CN101000326A - Method for investigating noise character of converter steelmaking blowing slag-making - Google Patents

Abstract

The invention discloses converter steel and slag making air refining noise feature detecting method. It includes the following steps: sampling air refining noise; calculating out the data reflected the slag making state to guide steel making; using Blackman window corrected Welch arithmetic to compute power spectrum; finding out each feature frequency band and its main one to confirm positive and negative ones; processing linear combination for each feature frequency band strength to gain the data reflected current slag making to guide frequency slag making, judging spurting, and dry returning. The result is more exact.

Description

Method for investigating noise character of converter steelmaking blowing slag-making

Technical field

The present invention relates to a kind ofly extract the method for slag making state, relate in particular to a kind of method for investigating noise character of converter steelmaking blowing slag-making by the blowing noise.

Technical background

Slag making is one of key operation of pneumatic steelmaking, and optimizing slag making is the precondition of optimizing pneumatic steelmaking, only forms good slag and could effectively remove harmful elements such as sulphur in the molten steel, phosphorus in steelmaking process, could improve steel quality.According to the height of slag liquid level in the stove, can judge come out of the stove in the quality of slag making: the slag liquid level is low excessively, shows that the solid matter ratio increases in the slag, and the liquid phase ratio reduces, and the possibility of " returning dried " is arranged, and should improve the rifle position or reduce the flow of oxygen; The slag liquid level is too high, and the possibility of splash is arranged, and should reduce the rifle position, improves the bottom blowing stirring intensity.

In the blowing slagging process of pneumatic steelmaking, should avoid splash as far as possible and return dried generation, particularly serious splash can bring equipment and personnel's loss.Therefore need in time carry out oxygen rifle height control according to the slag making state, prevent splash and return dried generation.Usually, be to judge by the blowing noise that sends in the converter for the online detection of slag, but people's unstable working condition can't realize automatic steel-making by skilled kiln worker.

Adopt computing machine to imitate the people and judge the slag making state, gather the noise that slagging process produces at the fire door place, analyze, be called the audio frequency slag practice by computing machine.Audio frequency slag practice in the past be analyze the fire door noise collect bulk strength over time, judge relatively that with empirical value slag making carries out state, the noise intensity of analysis at a specific frequency or frequency band further arranged, generally get the signal in the 180-200Hz scope.

These audio frequency slag practices have its relative rationality, can reflect the various states in the smelting process to a certain extent, but, also there are some problems, embody a concentrated expression of on the hysteresis quality of state-detection, return often and do or splash has taken place or imminent edge, could these states of perception, can't be in time or instant intervention as the adjusting of operation.In addition, the accuracy of detection is not high yet, has serious omission, flase drop phenomenon.

Reason is that this disposal route is too simple, do not consider the complicacy of slag making reaction fully, think that the source of blowing noise mainly is that the supersonic speed Oxygen Flow is impacted and slag making reflection equal excitation source produces metal bath, slag liquid and air system excitation in the body of heater.

Summary of the invention

The problem that purpose of the present invention is exactly can't be accurately in order to solve present employing audio frequency slag practice, in time detect, a kind of accuracy in detection height that has is provided, omission, flase drop probability are little, and the method for investigating noise character of converter steelmaking blowing slag-making of advantage such as slag making state comparatively accurately can be provided.

For achieving the above object, the present invention has adopted following technical scheme:

A kind of method for investigating noise character of converter steelmaking blowing slag-making, its method be,

(1) utilizes acoustic pickup to gather blowing noise in the slagging process in real time, send into computing machine then at converter mouth;

(2) computing machine carries out the initialization of system, determines each feature band distribution range of blowing noise;

(3) after the blowing beginning, gather the noise signal of t＜1 second time period by acoustic pickup;

(4) utilize the noise signal data that collect to utilize computed in software to go out power spectrum by computing machine;

(5) computing machine extracts corresponding power spectrum in each feature band scope then, and the power spectrum in each feature band is added up, and obtains the power level of this feature band;

(6) computing machine again the main feature band in the feature band and each in the same way the power level of feature band add up, deduct the power level of each opposite feature frequency band, obtain the slag making state parameter, judge the slag making state, finish the once analytical calculation of sampling;

(7) proceed sampling, carried out for (4) (5) (7) step once more and handle, obtain corresponding moment slag making state parameter; So circulation is up to finishing blowing.

In the described step (1), because the power of blowing noise mainly is distributed in below the 2KHz, for fear of the high frequency aliasing, sample frequency will be higher than 4KHz, and sampling precision is more than or equal to 16.

