CN108803555B - Sub-health online identification and diagnosis method based on performance monitoring data - Google Patents

Sub-health online identification and diagnosis method based on performance monitoring data Download PDF

Info

Publication number
CN108803555B
CN108803555B CN201810231114.5A CN201810231114A CN108803555B CN 108803555 B CN108803555 B CN 108803555B CN 201810231114 A CN201810231114 A CN 201810231114A CN 108803555 B CN108803555 B CN 108803555B
Authority
CN
China
Prior art keywords
sub
health
data
state
state data
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.)
Active
Application number
CN201810231114.5A
Other languages
Chinese (zh)
Other versions
CN108803555A (en
Inventor
石君友
郭绪浩
何庆杰
邓怡
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beihang University
Original Assignee
Beihang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beihang University filed Critical Beihang University
Priority to CN201810231114.5A priority Critical patent/CN108803555B/en
Publication of CN108803555A publication Critical patent/CN108803555A/en
Application granted granted Critical
Publication of CN108803555B publication Critical patent/CN108803555B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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 sub-health online identification and diagnosis method based on performance monitoring data, and belongs to the technical field of fault diagnosis. Firstly, establishing an initial model of probabilistic neural network state classification and calculating a threshold standard deviation, carrying out online monitoring and diagnosis classification on monitoring equipment by using a current model, further identifying and extracting sub-health state data, and putting the sub-health state data into a sub-health state data group; and if the sub-health data group to be identified reaches the storage tolerance or has a known state, suspending the storage work, carrying out K-means cluster analysis on all elements in the group to obtain a classification result, and emptying the storage space of the sub-health data group. Merging the sub-health state data set after the clustering analysis with the previous training sample, and updating the sub-health state data set into the initial model to obtain a new classification model; and finally, repeating the steps to identify the sub-health state, and maintaining in time when a fault state occurs. The invention can make timely and effective measures according to the equipment state, and reduce the loss caused by the fault.

Description

Sub-health online identification and diagnosis method based on performance monitoring data
Technical Field
The invention belongs to the technical field of fault diagnosis, and relates to a sub-health online identification and diagnosis method based on performance monitoring data.
Background
The traditional fault diagnosis method comprises a normal state and a fault state, and a diagnosis model is constructed by taking the normal state and the fault state as a division basis. However, in the running process of the equipment, the equipment does not always work efficiently and accurately, and a sub-health state also exists; therefore, the sub-health state can be misdiagnosed as a normal state or a fault state by the two-state model based on the normal state and the fault state, and the real state of the equipment cannot be reflected.
Currently, in engineering applications, the failure state of the equipment is generally identified by obtaining data of the failure state of the equipment through FMECA report; however, it is difficult to obtain the sub-health status of the device through the hardware structure and the working mechanism of the device, and only gradually obtain the status data of the device through monitoring the running status of the device and the data change condition of the monitoring point along with the time, and then identify the sub-health status according to the status data.
However, the method for identifying and diagnosing the sub-health state of the equipment according to the running state data of the equipment mainly adopts an off-line mode at present, namely, the sub-health state is diagnosed by analyzing the historical state data of the equipment, so that the working state of the equipment cannot be diagnosed in real time.
Disclosure of Invention
The invention provides a sub-health online identification and diagnosis method based on performance monitoring data, which aims to solve the problems.
The method comprises the following specific steps:
firstly, aiming at a certain monitoring device, establishing an initial model of probabilistic neural network state classification and calculating a threshold standard deviation;
firstly, acquiring measured data of the normal running state of the equipment, and acquiring measured data of the fault state of the equipment in a fault injection mode; then, part of measured data under normal and fault states are respectively selected as training samples, and an initial model of probabilistic neural network state classification is established.
Set of training samples
Figure BDA0001602810590000011
Wherein: k is a radical ofiRepresenting the number of the ith type training samples; m represents the total number of classes of training samples.
Figure BDA0001602810590000012
Represents the kth training sample in the ith classiAnd each training sample value is in a p-dimension.
Meanwhile, p-dimensional standard deviations of all data in a normal state and p-dimensional standard deviations of all data in a fault state are calculated respectively, the standard deviation values of all dimensions are compared pairwise correspondingly, the maximum value of all dimensions is selected as a threshold standard deviation, and the threshold standard deviation is p-dimensional.
Step two: carrying out online monitoring and diagnosis classification on the monitoring equipment by using the current model, further identifying and extracting sub-health state data by using a classification result, and putting the sub-health state data into a sub-health state data group;
the current model is initially an initial model; diagnosing the probability value of the classification result as normal or fault;
and 2.1, monitoring the running state of the equipment on line by using the current model, collecting p-dimensional state data in real time to serve as a group, and calculating the Euclidean distance between the current p-dimensional state data and each element in the training sample.
The current p-dimensional state data collected in real time are:
Figure BDA0001602810590000021
euclidean distance: e ═ d-xij)T(d-xij)；
xijRepresenting the jth training sample value in the ith training sample; j is 1,2, …, ki，i＝1,2,…,m；
Step 2.2, activating neurons of radial basis functions of the mode layer by combining Euclidean distance with Gaussian functions;
the gaussian function activation formula is as follows:
Figure BDA0001602810590000022
σiand the maximum value corresponding to each dimension in the standard deviation of the ith type training sample is shown. Pij(d) Representing the output of the training sample of the ith class corresponding to the equivalent neuron of the jth training sample;
step 2.3, the probability that the current p-dimensional state data belongs to the known class is obtained at a probability neural network summation layer:
Figure BDA0001602810590000023
fi,ki(d) representing the probability that the current p-dimensional state data d belongs to the known class i;
step 2.4, judging whether the sub-health data group is empty, and if so, entering step 2.5; otherwise, entering step 2.6;
the sub-health data set is initially null.
Step 2.5, judging whether the probability value is larger than the average value of the probability sum that the p-dimensional state data at the last moment belongs to various types according to the probability value that the current p-dimensional state data belongs to the known type, and if so, entering step 2.7; otherwise, entering step 2.8;
step 2.6, judging whether the probability value is larger than the probability mean value of the p-dimensional state data which can correctly identify the state at the last moment or not according to the probability value of the current p-dimensional state data belonging to the known category; if yes, go to step 2.7; otherwise, go to step 2.8.
And 2.7, correctly identifying the current p-dimensional state data, wherein the type with the maximum probability value is the identification result state of the current p-dimensional state data.
And 2.8, putting the current p-dimensional state data as the sub-health data to be identified into the sub-health data group to be identified.
Step three, judging whether the sub-health data group to be identified reaches a storage tolerance or a known state appears; if yes, entering the step four; otherwise, returning to the step two, and continuously acquiring the p-dimensional state data in real time by using the current model for analysis.
The known state at the beginning comprises a normal state and a fault state, so the detection of the fault state is stopped;
after the model is updated, the known states comprise a normal state, a sub-health state and a fault state, and any one of the states is detected to be stopped.
Sub-health status data set to be identified:
Figure BDA0001602810590000031
in the formula, n represents the number of extracted sub-health data, shnRepresenting an nth set of sub-health status data;
and fourthly, suspending the storage work of the sub-health data group, carrying out K-means cluster analysis on all elements in the group to obtain a classification result, and emptying the storage space of the sub-health data group.
The method comprises the following specific steps:
step 4.1, initializing the cluster number class _ k as 1, classifying all elements in the sub-health data group into one class, and calculating standard deviation in the class;
and 4.2, judging whether the standard deviation in each class is less than or equal to the threshold standard deviation, if so, marking the class data set and finishing the marking of the class data set, wherein the sub-health state data obtained by the class belongs to the class. Otherwise, entering step 4.3;
4.3, if the standard deviation in the class is larger than the threshold standard deviation, the obtained sub-health state data does not belong to the class;
step 4.4, increasing the number of the clusters class _ K by 1, adding one to the class formed by the elements in the sub-health data group by adopting a K-means method, and respectively calculating the standard deviation in each class;
step 4.5, returning to the step 4.2 until the standard deviation in all the categories is less than or equal to the threshold standard deviation, and obtaining cluster division results;
the cluster partitioning result is: c ═ C1,C2,…,Ck}；
4.6, dividing the sub-health data set according to the cluster division result;
Figure BDA0001602810590000032
wherein: t is ti'Representing the number of the i' th sub-health data; class _ k represents the total number of classes of the sub-health dataset.
Combining the sub-health state data set subjected to cluster analysis with the previous training sample, and updating the sub-health state data set into the initial model to obtain a new classification model;
the training sample data of the new classification model is as follows: x ═ U @;
and then, the probabilistic neural network classification model is created again by using the new training data, so that the model is updated.
And step six, returning to the step two, continuously iterating, updating and perfecting the classification model, identifying the sub-health state on line through the running state data of the monitoring equipment, and timely maintaining when the monitoring equipment is in a fault state.
The invention has the advantages that:
1) the sub-health online identification and diagnosis method based on the performance monitoring data makes up the defect that the traditional fault diagnosis method is easy to misdiagnose, and can reflect the real state of equipment.
2) The sub-health on-line identification and diagnosis method based on the performance monitoring data can identify the sub-health state of the equipment on line through the state data of the monitoring equipment during operation, so that timely and effective measures can be taken according to the state of the equipment, and the loss caused by faults is reduced.
3) The online sub-health recognition and diagnosis method based on the performance monitoring data provides a set of normalized and feasible new method for online diagnosis and recognition of the sub-health state of the equipment.
Drawings
FIG. 1 is a flow chart of a sub-health online identification and diagnosis method based on performance monitoring data according to the present invention;
FIG. 2 is a diagram of a state classification model for a probabilistic neural network constructed in accordance with the present invention;
FIG. 3 is a flow chart of a method of identifying and extracting sub-health status data and storing the same in accordance with the present invention;
FIG. 4 is a flowchart of a method for performing K-means cluster analysis on sub-health data in accordance with the present invention;
FIG. 5 is a distribution diagram of initial training samples extracted in an embodiment of the present invention;
FIG. 6 is a graph of a device monitoring data set and incorrectly identified sub-health data sets in accordance with an embodiment of the present invention;
FIG. 7 is a diagram illustrating a sub-health status data clustering result to be identified according to an embodiment of the present invention;
FIG. 8 is a circuit board status data set according to an embodiment of the present invention;
FIG. 9 is a graph of a training data distribution (cut-off to fault condition) in an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Firstly, establishing a state classification model by adopting a probabilistic neural network, carrying out online monitoring and diagnosis classification, and identifying and extracting sub-health state data; then, performing clustering analysis on the sub-health state data based on a K-means clustering method; and finally, updating the training data of the state classification model so as to obtain a new classification model. And (4) performing iteration, continuously improving the classification model, and if the equipment has a fault state, continuously performing fault maintenance, so as to identify all sub-health states occurring in the operation process of the equipment.
As shown in fig. 1, the specific steps are as follows:
firstly, aiming at a certain monitoring device, establishing an initial model of probabilistic neural network state classification and calculating a threshold standard deviation;
firstly, acquiring measured data of the normal running state of the equipment, and acquiring measured data of the fault state of the equipment in a fault injection mode; then, part of the measured data in normal and fault states is selected as training samples, and an initial model of probabilistic neural network state classification is established, as shown in fig. 2.
Set of training samples
Figure BDA0001602810590000041
Wherein: k is a radical ofiRepresenting the number of the ith type training samples; m represents the total number of classes of training samples.
Figure BDA0001602810590000042
Represents the kth training sample in the ith classiAnd each training sample value is in a p-dimension.
Meanwhile, p-dimensional standard deviations of all data in a normal state and p-dimensional standard deviations of all data in a fault state are calculated respectively, the standard deviation values of all dimensions are compared pairwise correspondingly, the maximum value of all dimensions is selected as a threshold standard deviation, and the threshold standard deviation is p-dimensional.
Step two: carrying out online monitoring and diagnosis classification on the monitoring equipment by using the current model, further identifying and extracting sub-health state data by using a classification result, and putting the sub-health state data into a sub-health state data group;
the current model is initially an initial model; diagnosing the probability value of the classification result as normal or fault;
as shown in fig. 3, the specific steps are as follows:
and 2.1, monitoring the running state of the equipment on line by using the current model, collecting p-dimensional state data in real time to serve as a group, and calculating the Euclidean distance between the current p-dimensional state data and each element in the training sample.
The current p-dimensional state data collected in real time are:
Figure BDA0001602810590000051
euclidean distance: e ═ d-xij)T(d-xij)；
xijRepresenting the jth training sample value in the ith training sample; j is 1,2, …, ki，i＝1,2,…,m；
Step 2.2, activating neurons of radial basis functions of the mode layer by combining Euclidean distance with Gaussian functions;
the gaussian function activation formula is as follows:
Figure BDA0001602810590000052
σiand the maximum value corresponding to each dimension in the standard deviation of the ith type training sample is shown. Pij(d) Representing the output of the training sample of the ith class corresponding to the equivalent neuron of the jth training sample;
step 2.3, the probability that the current p-dimensional state data belongs to the known class is obtained at a probability neural network summation layer:
Figure BDA0001602810590000053
fi,ki(d) representing a current p-dimensional stateProbability that data d belongs to a known class i;
step 2.4, judging whether the sub-health data group is empty, and if so, entering step 2.5; otherwise, entering step 2.6;
the sub-health data set is initially null.
Step 2.5, judging whether the probability value is larger than the average value of the probability sum that the p-dimensional state data at the last moment belongs to various types according to the probability value that the current p-dimensional state data belongs to the known type, and if so, entering step 2.7; otherwise, entering step 2.8;
step 2.6, judging whether the probability value is larger than the probability mean value of the p-dimensional state data which can correctly identify the state at the last moment or not according to the probability value of the current p-dimensional state data belonging to the known category; if yes, go to step 2.7; otherwise, go to step 2.8.
And 2.7, correctly identifying the current p-dimensional state data, wherein the type with the maximum probability value is the identification result state of the current p-dimensional state data.
And 2.8, putting the current p-dimensional state data as the sub-health data to be identified into the sub-health data group to be identified.
Step three, judging whether the sub-health data group to be identified reaches a storage tolerance or a known state appears; if yes, entering the step four; otherwise, returning to the step two, and continuously acquiring the p-dimensional state data in real time by using the current model for analysis.
The known state at the beginning comprises a normal state and a fault state, so the detection of the fault state is stopped;
after the model is updated, the known states comprise a normal state, a sub-health state and a fault state, and any one of the states is detected to be stopped.
Sub-health status data set to be identified:
Figure BDA0001602810590000061
in the formula, n represents the number of extracted sub-health data, shnRepresenting the nth group of sub-health status data；
And fourthly, suspending the storage work of the sub-health data group, carrying out K-means cluster analysis on all elements in the group to obtain a classification result, and emptying the storage space of the sub-health data group.
As shown in fig. 4, the specific steps are as follows:
step 4.1, initializing the cluster number class _ k as 1, classifying all elements in the sub-health data group into one class, and calculating standard deviation in the class;
and 4.2, judging whether the standard deviation in each class is less than or equal to the threshold standard deviation, if so, marking the class data set and finishing the marking of the class data set, wherein the sub-health state data obtained by the class belongs to the class. Otherwise, entering step 4.3;
