CN106563537A - Mill load detection method based on vibration signals of throwing-down area and sliding area of surface of barrel - Google Patents

Mill load detection method based on vibration signals of throwing-down area and sliding area of surface of barrel Download PDF

Info

Publication number
CN106563537A
CN106563537A CN201610976287.0A CN201610976287A CN106563537A CN 106563537 A CN106563537 A CN 106563537A CN 201610976287 A CN201610976287 A CN 201610976287A CN 106563537 A CN106563537 A CN 106563537A
Authority
CN
China
Prior art keywords
area
mill
vibration signal
signal
mill load
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201610976287.0A
Other languages
Chinese (zh)
Other versions
CN106563537B (en
Inventor
司刚全
石建全
李水旺
周舟
贾立新
张彦斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN201610976287.0A priority Critical patent/CN106563537B/en
Publication of CN106563537A publication Critical patent/CN106563537A/en
Application granted granted Critical
Publication of CN106563537B publication Critical patent/CN106563537B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B02—CRUSHING, PULVERISING, OR DISINTEGRATING; PREPARATORY TREATMENT OF GRAIN FOR MILLING
    • B02C—CRUSHING, PULVERISING, OR DISINTEGRATING IN GENERAL; MILLING GRAIN
    • B02C17/00—Disintegrating by tumbling mills, i.e. mills having a container charged with the material to be disintegrated with or without special disintegrating members such as pebbles or balls
    • B02C17/18—Details
    • B02C17/1805—Monitoring devices for tumbling mills
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B02—CRUSHING, PULVERISING, OR DISINTEGRATING; PREPARATORY TREATMENT OF GRAIN FOR MILLING
    • B02C—CRUSHING, PULVERISING, OR DISINTEGRATING IN GENERAL; MILLING GRAIN
    • B02C23/00—Auxiliary methods or auxiliary devices or accessories specially adapted for crushing or disintegrating not provided for in preceding groups or not specially adapted to apparatus covered by a single preceding group
    • B02C23/04—Safety devices
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B02—CRUSHING, PULVERISING, OR DISINTEGRATING; PREPARATORY TREATMENT OF GRAIN FOR MILLING
    • B02C—CRUSHING, PULVERISING, OR DISINTEGRATING IN GENERAL; MILLING GRAIN
    • B02C25/00—Control arrangements specially adapted for crushing or disintegrating

Landscapes

  • Engineering & Computer Science (AREA)
  • Food Science & Technology (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention discloses a mill load detection method based on vibration signals of a throwing-down area and a sliding area of the surface of a barrel. The method is used for solving the problem that the load of a power plant mill is difficult to accurately detect for a long time. By analyzing the grinding mechanism inside a mill, the vibration signals of the throwing-down area and the sliding area are selected as the research object; according to the power spectral centroid, characteristic frequency band selection and characteristic information extraction are conducted on a vibration spectrum; then a prediction model low in complexity and high in accuracy is established through the reconstructed support vector algorithm based on a least squares support vector machine; and finally, consistency check and data fusion are conducted on the predicted estimation value of each vibration signal, so that predicted value errors caused by fault of an individual vibration sensor are avoided. The mill load soft measurement method provided by the invention can ensure good accuracy and sensitivity in the measurement process and provides a guarantee for accurate monitoring of the mill.

