CN106124205A - A kind of decelerator health analysis system - Google Patents

Abstract

The invention provides a kind of decelerator health analysis system, including the sensing acquisition module for gathering tested retarder operation state mechanical signal；It is connected with described sensing acquisition module, for the data of collection being carried out the data preprocessing module of pretreatment；Being connected with described data preprocessing module, for receiving pretreated data and being uploaded to the communication control module of central processing module, described central processing module for carrying out after-treatment and comprehensively analyzing the running status of tested decelerator by the data received.The invention have the benefit that monitoring and the analysis achieving the health status to decelerator, the spot check of spot check personnel can be played directive function, greatly reduce the routine work amount of field maintenance person, improve work efficiency.

Description

A kind of decelerator health analysis system

Technical field

The present invention relates to industrial monitoring technical field, be specifically related to a kind of decelerator health analysis system.

Background technology

In correlation technique, decelerator uses in numerous industry spot along with motor or the equipment of driving of all size, with Country more and more higher for the attention degree of safety in production, national standard is strengthened again and again, and for the safety monitoring of decelerator Remain in and depend on the experimental check that spot check personnel are daily, demand can not be met the most far away, be badly in need of one the most at present and set The standby health status to decelerator is monitored and analyzed, and provides regular maintenance by data analysis modeling and instructs and fault report Alert.

Summary of the invention

For solving the problems referred to above, it is desirable to provide a kind of decelerator health analysis system.

The purpose of the present invention realizes by the following technical solutions:

A kind of decelerator health analysis system, including adopting for gathering the sensing of tested retarder operation state mechanical signal Collection module；It is connected with described sensing acquisition module, for the data of collection being carried out the data preprocessing module of pretreatment；With institute State data preprocessing module to connect, for receiving pretreated data and being uploaded to the Control on Communication mould of central processing module Block, described central processing module for carrying out after-treatment and comprehensively analyzing the operation shape of tested decelerator by the data received State.

The invention have the benefit that monitoring and the analysis achieving the health status to decelerator, can be to spot check personnel Spot check play directive function, greatly reduce the routine work amount of field maintenance person, improve work efficiency, thus solve Above-mentioned technical problem.

Accompanying drawing explanation

The invention will be further described to utilize accompanying drawing, but the embodiment in accompanying drawing does not constitute any limit to the present invention System, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to obtain according to the following drawings Other accompanying drawing.

Fig. 1 is present configuration schematic diagram；

Fig. 2 is the schematic diagram of inventive sensor fault diagnosis module.

Reference:

Including sensing acquisition module 1, data preprocessing module 2, communication control module 3, central processing module 4, sensor Fault diagnosis module 5, signals collecting filter unit 51, fault signature extraction unit 52, online feature extraction unit 53, feature to Amount preferred cell 54, failure modes recognition unit 55, failure mode updating block 56, health records unit 57.

Detailed description of the invention

The invention will be further described with the following Examples.

Application scenarios 1

See Fig. 1, Fig. 2, the decelerator health analysis system of an embodiment of this application scene, it is subject to including for collection Survey the sensing acquisition module 1 of retarder operation state mechanical signal；It is connected with described sensing acquisition module 1, for by collection Data carry out the data preprocessing module 2 of pretreatment；It is connected with described data preprocessing module 2, is used for receiving pretreated Data are also uploaded to the communication control module 3 of central processing module 4, and described central processing module 4 is for the data that will receive Carry out after-treatment and comprehensively analyze the running status of tested decelerator.

Preferably, described central processing module 4 uses with communication control module 3 one_to_one corresponding.

The above embodiment of the present invention achieves monitoring and the analysis of the health status to decelerator, can be to spot check personnel's Spot check plays directive function, greatly reduces the routine work amount of field maintenance person, improves work efficiency, thus solve Above-mentioned technical problem.

Preferably, described sensing acquisition module 1 includes some acceleration for gathering tested reductor acceleration signal Spend sensor and for gathering the temperature sensor of tested reductor temperature signal.

This preferred embodiment can the health status information of the more efficiently tested reductor of Real-time Collection.

Preferably, described decelerator health analysis system also includes the sensor fault diagnosis diagnosing each sensor Module 5, described sensor fault diagnosis module 5 includes signals collecting filter unit 51, fault signature extraction unit 52, online spy Levy extraction unit 53, characteristic vector preferred cell 54, failure modes recognition unit 55, failure mode updating block 56 and healthy note Record unit 57.

The above embodiment of the present invention arranges sensor fault diagnosis module 5 and achieves sensor fault diagnosis module 5 Fast construction, is conducive to monitoring each sensor, it is ensured that monitoring effectively performs.

Preferably, described signals collecting filter unit 51 is used for gathering historical sensor signal and on-line sensor test letter Number, and use combination form wave filter to be filtered signal processing；

This preferred embodiment arranges combination form wave filter, can effectively remove the various noise jamming of signal, preferably The primitive character information of stick signal.

Preferably, described fault signature extraction unit 52 is for carrying out integrated experience to filtered historical sensor signal Mode decomposition (EEMD) processes, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as training feature vector, including:

(1) the historical sensor signal by collection is divided into the fault-signal of nominal situation signal and plurality of classes；

(2) described historical sensor signal carries out integrated empirical mode decomposition (EEMD) process, it is thus achieved that described history passes The intrinsic mode function of sensor signal and remainder function；

(3) intrinsic mode function and the Energy-Entropy of remainder function of described historical sensor signal are calculated；

(4) Energy-Entropy of historical sensor signal is normalized, extracts the Energy-Entropy after normalization as instruction Practice characteristic vector；

Described online feature extraction unit 53 is for carrying out integrated Empirical Mode to filtered on-line sensor test signal State is decomposed (EEMD) and is processed, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as characteristic vector to be measured, including:

(1) described on-line sensor test signal is carried out EEMD process, it is thus achieved that described on-line sensor test signal Intrinsic mode function and remainder function；

(2) intrinsic mode function and the Energy-Entropy of remainder function of described on-line sensor test signal are calculated；

(3) Energy-Entropy of on-line sensor test signal is normalized, extracts the Energy-Entropy after normalization and make For characteristic vector to be measured.

This preferred embodiment carries out integrated empirical mode decomposition (EEMD) to the sensor signal gathered and processes, it is possible to effectively Elimination modal overlap phenomenon, the effect of decomposition is preferable.

Preferably, training feature vector is carried out similar with characteristic vector to be measured by described characteristic vector preferred cell 54 respectively Property tolerance, the characteristic vector high for similarity reject, including:

(1) two vector similarities function S (X, Y) are defined:

$S(X,Y)=\frac{\mathrm{cov}(X,Y)}{\sqrt{D\left(X\right)}\sqrt{D\left(Y\right)}}$

In formula, X, Y represent that two characteristic vectors, cov (X, Y) are the covariance of X and Y respectively,For X, Y standard deviation；

For any two training feature vector X₁、X₂, and any two characteristic vector D to be measured₁、D₂, it is respectively adopted similar Its similarity is measured by degree function, obtains S (X₁,X₂) and S (D₁,D₂)；

(2) for S (X₁,X₂) and S (D₁,D₂), if S is (X₁,X₂)>T₁, T₁∈ (0.9,1), only chooses X₁As training characteristics Vector, if S is (D₁,D₂)>T₂, T₂∈ (0.95,1), only chooses D₁As characteristic vector to be measured.