4.3, if the standard deviation in the class is larger than the threshold standard deviation, the obtained sub-health state data does not belong to the class;
step 4.4, increasing the number of the clusters class _ K by 1, adding one to the class formed by the elements in the sub-health data group by adopting a K-means method, and respectively calculating the standard deviation in each class;
step 4.5, returning to the step 4.2 until the standard deviation in all the categories is less than or equal to the threshold standard deviation, and obtaining cluster division results;
the cluster partitioning result is: c ═ C1,C2,…,Ck}；
4.6, dividing the sub-health data set according to the cluster division result;
Figure BDA0001602810590000062
wherein: t is ti'Representing the number of the i' th sub-health data; class _ k represents the total number of classes of the sub-health dataset.
Combining the sub-health state data set subjected to cluster analysis with the previous training sample, and updating the sub-health state data set into the initial model to obtain a new classification model;
the training sample data of the new classification model is as follows: x ═ U @;
and then, the probabilistic neural network classification model is created again by using the new training data, so that the model is updated.
And step six, returning to the step two, continuously iterating, updating and perfecting the classification model, identifying the sub-health state on line through the running state data of the monitoring equipment, and timely maintaining when the monitoring equipment is in a fault state.
Example (b):
the present embodiment selects a dc power conversion circuit, which includes three parts: the power supply circuit of 18V power supply, the power conversion circuit of 18V to 12V power supply and the power conversion circuit of 12V to 5V power supply are respectively.
The direct-current power supply conversion circuit is provided with 3 monitoring points which are respectively 18V power supply output voltage VOUT, 12V power supply output voltage S +12V and 5V power supply output voltage S + 5V. And respectively monitoring the voltage output of the 3 monitoring points, and collecting voltage data by using a data card so as to evaluate the health state of the circuit.
The 3 monitoring point voltages of VOUT, S +12V and S +5V all reflect the key functions of the circuit, and the voltage outputs of the 3 monitoring points are mutually independent, so that the voltage monitoring values of the 3 monitoring points are selected to evaluate the health state of the direct-current power supply conversion circuit; recording VOUT, S +12V and S +5V monitoring points as T respectively1，T2，T3。
In the embodiment, 1200 groups of normal and fault state data are selected as training samples to construct an initial model. Distribution of initial training data (normal state label is 1, fault state label is 9), as shown in fig. 5:
respectively determining standard deviations of the measured data in the normal state and the fault state, and taking the larger value of each dimension in the standard deviation as a threshold standard deviation of the equipment state data; the standard deviation calculation formula is as follows:
Figure BDA0001602810590000071
normal state standard deviation: 0.0150,0.0137,0.0074
Standard deviation of fault state: 0.0095,0.0050,0.0026
The threshold standard deviation is: 0.0150,0.0137,0.0074
The current model is used for carrying out on-line monitoring and diagnosis classification on the monitoring equipment, and the result shows that: and monitoring the running state of the equipment, wherein in the 2007 group of data, the probability that the data belongs to normal and fault is smaller than the mean probability of the data in the previous group, which indicates that the unknown state of the circuit board begins to appear at the moment. The storage tolerance of the circuit board selected in this case is 2000 groups of data, that is, 2000 groups of state data (that is, 2007-4006 groups of data) are accumulated from the occurrence of an unknown state, and a sub-health data set to be identified is extracted for cluster analysis. The device monitors the data set and the distribution of sub-health data sets that are not correctly identified, as shown in fig. 6.
2000 groups of extracted sub-health data to be identified have standard deviation as follows:
0.0085，0.0092，0.0211
and if the standard deviation is larger than the threshold standard deviation, the extracted data does not belong to the same class, and K-means cluster analysis is carried out.
As shown in FIG. 7, the K-means clustering results:
the number of clustering clusters: class _ k is 2;
mean vector: u. of1＝|17.9950,11.6779,4.5838|，u2＝|17.9913,11.6708,4.4727|
Standard deviation within class: s1＝|0.0086,0.0091,0.0042|，s2＝|0.0054,0.0104,0.0047|
Sub-health data set: { [3x1928double ], [3x72double ] }
Updating the sub-health state into the classification model to obtain a new classification model, wherein the new training data is shown in table 1:
TABLE 1
Figure BDA0001602810590000081
And monitoring the power panel on line, continuously updating and perfecting the classification model until the power panel is in a fault state, suspending working, and performing fault maintenance. The whole process collects a data 10615 group, and the data graph is shown in fig. 8.
Through the steps, the 3 types of sub-health states of the circuit board except the normal state and the fault state can be effectively extracted and identified, and therefore online identification and diagnosis of sub-health are achieved.
The data distribution of the training sample set obtained by the above process is shown in fig. 9:
{[3x1200double]，[3x1200double],[3x1928double],[3x72double],[3x2000double]}
identified sub-health data sets:
{[3x1928double],[3x72double],[3x2000double]}
the board states are shown in table 2:
TABLE 2
Figure BDA0001602810590000082
Figure BDA0001602810590000091