Description

The mill load detection method of area and sliding area vibration signal is left based on drum surface
Technical field
The present invention relates to a kind of flexible measurement method of kibbling mill load, and in particular to one kind leaves area based on drum surface With the mill load detection method of sliding area vibration signal.
Background technology
In thermal power plant, cartridge type low speed ball mill is the common equipment in pulverized coal preparation system.Due to barrel type steel ball coal-grinding The characteristics of machine is to the strong adaptability of coal, is still at present a kind of most wide coal pulverizer of domestic application.Coal pulverizer is pulverized coal preparation system Key equipment, can it be operated in optimum condition, be directly connected to the work efficiency of pulverized coal preparation system.And the accurate prison of mill load Survey is to realize real-time adjustment grinding machine operational factor, optimizes the necessary means of current operating condition.
Due to grinding machine poor working environment, dust pollution is big, and internal medium is severe, thus cannot direct measurement, can only pass through Indirect method is detected.At present, there are mill sound method, vibratory drilling method and power method using more mill load detection method.Existing mill sound Method is to judge its load by the single sound transducer easy detection grinding machine noise sound intensity, has the disadvantage that accuracy of detection is not high, not Background noise interference can be effectively removed, particularly simultaneously when multiple stage grinding machine runs in a workshop, when closing on grinding machine operation The noise for being sent can have a strong impact on the accuracy of cutting load testing.When vibratory drilling method is using mill running, abrasive body and material are inclined In the side of grinding machine, the rotating part of grinding machine is in serious unbalanced state, causes unbalanced centrifugal force, and makes grinding machine system System vibration, grinding machine its oscillation intensity and how many relevant this Characteristics Detections of grinding charge doses when rotating speed is constant；And base In vibratory drilling method load monitoring method more than using the relatively low bear vibration of sensitivity and cylinder vibration signal, cause vibratory drilling method to monitor Poor linearity, the problems such as accuracy is not high.The power that motor is consumed when the thinking of power method is by measuring grinding machine operation To judge the load in grinding machine.In actual applications, it is the operating current of measurement grinding machine.This kind of method is disadvantageous in that work Make electric current in whole work process, change is not very big, because the proportion of whole ball mill shared by ature of coal is less in grinding machine, institute To cause measurement sensitivity low.
In recent years, substantial amounts of research is had been carried out to the detection of mill load, such as《Thermal power plant's grinding machine based on ANFIS The soft-sensing model of cutting load testing》(department is just complete, Cao Hui, Zhang Yanbin etc., Chinese journal of scientific instrument, the 4th phase supplementary issue II, and 2007, Vol.28),《Heat engine plant canister type steel ball coal pulverizer load curve reversion based on combined type neutral net》(department is just complete, Cao Hui, Zhang Yan It is refined etc., thermal power generation, the 2007, the 5th phase).《Strategy of Ball Mill Coal Pulverizing System load curve reversion based on neuroid》(Wang Dongfeng, Song is flat).But the problem that these methods are present is to be all based on noise, bearing vibration signal or pressure signal as parameter It is modeled, the vibration signal of selection is mostly the low bearing vibration signal of sensitivity, and is used only to the auxiliary as modeling One of variable, fails to fully play effect of the vibration signal in mill load detection.Meanwhile, it is high, anti-dry based on sensitivity The mill load detection of the strong grinding mill barrel vibration signal of immunity also obtains application on Semi-Autogenous《Investigation on measuring the fill level of an industrial ball mill based on the vibration characteristics of the mill shell》(HUANG P, JIA M P, ZHONG B L.), but vibration used Signal is not to draw from the physical analysis of grinding machine vibration, and its modeling object for being adopted is the vibration letter of all frequencies Number, and most of frequency ranges of the vibration signal in practical application are nonsensical for mill load is detected.
The content of the invention
It is an object of the invention to provide a kind of mill load that area and sliding area vibration signal are left based on drum surface Detection method, with the defect for overcoming above-mentioned prior art to exist, the present invention is using characteristic spectra selection, feature information extraction and base In the least square method supporting vector machine of supporting vector reconstruct, and concordance monitoring and the technology such as data fusion, it is right so as to realize The Accurate Prediction of mill load.
To reach above-mentioned purpose, the present invention is adopted the following technical scheme that：
A kind of mill load detection method that area and sliding area vibration signal are left based on drum surface, including following step Suddenly：
1) some vibrating sensors positioned at same level are installed in drum surface, are gathered by controlling the sampling time Barrel type steel ball grinder leaves the vibration signal of area and sliding area；
2) characteristic spectra selection is carried out to the vibration signal for obtaining based on power spectrum, and according to power spectrum center of gravity and single Gauss Function calculates adaptive weighted coefficient and obtains characterizing the vibration performance information of kibbling mill load；
3) the vibration performance information for leaving area and sliding area to extracting is entered respectively using least square method supporting vector machine Row modeling, obtains initial model, and carries out rarefaction to characteristic information based on restructing algorithm, obtains complexity low high with stability Rarefaction model；
