CN109406117A - A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm - Google Patents
A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm Download PDFInfo
- Publication number
- CN109406117A CN109406117A CN201811137983.8A CN201811137983A CN109406117A CN 109406117 A CN109406117 A CN 109406117A CN 201811137983 A CN201811137983 A CN 201811137983A CN 109406117 A CN109406117 A CN 109406117A
- Authority
- CN
- China
- Prior art keywords
- data
- rbfnn
- module
- sample
- improved
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
-
- G—PHYSICS
- G08—SIGNALLING
- G08C—TRANSMISSION SYSTEMS FOR MEASURED VALUES, CONTROL OR SIMILAR SIGNALS
- G08C17/00—Arrangements for transmitting signals characterised by the use of a wireless electrical link
- G08C17/02—Arrangements for transmitting signals characterised by the use of a wireless electrical link using a radio link
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm, system includes epigynous computer section and multichannel remote oscillation signal monitor, multichannel remote oscillation signal monitor includes slave computer, GPRS wireless transport module and power module, slave computer acquires vibration signal when monitoring, and vibration signal is converted into electric signal, epigynous computer section is uploaded to by GPRS module, while vibration signal is shown slave computer by treated；Host computer is processed and displayed the signal received using improved RBFNN algorithm；Power module is that slave computer and GPRS wireless transport module are powered.The present invention with trained improved RBFNN carries out classification processing to test set sample data, it can be achieved that multichannel remote oscillation signal-data processing, analysis, prison show function.
Description
Technical field
The invention belongs to mechanical fault diagnosis field, especially a kind of multichannel based on improved RBFNN algorithm is remote
Journey vibration signal monitor system and monitoring method.
Background technique
Large rotating machinery structure, such as Turbo-generator Set, turbine-generator units and flue gas turbine expander group, it is common to use sliding
Dynamic bearing supports shaft, and protects system monitoring shaft to the Relative Vibration of bearing by vibration monitoring.Vibration monitoring protection system
System includes two eddy current sensors, is respectively used to the vibration in two channels of measurement.However since two current vortex sensors can only
The vibration signal of two positions of single-measurement mechanical structure, but the structure of different units, operating parameter and bearing film are rigid
The influence for spending anisotropy etc. in a circumferential direction, causes the vibration of both direction usually to have a certain difference, and such as one
A direction vibration is larger, and the vibration of another direction is smaller, this brings one to the alarm of vibration monitoring protection system and abnormal judgement
Fixed difficulty；And monitoring information has to progress real time monitoring and obtains, and can not be further transferred to remote terminal, this gives company
And the work of mechanical equipment actual monitoring is made troubles.
Therefore it needs to design a kind of different location by large rotating machine structure while monitoring its vibration signal, and realize
The multichannel remote oscillation signal monitor system of the live real-time display of monitoring information and the processing of long-range real-time online display.
Summary of the invention
Goal of the invention: a purpose be to provide it is a kind of based on improved RBFNN algorithm multichannel remote oscillation signal prison show
System and monitoring method extend vibration signal monitor function, improve to solve the prior art and apply the existing above problem
Vibration signal monitor monitors the vibration signal of a variety of mechanical structure different locations, and measurement is simple accurate, and discovery is mechanical in time
Monitoring data are simultaneously transmitted to host computer by the improved RBFNN algorithm in machine learning field to survey by the failure problems of structure appearance
The vibration signal data collection obtained carries out data processing and shows, realizes that remote oscillation signal-data processing, analysis, prison show function.
Technical solution: for achieving the above object, using following technical scheme:
A kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm, including epigynous computer section and more
Path remote vibration signal monitor, in which:
Multichannel remote oscillation signal monitor includes slave computer, GPRS wireless transport module and power module, slave computer
Vibration signal is acquired, and vibration signal is converted into electric signal, epigynous computer section, while slave computer is uploaded to by GPRS module
By treated, vibration signal is shown；Host computer using improved RBFNN algorithm to the signal received carry out processing and
Display；Power module is that slave computer and GPRS wireless transport module are powered.
Optionally, epigynous computer section includes Labview interface part and Python data processing section, host computer
Gui interface is designed by Labview in part, and by internal interface function call Python completed sample according to processing, first will
Sample data set is divided into test set and training set, improved RBFNN model is trained with training set, with trained improved RBFNN mould
Type carries out online classification processing to test set sample data, and designs gui interface and show to result, the sample data set
It is the data sets for uploading to the vibration signal data of host computer obtained by multiple measurements from GPRS wireless transport module and forming.
Optionally, Labview interface part mainly includes data reception module, data disaply moudle and data
Memory module, the data reception module listen to the company from client using TCPListen function in LabView means of communication
Request is connect, and waits GPRS wireless transport module to request by TCP connection in designated port, after establishing connection, utilizes TCP
The data of Read function reading terminals transmission, are standard by TCP Read function works mode setting, that is, are waited until reading
It takes all specified bytes or " overtime millisecond " to be finished, returns to the byte currently read, set if reading byte number and being less than
The fixed word joint number set, then report mistake；The data that the data disaply moudle will acquire are converted by format, with table shape
Formula shows data in front on the visualization interface of plate；The data memory module passes through Write To Spreadsheet
File function writes the data in electrical form for processed offline.