Claims (3)

1. A sub-health online identification and diagnosis method based on performance monitoring data is characterized by comprising the following specific steps:
firstly, aiming at a certain monitoring device, establishing an initial model of probabilistic neural network state classification and calculating a threshold standard deviation;
step two: carrying out online monitoring and diagnosis classification on the monitoring equipment by using the current model, further identifying and extracting sub-health state data by using a classification result, and putting the sub-health state data into a sub-health state data group;
the current model is initially an initial model; diagnosing the probability value of the classification result as normal or fault;
the process of identifying and extracting sub-health status data is as follows:
step 2.1, monitoring the running state of equipment on line by using a current model, collecting p-dimensional state data in real time as a group, and calculating the Euclidean distance between the current p-dimensional state data and each element in a training sample;
the current p-dimensional state data collected in real time are:
Figure FDA0002488779330000011
euclidean distance: e ═ d-xij)T(d-xij)；
xijRepresenting the jth training sample value in the ith training sample; j is 1,2, …, ki，i＝1,2,…,m；
Step 2.2, activating neurons of radial basis functions of the mode layer by combining Euclidean distance with Gaussian functions;
the gaussian function activation formula is as follows:
Figure FDA0002488779330000012
σirepresenting the maximum value corresponding to each dimension in the standard deviation of the ith type of training sample; pij(d) Representing the output of the training sample of the ith class corresponding to the equivalent neuron of the jth training sample;
step 2.3, the probability that the current p-dimensional state data belongs to the known class is obtained at a probability neural network summation layer:
Figure FDA0002488779330000013
Figure FDA0002488779330000014
representing the probability that the current p-dimensional state data d belongs to the known class i;
step 2.4, judging whether the sub-health state data set is empty, and if so, entering step 2.5; otherwise, entering step 2.6;
the initial value of the sub-health state data set is null;
step 2.5, judging whether the probability value is larger than the average value of the probability sum that the p-dimensional state data at the last moment belongs to various types according to the probability value that the current p-dimensional state data belongs to the known type, and if so, entering step 2.7; otherwise, entering step 2.8;
step 2.6, judging whether the probability value is larger than the probability mean value of the p-dimensional state data which can correctly identify the state at the last moment or not according to the probability value of the current p-dimensional state data belonging to the known category; if yes, go to step 2.7; otherwise, entering step 2.8;
step 2.7, the current p-dimensional state data can be correctly identified, and the type with the maximum probability value is the identification result state of the current p-dimensional state data;
and 2.8, putting the current p-dimensional state data which are the sub-health state data to be identified into the sub-health state data group to be identified.
Step three, judging whether the sub-health state data group to be identified reaches a storage tolerance or a known state appears; if yes, entering the step four; otherwise, returning to the step two, and continuously acquiring the p-dimensional state data in real time by using the current model for analysis;
the known state at the beginning comprises a normal state and a fault state, so the detection of the fault state is stopped;
after the model is updated, the known states comprise a normal state, a sub-health state and a fault state, and any one of the states is detected to be stopped;
sub-health status data set to be identified:
Figure FDA0002488779330000021
in the formula, n represents the number of extracted sub-health status data, shnRepresenting an nth set of sub-health status data;
suspending the storage work of the sub-health state data group, performing K-means cluster analysis on all elements in the group to obtain a classification result, and emptying the storage space of the sub-health state data group;
combining the sub-health state data group after the clustering analysis with the previous training sample, and updating the sub-health state data group into the initial model to obtain a new classification model;
the training sample data of the new classification model is as follows: x ═ U @;
x is a training sample set; sh is a sub-health state data group after cluster division;
then, a probabilistic neural network classification model is created again by using new training sample data, so that the model is updated;
and step six, returning to the step two, continuously iterating, updating and perfecting the classification model, identifying the sub-health state on line through the running state data of the monitoring equipment, and timely maintaining when the monitoring equipment is in a fault state.
2. The online sub-health identification and diagnosis method based on performance monitoring data as claimed in claim 1, wherein the first step is specifically:
firstly, acquiring measured data of the normal running state of the equipment, and acquiring measured data of the fault state of the equipment in a fault injection mode;
then, respectively selecting part of measured data in normal and fault states as training samples, and establishing an initial model of probabilistic neural network state classification;
set of training samples
Figure FDA0002488779330000022
Wherein: k is a radical ofiRepresenting the number of the ith type training samples; m represents the total number of classes of training samples;
Figure FDA0002488779330000023
represents the kth training sample in the ith classiTraining sample values, the value being in the p dimension;
meanwhile, p-dimensional standard deviations of all data in a normal state and p-dimensional standard deviations of all data in a fault state are calculated respectively, the standard deviation values of all dimensions are compared pairwise correspondingly, the maximum value of all dimensions is selected as a threshold standard deviation, and the threshold standard deviation is p-dimensional.
3. The on-line sub-health identification and diagnosis method based on performance monitoring data as claimed in claim 1, wherein the four specific steps are as follows:
step 4.1, initializing the cluster number class _ k as 1, classifying all elements in the sub-health state data set into one class, and calculating standard deviation in the class;
step 4.2, judging whether the standard deviation in each class is less than or equal to the threshold standard deviation, if so, marking the class data set and ending; otherwise, entering step 4.3;
4.3, if the standard deviation in the class is larger than the threshold standard deviation, the obtained sub-health state data does not belong to the class;
step 4.4, increasing the cluster number class _ K by 1, adding one to the class formed by the elements in the sub-health state data group by adopting a K-means method, and respectively calculating the standard deviation in each class;
step 4.5, returning to the step 4.2 until the standard deviation in all the categories is less than or equal to the threshold standard deviation, and obtaining cluster division results;
the cluster partitioning result is: c ═ 1,2, …, class _ k };
4.6, dividing the sub-health state data group according to the cluster division result;
Figure FDA0002488779330000031
wherein: t is ti'represents the number of i' th sub-health status data; class _ k represents the total class number of the sub-health status data set.
CN201810231114.5A 2018-03-20 2018-03-20 Sub-health online identification and diagnosis method based on performance monitoring data Active CN108803555B (en)