This preferred embodiment screens characteristic vector by measuring similarity, it is possible to reduce amount of calculation, improves efficiency.

Preferably, described failure modes recognition unit 55 is for using the least square method supporting vector machine of optimization to treat described Survey characteristic vector and carry out failure modes identification, select optimize submodule, training submodule and identify submodule, specifically including parameter For:

Described parameter selects the kernel function optimizing submodule for constructing least square method supporting vector machine, and to least square The structural parameters of support vector machine use multi-population to work in coordination with Chaos particle swarm optimization algorithm and are optimized；

Described training submodule, for using many classification sides of the least square support vector machines of the optimum binary tree structure of improvement Method, instructs the least square method supporting vector machine after structure parameter optimizing using the training feature vector obtained as training sample Practice, and build sensor fault diagnosis model；

Described identification submodule is used for using described sensor fault diagnosis model that described characteristic vector to be measured carries out event Barrier Classification and Identification；

Wherein, it is considered to Polynomial kernel function and the superiority of RBF kernel function, the core letter of described least square method supporting vector machine Number is configured to:

K=(1-δ) (xx_i+1)^p+δexp(-‖x-x_i‖²/σ²)

In formula, δ is the structure adjusting factor, and the span of δ is set as [0.45,0.55], and p is the rank of Polynomial kernel function Number, σ²For RBF kernel functional parameter.

Wherein, shown employing multi-population is worked in coordination with Chaos particle swarm optimization algorithm and is optimized, including:

(1) to main population and initialize from population respectively, randomly generate initial as particle of one group of parameter Position and initial velocity, definition fitness function is:

$S=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left|\frac{q_{i}W}{q_{i}W+(1-q_i)T}\right|\×100\%$

In formula, N is the total number of training sample, and W is that bug is classified number, and T is that fault is correctly classified number, q_iFor certainly The weight coefficient set, q_iSpan be set as [0.4,0.5]；

(2) renewal from population is carried out, in every generation renewal process, according to fitness function, from population respectively The speed of more new particle and position, then will be experienced in its history adaptive optimal control angle value and main population body each particle The fitness value of desired positions compares, if more preferably, then as current global optimum position；

(3) described global optimum position is carried out the optimum particle position in chaos optimization, and iteration current sequence and speed Degree, generates optimal particle sequence；

(4) in the main population of every generation, choose from population optimum particle, and the position of more new particle and speed, Until reaching maximum iteration time or meeting the error requirements of fitness function.

Wherein, many sorting techniques of the least square support vector machines of the optimum binary tree structure of described improvement specifically include:

(1) calculate the standard variance of all training samples and two classifications j,Between Separatory measure；

(2) output minimum separation estimate correspondence j,

(3) after being optimized the structural parameters of least square method supporting vector machine, the least square setting up two classification props up Hold vector machine in order to train jth class andThe training sample of class, forms optimum two classification least square method supporting vector machines, output The parameter of discriminant function,The training sample of class is merged in j class, constitutes new j class training sample；

(4) all of classification is circulated training according to (1)-(3), until the optimum root node of output；

(5) according to the categorised decision tree of above output result composition least square method supporting vector machine, then to remaining instruction Practice sample and carry out classifying quality test.

This preferred embodiment is in order to improve the precision of fault diagnosis, and employing training speed is fast, generalization ability strong and robustness Preferably least square support vector machines is as grader, and proposes the many sorting techniques improving optimum binary tree structure, with between class Separatory measure substitutes the weights in binary tree structure, the nicety of grading that improve and classification speed；In view of RBF kernel function it is Karyomerite function, Polynomial kernel function is overall situation kernel function, and karyomerite function learning ability is strong, and Generalization Capability is relatively weak, and Overall situation kernel function Generalization Capability is strong, and learning capacity is relatively weak, carries out on the basis of the advantage of summary two class kernel function The Kernel of least square method supporting vector machine, optimizes classification performance and the Generalization Capability of least square method supporting vector machine； The multi-population of design works in coordination with Chaos particle swarm optimization algorithm, has preferable convergence rate, and has preferable global and local Optimizing performance, it is possible to jump out Local Extremum timely, finds the optimal value of the overall situation, thus uses multi-population to work in coordination with chaotic particle The structural parameters of least square method supporting vector machine are optimized by colony optimization algorithm, and effect of optimization is good.

Preferably, described failure mode updating block 56, for being updated training set, continues to optimize sensor fault Diagnostic cast, including:

(1) when sensor fault diagnosis model cannot carry out effective failure modes to characteristic vector to be measured, by feature to be measured Vector is as new training feature vector；

(2) training sample is updated by new training feature vector, to the least square support after structure parameter optimizing Vector machine is trained, and builds the sensor fault diagnosis model made new advances；

(3) use new sensor fault diagnosis model that described characteristic vector to be measured is carried out failure modes identification, complete Failure mode updates.

This preferred embodiment arranges failure mode updating block 56, to improve adaptation ability and the range of application of model.

Preferably, described health records unit 57 includes sub module stored and secure access submodule, described storage submodule Block uses storage model based on cloud storage, specifically, is encrypted after being compressed by fault message, is uploaded to cloud storage Device, described secure access submodule, for conducting interviews information, specifically, corresponding to sub module stored, downloads data to This locality, after using corresponding secret key to be unlocked, then carries out decompressing to read information.

This preferred embodiment arranges health records unit 57, on the one hand ensure that information security, on the other hand can be at any time Fault is conducted interviews, it is simple to search problem.

In this application scenarios, set threshold value T₁Value be 0.96, the monitoring velocity phase of sensor fault diagnosis module 5 To improve 10%, the monitoring accuracy of sensor fault diagnosis module 5 improves 12% relatively.

Application scenarios 2

See Fig. 1, Fig. 2, the decelerator health analysis system of an embodiment of this application scene, it is subject to including for collection Survey the sensing acquisition module 1 of retarder operation state mechanical signal；It is connected with described sensing acquisition module 1, for by collection Data carry out the data preprocessing module 2 of pretreatment；It is connected with described data preprocessing module 2, is used for receiving pretreated Data are also uploaded to the communication control module 3 of central processing module 4, and described central processing module 4 is for the data that will receive Carry out after-treatment and comprehensively analyze the running status of tested decelerator.

Preferably, described central processing module 4 uses with communication control module 3 one_to_one corresponding.