Optionally, slave computer include single-chip microcontroller, data acquisition module and control display module, the data acquisition module and
Control display module is connect with single-chip microcontroller respectively, and single-chip microcontroller includes control unit, outside RTC, external crystal-controlled oscillation and EEPROM module,
External RTC and external crystal-controlled oscillation are respectively that control unit provides clock and frequency signal, double between EEPROM module and control unit
To communication；Data acquisition module includes multiple vibration signal sensors, signal condition amplifying circuit and multichannel ADC modulus of conversion
Block；Multiple vibration signal sensors multiple vibration signals of collection machinery device and are converted to electric signal respectively, and electric signal passes through
Signal condition amplifying circuit amplify after by multichannel ADC conversion module input control unit, control unit to signal into
It is shown after row analysis processing by controlling display module, and host computer is uploaded to by GPRS wireless transport module and is divided
Analysis processing and display；Wherein, vibration signal includes radial vibration signal, axial vibration signal and the case vibration letter of mechanical device
Number.
In another embodiment of the present invention, a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm
Monitoring method, comprising:
(1) slave computer part controls multiple vibration signal sensors in data acquisition module by single-chip microcontroller and cooperates with completion
The acquisition of multiple spot vibration signal data；
(2) data of slave computer part acquisition are uploaded to epigynous computer section by GPRS wireless transport module；
(3) epigynous computer section passes through internal interface function call Python completed sample according to processing, first by sample data
Collection is divided into test set and training set, improved RBFNN model is trained with training set, with trained improved RBFNN model to test
Collect sample data and carry out offline classification processing, and designs gui interface and result is shown.
Further, in step (3) with training set training improved RBFNN model, and in advance trained improved RBFNN
Model to test set sample data carry out online classification processing method include:
(31) cluster centre c is adjusted with least square error criterionj, traditional two step method RBFNN is optimized；
(32) U (c is uniformly adjusted with gradient descent methodj,σj,ωij) value, the RBFNN finally trained, to traditional two steps
Method RBFNN carries out double optimization, c in formulajIndicate cluster centre, σjIndicate diameter sound stage width degree, ωijOutput layer weight is indicated, by three
One vector set U (c of a parameterj,σj,ωij) indicate；
(33) by standardizing to existing vibration signal sample data set, index system is established, i.e., by vibration signal sample
Data are divided into according to the difference of measurement object: shell vibration, axis vibration, oscillation crosswise and extensional vibration；
(34) sample data set that host computer is uploaded to from GPRS wireless transport module is divided into 1/3 by reserving method
Test set and 2/3 training set, and according to the parameter value measured to data in training set；Addition shaft probability of malfunction mark manually
Label；
(35) improved RBFNN is trained with training set, and is tested on test set, test result can be according to the difference of shaft
Vibration parameters characteristic shows different shaft probabilities of malfunction automatically, and the bigger shaft for representing the location point of probability is more possible to send out
Raw failure, and the results are shown on gui interface.
Further, cluster centre c is adjusted with least square error criterion in step (31)j, to traditional two step method RBFNN
The process optimized are as follows:
(310) using the initial configuration for determining hidden layer center based on the clustering algorithm of dynamic attenuation radius, while sample is utilized
This information dynamic control min cluster radius；
(311) it is based on error sum of squares criterion, it is influenced by investigating sample movement, adjusts center point value, simultaneously
Diameter sound stage width degree is determined using class spacing combination sample actual distribution in class；
(312) weight that hidden layer and output layer are determined using pseudoinverse technique, can finally obtain ωij；
(313) after calculating by the optimization of first time, the basic network topology and overall network for having obtained RBFNN are joined
Number, network parameter include cluster centre, diameter sound stage width degree and output layer weight, by three parameters with a vector set U (cj,σj,
ωij) indicate；Then pass through the parameter of the unified training adjustment RBFNN model of gradient descent method.
Further, in step (310) dynamic attenuation radius clustering algorithm are as follows:
(a) initialization sample data set, including training set or test set；
(b) each sample point that selection sample data is concentrated, calculates separately at a distance from existing cluster centre, finds it most
The cluster centre of neighbour；
(c) cluster radius is calculated, judges whether to meet the Gauss distance between sample point and the cluster centre and is less than cluster
Radius, if it is less, the sample point is added in this cluster；
(d) sample point is calculated every time it is necessary to subtract this sample point in original data set, until original
Sample point in data set all calculates one time, and otherwise terminator goes to second step, continue to calculate next sample point.
Further, the method for error sum of squares criterion adjustment center point value is utilized in step (311) are as follows:
(a) dynamic attenuation cluster result is utilized, initial error quadratic sum is calculatedFormula
In, xtIndicate sample data, cjIndicate current cluster centre；
(b) it to each sample in each cluster, calculates Enable pil=minj≠i{pij, piiIndicate spacing in class, pijIndicate this
Spacing between sample point and other cluster centres；
If (c) pil< pij, then sample is moved into l class, modifies cluster centre, and calculate new error sum of squares:
Jc+1=Jc-(pii-pil)；
If (d) Jc+1< Jc, then (2) step is returned to, otherwise, algorithm terminates.
Further, in step (313) gradient descent method specific calculating step are as follows:
(1) mean square deviation between training result and legitimate reading is calculated:M is in formula
Input sample number, L are output node sum,For training result,For legitimate reading, RBFNN training goal is exactly as far as possible
The value of ground reduction E；
(2) when the value of E is also not up to training requirement index and the frequency of training also not up to setting upper limit, following formula is calculated:η is the step-length in training gradient descent method, and η is bigger, and gradient declines faster, training result
Easily convergence, but η value should be reduced when error change is little, prevent over-fitting and reduce fluctuating error,For the ladder of algorithm
Descent direction is spent, is determined by E, U(τ)With U(τ+1)The respectively τ and U (C of τ+1 time calculatingj,σj,ωij) value.