In the described step (3), in slagging process, the blowing noise is subjected to the influence of slag making state and changes, and therefore need be divided into slagging process the time period of a large amount of t＜1 second, think that the characteristic of noise remains unchanged in each time period, therefore calculate power spectrum by sampled data in each time period.

In the described step (4), the power spectrum computing method are, adopt the Blackman window function to combine with the Welch algorithm, the Welch algorithm of formation Blackman function correction, and its specific algorithm is as follows:

A, at first be to be hits that blowing noise samples value x (t) sequence of N is divided into the L section, every section has M sampled value, and Noverlap overlap sampling value arranged between the adjacent segment, has

$L=\mathrm{fix}\left(\frac{N-\mathrm{Noverlap}}{M-\mathrm{Noverlap}}\right)$

Wherein the effect of fix function is to round to zero direction, and t is the sequence number that sampled value was arranged according to the time.

I section sample sequence is

x ⁱ(t)＝x(t+iM-M)， 1≤t≤M， 1≤i≤L

B, be with Blackman window function correction Welch algorithm computation power spectrum Blackman window function

$w\left(t\right)=0.42-0.5\cos \left(\frac{2\mathrm{\π}}{M}t\right)+0.08\cos \left(\frac{2\mathrm{\π}}{M}2t\right)$

Obtain the power spectrum of each section

${\hat{P}}_{\mathrm{PER}}^i\left(f\right)=\frac{1}{\mathrm{MU}}{\left|\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{x}^i\left(t\right)w\left(t\right)\exp (-j2\mathrm{\πft}/M)\right|}^2$

Wherein $U=\frac{1}{M}\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{w}^2\left(t\right)$ Be normalized factor;

Then the power spectrum in this moment is adding up of every section power spectrum

${\hat{P}}_{\mathrm{PER}}\left(f\right)=\frac{1}{\mathrm{MUL}}\underset{m=1}{\overset{L}{\mathrm{\Σ}}}{\left|\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{x}^i\left(t\right)w\left(t\right)\exp (-j2\mathrm{\πft}/M)\right|}^2,$

From the power spectrum that calculates as can be seen, the energy of blowing noise mainly concentrates on the several frequency bands, and the position of these frequency bands remains unchanged in whole slagging process, the characteristics of noise frequency band is referred to as to blow, and the highest feature band of one of them intensity is referred to as main feature band, the evolution trend feature band similar to main feature band is feature band in the same way At All Other Times, is the opposite feature frequency band with main feature band is opposite trend.

Feature band adopts following method to carry out the distribution range of feature band:

Each power spectrum constantly of A, the whole converting process of calculating

B, the power spectrum in the whole converting process is added up at time orientation, obtain P (f)

$P\left(f\right)=\underset{t}{\mathrm{\Σ}}{\hat{P}}_{\mathrm{PER}}(f,t)$

The halfwidth of C, each peak value of calculating P (f) can obtain the position of corresponding each extraordinary frequency band.

The present invention by the blowing noise is sampled, calculates the data of reflection slag making state in real time in steel-making blowing slagging process, in order to instruct the steel-making operation.It adopts the Welch algorithm computation power spectrum of Blackman window correction, find out each feature band and main feature band wherein, determine feature band and opposite feature frequency band in the same way, the intensity of each feature band is carried out linear combination, obtain reflecting the data of current time slag making state, instruct the audio frequency slag making to handle and judge splash and return dried trend, the result is more accurate.

The invention has the beneficial effects as follows:

1, result of calculation can reflect pneumatic steelmaking slag making state more accurately, instructs the steel-making operation on this basis, can shorten the steel-making cycle, improves steel-making efficient;

2, result of calculation can reflect splash more accurately and return dried, can in time forecast according to result of calculation in splash or before returning dried the generation, take to adjust measures such as oxygen rifle height in advance, reduce the splash incidence, thereby significantly alleviated the molten iron that splash causes, equipment or even personal casualty loss;

3, by at fire door collection blowing noise, make and fall the cover slag making and become possibility, can realize that furnace gas reclaims, flue dust collecting can bring huge economic benefit and environmental benefit.

Description of drawings

Fig. 1 is the power spectrum of converter slag-making blowing noise;

Fig. 2 is power spectrum-time distribution map;

Fig. 3 is a power spectrum time accumulation curve;

The Welch algorithm computation result of Fig. 4 Blackman window correction;

Fig. 5 feature band power spectrum intensity time evolution.