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 CN108803555A (en) 2018-11-13
CN108803555B true 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)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN110070076B (en) * 2019-05-08 2021-05-18 北京字节跳动网络技术有限公司 Method and device for selecting training samples
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
CN117725538B (en) * 2024-02-07 2024-04-16 河海大学 Water gate deformation multi-measuring-point joint diagnosis system and method

Family Cites Families (5)

* Cited by examiner, † Cited by third party
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
CN103927446B (en) * 2014-04-18 2017-02-15 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
CN105782071B (en) * 2016-03-04 2018-07-13 长沙有色冶金设计研究院有限公司 A kind of water-isolation slurry pump method for diagnosing faults based on probabilistic neural network
CN107153841B (en) * 2017-04-24 2020-02-18 南京康尼机电股份有限公司 Sub-health prediction method for urban rail transit vehicle door system

Also Published As

Publication number Publication date
CN108803555A (en) 2018-11-13

Similar Documents

Publication Publication Date Title
CN108803555B (en) Sub-health online identification and diagnosis method based on performance monitoring data
CN114065613B (en) Multi-working-condition process industrial fault detection and diagnosis method based on deep migration learning
CN107505133B (en) The probability intelligent diagnosing method of rolling bearing fault based on adaptive M RVM
CN105653444B (en) Software defect fault recognition method and system based on internet daily record data
CN110609524B (en) Industrial equipment residual life prediction model and construction method and application thereof
CN109086793A (en) A kind of abnormality recognition method of wind-driven generator
CN104596780B (en) Diagnosis method for sensor faults of motor train unit braking system
CN110757510B (en) Method and system for predicting remaining life of robot
CN111538311B (en) Flexible multi-state self-adaptive early warning method and device for mechanical equipment based on data mining
CN113111591B (en) Automatic diagnosis method, device and equipment based on internal fault of modular power distribution terminal
CN111126820A (en) Electricity stealing prevention method and system
CN108304567A (en) High-tension transformer regime mode identifies and data classification method and system
CN109298633A (en) Chemical production process fault monitoring method based on adaptive piecemeal Non-negative Matrix Factorization
CN112070180B (en) Power grid equipment state judging method and device based on information physical bilateral data
CN114118219A (en) Data-driven real-time abnormal detection method for health state of long-term power-on equipment
CN111398798A (en) Circuit breaker energy storage state identification method based on vibration signal interval characteristic extraction
CN111271183B (en) Method and system for self-adaptive online prediction of state of diesel engine
CN115112372A (en) Bearing fault diagnosis method and device, electronic equipment and storage medium
Wen et al. A new method for identifying the ball screw degradation level based on the multiple classifier system
CN114817856B (en) Beam-pumping unit fault diagnosis method based on structural information retention domain adaptation network
CN115859777A (en) Method for predicting service life of product system in multiple fault modes
CN108827671A (en) A kind of Trouble Diagnostic Method of Machinery Equipment
CN113093695A (en) Data-driven SDN controller fault diagnosis system
CN117633688A (en) Large-scale power data anomaly detection method based on ridge regression-k-means clustering-LOF-LSTM fusion algorithm
Dui et al. Reliability Evaluation and Prediction Method with Small Samples.

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