The utility model has the advantages that compared with prior art, the present invention realizes that multi-channel testing rotating machinery is different using STM32
The vibration signal of position, monitoring is more accurate quickly, can more find the mechanical breakdown information of different location in time；Introduce TFTLCD
Touch screen module is more convenient to carry out human-computer interaction, implements manipulation detection process and scene display monitoring information；It is wireless to introduce GPRS
Hop and host computer prison show GUI module, by SIM900 module, SIM card link block, realize the long-range reality of vibration signal
When show, vibration signal data is carried out at the algorithm based on improved RBFNN model by LabView software and Python software
Reason realizes that the host computer of monitoring data is handled in real time online, analyzed, prison shows, makes the malfunction monitoring to large rotating machinery equipment
And diagnosis is more convenient；Help further to improve mechanical Fault Monitoring of HV diagnosis problem.
Detailed description of the invention
Fig. 1 is monitor circuit module connection diagram of the present invention；
Fig. 2 is monitor signal transmission structure schematic diagram of the present invention；
Fig. 3 is vibration signal sensing installation site structural schematic diagram of the invention；
Fig. 4 is vibration signal sample data sorting algorithm flow chart of the invention；
Fig. 5 is the algorithm flow chart of improved RBFNN of the invention.
Specific embodiment
Technical solution of the present invention is described in detail in the following with reference to the drawings and specific embodiments.
The present invention implements a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm and monitoring side
Method, can be realized multi-channel testing large rotating machinery equipment vibrating signal, TFTLCD touch screen control show monitoring process and
As a result, long-range host computer is handled in real time online, analyzed, prison shows vibration signal, wherein STM32F103RBT6 is that master chip (is controlled
Unit processed, similarly hereinafter), coordinate each module and orderly works.
A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm of the invention,
Monitor system includes epigynous computer section and multichannel remote oscillation signal monitor, and epigynous computer section includes that the interface Labview is set
Meter part and Python data processing section, monitor include slave computer, GPRS wireless transport module and power module, upper
Machine is connected with slave computer by GPRS wireless transport module, and power module provides electric energy for monitor；Method includes slave computer portion
The multiple vibrating sensors collaboration passed through in single-chip microcontroller control part of data acquisition is divided to complete the acquisition of multi-channel Vibration Signal data；
The data of slave computer part acquisition are uploaded to epigynous computer section by GPRS wireless transport module；Epigynous computer section passes through
Labview designs gui interface, and by internal interface function call Python completed sample according to processing.Present invention training
Good improved RBFNN carries out classification processing to test set sample data, it can be achieved that multichannel remote oscillation signal-data processing, point
Analysis, prison show function.
Referring to Fig. 1, in an embodiment of the present invention, a kind of multichannel remote oscillation signal based on improved RBFNN algorithm
Monitor system, including epigynous computer section and multichannel remote oscillation signal monitor, epigynous computer section include that the interface Labview is set
Meter part and Python data processing section, epigynous computer section designs gui interface by Labview, and passes through internal interface
Sample data set is first divided into test set and training set according to processing by function call Python completed sample, with training set training
Improved RBFNN model carries out online classification processing to test set sample data with trained improved RBFNN model, and designs
Gui interface shows result, the sample data set be uploaded to from GPRS wireless transport module host computer by multiple
Measure the data set of obtained vibration signal data composition；
Multichannel remote oscillation signal monitor includes slave computer, GPRS wireless transport module and power module, under described
Position machine includes single-chip microcontroller, data acquisition module and control display module, and the data acquisition module and control display module are distinguished
It is connect with single-chip microcontroller, single-chip microcontroller includes control unit, outside RTC, external crystal-controlled oscillation and EEPROM module；Data acquisition module includes
Multiple vibration signal sensors, signal condition amplifying circuit and multichannel ADC conversion module；Controlling display module includes TFTLCD
Touch screen module；
GPRS wireless transport module includes SIM900 module and SIM card link block, the SIM900 module and single-chip microcontroller
Connection establishes connection by the server on GPRS network and Internet, collected data is sent to clothes by antenna
Business device；The SIM card link block is electrically connected with SIM900 module, is powered by SIM900 module through overvoltage conditioning circuit；
Power module includes DC power supply drive module and voltage amplification module, and DC power supply drive module exports and voltage
Amplification module input connection, voltage amplification module output end are inputted with control unit power input, TFTLCD touch screen respectively
End is connected with SIM900 modular power source input terminal.
Described control unit selects the STM32F103RBT6 chip of high-performance low-power-consumption, is responsible for coordinating monitor device entirety
Operation, external RTC is responsible for control unit and provides time scale, and external crystal-controlled oscillation provides 72MHz working frequency for control unit,
EEPROM module is responsible for the data buffer storage of monitor.