Embodiment

The invention will be further described below in conjunction with accompanying drawing and embodiment.

Because blowing noise right and wrong stochastic process stably produce, cause frequency spectrum to have strong random character, particularly phase place wherein is a completely random, be difficult to analyze its trend, so the present invention adopts the mode of calculating its power spectrum to analyze intensity distributions and the differentiation characteristic thereof of noise on different frequency.

In steelmaking process, converter is a time-varying system, therefore must be divided into many very short time intervals to whole steelmaking process, can think that the state of converter system remains unchanged in each time interval.In each time interval, calculate the power spectrum of blowing noise,, be referred to as power spectrum, as shown in Figure 1 to reflect the slag making characteristic in this time period.In 500Hz, the power spectrum of noise on frequency direction, exist a plurality of maximum value ( 110, 120, 130, 140, 150), illustrating that the blowing noise is not a white noise, each power spectrum maximum value should reflect some characteristic of slagging process, is characteristic frequency.But because its strong random character, the broadening phenomenon can appear in characteristic frequency, constitutes a plurality of feature bands.

Further calculate the power spectrum of whole slagging process, obtain power spectrum distribution plan in time, as shown in Figure 2, can obviously find out the blowing noise a plurality of feature bands ( 210, 220, 230, 240, 250, 260).Particularly the power spectrum intensity of feature band changes with the progress of slagging process, should be the variation that has reflected the slag making state therefore.

Influenced by random character, the extreme value of the power spectrum of each time period that calculates and distribution all have difference, and directly the result from the single time period is difficult to obtain the distribution of feature band accurately.Therefore power spectrum is carried out integration along time orientation, eliminates the influence of its random character, the power spectrum that the obtains whole slagging process smooth curve of distribution that adds up, as shown in Figure 3, thus can extract more exactly each feature band center frequency point ( 310, 320, 330, 340) and distribution range ( 315, 325, 335, 345).Wherein the frequency band that intensity is the highest is referred to as main feature band (325).

After determining feature band, tackle each feature band power spectrum intensity evolution rule in time and analyze.The precision that will determine the evolution law-analysing to the accuracy and the stability of the band power spectrum result of calculation of random signal, and the front directly rises and falls very big by the power spectrum that Fourier transform obtains, still with very big random character, be difficult to carry out subsequent analysis processing.Therefore adopt Welch direct method estimated power spectrum, and propose further to use the Blackman window function on this basis and revise, come the rated output spectrum.

When rated output was composed, the hits of each time period was N, and promptly calculation window is N.In order to improve feature band continuity in time, and improve the promptness of result of calculation, adjacent 2 times calculation window keeps most of and overlaps, i.e. window shifts Δ N, and Δ N＜N, generalized case Δ N are 1/4,1/8 or 1/16 of N.

After the estimated value that calculates more stable power spectrum, analyze the variation tendency of power spectrum in slagging process.Each feature band remains unchanged on frequency, but the power spectrum intensity of different characteristic frequency band has different variation tendencies in time, as shown in Figure 5, can be divided into 2 classes basically, wherein the evolving trend of Partial Feature frequency band is similar to main feature channel, be referred to as in the same way feature band ( 510), in addition be opposite trend with main feature band, be referred to as the opposite feature frequency band ( 530, 540).The power level addition of the feature band in the same way in the identical moment, and deduct the power level of each opposite feature frequency band, the result can reflect the slag making state exactly.This is because the counteracting factor between the opposite feature band of the trend of having eliminated, and data are sensitiveer to the reflection of slag making state, use original audio frequency slag making disposal route on this basis and judge splash and return dried trend that the result is more accurate.

Concrete steps of the present invention are as follows:

(1) becomes in the process of molten steel in stove molten iron refining, the noise between the blowing process is sampled.Because the distribution of feature band all below 1KHz, particularly mainly concentrates on below the 500Hz, so sample frequency generally adopts 11.025Khz or 22.05KHz, and higher sample frequency can not improve data precision.Sampling precision is not less than 16.

(2) when hits reaches N, carry out the Welch method of Blackman window correction and calculate, the result is as the power spectrum of current time, and the power spectrum to each time period carries out integration on time orientation at last, obtains the power spectrum time accumulation curve as Fig. 3.

The concrete computing method that power spectrum calculates are as follows:

At first be to be length that the blowing noise samples point sequence of N is divided into the L section, every section has M sample, and Noverlap overlapping sample arranged between the adjacent segment, has

$L=\mathrm{fix}\left(\frac{N-\mathrm{Noverlap}}{M-\mathrm{Noverlap}}\right)$

I section sample sequence is

x ⁱ(t)＝x(t+iM-M)， 1≤n≤M， 1≤l≤L

Adopt Welch algorithm estimated power spectrum, and introduce the Blackman window function

$w\left(t\right)=0.42-0.5\cos \left(\frac{2\mathrm{\π}}{M}t\right)+0.08\cos \left(\frac{2\mathrm{\π}}{M}2t\right)$

Revise the Welch algorithm, obtain the power spectrum of each section

${\hat{P}}_{\mathrm{PER}}^i\left(f\right)=\frac{1}{\mathrm{MU}}{\left|\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{x}^i\left(t\right)w\left(t\right)\exp (-j2\mathrm{\πft}/M)\right|}^2$

Wherein $U=\frac{1}{M}\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{w}^2\left(t\right)$ Be normalized factor;

Then the power spectrum in this moment is adding up of every section power spectrum

${\hat{P}}_{\mathrm{PER}}\left(f\right)=\frac{1}{\mathrm{MUL}}\underset{m=1}{\overset{L}{\mathrm{\Σ}}}{\left|\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{x}^i\left(t\right)w\left(t\right)\exp (-j2\mathrm{\πft}/M)\right|}^2$

The Welch algorithm computation result of Blackman window correction ( 400) flatness very good, and all be nonnegative value.

On this basis, select 4 or more a plurality of extreme point of intensity maximum, as the centre frequency of each feature band.With the halfwidth of extreme point intensity distribution range as this feature band.Wherein the maximal value correspondence main feature band ( 325).

For each power spectrum constantly, the power spectrum intensity level of each frequency correspondence in each feature band scope is added up, as the intensity of character pair frequency band in this moment.As Fig. 5, the power spectrum intensity of each feature band that draws along the evolution curve of time ( 500), therefrom find out with main feature necktie temporal evolution rule ( 520) roughly the same feature band ( 510), be called feature band in the same way, remaining evolution rule is roughly opposite with main feature band, be referred to as the opposite feature frequency band ( 530, 540).Initialization is finished.(3) numerical value in each feature band is added up the power level of each character pair frequency band of attaining the Way.Main feature band and in the same way the power level of feature band add up, deduct the power level of each opposite feature frequency band then, as the parameter of reflection slag making state.

(4) audio frequency slag making system judges according to this result of calculation whether the slag making state is normal.

(5) proceed sampling, obtain corresponding moment slag making state parameter; Carrying out for (4) step once more handles; So circulation is up to finishing blowing.

In actual computation, the size of N value can influence precision of calculation results and accuracy.Value is big, and the stability of the power spectrum that calculates can improve, and possesses higher frequency resolution, but time-delay increases, and has reduced real-time; Value is little, and real-time performance improves, but is subjected to the influence of random character very big, and frequency resolution is very low.Therefore according to the difference of noise signal sample frequency, N generally gets intermediate value, make time-delay about 0.2s, as when the sampling rate 11.025KHz, N generally is taken as 2048, and Δ N is 128 or 256, maximum delay was less than 0.2 second, when sampling rate was 22.05KHz, N generally was taken as 2048 or 4096, and Δ N is 256 or 512.

In long-term steelmaking process, reasons such as erosion along with converter inside, can cause the feature band of converter to be offset gradually, for the accuracy that keeps calculating, need revise the distributing position of feature band: after every one heat steel finishing blowing, each power spectrum constantly of converting process is added up in time, obtain power spectrum time accumulation curve (300), redefine the distribution range of each feature band on this basis as Fig. 3.After converter carried out large repairs, characteristic variations was bigger, need carry out initialization again.

In steel-making blowing slagging process, by the blowing noise is sampled, calculate the data of reflection slag making state in real time, in order to instruct the steel-making operation.

Because the volume of different converters, structure composition etc. have nothing in common with each other, the distribution of their feature band and evolution rule are also had nothing in common with each other.When therefore in a converter, using slag making noise monitoring method, at first to carry out initialization to the feature band of this converter.

Claims (5)

1, a kind of method for investigating noise character of converter steelmaking blowing slag-making is characterized in that, may further comprise the steps:

(1) utilizes acoustic pickup to gather blowing noise in the slagging process in real time, send into computing machine then at converter mouth;

(2) computing machine carries out system initialization, determines each feature band distribution range of blowing noise;

(3) after the blowing beginning, gather the noise signal of t＜1 second time period by acoustic pickup;

(4) utilize the noise signal data that collect to utilize computed in software to go out power spectrum by computing machine;

(5) computing machine extracts corresponding power spectrum in each feature band scope then, and the power spectrum in each feature band is added up, and obtains the power level of this feature band;

(6) with the main feature band in the feature band and each in the same way the power level of feature band add up, deduct the power level of each opposite feature frequency band, obtain the slag making state parameter, judge the slag making state, finish the once analytical calculation of sampling;

(7) proceed sampling, carried out for (4) (5) (7) step once more and handle, obtain corresponding moment slag making state parameter; So circulation is up to finishing blowing.

2, method for investigating noise character of converter steelmaking blowing slag-making according to claim 1 is characterized in that, sample frequency is greater than 4KHz in the described step (1), and sampling precision is more than or equal to 16.

3, method for investigating noise character of converter steelmaking blowing slag-making according to claim 1 is characterized in that, in the described step (2), the distribution range of feature band adopts following method to determine:

Each power spectrum constantly of A, the whole converting process of calculating

B, the power spectrum in the whole converting process is added up at time orientation, obtain P (f)

$P\left(f\right)=\underset{t}{\mathrm{\Σ}}{\hat{P}}_{\mathrm{PER}}(f,t);$

The halfwidth of C, each peak value of calculating P (f) can obtain the position of corresponding each extraordinary frequency band.

4, according to claim 1 or 3 described method for investigating noise character of converter steelmaking blowing slag-making, it is characterized in that described step

(2) system initialization is that power spectrum to whole converting process adds up in time in, according to the centre frequency of each power peak as each feature band, determine the width of feature band by the halfwidth of each peak value, and determine pearl feature band, feature band and opposite feature frequency band in the same way, concrete grammar is as follows:

A, in a converting process, calculate each power spectrum constantly;

B, each power spectrum is constantly added up in time, obtain level and smooth power spectrum curve;

C, previous step obtained the extreme point frequency at each power peak of curve as the center frequency point of feature band;

D, by the frequency range of the halfwidth correspondence at each power peak distribution range as each feature band;

Maximal value characteristic of correspondence frequency band is as main feature band in the curve that E, selection step (B) obtain;

F, draw the power level of each feature band;

G, the power level curve is close with main feature band trend in slagging process feature band are feature band in the same way, and all the other are the opposite feature frequency band;

The working of a furnace can gradually change in the steelmaking converter use, and feature band can be moved, and therefore all recomputates the distribution of each feature band according to said process in each converting process.

5, method for investigating noise character of converter steelmaking blowing slag-making according to claim 1, it is characterized in that, power spectrum is calculated as in the described step (4), adopt the Blackman window function to combine with the Welch algorithm, form the Welch algorithm of Blackman function correction, its specific algorithm is as follows:

A, at first be to be hits that blowing noise samples value x (t) sequence of N is divided into the L section, every section has M sampled value, and Noverlap overlap sampling value arranged between the adjacent segment, has

$L=\mathrm{fix}\left(\frac{N-\mathrm{Noverlap}}{M-\mathrm{Noverlap}}\right)$

Wherein the effect of fix function is to round to zero direction, and t is the sequence number that sampled value was arranged according to the time.

I section sample sequence is

x ⁱ(t)＝x(t+iM-M)， 1≤t≤M 1≤i≤L

B, usefulness Blackman window function correction Welch algorithm computation power spectrum

The Blackman window function is

$w\left(t\right)=0.42-0.5\cos \left(\frac{2\mathrm{\π}}{M}t\right)+0.08\cos \left(\frac{2\mathrm{\π}}{M}2t\right)$

Obtain the power spectrum of each section

${\hat{P}}_{\mathrm{PER}}^i\left(f\right)=\frac{1}{\mathrm{MU}}|\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{x}^i\left(t\right)w\left(t\right)\exp (-j2\mathrm{\πft}/M){|}^2$

Wherein $U=\frac{1}{M}\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{w}^2\left(t\right)$ Be normalized factor;

Then the power spectrum in this moment is adding up of every section power spectrum

${\hat{P}}_{\mathrm{PER}}\left(f\right)=\frac{1}{\mathrm{MUL}}\underset{m=1}{\overset{L}{\mathrm{\Σ}}}|\underset{t=1}{\overset{M}{\mathrm{\Σ}}}{x}^i\left(t\right)w\left(t\right)\exp (-j2\mathrm{\πft}/M){|}^2,$

Wherein the evolving trend feature band similar to main feature channel is feature band in the same way, is the opposite feature frequency band with main feature band is opposite trend.

Images