4) rarefaction model is set up to each vibration signal for leaving area and sliding area, carries out load prediction, then to obtaining Multigroup mill load predictive value for leaving area and sliding area carried out based on the two-sided test of hypothesis method of the equal value difference of two normal distributions Homogeneity measure, rejects error prediction value；
5) after the concordance for ensureing the mill load predictive value between each signal, using optimal weighted factor algorithm pair Mill load predictive value carries out data fusion, finally obtains the overall load predictive value of barrel type steel ball grinder.
Further, described vibrating sensor is respectively arranged in apart from the position of grinding mill barrel porch 1/3,1/2,2/3, The area that leaves recognizes that the recognition methodss of the sliding area are to leave area by the position identification device installed in grinding machine bottom For initial time, retardation time t, t=1/3 × T, T are the swing circles of cylinder, in a cycle, each vibrating sensor Collection two-time vibration signal, each vibration signal is expressed as leaving area signal Vib at 1/3 respectivelyp1, sliding area signal at 1/3 Vibh1；Area signal Vib is left at 1/2p2, signal Vib in sliding area at 1/2h2；Area signal Vib is left at 2/3p3, sliding area at 2/3 Signal Vibh3。
Further, described vibration performance information extracting method is based on power spectrum center of gravity and single Gaussian function weighting system What number was carried out, specifically include following steps：
1) power spectral energies are calculated, and mill load ML is gathered first and is gradually believed to the vibration under full mill state from empty mill state Number sequence { xML(i)(n)；N=1,2 ..., N；I=1,2 ..., I }, wherein xML(i)N () represents that in mill load be ML (i) operating modes The nth data of lower collection, N is sample length, and I is the different operating mode quantity of mill load；Jing after discrete Fourier transform (DFT), meter Calculate in frequency range [ωi,ωj] in energy beIt is a certain on frequency band that wherein ω is corresponding Frequency, and ωi,ωj∈ [0, π], to XML(i)(ejω) be sampled,Therefore definition Power spectrum in a certain small frequency range isFrequency information is described with k；
2) identification feature frequency range, sets first from i-th operating mode ML (i) to the spirit in j-th operating mode ML (j) transformation process Sensitivity threshold value is θi→j, then calculate sensitivity of each frequency content in mill load transformation process from ML (i) to ML (j)For si→j(k) ＞ θi→jCorresponding frequency range is screened, and composition is special Levy frequency range collection Kf, its corresponding power spectrum is designated as Pf(k), so as to reject signal in be not included in KfInterior frequency content, it is right to realize The selection of characteristic spectra；
3) energy feature, the gravity frequency to the signal of change power spectrum in the characteristic spectra that obtains are extractedWherein f be sample frequency, ρ1For the first-order autocorrelation coefficient of signal, i.e., The position of centre of gravity of feature power spectrum isΔ f is represented most Small frequency is spaced, and the characteristic energy of vibration signal to each frequency content energy in feature power spectrum by carrying out adaptive weighted obtaining ：Wherein μ (k) is adaptive weighted coefficient, is obtained by single Gaussian functionWherein parameter δ is obtained by following formula, Wherein a1, a2For threshold value, the size of energy attenuation amplitude is affected, so as to the sensitivity of effect characteristicses energy characterization mill load.
Further, described rarefaction model is to replace vibration based on reconstruct supporting vector algorithm minority supporting vector Signal characteristic information sequence, low and improve model robustness so as to substantially reduce model complexity, its method for building up is as follows：
1) original training set is builtX represents after collection and carries out the vibration signal of feature extraction, Y tables Show true mill load, M represents number of samples, with least square method supporting vector machine initial model Model0 is set up, so as to obtain Prediction value setAnd calculate the maximum max_Dist of the Euclidean distance between any two of sample in sample set S, One distance proportion coefficient r of setting, tries to achieve distance radius R=r × max_Dist, and a density is randomly choosed on sample set S Center C=(Xc,Yc)c∈[1,M], initialize supporting vector collection Sv=φ, mark collection U={ (Xj,Yj)∈S,j≠c}；
2) replaces original training set S with the supporting vector of reconstruct, calculate the Euclidean distance of sample and density center C in U to Amount D-c, chooses sample the L={ (X in R distance rangesi,Yi) | D-c (c, j)≤R }, supporting vector collection is updated to mark collectionU=U- { (Xj,Yj)∈L}-(Xc,Yc), if L=is φ, density center C is nearest for chosen distance Point is used as new density center；Otherwise,New density center C is updated toWork as U During=φ, illustrate that whole data set traversal is completed, otherwise recalculate the step, until traversal is completed, the support for finally obtaining Vector set Sv is the supporting vector of reconstruct；
3) using RBF as least square method supporting vector machine kernel function, hyper parameter therein pass through leaving-one method Obtain, least square method supporting vector machine is re-established on supporting vector collection Sv, so as to obtain final soft-sensing model.
Further, first six vibration signals are set up with nine soft-sensing models, then based on the equal value difference of two normal distributions Two-sided test of hypothesis method homogeneity measure is carried out to the predicted load of nine models, concrete grammar is as follows：
1) according to the vibration signal of collection, based on supporting vector restructing algorithm respectively to Vibp1, Vibh1, Vibp2, Vibh2, Vibp3, Vibh3Set up and there is openness least square method supporting vector machine model, be abbreviated as respectivelyAgain to same The vibration signal modeling simultaneously for leaving area and sliding area of one vibrating sensor, respectively obtainsObtain right simultaneously The predictive value answered