Data acquisition module is responsible for being placed on the vibration that the vibration signal sensor of rotating machinery each section monitors
Dynamic signal is converted to inside digital data transmission to control unit.The output of multiple vibration signal sensors and signal condition amplify
The input of circuit connects, and then connect through multichannel ADC conversion module with control unit.The vibration signal sensor mainly by
Piezoelectric acceleration transducer composition, piezoelectric acceleration transducer are most commonly seen the most practical a type of on the market
Vibration acceleration sensor, it has lesser volume, higher cost performance, stronger noiseproof feature, and side the most practical
Just measurement and calibrating mode, piezoelectric acceleration transducer using contact measurement method, can be convenient handle
Multiple sensors are attached to rotating machinery surface, measure its vibration signal；The signal condition amplifying circuit mainly by
The voltage amplifier circuit and rc filter circuit composition of OPA2336UA chip composition, are responsible for the amplification and filtering of analog signal；Institute
It states multichannel ADC conversion module and is equipped with AD7705 high precision analogue conversion chip, acquisition precision is up to 16, using the side SPI
Formula is communicated with control unit, and uses dma mode, realizes that multi channel signals acquire simultaneously.
The onboard control chip of the TFTLCD touch screen module is XPT2046, and XPT2046 is that a piece of 4 conducting wire system touches
Screen controller includes 12 bit resolution 125KHZ conversion rate Step wise approximation type A/D converters, connects the 5v of voltage amplification module
Voltage output monitors the work of input control monitoring instrument according to touch screen module in measurement process, and is specifically supervised according to instrument
Survey process shows different monitoring contents.
Carried out data transmission between SIM900 module and control unit by serial ports, provides 4v electricity by voltage amplification module
Pressure establishes connection by the server on GPRS network and Internet, collected data is sent to service by antenna
Device；The SIM card link block is electrically connected with SIM900 module, by SIM900 module through overvoltage conditioning circuit provide 3v or
1.8v voltage is automatically selected according to the sim module that communication common carrier provides.
Power module include by two section 1.5v dry cell batteries at DC power supply drive module and voltage amplification module, voltage
3v voltage is converted 3.3v, 4v and 5v output, the 3.3v output end and control unit power supply of voltage amplification module by amplification module
Input terminal connection, provides chip operation normal voltage；The 5v output end and TFTLCD touch screen power input of voltage amplification module
End connection, the process measured for realizing human-computer interaction, real-time control；The 4v output end and SIM900 module of voltage amplification module
Power input connection.Voltage amplification module fully demonstrates the low power consumption characteristic of instrument entirety, and all modules are by same voltage source
Power supply separately designs amplifying circuit according to multiple PT1301 chips, and respectively disparate modules provide operating voltage.
Referring to fig. 2, one kind of the invention is based on improved RBFNN algorithm multichannel remote oscillation signal monitor system host computer
Gui interface is designed by Labview in part, and is used for by internal interface function call Python completed sample according to processing
It imports, processing, analyze and show data；Epigynous computer section can be connected with slave computer by GPRS wireless transport module, GPRS
Wireless transport module is made of SIM900 module and SIM card module, to realize the long range radio transmissions of data.Wherein, slave computer
Part can be measured by being placed in the vibration signal sensor of large rotating machinery equipment different location mechanical equipment laterally, it is longitudinal,
Axial, shell vibration information, and handled by the Data collecting conversion of multichannel ADC conversion module, in TFTLCD touch screen
Control show down, by GPRS wireless transport module, monitoring data information is transmitted to PC machine (host computer), is passed through
Labview designs gui interface, and by internal interface function call Python completed sample according to processing, for importing, locating
Reason, analysis and display data, wherein Labview interface part can utilize the frequency spectrum in LabView data analysis tool
The data processing tools such as conversion, digital filtering and Fast Fourier Transform (FFT), online processing analysis data simultaneously show vibration signal
Amplitude-frequency response signal and spectrum analysis figure.
Labview interface part mainly includes data reception module, data disaply moudle and data memory module, institute
It states data reception module and listens to the connection request from client using TCPListen function in LabView means of communication, and
Designated port waits GPRS module to request by TCP connection, after establishing connection, utilizes the transmission of TCP Read function reading terminals
Data, are standard by TCP Read function works mode setting, that is, are waited until reading all specified bytes or " time-out
Millisecond " is finished, and returns to the byte currently read, if reading fixed word joint number of the byte number less than setting, then reports mistake
Accidentally；The data that the data disaply moudle will acquire by format convert, show in a tabular form data in front plate can
Depending on changing on interface；The data memory module writes the data to electronic watch by Write To Spreadsheet File function
For processed offline in lattice.
Referring to Fig. 3, data acquisition module mainly uses multiple vibration signal sensors, is respectively placed in mechanical cover side
Face, be placed in shaft it is radial with horizontal axis and longitudinal axis cross-distribution, to be placed in shaft cross section central, be respectively used for measuring mechanical equipment
Case vibration signal, radial oscillation crosswise signal and extensional vibration signal, axial vibration signal, and turned by multichannel ADC
All vibration signals are transmitted to control unit simultaneously to mold changing block and host computer remotely monitors gui interface, and is connect by inside
Mouth function call Python completed sample carries out the comprehensive of signal for importing, handling, analyze and showing data according to processing
Conjunction analyze, compares, handles and is shown, the working order of large rotating machinery equipment is shown with this accurate quickly prison, when it is faulty out
It can accomplish to judge in time, show and protect now.
In another embodiment, a kind of monitoring of the multichannel remote oscillation signal monitor system based on improved RBFNN algorithm
Method, comprising:
(1) slave computer part controls multiple vibration signal sensors in data acquisition module by single-chip microcontroller and cooperates with completion
The acquisition of multiple spot vibration signal data；
(2) data of slave computer part acquisition are uploaded to epigynous computer section by GPRS wireless transport module；
(3) epigynous computer section passes through internal interface function call Python completed sample according to processing, first by sample data
Collection is divided into test set and training set, improved RBFNN model is trained with training set, with trained improved RBFNN model to test
Collect sample data and carry out offline classification processing, and designs gui interface and result is shown.By improved RBFNN algorithm process
Later test set sample data can show that automatically different shaft failures is general according to the different vibration parameters characteristics of shaft
Rate, the bigger shaft for representing the location point of probability are more likely to occur failure, and user can generate according in shaft different location point
Different shaft probabilities of malfunction the vibration signal data of difference is distinguished, the point in each shaft of distinguishing judgement
Vibration Condition, can also individually be analysed in depth by the sample data to the problematic point of vibration signal, find shaft
On this location point whether really exist mechanical breakdown or according to the judgement of vibration signal sample data be belong to which kind of it is mechanical therefore
Barrier.
Improved RBFNN model is trained with training set in step (3), and with prior trained improved RBFNN model to survey
Examination collects the method that sample data carries out online classification processing
Cluster centre c is adjusted with least square error criterionj, traditional two step method RBFNN is optimized；Use gradient descent method
Unified adjustment U (cj,σj,ωij) value, the RBFNN finally trained carries out double optimization to traditional two step method RBFNN, in formula
cjIndicate cluster centre, σjIndicate diameter sound stage width degree, ωijOutput layer weight is indicated, by three parameters with a vector set U (cj,
σj,ωij) indicate；By standardizing to existing vibration signal sample data set, index system is established, i.e., by vibration signal sample
Data are divided into according to the difference of measurement object: the inhomogeneities such as shell vibration, axis vibration, oscillation crosswise, extensional vibration；It will wirelessly be passed from GPRS
Defeated module uploads to the sample data set of host computer and is divided into 1/3 test set and 2/3 training set by reserving method, and according to
The parameter value measured adds shaft probability of malfunction label to data in training set manually；Improved RBFNN is trained with training set, and
It is tested on test set, test result can show that automatically different shaft failures is general according to the different vibration parameters characteristics of shaft
Rate, the bigger shaft for representing the location point of probability is more likely to occur failure, and the results are shown on gui interface；With improvement
RBFNN the algorithm available classification data and shaft event more more accurate than traditional clustering algorithm or traditional two step method RBFNN
Hinder probability tag, obtain the higher classification results of confidence level, according to the result accurate judgement rotating shaft position event can occur for user
The probability size of barrier, eliminates the process artificially judged, avoid the randomness of artificial judging result, blindness, subjectivity and
Inaccuracy ensure that result is reliable, accurate, timely.
As shown in figure 4, vibration signal sample data sorting algorithm of the invention the following steps are included:
(1) vibration signal sample data set is imported, because clustering method can be used for the classification problem of data untagged collection, institute
It can be no label data collection with data set.
(2) data set standardizes, and can handle data set using the methods of min-max, i.e., with some sample in data set
Data are divided by the difference of maximum number and minimum number in data set as the data after the sample standardization.
(3) use reserves method and data set is divided into 1/3 test set and 2/3 training set,
(4) using improved RBFNN training method training RBFNN model, and with trained improved RBFNN model to survey
Examination collection carries out online classification processing.
(5) the shaft probability of malfunction label of different test set vibration signal datas is obtained.It can be obtained with improved RBFNN algorithm
To the classification data and shaft probability of malfunction label more more accurate than traditional clustering algorithm or traditional two step method RBFNN, obtain
The higher classification results of confidence level, the probability size that user can break down according to the result accurate judgement rotating shaft position, save
The process for having gone artificial judgement avoids randomness, blindness, subjectivity and the inaccuracy of artificial judging result, ensure that
As a result reliable, accurate, timely.
It preferably, is min-max method to the standardized standardized method of available data collection, circular isWherein, x is new data, x1For former data, min is the minimum value in data set, max be in data set most
Big value.
Referring to Fig. 5, cluster centre c is adjusted with least square error criterionj, mistake that traditional two step method RBFNN is optimized
Journey are as follows:
(1) it firstly, determining the initial configuration at hidden layer center using the clustering algorithm based on dynamic attenuation radius, utilizes simultaneously
Sample data dynamic control min cluster radius.This improvement had both prevented radii fixus to cluster poor adaptivity, avoided
Multiple explorations of decaying least radius empirical value determines, and effectively reduces the caused overfitting phenomenon of infinite radius reduction
It generates.
(2) then, it is based on error sum of squares criterion, it is influenced by investigating sample movement, adjusts center point value, together
Class spacing combination sample actual distribution determines diameter sound stage width degree in Shi Liyong class.This process has fully considered that class spacing is poly- to sample
It is excessively overlapped to avoid the specification area that unified diameter sound stage width degree may cause for the influence of class.
(3) finally, determining the weight of hidden layer and output layer using pseudoinverse technique, ω can finally be obtainedij。
(4) after calculating by the optimization of first time, the basic network topology and overall network parameter of RBFNN has been obtained
(cluster centre, diameter sound stage width degree, output layer weight), by three parameters with a vector set U (cj,σj,ωij) indicate.Then
Pass through the parameter of the unified training adjustment RBFNN model of gradient descent method.
Preferably, the clustering algorithm of the dynamic attenuation radius are as follows:
(1) initialization sample data set (training set or test set).
(2) each sample point that selection sample data is concentrated, calculates separately at a distance from existing cluster centre, finds it most
The cluster centre of neighbour.
(3) cluster radius is calculated, judges whether to meet the Gauss distance between sample point and the cluster centre and is less than cluster
Radius, if it is less, the sample point is added in this cluster.
(4) sample point is calculated every time it is necessary to subtract this sample point in original data set, until original
Sample point in data set all calculates one time, and otherwise terminator goes to second step, continue to calculate next sample point.
Preferably, the method for the fine tuning that hidden layer center is carried out using error sum of squares are as follows:
(1) dynamic attenuation cluster result is utilized, initial error quadratic sum is calculated:Formula
In, xtIndicate sample data, cjIndicate current cluster centre.
(2) it to each sample in each cluster, calculates: Enable pil=minj≠i{pij, piiIndicate spacing in class, pijIndicate this
Spacing between sample point and other cluster centres.
(3) if pil< pij, then sample is moved into l class, modifies cluster centre, and calculate new error sum of squares:
Jc+1=Jc-(pii-pil)。
(4) if Jc+1< Jc, then (2) step is returned to, otherwise, algorithm terminates.
Preferably, specific calculating step of the gradient descent method in this improved RBFNN algorithm are as follows:
(1) mean square deviation between training result and legitimate reading is calculatedM is in formula
Input sample number, L are output node sum,For training result,For legitimate reading, RBFNN training goal is exactly as far as possible
The value of ground reduction E.
(2) when the value of E is also not up to training requirement index and the frequency of training also not up to setting upper limit, following formula is calculated:η is the step-length in training gradient descent method, and η is bigger, and gradient declines faster, training result
Easily convergence, but η value should be reduced when error change is little, prevent over-fitting and reduce fluctuating error, therefore preferably with
Newton method adds momentum method to improve algorithm, so that η value is variable,For the gradient descent direction of algorithm, determined by E, U(τ)With U(τ+1)The respectively τ and U (c of τ+1 time calculatingj,σj,ωij) value.
In short, the present invention realizes the vibration signal of multi-channel testing rotating machinery different location, monitoring using STM32
It is more accurate quick, the mechanical breakdown information of different location can be more found in time；Introduce TFTLCD touch screen module, be more convenient into
Manipulation detection process and scene display monitoring information are implemented in row human-computer interaction；It introduces GPRS wireless transmission section and host computer is supervised
Show GUI module, by SIM900 module, SIM card link block, the long-range real-time display of vibration signal is realized, by LabView
Software and Python software carry out vibration signal data to realize monitoring data based on the algorithm process of improved RBFNN model
Host computer is handled in real time online, is analyzed, prison shows, is made more convenient to the malfunction monitoring of large rotating machinery equipment and diagnosis fast
It is prompt；Help further to improve mechanical Fault Monitoring of HV diagnosis problem.
Claims (10)
1. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm, it is characterised in that: including host computer
Part and multichannel remote oscillation signal monitor, in which:
Multichannel remote oscillation signal monitor includes slave computer, GPRS wireless transport module and power module, slave computer acquisition
Vibration signal, and vibration signal is converted into electric signal, epigynous computer section is uploaded to by GPRS module, while slave computer will be located
Vibration signal after reason is shown；Host computer is processed and displayed the signal received using improved RBFNN algorithm；
Power module is that slave computer and GPRS wireless transport module are powered.
2. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm according to claim 1,
Be characterized in that: epigynous computer section includes Labview interface part and Python data processing section, and epigynous computer section is logical
Labview design gui interface is crossed, and by internal interface function call Python completed sample according to processing, first by sample number
It is divided into test set and training set according to collection, improved RBFNN model is trained with training set, with trained improved RBFNN model to survey
Examination collection sample data carries out online classification processing, and designs gui interface and shows to result, the sample data set be from
GPRS wireless transport module uploads to the data sets that the vibration signal data of host computer obtained by multiple measurements forms.
3. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm according to claim 2,
Be characterized in that: Labview interface part mainly includes data reception module, data disaply moudle and data storage mould
Block, the data reception module are listened to the connection from client using TCPListen function in LabView means of communication and are asked
It asks, and waits GPRS wireless transport module to request by TCP connection in designated port, after establishing connection, utilize TCP Read letter
The data of number reading terminals transmission, are standard by TCP Read function works mode setting, that is, are waited until reading all
Specified byte or " overtime millisecond " is finished, and returns to the byte currently read, if reading byte number consolidating less than setting
Determine byte number, then reports mistake；The data that the data disaply moudle will acquire are converted by format, are shown in a tabular form
Data are in front on the visualization interface of plate；The data memory module passes through Write To Spreadsheet File function,
It writes the data in electrical form for processed offline.
4. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm according to claim 1,
Be characterized in that: slave computer includes single-chip microcontroller, data acquisition module and control display module, and the data acquisition module and control are aobvious
Show that module is connect with single-chip microcontroller respectively, single-chip microcontroller includes control unit, outside RTC, external crystal-controlled oscillation and EEPROM module, outside
RTC and external crystal-controlled oscillation are respectively that control unit provides clock and frequency signal, two-way between EEPROM module and control unit
Letter；Data acquisition module includes multiple vibration signal sensors, signal condition amplifying circuit and multichannel ADC conversion module；It is more
A vibration signal sensor multiple vibration signals of collection machinery device and is converted to electric signal respectively, and electric signal passes through signal tune
Reason amplifying circuit passes through multichannel ADC conversion module input control unit after amplifying, and control unit analyzes signal
It is shown after processing by controlling display module, and host computer is uploaded to by GPRS wireless transport module and is analyzed and processed
And display；Wherein, vibration signal includes the radial vibration signal, axial vibration signal and case vibration signal of mechanical device.
5. a kind of described in any item multichannel remote oscillation signal prisons based on improved RBFNN algorithm of claim 1-4 show and are
The monitoring method of system characterized by comprising
(1) slave computer part controls multiple vibration signal sensors collaboration in data acquisition module by single-chip microcontroller and completes multiple spot
Vibration signal data acquisition；
(2) data of slave computer part acquisition are uploaded to epigynous computer section by GPRS wireless transport module；
(3) epigynous computer section first divides sample data set by internal interface function call Python completed sample according to processing
For test set and training set, improved RBFNN model is trained with training set, with trained improved RBFNN model to test set sample
Notebook data carries out offline classification processing, and designs gui interface and show to result.
6. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm according to claim 5
Monitoring method, which is characterized in that in step (3) with training set training improved RBFNN model, and in advance trained improvement
RBFNN model to test set sample data carry out online classification processing method include:
(31) cluster centre c is adjusted with least square error criterionj, traditional two step method RBFNN is optimized；
(32) U (c is uniformly adjusted with gradient descent methodj,σj,ωij) value, the RBFNN finally trained, to traditional two step method
RBFNN carries out double optimization, c in formulajIndicate cluster centre, σjIndicate diameter sound stage width degree, ωijOutput layer weight is indicated, by three
One vector set U (c of parameterj,σj,ωij) indicate；
(33) by standardizing to existing vibration signal sample data set, index system is established, i.e., by vibration signal sample data
It is divided into according to the difference of measurement object: shell vibration, axis vibration, oscillation crosswise and extensional vibration；
(34) test that the sample data set that host computer is uploaded to from GPRS wireless transport module is divided into 1/3 by reserving method
Collect the training set with 2/3, and according to the parameter value measured to data in training set；Addition shaft probability of malfunction label manually；
(35) improved RBFNN is trained with training set, and is tested on test set, test result can be vibrated according to the difference of shaft
Parameter characteristic shows different shaft probabilities of malfunction automatically, and the bigger shaft for representing the location point of probability is more likely to occur event
Barrier, and the results are shown on gui interface.
7. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm according to claim 6
Monitoring method, it is characterised in that: adjust cluster centre c with least square error criterion in step (31)j, to traditional two step method
The process that RBFNN is optimized are as follows:
(310) using the initial configuration for determining hidden layer center based on the clustering algorithm of dynamic attenuation radius, while being believed using sample
Breath dynamic control min cluster radius；
(311) it is based on error sum of squares criterion, it is influenced by investigating sample movement, center point value is adjusted, utilizes simultaneously
Class spacing combination sample actual distribution determines diameter sound stage width degree in class；
(312) weight that hidden layer and output layer are determined using pseudoinverse technique, can finally obtain ωij；
(313) after calculating by the optimization of first time, the basic network topology and overall network parameter of RBFNN, net have been obtained
Network parameter includes cluster centre, diameter sound stage width degree and output layer weight, by three parameters with a vector set U (cj,σj,ωij)
It indicates；Then pass through the parameter of the unified training adjustment RBFNN model of gradient descent method.
8. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm according to claim 7
Monitoring method, which is characterized in that the clustering algorithm of dynamic attenuation radius in step (310) are as follows:
(a) initialization sample data set, including training set or test set；
(b) each sample point that selection sample data is concentrated, calculates separately at a distance from existing cluster centre, finds its arest neighbors
Cluster centre；
(c) cluster radius is calculated, judges whether to meet the Gauss distance between sample point and the cluster centre and is less than cluster radius,
If it is less, the sample point is added in this cluster；
(d) sample point is calculated every time it is necessary to subtract this sample point in original data set, until original data
The sample point of concentration all calculates one time, and otherwise terminator goes to second step, continue to calculate next sample point.
9. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm according to claim 7
Monitoring method, it is characterised in that: the method for error sum of squares criterion adjustment center point value is utilized in step (311) are as follows:
(a) dynamic attenuation cluster result is utilized, initial error quadratic sum is calculatedIn formula, xtTable
Sample notebook data, cjIndicate current cluster centre；
(b) it to each sample in each cluster, calculates Enable pil=minj≠i{pij, piiIndicate spacing in class, pijIndicate this
Spacing between sample point and other cluster centres；
If (c) pil< pij, then sample is moved into l class, modifies cluster centre, and calculate new error sum of squares: Jc+1=
Jc-(pii-pil)；
If (d) Jc+1< Jc, then (2) step is returned to, otherwise, algorithm terminates.
10. a kind of multichannel remote oscillation signal monitor system based on improved RBFNN algorithm according to claim 7
Monitoring method, it is characterised in that: the specific calculating step of gradient descent method in step (313) are as follows:
(1) mean square deviation between training result and legitimate reading is calculated:M is input in formula
Sample number, L are output node sum,For training result,For legitimate reading, RBFNN training goal is exactly to subtract as much as possible
The value of small E；
(2) when the value of E is also not up to training requirement index and the frequency of training also not up to setting upper limit, following formula is calculated:η is the step-length in training gradient descent method, and η is bigger, and gradient declines faster, training result
Easily convergence, but η value should be reduced when error change is little, prevent over-fitting and reduce fluctuating error,For the ladder of algorithm
Descent direction is spent, is determined by E, U(4)With U(4+1)The respectively τ and U (c of τ+1 time calculatingj,σj,ωij) value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811137983.8A CN109406117A (en) | 2018-09-28 | 2018-09-28 | A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811137983.8A CN109406117A (en) | 2018-09-28 | 2018-09-28 | A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109406117A true CN109406117A (en) | 2019-03-01 |
Family
ID=65465530
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811137983.8A Pending CN109406117A (en) | 2018-09-28 | 2018-09-28 | A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109406117A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112216085A (en) * | 2020-09-15 | 2021-01-12 | 青岛科技大学 | Equipment key load-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples |
| CN112665837A (en) * | 2020-12-03 | 2021-04-16 | 江阴众和电力仪表有限公司 | Multi-channel vibration monitoring and protecting instrument for rotary machinery |
| TWI758824B (en) * | 2020-08-18 | 2022-03-21 | 神通資訊科技股份有限公司 | Abnormality detection and breakage detection system for mechanical operation and the method thereof |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102063109A (en) * | 2010-11-29 | 2011-05-18 | 株洲南车时代电气股份有限公司 | Neural network-based subway train fault diagnosis device and method |
| US20110125685A1 (en) * | 2009-11-24 | 2011-05-26 | Rizvi Syed Z | Method for identifying Hammerstein models |
| CN104865075A (en) * | 2014-02-26 | 2015-08-26 | 南京理工大学 | Marine diesel engine vibration signal analysis system and method thereof |
| CN105930901A (en) * | 2016-07-18 | 2016-09-07 | 河海大学 | RBPNN-based transformer fault diagnosis method |
-
2018
- 2018-09-28 CN CN201811137983.8A patent/CN109406117A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110125685A1 (en) * | 2009-11-24 | 2011-05-26 | Rizvi Syed Z | Method for identifying Hammerstein models |
| CN102063109A (en) * | 2010-11-29 | 2011-05-18 | 株洲南车时代电气股份有限公司 | Neural network-based subway train fault diagnosis device and method |
| CN104865075A (en) * | 2014-02-26 | 2015-08-26 | 南京理工大学 | Marine diesel engine vibration signal analysis system and method thereof |
| CN105930901A (en) * | 2016-07-18 | 2016-09-07 | 河海大学 | RBPNN-based transformer fault diagnosis method |
Non-Patent Citations (3)
| Title |
|---|
| 晋敏: "基于虚拟仪器和物联网的设备故障监测系统", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 田津等: "基于三阶段RBFNN 学习算法的复杂样本分类研究", 《系统工程与电子技术》 * |
| 肖清湄: "基于LM改进的RBF神经网络算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI758824B (en) * | 2020-08-18 | 2022-03-21 | 神通資訊科技股份有限公司 | Abnormality detection and breakage detection system for mechanical operation and the method thereof |
| CN112216085A (en) * | 2020-09-15 | 2021-01-12 | 青岛科技大学 | Equipment key load-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples |
| CN112665837A (en) * | 2020-12-03 | 2021-04-16 | 江阴众和电力仪表有限公司 | Multi-channel vibration monitoring and protecting instrument for rotary machinery |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108871434B (en) | Online monitoring system and method for rotating equipment | |
| CN105338489B (en) | A kind of intelligent terminal and bluetooth indoor locating system for indoor positioning | |
| CN109406117A (en) | A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm | |
| CN103108412A (en) | Remote monitoring system on parameters of aquafarm water quality and control method | |
| CN110046834A (en) | Workplace pernicious gas health risk quantitative evaluating system and its appraisal procedure | |
| CN105911499A (en) | Ultrasonically local discharging metering system under onsite environment and method | |
| CN100516856C (en) | Intelligent testing system for moisture of cotton fiber | |
| CN104154987A (en) | Vibration wire type sensor reader measuring meter | |
| CN117641157A (en) | Meter reading method of electricity consumption information acquisition terminal | |
| CN106959688A (en) | A kind of vehicle trouble messages acquisition method and device | |
| CN115931665A (en) | Dust monitoring system and method based on underground WiFi6 communication network | |
| CN109857025A (en) | Express item in-transit state monitoring system | |
| CN205983818U (en) | Based on automatic transmission device of sampling monitoring equipment information | |
| CN117572837B (en) | Intelligent power plant AI active operation and maintenance method and system | |
| CN204989451U (en) | On --spot check -up communication tester of electric energy meter | |
| CN203069733U (en) | A vibration-detection-based partial discharging point positioning device used for GIS tests | |
| CN108955867A (en) | A kind of multichannel remote oscillation signal monitor and monitor system | |
| CN201707221U (en) | Mechanical equipment remote detecting service device based on Internet | |
| CN109271948A (en) | A kind of multichannel remote oscillation signal monitor system and monitoring method based on clustering algorithm | |
| CN114415136B (en) | Method and system for online calibrating echo intensity by continuous wave weather radar | |
| CN201561863U (en) | Data quality assurance device for mining equipment fault diagnosis | |
| CN104614015A (en) | Portable field environment data comprehensive acquisition and analysis system | |
| CN203083612U (en) | Comprehensive testing system | |
| CN206975208U (en) | The positioner of crusing robot | |
| CN111855520A (en) | Laser dust particle counter system based on data cloud |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190301 |