The above embodiment of the present invention achieves monitoring and the analysis of the health status to decelerator, can be to spot check personnel's Spot check plays directive function, greatly reduces the routine work amount of field maintenance person, improves work efficiency, thus solve Above-mentioned technical problem.

Preferably, described sensing acquisition module 1 includes some acceleration for gathering tested reductor acceleration signal Spend sensor and for gathering the temperature sensor of tested reductor temperature signal.

This preferred embodiment can the health status information of the more efficiently tested reductor of Real-time Collection.

Preferably, described decelerator health analysis system also includes the sensor fault diagnosis diagnosing each sensor Module 5, described sensor fault diagnosis module 5 includes signals collecting filter unit 51, fault signature extraction unit 52, online spy Levy extraction unit 53, characteristic vector preferred cell 54, failure modes recognition unit 55, failure mode updating block 56 and healthy note Record unit 57.

The above embodiment of the present invention arranges sensor fault diagnosis module 5 and achieves sensor fault diagnosis module 5 Fast construction, is conducive to monitoring each sensor, it is ensured that monitoring effectively performs.

Preferably, described signals collecting filter unit 51 is used for gathering historical sensor signal and on-line sensor test letter Number, and use combination form wave filter to be filtered signal processing；

This preferred embodiment arranges combination form wave filter, can effectively remove the various noise jamming of signal, preferably The primitive character information of stick signal.

Preferably, described fault signature extraction unit 52 is for carrying out integrated experience to filtered historical sensor signal Mode decomposition (EEMD) processes, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as training feature vector, including:

(1) the historical sensor signal by collection is divided into the fault-signal of nominal situation signal and plurality of classes；

(2) described historical sensor signal carries out integrated empirical mode decomposition (EEMD) process, it is thus achieved that described history passes The intrinsic mode function of sensor signal and remainder function；

(3) intrinsic mode function and the Energy-Entropy of remainder function of described historical sensor signal are calculated；

(4) Energy-Entropy of historical sensor signal is normalized, extracts the Energy-Entropy after normalization as instruction Practice characteristic vector；

Described online feature extraction unit 53 is for carrying out integrated Empirical Mode to filtered on-line sensor test signal State is decomposed (EEMD) and is processed, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as characteristic vector to be measured, including:

(1) described on-line sensor test signal is carried out EEMD process, it is thus achieved that described on-line sensor test signal Intrinsic mode function and remainder function；

(2) intrinsic mode function and the Energy-Entropy of remainder function of described on-line sensor test signal are calculated；

(3) Energy-Entropy of on-line sensor test signal is normalized, extracts the Energy-Entropy after normalization and make For characteristic vector to be measured.

This preferred embodiment carries out integrated empirical mode decomposition (EEMD) to the sensor signal gathered and processes, it is possible to effectively Elimination modal overlap phenomenon, the effect of decomposition is preferable.

Preferably, training feature vector is carried out similar with characteristic vector to be measured by described characteristic vector preferred cell 54 respectively Property tolerance, the characteristic vector high for similarity reject, including:

(1) two vector similarities function S (X, Y) are defined:

$S(X,Y)=\frac{\mathrm{cov}(X,Y)}{\sqrt{D\left(X\right)}\sqrt{D\left(Y\right)}}$

In formula, X, Y represent that two characteristic vectors, cov (X, Y) are the covariance of X and Y respectively,For X, Y standard deviation；

For any two training feature vector X₁、X₂, and any two characteristic vector D to be measured₁、D₂, it is respectively adopted similar Its similarity is measured by degree function, obtains S (X₁,X₂) and S (D₁,D₂)；

(2) for S (X₁,X₂) and S (D₁,D₂), if S is (X₁,X₂)>T₁, T₁∈ (0.9,1), only chooses X₁As training characteristics Vector, if S is (D₁,D₂)>T₂, T₂∈ (0.95,1), only chooses D₁As characteristic vector to be measured.

This preferred embodiment screens characteristic vector by measuring similarity, it is possible to reduce amount of calculation, improves efficiency.

Preferably, described failure modes recognition unit 55 is for using the least square method supporting vector machine of optimization to treat described Survey characteristic vector and carry out failure modes identification, select optimize submodule, training submodule and identify submodule, specifically including parameter For:

Described parameter selects the kernel function optimizing submodule for constructing least square method supporting vector machine, and to least square The structural parameters of support vector machine use multi-population to work in coordination with Chaos particle swarm optimization algorithm and are optimized；

Described training submodule, for using many classification sides of the least square support vector machines of the optimum binary tree structure of improvement Method, instructs the least square method supporting vector machine after structure parameter optimizing using the training feature vector obtained as training sample Practice, and build sensor fault diagnosis model；

Described identification submodule is used for using described sensor fault diagnosis model that described characteristic vector to be measured carries out event Barrier Classification and Identification；

Wherein, it is considered to Polynomial kernel function and the superiority of RBF kernel function, the core letter of described least square method supporting vector machine Number is configured to:

K=(1-δ) (xx_i+1)^p+δexp(-‖x-x_i‖²/σ²)

In formula, δ is the structure adjusting factor, and the span of δ is set as [0.45,0.55], and p is the rank of Polynomial kernel function Number, σ²For RBF kernel functional parameter.

Wherein, shown employing multi-population is worked in coordination with Chaos particle swarm optimization algorithm and is optimized, including:

(1) to main population and initialize from population respectively, randomly generate initial as particle of one group of parameter Position and initial velocity, definition fitness function is:

$S=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left|\frac{q_{i}W}{q_{i}W+(1-q_i)T}\right|\×100\%$

In formula, N is the total number of training sample, and W is that bug is classified number, and T is that fault is correctly classified number, q_iFor certainly The weight coefficient set, q_iSpan be set as [0.4,0.5]；

(2) renewal from population is carried out, in every generation renewal process, according to fitness function, from population respectively The speed of more new particle and position, then will be experienced in its history adaptive optimal control angle value and main population body each particle The fitness value of desired positions compares, if more preferably, then as current global optimum position；

(3) described global optimum position is carried out the optimum particle position in chaos optimization, and iteration current sequence and speed Degree, generates optimal particle sequence；

(4) in the main population of every generation, choose from population optimum particle, and the position of more new particle and speed, Until reaching maximum iteration time or meeting the error requirements of fitness function.

Wherein, many sorting techniques of the least square support vector machines of the optimum binary tree structure of described improvement specifically include:

(1) calculate the standard variance of all training samples and two classifications j,Between Separatory measure；

(2) output minimum separation estimate correspondence j,

(3) after being optimized the structural parameters of least square method supporting vector machine, the least square setting up two classification props up Hold vector machine in order to train jth class andThe training sample of class, forms optimum two classification least square method supporting vector machines, output The parameter of discriminant function,The training sample of class is merged in j class, constitutes new j class training sample；

(4) all of classification is circulated training according to (1)-(3), until the optimum root node of output；

(5) according to the categorised decision tree of above output result composition least square method supporting vector machine, then to remaining instruction Practice sample and carry out classifying quality test.

This preferred embodiment is in order to improve the precision of fault diagnosis, and employing training speed is fast, generalization ability strong and robustness Preferably least square support vector machines is as grader, and proposes the many sorting techniques improving optimum binary tree structure, with between class Separatory measure substitutes the weights in binary tree structure, the nicety of grading that improve and classification speed；In view of RBF kernel function it is Karyomerite function, Polynomial kernel function is overall situation kernel function, and karyomerite function learning ability is strong, and Generalization Capability is relatively weak, and Overall situation kernel function Generalization Capability is strong, and learning capacity is relatively weak, carries out on the basis of the advantage of summary two class kernel function The Kernel of least square method supporting vector machine, optimizes classification performance and the Generalization Capability of least square method supporting vector machine； The multi-population of design works in coordination with Chaos particle swarm optimization algorithm, has preferable convergence rate, and has preferable global and local Optimizing performance, it is possible to jump out Local Extremum timely, finds the optimal value of the overall situation, thus uses multi-population to work in coordination with chaotic particle The structural parameters of least square method supporting vector machine are optimized by colony optimization algorithm, and effect of optimization is good.

Preferably, described failure mode updating block 56, for being updated training set, continues to optimize sensor fault Diagnostic cast, including:

(1) when sensor fault diagnosis model cannot carry out effective failure modes to characteristic vector to be measured, by feature to be measured Vector is as new training feature vector；

(2) training sample is updated by new training feature vector, to the least square support after structure parameter optimizing Vector machine is trained, and builds the sensor fault diagnosis model made new advances；

(3) use new sensor fault diagnosis model that described characteristic vector to be measured is carried out failure modes identification, complete Failure mode updates.

This preferred embodiment arranges failure mode updating block 56, to improve adaptation ability and the range of application of model.

Preferably, described health records unit 57 includes sub module stored and secure access submodule, described storage submodule Block uses storage model based on cloud storage, specifically, is encrypted after being compressed by fault message, is uploaded to cloud storage Device, described secure access submodule, for conducting interviews information, specifically, corresponding to sub module stored, downloads data to This locality, after using corresponding secret key to be unlocked, then carries out decompressing to read information.

This preferred embodiment arranges health records unit 57, on the one hand ensure that information security, on the other hand can be at any time Fault is conducted interviews, it is simple to search problem.

In this application scenarios, set threshold value T₁Value be 0.95, the monitoring velocity phase of sensor fault diagnosis module 5 To improve 11%, the monitoring accuracy of sensor fault diagnosis module 5 improves 11% relatively.

Application scenarios 3

See Fig. 1, Fig. 2, the decelerator health analysis system of an embodiment of this application scene, it is subject to including for collection Survey the sensing acquisition module 1 of retarder operation state mechanical signal；It is connected with described sensing acquisition module 1, for by collection Data carry out the data preprocessing module 2 of pretreatment；It is connected with described data preprocessing module 2, is used for receiving pretreated Data are also uploaded to the communication control module 3 of central processing module 4, and described central processing module 4 is for the data that will receive Carry out after-treatment and comprehensively analyze the running status of tested decelerator.

Preferably, described central processing module 4 uses with communication control module 3 one_to_one corresponding.

The above embodiment of the present invention achieves monitoring and the analysis of the health status to decelerator, can be to spot check personnel's Spot check plays directive function, greatly reduces the routine work amount of field maintenance person, improves work efficiency, thus solve Above-mentioned technical problem.

Preferably, described sensing acquisition module 1 includes some acceleration for gathering tested reductor acceleration signal Spend sensor and for gathering the temperature sensor of tested reductor temperature signal.

This preferred embodiment can the health status information of the more efficiently tested reductor of Real-time Collection.

Preferably, described decelerator health analysis system also includes the sensor fault diagnosis diagnosing each sensor Module 5, described sensor fault diagnosis module 5 includes signals collecting filter unit 51, fault signature extraction unit 52, online spy Levy extraction unit 53, characteristic vector preferred cell 54, failure modes recognition unit 55, failure mode updating block 56 and healthy note Record unit 57.

The above embodiment of the present invention arranges sensor fault diagnosis module 5 and achieves sensor fault diagnosis module 5 Fast construction, is conducive to monitoring each sensor, it is ensured that monitoring effectively performs.

Preferably, described signals collecting filter unit 51 is used for gathering historical sensor signal and on-line sensor test letter Number, and use combination form wave filter to be filtered signal processing；

This preferred embodiment arranges combination form wave filter, can effectively remove the various noise jamming of signal, preferably The primitive character information of stick signal.

Preferably, described fault signature extraction unit 52 is for carrying out integrated experience to filtered historical sensor signal Mode decomposition (EEMD) processes, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as training feature vector, including:

(1) the historical sensor signal by collection is divided into the fault-signal of nominal situation signal and plurality of classes；

(2) described historical sensor signal carries out integrated empirical mode decomposition (EEMD) process, it is thus achieved that described history passes The intrinsic mode function of sensor signal and remainder function；

(3) intrinsic mode function and the Energy-Entropy of remainder function of described historical sensor signal are calculated；

(4) Energy-Entropy of historical sensor signal is normalized, extracts the Energy-Entropy after normalization as instruction Practice characteristic vector；

Described online feature extraction unit 53 is for carrying out integrated Empirical Mode to filtered on-line sensor test signal State is decomposed (EEMD) and is processed, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as characteristic vector to be measured, including:

(1) described on-line sensor test signal is carried out EEMD process, it is thus achieved that described on-line sensor test signal Intrinsic mode function and remainder function；

(2) intrinsic mode function and the Energy-Entropy of remainder function of described on-line sensor test signal are calculated；

(3) Energy-Entropy of on-line sensor test signal is normalized, extracts the Energy-Entropy after normalization and make For characteristic vector to be measured.

This preferred embodiment carries out integrated empirical mode decomposition (EEMD) to the sensor signal gathered and processes, it is possible to effectively Elimination modal overlap phenomenon, the effect of decomposition is preferable.

Preferably, training feature vector is carried out similar with characteristic vector to be measured by described characteristic vector preferred cell 54 respectively Property tolerance, the characteristic vector high for similarity reject, including:

(1) two vector similarities function S (X, Y) are defined:

$S(X,Y)=\frac{\mathrm{cov}(X,Y)}{\sqrt{D\left(X\right)}\sqrt{D\left(Y\right)}}$

In formula, X, Y represent that two characteristic vectors, cov (X, Y) are the covariance of X and Y respectively,For X, Y standard deviation；

For any two training feature vector X₁、X₂, and any two characteristic vector D to be measured₁、D₂, it is respectively adopted similar Its similarity is measured by degree function, obtains S (X₁,X₂) and S (D₁,D₂)；

(2) for S (X₁,X₂) and S (D₁,D₂), if S is (X₁,X₂)>T₁, T₁∈ (0.9,1), only chooses X₁As training characteristics Vector, if S is (D₁,D₂)>T₂, T₂∈ (0.95,1), only chooses D₁As characteristic vector to be measured.

This preferred embodiment screens characteristic vector by measuring similarity, it is possible to reduce amount of calculation, improves efficiency.

Preferably, described failure modes recognition unit 55 is for using the least square method supporting vector machine of optimization to treat described Survey characteristic vector and carry out failure modes identification, select optimize submodule, training submodule and identify submodule, specifically including parameter For:

Described parameter selects the kernel function optimizing submodule for constructing least square method supporting vector machine, and to least square The structural parameters of support vector machine use multi-population to work in coordination with Chaos particle swarm optimization algorithm and are optimized；

Described training submodule, for using many classification sides of the least square support vector machines of the optimum binary tree structure of improvement Method, instructs the least square method supporting vector machine after structure parameter optimizing using the training feature vector obtained as training sample Practice, and build sensor fault diagnosis model；

Described identification submodule is used for using described sensor fault diagnosis model that described characteristic vector to be measured carries out event Barrier Classification and Identification；

Wherein, it is considered to Polynomial kernel function and the superiority of RBF kernel function, the core letter of described least square method supporting vector machine Number is configured to:

K=(1-δ) (xx_i+1)^p+δexp(-‖x-x_i‖²/σ²)

In formula, δ is the structure adjusting factor, and the span of δ is set as [0.45,0.55], and p is the rank of Polynomial kernel function Number, σ²For RBF kernel functional parameter.

Wherein, shown employing multi-population is worked in coordination with Chaos particle swarm optimization algorithm and is optimized, including:

(1) to main population and initialize from population respectively, randomly generate initial as particle of one group of parameter Position and initial velocity, definition fitness function is:

$S=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left|\frac{q_{i}W}{q_{i}W+(1-q_i)T}\right|\×100\%$

In formula, N is the total number of training sample, and W is that bug is classified number, and T is that fault is correctly classified number, q_iFor certainly The weight coefficient set, q_iSpan be set as [0.4,0.5]；

(2) renewal from population is carried out, in every generation renewal process, according to fitness function, from population respectively The speed of more new particle and position, then will be experienced in its history adaptive optimal control angle value and main population body each particle The fitness value of desired positions compares, if more preferably, then as current global optimum position；

(3) described global optimum position is carried out the optimum particle position in chaos optimization, and iteration current sequence and speed Degree, generates optimal particle sequence；

(4) in the main population of every generation, choose from population optimum particle, and the position of more new particle and speed, Until reaching maximum iteration time or meeting the error requirements of fitness function.

Wherein, many sorting techniques of the least square support vector machines of the optimum binary tree structure of described improvement specifically include:

(1) calculate the standard variance of all training samples and two classifications j,Between Separatory measure；

(2) output minimum separation estimate correspondence j,

(3) after being optimized the structural parameters of least square method supporting vector machine, the least square setting up two classification props up Hold vector machine in order to train jth class andThe training sample of class, forms optimum two classification least square method supporting vector machines, output The parameter of discriminant function,The training sample of class is merged in j class, constitutes new j class training sample；

(4) all of classification is circulated training according to (1)-(3), until the optimum root node of output；

(5) according to the categorised decision tree of above output result composition least square method supporting vector machine, then to remaining instruction Practice sample and carry out classifying quality test.

This preferred embodiment is in order to improve the precision of fault diagnosis, and employing training speed is fast, generalization ability strong and robustness Preferably least square support vector machines is as grader, and proposes the many sorting techniques improving optimum binary tree structure, with between class Separatory measure substitutes the weights in binary tree structure, the nicety of grading that improve and classification speed；In view of RBF kernel function it is Karyomerite function, Polynomial kernel function is overall situation kernel function, and karyomerite function learning ability is strong, and Generalization Capability is relatively weak, and Overall situation kernel function Generalization Capability is strong, and learning capacity is relatively weak, carries out on the basis of the advantage of summary two class kernel function The Kernel of least square method supporting vector machine, optimizes classification performance and the Generalization Capability of least square method supporting vector machine； The multi-population of design works in coordination with Chaos particle swarm optimization algorithm, has preferable convergence rate, and has preferable global and local Optimizing performance, it is possible to jump out Local Extremum timely, finds the optimal value of the overall situation, thus uses multi-population to work in coordination with chaotic particle The structural parameters of least square method supporting vector machine are optimized by colony optimization algorithm, and effect of optimization is good.

Preferably, described failure mode updating block 56, for being updated training set, continues to optimize sensor fault Diagnostic cast, including:

(1) when sensor fault diagnosis model cannot carry out effective failure modes to characteristic vector to be measured, by feature to be measured Vector is as new training feature vector；

(2) training sample is updated by new training feature vector, to the least square support after structure parameter optimizing Vector machine is trained, and builds the sensor fault diagnosis model made new advances；

(3) use new sensor fault diagnosis model that described characteristic vector to be measured is carried out failure modes identification, complete Failure mode updates.

This preferred embodiment arranges failure mode updating block 56, to improve adaptation ability and the range of application of model.

Preferably, described health records unit 57 includes sub module stored and secure access submodule, described storage submodule Block uses storage model based on cloud storage, specifically, is encrypted after being compressed by fault message, is uploaded to cloud storage Device, described secure access submodule, for conducting interviews information, specifically, corresponding to sub module stored, downloads data to This locality, after using corresponding secret key to be unlocked, then carries out decompressing to read information.

This preferred embodiment arranges health records unit 57, on the one hand ensure that information security, on the other hand can be at any time Fault is conducted interviews, it is simple to search problem.

In this application scenarios, set threshold value T₁Value be 0.94, the monitoring velocity phase of sensor fault diagnosis module 5 To improve 12%, the monitoring accuracy of sensor fault diagnosis module 5 improves 10% relatively.

Application scenarios 4

See Fig. 1, Fig. 2, the decelerator health analysis system of an embodiment of this application scene, it is subject to including for collection Survey the sensing acquisition module 1 of retarder operation state mechanical signal；It is connected with described sensing acquisition module 1, for by collection Data carry out the data preprocessing module 2 of pretreatment；It is connected with described data preprocessing module 2, is used for receiving pretreated Data are also uploaded to the communication control module 3 of central processing module 4, and described central processing module 4 is for the data that will receive Carry out after-treatment and comprehensively analyze the running status of tested decelerator.

Preferably, described central processing module 4 uses with communication control module 3 one_to_one corresponding.

The above embodiment of the present invention achieves monitoring and the analysis of the health status to decelerator, can be to spot check personnel's Spot check plays directive function, greatly reduces the routine work amount of field maintenance person, improves work efficiency, thus solve Above-mentioned technical problem.

Preferably, described sensing acquisition module 1 includes some acceleration for gathering tested reductor acceleration signal Spend sensor and for gathering the temperature sensor of tested reductor temperature signal.

This preferred embodiment can the health status information of the more efficiently tested reductor of Real-time Collection.

Preferably, described decelerator health analysis system also includes the sensor fault diagnosis diagnosing each sensor Module 5, described sensor fault diagnosis module 5 includes signals collecting filter unit 51, fault signature extraction unit 52, online spy Levy extraction unit 53, characteristic vector preferred cell 54, failure modes recognition unit 55, failure mode updating block 56 and healthy note Record unit 57.

The above embodiment of the present invention arranges sensor fault diagnosis module 5 and achieves sensor fault diagnosis module 5 Fast construction, is conducive to monitoring each sensor, it is ensured that monitoring effectively performs.

Preferably, described signals collecting filter unit 51 is used for gathering historical sensor signal and on-line sensor test letter Number, and use combination form wave filter to be filtered signal processing；

This preferred embodiment arranges combination form wave filter, can effectively remove the various noise jamming of signal, preferably The primitive character information of stick signal.

Preferably, described fault signature extraction unit 52 is for carrying out integrated experience to filtered historical sensor signal Mode decomposition (EEMD) processes, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as training feature vector, including:

(1) the historical sensor signal by collection is divided into the fault-signal of nominal situation signal and plurality of classes；

(2) described historical sensor signal carries out integrated empirical mode decomposition (EEMD) process, it is thus achieved that described history passes The intrinsic mode function of sensor signal and remainder function；

(3) intrinsic mode function and the Energy-Entropy of remainder function of described historical sensor signal are calculated；

(4) Energy-Entropy of historical sensor signal is normalized, extracts the Energy-Entropy after normalization as instruction Practice characteristic vector；

Described online feature extraction unit 53 is for carrying out integrated Empirical Mode to filtered on-line sensor test signal State is decomposed (EEMD) and is processed, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as characteristic vector to be measured, including:

(1) described on-line sensor test signal is carried out EEMD process, it is thus achieved that described on-line sensor test signal Intrinsic mode function and remainder function；

(2) intrinsic mode function and the Energy-Entropy of remainder function of described on-line sensor test signal are calculated；

(3) Energy-Entropy of on-line sensor test signal is normalized, extracts the Energy-Entropy after normalization and make For characteristic vector to be measured.

This preferred embodiment carries out integrated empirical mode decomposition (EEMD) to the sensor signal gathered and processes, it is possible to effectively Elimination modal overlap phenomenon, the effect of decomposition is preferable.

Preferably, training feature vector is carried out similar with characteristic vector to be measured by described characteristic vector preferred cell 54 respectively Property tolerance, the characteristic vector high for similarity reject, including:

(1) two vector similarities function S (X, Y) are defined:

$S(X,Y)=\frac{\mathrm{cov}(X,Y)}{\sqrt{D\left(X\right)}\sqrt{D\left(Y\right)}}$

In formula, X, Y represent that two characteristic vectors, cov (X, Y) are the covariance of X and Y respectively,For X, Y standard deviation；

For any two training feature vector X₁、X₂, and any two characteristic vector D to be measured₁、D₂, it is respectively adopted similar Its similarity is measured by degree function, obtains S (X₁,X₂) and S (D₁,D₂)；

(2) for S (X₁,X₂) and S (D₁,D₂), if S is (X₁,X₂)>T₁, T₁∈ (0.9,1), only chooses X₁As training characteristics Vector, if S is (D₁,D₂)>T₂, T₂∈ (0.95,1), only chooses D₁As characteristic vector to be measured.

This preferred embodiment screens characteristic vector by measuring similarity, it is possible to reduce amount of calculation, improves efficiency.

Preferably, described failure modes recognition unit 55 is for using the least square method supporting vector machine of optimization to treat described Survey characteristic vector and carry out failure modes identification, select optimize submodule, training submodule and identify submodule, specifically including parameter For:

Described parameter selects the kernel function optimizing submodule for constructing least square method supporting vector machine, and to least square The structural parameters of support vector machine use multi-population to work in coordination with Chaos particle swarm optimization algorithm and are optimized；

Described training submodule, for using many classification sides of the least square support vector machines of the optimum binary tree structure of improvement Method, instructs the least square method supporting vector machine after structure parameter optimizing using the training feature vector obtained as training sample Practice, and build sensor fault diagnosis model；

Described identification submodule is used for using described sensor fault diagnosis model that described characteristic vector to be measured carries out event Barrier Classification and Identification；

Wherein, it is considered to Polynomial kernel function and the superiority of RBF kernel function, the core letter of described least square method supporting vector machine Number is configured to:

K=(1-δ) (xx_i+1)^p+δexp(-‖x-x_i‖²/σ²)

In formula, δ is the structure adjusting factor, and the span of δ is set as [0.45,0.55], and p is the rank of Polynomial kernel function Number, σ²For RBF kernel functional parameter.

Wherein, shown employing multi-population is worked in coordination with Chaos particle swarm optimization algorithm and is optimized, including:

(1) to main population and initialize from population respectively, randomly generate initial as particle of one group of parameter Position and initial velocity, definition fitness function is:

$S=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left|\frac{q_{i}W}{q_{i}W+(1-q_i)T}\right|\×100\%$

In formula, N is the total number of training sample, and W is that bug is classified number, and T is that fault is correctly classified number, q_iFor certainly The weight coefficient set, q_iSpan be set as [0.4,0.5]；

(2) renewal from population is carried out, in every generation renewal process, according to fitness function, from population respectively The speed of more new particle and position, then will be experienced in its history adaptive optimal control angle value and main population body each particle The fitness value of desired positions compares, if more preferably, then as current global optimum position；

(3) described global optimum position is carried out the optimum particle position in chaos optimization, and iteration current sequence and speed Degree, generates optimal particle sequence；

(4) in the main population of every generation, choose from population optimum particle, and the position of more new particle and speed, Until reaching maximum iteration time or meeting the error requirements of fitness function.

Wherein, many sorting techniques of the least square support vector machines of the optimum binary tree structure of described improvement specifically include:

(1) calculate the standard variance of all training samples and two classifications j,Between Separatory measure；

(2) output minimum separation estimate correspondence j,

(3) after being optimized the structural parameters of least square method supporting vector machine, the least square setting up two classification props up Hold vector machine in order to train jth class andThe training sample of class, forms optimum two classification least square method supporting vector machines, output The parameter of discriminant function,The training sample of class is merged in j class, constitutes new j class training sample；

(4) all of classification is circulated training according to (1)-(3), until the optimum root node of output；

(5) according to the categorised decision tree of above output result composition least square method supporting vector machine, then to remaining instruction Practice sample and carry out classifying quality test.

This preferred embodiment is in order to improve the precision of fault diagnosis, and employing training speed is fast, generalization ability strong and robustness Preferably least square support vector machines is as grader, and proposes the many sorting techniques improving optimum binary tree structure, with between class Separatory measure substitutes the weights in binary tree structure, the nicety of grading that improve and classification speed；In view of RBF kernel function it is Karyomerite function, Polynomial kernel function is overall situation kernel function, and karyomerite function learning ability is strong, and Generalization Capability is relatively weak, and Overall situation kernel function Generalization Capability is strong, and learning capacity is relatively weak, carries out on the basis of the advantage of summary two class kernel function The Kernel of least square method supporting vector machine, optimizes classification performance and the Generalization Capability of least square method supporting vector machine； The multi-population of design works in coordination with Chaos particle swarm optimization algorithm, has preferable convergence rate, and has preferable global and local Optimizing performance, it is possible to jump out Local Extremum timely, finds the optimal value of the overall situation, thus uses multi-population to work in coordination with chaotic particle The structural parameters of least square method supporting vector machine are optimized by colony optimization algorithm, and effect of optimization is good.

Preferably, described failure mode updating block 56, for being updated training set, continues to optimize sensor fault Diagnostic cast, including:

(1) when sensor fault diagnosis model cannot carry out effective failure modes to characteristic vector to be measured, by feature to be measured Vector is as new training feature vector；

(2) training sample is updated by new training feature vector, to the least square support after structure parameter optimizing Vector machine is trained, and builds the sensor fault diagnosis model made new advances；

(3) use new sensor fault diagnosis model that described characteristic vector to be measured is carried out failure modes identification, complete Failure mode updates.

This preferred embodiment arranges failure mode updating block 56, to improve adaptation ability and the range of application of model.

Preferably, described health records unit 57 includes sub module stored and secure access submodule, described storage submodule Block uses storage model based on cloud storage, specifically, is encrypted after being compressed by fault message, is uploaded to cloud storage Device, described secure access submodule, for conducting interviews information, specifically, corresponding to sub module stored, downloads data to This locality, after using corresponding secret key to be unlocked, then carries out decompressing to read information.

This preferred embodiment arranges health records unit 57, on the one hand ensure that information security, on the other hand can be at any time Fault is conducted interviews, it is simple to search problem.

In this application scenarios, set threshold value T₁Value be 0.93, the monitoring velocity phase of sensor fault diagnosis module 5 To improve 13%, the monitoring accuracy of sensor fault diagnosis module 5 improves 9% relatively.

Application scenarios 5

See Fig. 1, Fig. 2, the decelerator health analysis system of an embodiment of this application scene, it is subject to including for collection Survey the sensing acquisition module 1 of retarder operation state mechanical signal；It is connected with described sensing acquisition module 1, for by collection Data carry out the data preprocessing module 2 of pretreatment；It is connected with described data preprocessing module 2, is used for receiving pretreated Data are also uploaded to the communication control module 3 of central processing module 4, and described central processing module 4 is for the data that will receive Carry out after-treatment and comprehensively analyze the running status of tested decelerator.

Preferably, described central processing module 4 uses with communication control module 3 one_to_one corresponding.

The above embodiment of the present invention achieves monitoring and the analysis of the health status to decelerator, can be to spot check personnel's Spot check plays directive function, greatly reduces the routine work amount of field maintenance person, improves work efficiency, thus solve Above-mentioned technical problem.

Preferably, described sensing acquisition module 1 includes some acceleration for gathering tested reductor acceleration signal Spend sensor and for gathering the temperature sensor of tested reductor temperature signal.

This preferred embodiment can the health status information of the more efficiently tested reductor of Real-time Collection.

Preferably, described decelerator health analysis system also includes the sensor fault diagnosis diagnosing each sensor Module 5, described sensor fault diagnosis module 5 includes signals collecting filter unit 51, fault signature extraction unit 52, online spy Levy extraction unit 53, characteristic vector preferred cell 54, failure modes recognition unit 55, failure mode updating block 56 and healthy note Record unit 57.

The above embodiment of the present invention arranges sensor fault diagnosis module 5 and achieves sensor fault diagnosis module 5 Fast construction, is conducive to monitoring each sensor, it is ensured that monitoring effectively performs.

Preferably, described signals collecting filter unit 51 is used for gathering historical sensor signal and on-line sensor test letter Number, and use combination form wave filter to be filtered signal processing；

This preferred embodiment arranges combination form wave filter, can effectively remove the various noise jamming of signal, preferably The primitive character information of stick signal.

Preferably, described fault signature extraction unit 52 is for carrying out integrated experience to filtered historical sensor signal Mode decomposition (EEMD) processes, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as training feature vector, including:

(1) the historical sensor signal by collection is divided into the fault-signal of nominal situation signal and plurality of classes；

(2) described historical sensor signal carries out integrated empirical mode decomposition (EEMD) process, it is thus achieved that described history passes The intrinsic mode function of sensor signal and remainder function；

(3) intrinsic mode function and the Energy-Entropy of remainder function of described historical sensor signal are calculated；

(4) Energy-Entropy of historical sensor signal is normalized, extracts the Energy-Entropy after normalization as instruction Practice characteristic vector；

Described online feature extraction unit 53 is for carrying out integrated Empirical Mode to filtered on-line sensor test signal State is decomposed (EEMD) and is processed, and extracts the Energy-Entropy of integrated empirical mode decomposition (EEMD) as characteristic vector to be measured, including:

(1) described on-line sensor test signal is carried out EEMD process, it is thus achieved that described on-line sensor test signal Intrinsic mode function and remainder function；

(2) intrinsic mode function and the Energy-Entropy of remainder function of described on-line sensor test signal are calculated；

(3) Energy-Entropy of on-line sensor test signal is normalized, extracts the Energy-Entropy after normalization and make For characteristic vector to be measured.

This preferred embodiment carries out integrated empirical mode decomposition (EEMD) to the sensor signal gathered and processes, it is possible to effectively Elimination modal overlap phenomenon, the effect of decomposition is preferable.

Preferably, training feature vector is carried out similar with characteristic vector to be measured by described characteristic vector preferred cell 54 respectively Property tolerance, the characteristic vector high for similarity reject, including:

(1) two vector similarities function S (X, Y) are defined:

$S(X,Y)=\frac{\mathrm{cov}(X,Y)}{\sqrt{D\left(X\right)}\sqrt{D\left(Y\right)}}$

In formula, X, Y represent that two characteristic vectors, cov (X, Y) are the covariance of X and Y respectively,For X, Y standard deviation；

For any two training feature vector X₁、X₂, and any two characteristic vector D to be measured₁、D₂, it is respectively adopted similar Its similarity is measured by degree function, obtains S (X₁,X₂) and S (D₁,D₂)；

(2) for S (X₁,X₂) and S (D₁,D₂), if S is (X₁,X₂)>T₁, T₁∈ (0.9,1), only chooses X₁As training characteristics Vector, if S is (D₁,D₂)>T₂, T₂∈ (0.95,1), only chooses D₁As characteristic vector to be measured.

This preferred embodiment screens characteristic vector by measuring similarity, it is possible to reduce amount of calculation, improves efficiency.

Preferably, described failure modes recognition unit 55 is for using the least square method supporting vector machine of optimization to treat described Survey characteristic vector and carry out failure modes identification, select optimize submodule, training submodule and identify submodule, specifically including parameter For:

Described parameter selects the kernel function optimizing submodule for constructing least square method supporting vector machine, and to least square The structural parameters of support vector machine use multi-population to work in coordination with Chaos particle swarm optimization algorithm and are optimized；

Described training submodule, for using many classification sides of the least square support vector machines of the optimum binary tree structure of improvement Method, instructs the least square method supporting vector machine after structure parameter optimizing using the training feature vector obtained as training sample Practice, and build sensor fault diagnosis model；

Described identification submodule is used for using described sensor fault diagnosis model that described characteristic vector to be measured carries out event Barrier Classification and Identification；

Wherein, it is considered to Polynomial kernel function and the superiority of RBF kernel function, the core letter of described least square method supporting vector machine Number is configured to:

K=(1-δ) (xx_i+1)^p+δexp(-‖x-x_i‖²/σ²)

In formula, δ is the structure adjusting factor, and the span of δ is set as [0.45,0.55], and p is the rank of Polynomial kernel function Number, σ²For RBF kernel functional parameter.

Wherein, shown employing multi-population is worked in coordination with Chaos particle swarm optimization algorithm and is optimized, including:

(1) to main population and initialize from population respectively, randomly generate initial as particle of one group of parameter Position and initial velocity, definition fitness function is:

$S=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left|\frac{q_{i}W}{q_{i}W+(1-q_i)T}\right|\×100\%$

In formula, N is the total number of training sample, and W is that bug is classified number, and T is that fault is correctly classified number, q_iFor certainly The weight coefficient set, q_iSpan be set as [0.4,0.5]；

(2) renewal from population is carried out, in every generation renewal process, according to fitness function, from population respectively The speed of more new particle and position, then will be experienced in its history adaptive optimal control angle value and main population body each particle The fitness value of desired positions compares, if more preferably, then as current global optimum position；

(3) described global optimum position is carried out the optimum particle position in chaos optimization, and iteration current sequence and speed Degree, generates optimal particle sequence；

(4) in the main population of every generation, choose from population optimum particle, and the position of more new particle and speed, Until reaching maximum iteration time or meeting the error requirements of fitness function.

Wherein, many sorting techniques of the least square support vector machines of the optimum binary tree structure of described improvement specifically include:

(1) calculate the standard variance of all training samples and two classifications j,Between Separatory measure；

(2) output minimum separation estimate correspondence j,

(3) after being optimized the structural parameters of least square method supporting vector machine, the least square setting up two classification props up Hold vector machine in order to train jth class andThe training sample of class, forms optimum two classification least square method supporting vector machines, output The parameter of discriminant function,The training sample of class is merged in j class, constitutes new j class training sample；

(4) all of classification is circulated training according to (1)-(3), until the optimum root node of output；

(5) according to the categorised decision tree of above output result composition least square method supporting vector machine, then to remaining instruction Practice sample and carry out classifying quality test.

This preferred embodiment is in order to improve the precision of fault diagnosis, and employing training speed is fast, generalization ability strong and robustness Preferably least square support vector machines is as grader, and proposes the many sorting techniques improving optimum binary tree structure, with between class Separatory measure substitutes the weights in binary tree structure, the nicety of grading that improve and classification speed；In view of RBF kernel function it is Karyomerite function, Polynomial kernel function is overall situation kernel function, and karyomerite function learning ability is strong, and Generalization Capability is relatively weak, and Overall situation kernel function Generalization Capability is strong, and learning capacity is relatively weak, carries out on the basis of the advantage of summary two class kernel function The Kernel of least square method supporting vector machine, optimizes classification performance and the Generalization Capability of least square method supporting vector machine； The multi-population of design works in coordination with Chaos particle swarm optimization algorithm, has preferable convergence rate, and has preferable global and local Optimizing performance, it is possible to jump out Local Extremum timely, finds the optimal value of the overall situation, thus uses multi-population to work in coordination with chaotic particle The structural parameters of least square method supporting vector machine are optimized by colony optimization algorithm, and effect of optimization is good.

Preferably, described failure mode updating block 56, for being updated training set, continues to optimize sensor fault Diagnostic cast, including:

(1) when sensor fault diagnosis model cannot carry out effective failure modes to characteristic vector to be measured, by feature to be measured Vector is as new training feature vector；

(2) training sample is updated by new training feature vector, to the least square support after structure parameter optimizing Vector machine is trained, and builds the sensor fault diagnosis model made new advances；

(3) use new sensor fault diagnosis model that described characteristic vector to be measured is carried out failure modes identification, complete Failure mode updates.

This preferred embodiment arranges failure mode updating block 56, to improve adaptation ability and the range of application of model.

Preferably, described health records unit 57 includes sub module stored and secure access submodule, described storage submodule Block uses storage model based on cloud storage, specifically, is encrypted after being compressed by fault message, is uploaded to cloud storage Device, described secure access submodule, for conducting interviews information, specifically, corresponding to sub module stored, downloads data to This locality, after using corresponding secret key to be unlocked, then carries out decompressing to read information.

This preferred embodiment arranges health records unit 57, on the one hand ensure that information security, on the other hand can be at any time Fault is conducted interviews, it is simple to search problem.

In this application scenarios, set threshold value T₁Value be 0.92, the monitoring velocity phase of sensor fault diagnosis module 5 To improve 14%, the monitoring accuracy of sensor fault diagnosis module 5 improves 8% relatively.

Last it should be noted that, above example is only in order to illustrate technical scheme, rather than the present invention is protected Protecting the restriction of scope, although having made to explain to the present invention with reference to preferred embodiment, those of ordinary skill in the art should Work as understanding, technical scheme can be modified or equivalent, without deviating from the reality of technical solution of the present invention Matter and scope.

Claims (3)

1. a decelerator health analysis system, is characterized in that, including for gathering tested retarder operation state mechanical signal Sensing acquisition module；It is connected with described sensing acquisition module, for the data of collection being carried out the data prediction of pretreatment Module；It is connected with described data preprocessing module, for receiving pretreated data and being uploaded to the logical of central processing module Letter control module, described central processing module for carrying out after-treatment and comprehensively analyzing tested decelerator by the data received Running status.

A kind of decelerator health analysis system the most according to claim 1, is characterized in that, described central processing module with Communication control module one_to_one corresponding uses.

A kind of decelerator health analysis system the most according to claim 2, is characterized in that, described sensing acquisition module bag Include some acceleration transducers for gathering tested reductor acceleration signal and for gathering tested reductor temperature letter Number temperature sensor.