2) for the mill load predictive value of each vibration signal sequence regards Normal Distribution as, it is designated as respectivelyWherein, μaFor average,For variance, and for two normal distribution totality, when known to population variance When, the equal value difference of two totality, from statistic of testδ=μi-μj, i, j ∈ [1,9], liFor the number of samples of i-th oscillating sequence, ljFor the number of samples of j-th oscillating sequence, therefore, it is fixed under confidence level α A kind of Confidence distance of justiceWhereinIt is i-th vibration signal The variance of prediction value sequence,It is the variance of j-th vibration signal prediction value sequence, in order to simplify calculating, by giving one Confidence level ε0Come the concordance between the mill load predictive value for judging each signal, if making the statistic of test be againI, j ∈ [1,9], then judge that i-th mill load predictive value and j-th mill load are pre- Measured value has concordance, conversely, not having concordance.
Further, step 5) in carry out data fusion to mill load predictive value using optimal weighted factor algorithm concrete For：
There is the weighter factor of concordance, wherein T≤9, each forecast model to be for the load prediction that hypothesis has T model T=1,2 ... T, and meetMeasurement result after fusion isAfter obtaining its fusion Mean square error is
Due to σ '2Be with regard toQuadratic function, therefore σ '2Certainly exist ratioLittle minima, by solving ConstraintsExtreme value, draw in σ '2Obtaining the weighter factor corresponding to minima is
May certify that simultaneously, the estimated result after fusion is better than any single model estimated result, because after fusion most Little mean square error
Compared with prior art, the present invention has following beneficial technique effect：
The present invention proposes a kind of mill load detection method that area and sliding area vibration signal are left based on drum surface, first First from the vibration mechanism of grinding machine, selection leaves the vibration signal of area and sliding area as object of study；According to power spectrum weight The heart, characteristic spectra selection and feature information extraction are carried out to rumble spectrum, and the present invention is to leave area and slip based on drum surface Multiple vibration signals in area carry out feature information extraction, so as to filter the frequency of vibration unrelated for mill load；Then utilize Reconstruct supporting vector algorithm simultaneously sets up the forecast model for having that complexity is low, degree of accuracy is high based on least square method supporting vector machine, Significantly reduce model complexity, reduced-order models predicted time；Finally, concordance is carried out to the predicted estimate value of each vibration signal Inspection and data fusion, it is most multiple correct negative at last so as to the predictive value mistake for avoiding indivedual vibrating sensor failures from causing Lotus predictive value carries out the load estimation that data fusion obtains optimum, and the present invention is able to ensure that and good standard is obtained in measurement process Exactness and sensitivity, the accurate measurements for grinding machine provide guarantee.
Description of the drawings
Fig. 1 is the system block diagram based on the mill load hard measurement for leaving the multiple vibration signals in area and sliding area；
Fig. 2 is heat engine plant canister type mill load hard measurement hardware system configuration figure；Label in figure is represented respectively：1st, to coal Machine controller, 2, cold-air flap, 3, hot air disperser, 4, recirculation air door, 5, vibrating sensor, 6, position locator, 7, gateway it is poor Pressure sensor, 8, mill exhauster inlet baffle, 9, data acquisition unit and computer, 10, soft-sensing model；It is related to pulverized coal preparation system in figure The label of equipment is represented respectively：11st, run coal bin, 12, feeder, 13, coal pulverizer, 14, mill separator, 15, pulverized-coal collector, 16th, Pulverized Coal Bin, 17, mill exhauster.
Fig. 3 (a) is the position installation diagram of vibrating sensor；
The data sets figure of Fig. 3 (b) vibrating sensors；
Fig. 4 (a) is kibbling mill feature information extraction flow chart；
Fig. 4 (b) is that kibbling mill characteristic spectra extracts flow chart；
Fig. 5 is the least square method supporting vector machine flow chart based on supporting vector reconstruct.
Specific embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings：
By taking certain thermal power plant's the ball type pulverizer system as an example, a concrete application of the present invention is given.As shown in Fig. 2 in this Storage formula pulverized coal preparation system is equipped with a coal pulverizer 13, and model DTM350/700, rotating speed is 17.57r/min, and design rating is 60t/ H, is ground using cartridge type low-speed coal mill.Its workflow is：Raw coal in run coal bin 11 is sent into coal-grinding by feeder 12 In machine 13, while hot blast, cold wind, reduced air are respectively by control cold-air flap 2, hot air disperser 3 and recirculation air door 4 also by amount Into coal pulverizer 13, raw coal is transferred out through crushing grinding, the coal dust for grinding by air-flow, from coal pulverizer 13 out be gas Powder mixture, Jing after mill separator 14, excessively thick coal dust returns to the entrance of coal pulverizer 13 and is ground again, qualified coal dust It is brought into pulverized-coal collector 15 and enters the separation of circulation of qi promoting powder, again qualified coal dust falls into Pulverized Coal Bin 16.
First it is collection of the soft sensor node to vibration signal.Vibrating sensor is respectively arranged in apart from grinding mill barrel On the drum surface of the same level of porch 1/3,1/2,2/3, three vibrating sensors are installed altogether, sliding area is believed with area is left Number collection be that a sensor is gathered by control time, that is, the sensor of a position gathers the position simultaneously It is corresponding to leave area and sliding area vibration signal, its position identification device of identification by installation to leaving area, to sliding area Recognition methodss be to leave area as initial time, retardation time t, t=1/3 × T, T are the swing circles of cylinder.Described position Identifying device is exactly a discriminating gear based on grinding mill barrel surface noise, and in grinding machine running, zones of different is made an uproar Message number makes a big difference, and we gather in advance 360 degree of noises for going out around cylinder, Jing after Fourier transform, by analysis The corresponding area that leaves can be obtained in which angle, so as to obtain leaving zone position.Leaving three, area vibrating sensor simultaneously Vibration signal of collection, after the 0.1s times, then simultaneously in vibration signal of sliding area collection, position identification device is every The positioning updated to leaving area every 5 minutes.The present invention is wirelessly transmitted at data from PTR4000 to the vibration data for collecting In the MSPF2812 of reason center, process is filtered using adaptive weighted filter algorithm in single-chip microcomputer and obtains each vibration letter Number.The position installation diagram and data set constitutional diagram of vibrating sensor is respectively as shown in Fig. 3 (a) and Fig. 3 (b).
Vibration signal to collecting carries out again characteristic spectra selection and feature information extraction, its method such as Fig. 4 (a) and figure Shown in 4 (b), following steps are specifically included：
1) power spectral energies are calculated, and mill load ML is gathered first and is gradually believed to the vibration under full mill state from empty mill state Number sequence { xML(i)(n)；N=1,2 ..., N；I=1,2 ..., I }, wherein xML(i)N () represents that in mill load be ML (i) operating modes The nth data of lower collection, N is sample length, and I is the different operating mode quantity of mill load；Jing discrete Fourier transform (DFT) DTFT Afterwards, can calculate in frequency range [ωi,ωj] in energy beIt is frequency that wherein ω is corresponding The a certain frequency for taking, and ωi,ωj∈ [0, π], to XML(i)(ejw) be sampled, Therefore the power spectrum in a certain small frequency range of definable isFrequency information is described with k.
2) identification feature frequency range, sets first from i-th operating mode ML (i) to the spirit in j-th operating mode ML (j) transformation process Sensitivity threshold value is θi→j=0.05.Sensitivity of each frequency content in mill load transformation process from ML (i) to ML (j) is calculated againFor si→j(k) ＞ θi→jCorresponding frequency range is screened, and composition is special Levy frequency range collection Kf, its corresponding power spectrum is designated as Pf(k).K is not included in so as to reject signalfInterior frequency content, it is right to realize The selection of characteristic spectra.
3) energy feature is extracted, the present invention adopts a kind of Energy extraction algorithm based on power spectrum gravity frequency.To what is obtained The gravity frequency of the signal of change power spectrum in characteristic spectraWherein f be sample frequency, ρ1For signal First-order autocorrelation coefficient, you can the position of centre of gravity of feature power spectrum is Δ f represents minimum frequency space.The characteristic energy of vibration signal is by each frequency content energy in feature power spectrum Carry out adaptive weighted acquisitionWherein μ (k) is adaptive weighted coefficient, by list Gaussian function is obtainedWherein parameter δ is obtained by following formula,Wherein a1=0.3, a2=0.8.
The characteristic information of vibration signal that then will be to extracting is modeled, and its model flow figure is as shown in figure 5, it is built Cube method is as follows：
1) original training set is builtX represents after collection and carries out the vibration signal of feature extraction, Y tables Show true mill load, M represents number of samples, with least square method supporting vector machine (LSSVM) initial model Model0 is set up, from And obtain predicting value setAnd calculate the maximum max_ of the Euclidean distance between any two of sample in sample set S Dist, tries to achieve distance radius R=r × max_Dist, r=0.2.A density center C=(X is randomly choosed on sample set Sc, Yc)c∈[1, M], initializes supporting vector collection Sv=φ, mark collection U={ (Xj,Yj)∈S,j≠c}。
2) replaces original training set S with the supporting vector of reconstruct, calculate the Euclidean distance of sample and density center C in U to Amount D-C, chooses sample the L={ (X in R distance rangesi,Yi)|D-c(c,j)≤R}.Supporting vector collection and mark collection are updated toU=U- { (Xj,Yj)∈L}-(Xc,Yc).If L=is φ, density center C is nearest for chosen distance Point is used as new density center；Otherwise,New density center C is updated toWork as U During=φ, illustrate that whole data set traversal is completed, otherwise recalculate the step, until traversal is completed.The support for finally obtaining Vector set Sv is the supporting vector of reconstruct.
3) using RBF as LSSVM kernel function, hyper parameter therein by leaving-one method obtain, support to Least square method supporting vector machine is re-established on quantity set Sv, so as to obtain final soft-sensing model.Due to the number of samples of Sv It is far smaller than the number of samples in S, and some exceptional values will not be selected as density as during the selection of supporting vector after reconstructing Center, therefore model has the characteristics of model complexity is low, robustness is high.
Finally soft-sensing model is set up to the characteristic information of all of oscillating sequence, totally nine, while obtaining corresponding pre- Measured valueAnd the predictive value to this nine models carries out consistency detection, first, we define a kind of Confidence distanceWhereinIt is the side of i-th vibration signal prediction value sequence Difference,It is the variance of j-th vibration signal prediction value sequence.And pass throughi,j∈[1, 9] there is concordance judging i-th mill load predictive value and j-th mill load predictive value.It is pre- for consistent Measured value, data fusion is carried out using optimal weighted factor algorithm to mill load predictive valueWhereinMay certify that the estimated result after fusion is better than any single model estimated result.

Claims (6)

1. a kind of mill load detection method that area and sliding area vibration signal are left based on drum surface, it is characterised in that bag Include following steps：
1) some vibrating sensors positioned at same level are installed in drum surface, by controlling the sampling time cartridge type is gathered Steel ball grinder leaves the vibration signal of area and sliding area；
2) characteristic spectra selection is carried out to the vibration signal for obtaining based on power spectrum, and according to power spectrum center of gravity and single Gaussian function Calculate adaptive weighted coefficient to obtain characterizing the vibration performance information of kibbling mill load；
3) the vibration performance information for leaving area and sliding area to extracting is built respectively using least square method supporting vector machine Mould, obtains initial model, and carries out rarefaction to characteristic information based on restructing algorithm, obtains that complexity is low and that stability is high is dilute Thinization model；
4) rarefaction model is set up to each vibration signal for leaving area and sliding area, carries out load prediction, then it is many to what is obtained The two-sided test of hypothesis method that group leaves area with the mill load predictive value of sliding area is based on the equal value difference of two normal distributions carries out consistent Property estimate, reject error prediction value；
5) after the concordance for ensureing the mill load predictive value between each signal, using optimal weighted factor algorithm to grinding machine Predicted load carries out data fusion, finally obtains the overall load predictive value of barrel type steel ball grinder.
2. a kind of mill load detection that area and sliding area vibration signal are left based on drum surface according to claim 1 Method, it is characterised in that described vibrating sensor is respectively arranged in apart from the position of grinding mill barrel porch 1/3,1/2,2/3, The area that leaves recognizes that the recognition methodss of the sliding area are to leave area by the position identification device installed in grinding machine bottom For initial time, retardation time t, t=1/3 × T, T are the swing circles of cylinder, in a cycle, each vibrating sensor Collection two-time vibration signal, each vibration signal is expressed as leaving area signal Vib at 1/3 respectivelyp1, sliding area signal at 1/3 Vibh1；Area signal Vib is left at 1/2p2, signal Vib in sliding area at 1/2h2；Area signal Vib is left at 2/3p3, sliding area at 2/3 Signal Vibh3。
3. a kind of mill load detection that area and sliding area vibration signal are left based on drum surface according to claim 1 Method, it is characterised in that described vibration performance information extracting method is based on power spectrum center of gravity and single Gaussian function weighting system What number was carried out, specifically include following steps：
1) power spectral energies are calculated, and mill load ML is gathered first from empty mill state gradually to the vibration signal sequence under full mill state Row { xML(i)(n)；N=1,2 ..., N；I=1,2 ..., I }, wherein xML(i)N () represents and is adopted in the case where mill load is ML (i) operating modes The nth data of collection, N is sample length, and I is the different operating mode quantity of mill load；Jing after discrete Fourier transform (DFT), calculate Frequency range [ωi,ωj] in energy beIt is a certain frequency on frequency band that wherein ω is corresponding Rate, and ωi,ωj∈ [0, π], to XML(i)(ejω) be sampled,Therefore define certain Power spectrum in one small frequency range isFrequency information is described with k；
2) identification feature frequency range, sets first from i-th operating mode ML (i) to the sensitivity in j-th operating mode ML (j) transformation process Threshold value is θi→j, then calculate sensitivity of each frequency content in mill load transformation process from ML (i) to ML (j)For si→j(k) ＞ θi→jCorresponding frequency range is screened, and composition is special Levy frequency range collection Kf, its corresponding power spectrum is designated as Pf(k), so as to reject signal in be not included in KfInterior frequency content, it is right to realize The selection of characteristic spectra；
3) energy feature, the gravity frequency to the signal of change power spectrum in the characteristic spectra that obtains are extractedWherein f be sample frequency, ρ1For the single order auto-correlation system of signal Count, the position of centre of gravity for obtaining final product feature power spectrum isΔ f tables Show minimum frequency space, the characteristic energy of vibration signal adds by carrying out self adaptation to each frequency content energy in feature power spectrum Power is obtained：Wherein μ (k) is adaptive weighted coefficient, is obtained by single Gaussian functionWherein parameter δ is obtained by following formula, Wherein a1, a2For threshold value, the size of energy attenuation amplitude is affected, so as to the sensitivity of effect characteristicses energy characterization mill load.
4. a kind of mill load detection that area and sliding area vibration signal are left based on drum surface according to claim 1 Method, it is characterised in that described rarefaction model is to replace vibration based on reconstruct supporting vector algorithm minority supporting vector Signal characteristic information sequence, low and improve model robustness so as to substantially reduce model complexity, its method for building up is as follows：
1) original training set is builtX is represented after collection and is carried out the vibration signal of feature extraction, and Y represents true Real mill load, M represents number of samples, and with least square method supporting vector machine initial model Model0 is set up, so as to be predicted Value setAnd calculate the maximum max_Dist of the Euclidean distance between any two of sample in sample set S, setting One distance proportion coefficient r, tries to achieve distance radius R=r × max_Dist, and a density center C is randomly choosed on sample set S =(Xc,Yc)c∈[1,M], initialize supporting vector collection Sv=φ, mark collection U={ (Xj,Yj)∈S,j≠c}；
2) replace original training set S with the supporting vector of reconstruct, calculate the Euclidean distance vector D_ of sample and density center C in U C, chooses sample the L={ (X in R distance rangesi,Yi) | D_c (c, j)≤R }, supporting vector collection is updated to mark collectionU=U- { (Xj,Yj)∈L}-(Xc,Yc), if L=is φ, density center C is nearest for chosen distance Point is used as new density center；Otherwise,New density center C is updated toWork as U During=φ, illustrate that whole data set traversal is completed, otherwise recalculate the step, until traversal is completed, the support for finally obtaining Vector set Sv is the supporting vector of reconstruct；
3) using RBF as least square method supporting vector machine kernel function, hyper parameter therein obtained by leaving-one method , least square method supporting vector machine is re-established on supporting vector collection Sv, so as to obtain final soft-sensing model.
5. a kind of mill load detection that area and sliding area vibration signal are left based on drum surface according to claim 1 Method, it is characterised in that first six vibration signals are set up with nine soft-sensing models, then based on the equal value difference of two normal distributions Two-sided test of hypothesis method carries out homogeneity measure to the predicted load of nine models, and concrete grammar is as follows：
1) according to the vibration signal of collection, based on supporting vector restructing algorithm respectively to Vibp1, Vibh1, Vibp2, Vibh2, Vibp3, Vibh3Set up and there is openness least square method supporting vector machine model, be abbreviated as respectivelyShake to same again The vibration signal modeling simultaneously for leaving area and sliding area of dynamic sensor, respectively obtainsObtain corresponding simultaneously Predictive value
2) for the mill load predictive value of each vibration signal sequence regards Normal Distribution as, it is designated as respectivelyWherein, μaFor average,For variance, and for two normal distribution totality, when known to population variance When, the equal value difference of two totality, from statistic of testδ=μi-μj, i, j ∈ [1,9], liFor the number of samples of i-th oscillating sequence, ljFor the number of samples of j-th oscillating sequence, therefore, it is fixed under confidence level α A kind of Confidence distance of justiceWhereinIt is i-th vibration signal The variance of prediction value sequence,It is the variance of j-th vibration signal prediction value sequence, in order to simplify calculating, by giving one Confidence level ε0Come the concordance between the mill load predictive value for judging each signal, if making the statistic of test be againI, j ∈ [1,9], then judge that i-th mill load predictive value and j-th mill load are pre- Measured value has concordance, conversely, not having concordance.
6. a kind of mill load detection that area and sliding area vibration signal are left based on drum surface according to claim 1 Method, it is characterised in that step 5) in carry out data fusion to mill load predictive value using optimal weighted factor algorithm concrete For：
There is the weighter factor of concordance, wherein T≤9, each forecast model to be for the load prediction that hypothesis has T modelT= 1,2 ... T, and meetMeasurement result after fusion isObtain mean square after its fusion Error is
Due to σ '2Be with regard toQuadratic function, therefore σ '2Certainly exist ratioLittle minima, by solving constraint ConditionExtreme value, draw in σ '2Obtaining the weighter factor corresponding to minima is
May certify that simultaneously, the estimated result after fusion is better than any single model estimated result, because the minimum after fusion is Square error
CN201610976287.0A 2016-11-07 2016-11-07 The mill load detection method with sliding area vibration signal is left based on drum surface Active CN106563537B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610976287.0A CN106563537B (en) 2016-11-07 2016-11-07 The mill load detection method with sliding area vibration signal is left based on drum surface

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610976287.0A CN106563537B (en) 2016-11-07 2016-11-07 The mill load detection method with sliding area vibration signal is left based on drum surface

Publications (2)

Publication Number Publication Date
CN106563537A true CN106563537A (en) 2017-04-19
CN106563537B CN106563537B (en) 2019-02-05

Family

ID=58540304

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610976287.0A Active CN106563537B (en) 2016-11-07 2016-11-07 The mill load detection method with sliding area vibration signal is left based on drum surface

Country Status (1)

Country Link
CN (1) CN106563537B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107597313A (en) * 2017-10-27 2018-01-19 江西理工大学 Contact detecting and analysing system and its method of testing suitable for rotating grinding mill barrel
CN111089740A (en) * 2019-11-25 2020-05-01 北京工业大学 Multi-layer correlation-based multi-channel mechanical spectrum sub-band characteristic evaluation method
CN111611533A (en) * 2020-05-25 2020-09-01 湖南大学 Mill load characteristic extraction method based on self-adaptive VMD and improved power spectrum
CN112197817A (en) * 2020-12-01 2021-01-08 矿冶科技集团有限公司 Method and device for measuring grinding state of grinding machine and electronic equipment
CN112317109A (en) * 2020-09-27 2021-02-05 鞍钢集团矿业有限公司 Cone crusher fault pre-judging method
CN113182023A (en) * 2021-04-21 2021-07-30 南京工程学院 On-line detection method for mill load of non-measurable disturbance self-adaptive monitoring and compensation
CN113190983A (en) * 2021-04-21 2021-07-30 南京工程学院 Thermal power plant mill load prediction method based on composite soft measurement
CN113953056A (en) * 2021-09-28 2022-01-21 宁波圣燕工程设备有限公司 Utilize inside quantity control of vibration response to smash complete cement manufacture equipment
CN114522776A (en) * 2022-04-23 2022-05-24 南通腾宇环保设备有限公司 Zinc dross screening-grinding process

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN2129649Y (en) * 1991-04-10 1993-04-14 华北电力学院 Vibration frequency energy type ball mill load measurer
CN101024210A (en) * 2006-02-21 2007-08-29 上海重型机器厂有限公司 Dynamic separator
CN101126680A (en) * 2007-09-11 2008-02-20 西安交通大学 Thermal power plant ball mill load soft-sensing method
US20120111978A1 (en) * 2010-11-08 2012-05-10 Alstom Technology Ltd. System and method for monitoring operational characteristics of pulverizers
CN103150413A (en) * 2011-12-07 2013-06-12 华北电力科学研究院有限责任公司 Determining method and device for RB (Runback) project parameters in power system
CN203082908U (en) * 2013-01-29 2013-07-24 福州福大自动化科技有限公司 Energy conservation optimization system of direct-blow-type double-inlet and double-outlet steel ball mill
CN103902776A (en) * 2014-04-02 2014-07-02 沈阳化工大学 Wet type ball grinder load parameter integrated modeling method based on EEMD (ensemble empirical mode decomposition)

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN2129649Y (en) * 1991-04-10 1993-04-14 华北电力学院 Vibration frequency energy type ball mill load measurer
CN101024210A (en) * 2006-02-21 2007-08-29 上海重型机器厂有限公司 Dynamic separator
CN101126680A (en) * 2007-09-11 2008-02-20 西安交通大学 Thermal power plant ball mill load soft-sensing method
US20120111978A1 (en) * 2010-11-08 2012-05-10 Alstom Technology Ltd. System and method for monitoring operational characteristics of pulverizers
CN103150413A (en) * 2011-12-07 2013-06-12 华北电力科学研究院有限责任公司 Determining method and device for RB (Runback) project parameters in power system
CN203082908U (en) * 2013-01-29 2013-07-24 福州福大自动化科技有限公司 Energy conservation optimization system of direct-blow-type double-inlet and double-outlet steel ball mill
CN103902776A (en) * 2014-04-02 2014-07-02 沈阳化工大学 Wet type ball grinder load parameter integrated modeling method based on EEMD (ensemble empirical mode decomposition)

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
司刚全, 曹晖等: "基于复合式神经网络的火电厂", 《热力发电》 *

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107597313A (en) * 2017-10-27 2018-01-19 江西理工大学 Contact detecting and analysing system and its method of testing suitable for rotating grinding mill barrel
CN107597313B (en) * 2017-10-27 2019-04-09 江西理工大学 Suitable for rotating the contact detecting and analysing system and its test method of grinding mill barrel
CN111089740A (en) * 2019-11-25 2020-05-01 北京工业大学 Multi-layer correlation-based multi-channel mechanical spectrum sub-band characteristic evaluation method
CN111611533A (en) * 2020-05-25 2020-09-01 湖南大学 Mill load characteristic extraction method based on self-adaptive VMD and improved power spectrum
CN112317109B (en) * 2020-09-27 2022-02-18 鞍钢集团矿业有限公司 Cone crusher fault pre-judging method
CN112317109A (en) * 2020-09-27 2021-02-05 鞍钢集团矿业有限公司 Cone crusher fault pre-judging method
CN112197817B (en) * 2020-12-01 2021-03-02 矿冶科技集团有限公司 Method and device for measuring grinding state of grinding machine and electronic equipment
CN112197817A (en) * 2020-12-01 2021-01-08 矿冶科技集团有限公司 Method and device for measuring grinding state of grinding machine and electronic equipment
CN113182023A (en) * 2021-04-21 2021-07-30 南京工程学院 On-line detection method for mill load of non-measurable disturbance self-adaptive monitoring and compensation
CN113190983A (en) * 2021-04-21 2021-07-30 南京工程学院 Thermal power plant mill load prediction method based on composite soft measurement
CN113182023B (en) * 2021-04-21 2022-06-03 南京工程学院 On-line detection method for mill load of non-measurable disturbance self-adaptive monitoring and compensation
CN113190983B (en) * 2021-04-21 2024-03-01 南京工程学院 Thermal power plant mill load prediction method based on composite soft measurement
CN113953056A (en) * 2021-09-28 2022-01-21 宁波圣燕工程设备有限公司 Utilize inside quantity control of vibration response to smash complete cement manufacture equipment
CN114522776A (en) * 2022-04-23 2022-05-24 南通腾宇环保设备有限公司 Zinc dross screening-grinding process
CN114522776B (en) * 2022-04-23 2022-06-24 南通腾宇环保设备有限公司 Zinc dross screening-grinding process

Also Published As

Publication number Publication date
CN106563537B (en) 2019-02-05

Similar Documents

Publication Publication Date Title
CN106563537B (en) The mill load detection method with sliding area vibration signal is left based on drum surface
CN106568503B (en) A kind of mill load detection method based on drum surface multiple spot vibration signal
CN104614179B (en) A kind of gearbox of wind turbine state monitoring method
Zhao et al. Fault diagnosis of wind turbine bearing based on stochastic subspace identification and multi-kernel support vector machine
CN100538316C (en) A kind of thermal power plant ball mill load soft-sensing method
CN105268536B (en) A kind of thermal power plant ball mill load soft-sensing method of adaptive on-line correction
CN105021334B (en) Fan energy consumption monitoring recognition methods based on multiple weighing value neutral net
CN109101768A (en) Leaf end timing sensor layout optimization design method based on compressed sensing
CN103149029B (en) Cepstrum identification aeromotor is utilized to turn quiet method of touching position of rubbing
CN109000921B (en) Method for diagnosing main shaft fault of wind turbine generator
CN102944418A (en) Wind turbine generator group blade fault diagnosis method
CN102889987A (en) Gear fault diagnosis platform and gear fault diagnosis method
CN103902776A (en) Wet type ball grinder load parameter integrated modeling method based on EEMD (ensemble empirical mode decomposition)
CN104215323A (en) Method for determining sensitivity of each sensor in mechanical equipment vibrating sensor network
Sun et al. Fault diagnosis of mechanical equipment in high energy consumption industries in China: A review
CN108444696A (en) A kind of gearbox fault analysis method
CN106404377A (en) Transformer mechanical fault diagnosis method based on acoustic imaging technology
CN102721462B (en) Method for quickly computing Bode plot and Nyquist plot of rotary mechanical vehicle starting and parking processes
CN112163350B (en) Real-time on-line soft measurement system and method for pulverized coal fineness of double models of pulverizing system
CN106018730A (en) Coal water content measurement device and method based on coal mill inlet primary air correction
CN107101834A (en) The turbo-charger surge prediction meanss and Forecasting Methodology of feature based frequency
CN108038079A (en) A kind of multi-source mechanical signal analysis and optimization combined method
CN114705432B (en) Method and system for evaluating health state of explosion-proof motor bearing
CN115542099A (en) Online GIS partial discharge detection method and device
CN108844620B (en) gravity type propeller noise measuring device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant