi-measuring-point joint diagnosis system and method |
Also Published As
Publication number | Publication date |
---|---|
CN108803555B (en) | 2020-08-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108803555A (en) | A kind of inferior health online recognition and diagnostic method based on performance monitoring data | |
Cao et al. | A novel false data injection attack detection model of the cyber-physical power system | |
CN108897954A (en) | Wind turbines temperature pre-warning method and its system based on BootStrap confidence calculations | |
CN107340766B (en) | Power scheduling alarm signal text based on similarity sorts out and method for diagnosing faults | |
CN108985380B (en) | Point switch fault identification method based on cluster integration | |
CN107132310B (en) | Transformer equipment health status method of discrimination based on gauss hybrid models | |
CN104091070A (en) | Rail transit fault diagnosis method and system based on time series analysis | |
Zhang et al. | Power consumption predicting and anomaly detection based on transformer and K-means | |
CN107065834B (en) | The method for diagnosing faults of concentrator in hydrometallurgy process | |
CN109597396B (en) | A kind of distribution transforming on-line fault diagnosis method based on high amount of traffic and transfer learning | |
CN114676742A (en) | Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network | |
CN111190349A (en) | Method, system and medium for monitoring state and diagnosing fault of ship engine room equipment | |
CN109389325A (en) | Transformer substation electronic transducer state evaluating method based on wavelet neural network | |
CN101738998A (en) | System and method for monitoring industrial process based on local discriminatory analysis | |
CN116664015A (en) | Intelligent charging pile management system and method thereof | |
CN110569888A (en) | transformer fault diagnosis method and device based on directed acyclic graph support vector machine | |
CN109739197A (en) | A kind of multi-state failure prediction method of chemical spent material processing equipment | |
CN113093695A (en) | Data-driven SDN controller fault diagnosis system | |
Shi et al. | An unsupervised anomaly detection method based on density peak clustering for rail vehicle door system | |
CN110598750B (en) | Working condition identification method based on switch machine action curve similarity characteristics | |
Huang et al. | Static and dynamic joint analysis for operation condition division of industrial process with incremental learning | |
Min et al. | Fault prediction for distribution network based on CNN and LightGBM algorithm | |
CN114167837A (en) | Intelligent fault diagnosis method and system for railway signal system | |
CN110874584B (en) | Blade fault diagnosis method based on improved prototype clustering | |
CN113496440A (en) | User abnormal electricity utilization detection method and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |