Background technology:
The quality of continuous casting steel billet inside plays an important role to the quality of iron and steel final products (for example hot rolled plate, cold-reduced sheet etc.).Because the relation of equipment, technology and new steel grade, bad change may take place in the internal soundness of strand, if untimely discovery then may produce a large amount of defectives, thereby has a strong impact on product quality.Therefore the quality information of continuous casting steel billet inside is monitored and just seemed extremely important.
The common method of checking the strand internal soundness is to utilize the quick sulfur print photo.The procurement process of this photo is as follows: from next piece sample of strand cutting of online production, the silver bromide glossy paper identical with sample size put into the dilute sulfuric acid immersion pick up after 5 minutes earlier, then printing paper is covered on the sample.Chemical reaction takes place in sulfuric acid on the printing paper and the sulfide on the sample (FeS, MnS), forms black or brown spot.Then, promptly obtain the sufur printing photo with printing paper flushing and photographic fixing.The advantage of sufur printing photo is that reliability height, intuitive are good.
The mass defect of slab inside mainly is center segregation and underbead crack, they result from the secondary cooling zone process of setting of strand, chemical constitution is mainly element sulphur and P elements, therefore shape and area are then different because of the continuous casting process condition, the quality by can the quantitative test strand to the grading of center segregation and underbead crack.Each tame steel mill all formulates certain grading standard so that operating personnel grade to slab quality according to classification standard for this reason.But because each one has deviation to the understanding of standard, often exist subjective randomly so declare grade result, this is unfavorable for the quality of objective evaluation strand.
Japanese patent application " evaluating apparatus of casting sheet part quality " (referring to the flat 2-24542 of day disclosure special permission communique (A)) has disclosed a kind of device of the sufur printing image being declared automatically level, this device, utilizes certain algorithm identified to go out V-type segregation, center segregation and the underbead crack in the photo then and carries out the quality grading sufur printing photo input computing machine by digital camera.
Below the concise and to the point method of describing its identification and grading center segregation and underbead crack.This method at first will be made binary conversion treatment to the zone that may have defective, at first sets the higher regional A of the center segregation or the underbead crack frequency of occurrences for this reason and calculate the mean concentration of this zone interior pixel on the sufur printing photo; Set the standard deviation of lower area B of the center segregation or the underbead crack frequency of occurrences and zoning interior pixel CONCENTRATION DISTRIBUTION then; Thereby follow according to mean concentration and standard deviation setting threshold concentration and the pixel in the regional A is made binary conversion treatment and obtain the deep or light binary image that is certain variation.
In order to identify center segregation, the black region that at first will isolate distribution extracts, and this zone is corresponding possible center segregation is hereinafter referred to as particle.Calculate area, diameter and the mean diameter of each particle then.The less particle of area is that noise causes, therefore can utilize based on the threshold value of mean diameter they are removed, and remaining particle is regarded as center segregation.Calculate the area sum of these particles then, and then itself and regional A area be divided by draw the area occupation ratio of center segregation.Can do to declare level analysis to center segregation according to above-mentioned area occupation ratio.
In order to identify underbead crack, the black region that equally also at first will isolate distribution extracts, and black region that will be more close carries out connection processing then, removes other black regions, and is corresponding with underbead crack through this zone of connection processing.Then calculate each underbead crack along the length of X, Y direction and calculate the slope value Y/X of underbead crack.Calculate the area and the summation of underbead crack subsequently, and then area sum and regional A area be divided by draw the area occupation ratio of underbead crack.Can do to declare level analysis to underbead crack according to slope value Y/X that calculates and area occupation ratio etc.
It is worthy of note that the setting to regional A and B in said method need manually be finished, be doped with artificial factor so draw the area size of getting the zone, this introduces some subjective randomness for inevitably the evaluation of back strand internal soundness.In addition, when determining the connection processing of underbead crack, if the resolution of strand image is not enough, then also need the pixel in the black region is carried out interpolation, the workload that this has increased Flame Image Process has greatly reduced and has declared a grade processing speed.
Embodiment:
Referring to Fig. 1, it shows the process flow diagram of declaring grade process according to the casting ingot cross section quality of the embodiment of the invention automatically.In this embodiment, the sufur printing photo size that takes off from the strand sample is 30 millimeters * 1000 millimeters, and it is 256 grades image file that the resolution scan that scanner can 200DPI generates gray shade scale.Image file can the input image data storehouse, as the data source in sufur printing picture data storehouse, also can export other equipment to through the external module interface.
Because the noise of scanner is very big to the influence of picture quality, so need carry out pre-service to image by computing machine before the defective in recognition image, these noises of filtering.The noise of scanner can be divided into two kinds, a kind of sensor that comes from scanner, and when scanner carried out the multirow parallel sweep, because the disturbance of transducer sensitivity, the gray scale between the adjacent lines was inhomogeneous, disturbed thereby form striped; Another kind of thermonoise from electron device, it on image, show as mixed and disorderly distribution stain.These noises have caused the distortion of casting blank section image, cause defectives such as center segregation, implosion identification, declare staging error, therefore must effectively remove.For first kind of noise, has approximate gray-scale value in the horizontal direction and than the shallow characteristics of sufur printing data gray scale, the design threshold wave filter gives filtering according to noise.For second kind of noise,, can adopt the level and smooth method of spatial domain to remove according to the characteristics of its stochastic distribution.
Pretreated image will be used to estimate the quality of casting blank section, and this mainly is divided into following two main processes: the identification of defectives such as (1) center segregation, underbead crack; (2) utilize the grading standard that defectives such as center segregation, underbead crack are declared level analysis, thereby draw the evaluation result of casting ingot cross section quality.In addition,, improve and estimate accuracy, evaluation result can be imported knowledge base for casting ingot cross section quality automated decision system study reference in order to improve the standard of quality assessment.
Below by Fig. 2 and Fig. 3 the identifying of center segregation and underbead crack defective is described respectively.As shown in Figure 2, at first adopt clustering algorithm to extract pattern primitive.The inventor finds that by the observation to the sufur printing photo center segregation has identical distribution characteristics with underbead crack on image, and promptly they all are made up of dense distribution dark pixels together.Therefore can with dense distribution together and gray-scale value be included in the identity set greater than the pixel of certain threshold value, such set is called pattern primitive.By top selection to pattern primitive as seen, the extraction of pattern primitive can be summed up as a clustering problem in fact.For clustering problem, multiple algorithm can be arranged, but because the sufur printing image data amount is very huge (in the present embodiment, if image file is the BMP form, then file size will reach 30 megabyte), so if the algorithm complexity, then with consuming time very big, according to experiment, different clustering algorithms may differ several times to tens of times to the execution time of same problem, therefore must seek the quick clustering algorithm at this problem.Through experiment, the inventor has found a kind of preferable clustering algorithm, below it is described specifically.
(a) be threshold value D at first, extract all pixel X greater than threshold value D with a certain gray-scale value
1, X
2... X
n(b) selected pixel X
1Be class C
1(c) with X
1For finding out all and X in the center
1Distance less than the pixel Z of threshold value T1
1, Z
2Z
n, and they are included into class C
1, T1 is a positive number here; (d) successively with pixel Z
1, Z
2Z
nBe the center, repeat above-mentioned steps (c); (e) as fruit C
1In pixel count less than threshold value T2, then delete class C
1, as fruit C
1In pixel count more than or equal to threshold value T2 but less than 2 * T2, reserved category C then
1Constant, as fruit C
1In pixel count more than or equal to 2 * T2, then with class C
1In the Region Segmentation that occupies of pixel be the m sub regions, the pixel count in each subregion and is included into subclasses C respectively with the pixel in the subregion between T2 and 2 * T2
11, C
12... C
1m, T2 here and m are not less than 2 integer; And (f) to pixel X
2, X
3... X
nRepeat above-mentioned steps (b)~(e), obtain class C
2, C
3..., C
nAnd subclass separately, above-mentioned class and subclass are pattern primitive.
After extracting pattern primitive, can discern center segregation and underbead crack defective.For the center segregation defective, can take following recognizer: at first according to its characteristics of linear distribution in vertical direction, the zone of " moving window " method of utilization searching modes primitive quantity maximum, this zone is that the center segregation probability of occurrence is the highest, its defective also is the most serious, and therefore in fact it is the main analyzed area of quality assessment.So-called " moving window " method refers to the method for when along continuous straight runs moves the long and narrow zone that a width is fixed and height equates with photo continuously pattern primitive quantity being added up, and the width in long and narrow zone should be determined by experiment according to actual conditions.Next is to utilize following formula to calculate each pattern primitive H in this zone
iCenter of gravity:
X
i=(∑x
j)/n
i,Y
i=(∑y
j)/n
i,
X wherein
iWith Y
iBe the horizontal ordinate and the ordinate of a certain pattern primitive center of gravity, x
jWith y
jBe the horizontal ordinate and the ordinate of a certain pixel in the pattern primitive, n
iBe pattern primitive H
iThe quantity of interior pixel.Then to all pattern primitive H in the zone
iCenter of gravity (X
i, Y
i) make linear fit, obtain fitting a straight line y=ax+b, and calculate each pattern primitive center of gravity (X
i, Y
i) with straight line y apart from S
iAt last, if S
i<Sa then is identified as center segregation with this pattern primitive, and here Sa is the constant greater than zero.
Below by Fig. 3 the identifying of underbead crack is described.At first adopt clustering algorithm to extract pattern primitive, calculate the center of gravity of each mode computation pattern primitive then, then the center of gravity of pattern primitive is made linear fit and calculate distance between each pattern primitive center of gravity and the fitting a straight line.Therefore just the same in above-mentioned these steps and the center segregation identifying repeat no more herein.Through research, the inventor finds that underbead crack has the advantages that to be distributed in the center segregation both sides and to be the horizontal crackle shape, and the angle (in fact it be equivalent to the slope of pattern primitive) that therefore can adopt pattern primitive distribution arrangement and horizontal direction is as one of criterion of discerning underbead crack.The step of computation schema primitive slope comprises: the difference L that calculates interior maximum horizontal ordinate of each pattern primitive and minimum horizontal ordinate
iDifference W with maximum ordinate and minimum ordinate
iThereby, obtain slope W
i/ L
IIf Sa≤S
i<Sb, and W
i/ L
i<α then is included into the underbead crack defective with this pattern primitive, and Sb here and α are all the constant greater than 0.It is worthy of note, because Sa≤S
i<Sb is the necessary condition that pattern primitive is included into underbead crack, so can only calculate Sa≤S
iThe slope of the pattern primitive of<Sb so just can be saved the workload that underbead crack is discerned.
" People's Republic of China's iron and steel industry industry standard (YB/T 4003-1997) " provided one group of sufur printing photo to the level of declaring of center segregation and underbead crack, and it is relevant with their features such as length, size and gray scale to indicate the rank of center segregation and underbead crack.But this only is to declaring fuzzy a, explanation qualitatively of level, do not provide the rule of quantification, can't declaring directly employing in the level automatically.By analysis, statistics to the sufur printing photo, the inventor finds, shape, continuity, length, area and gray scale etc. have stronger correlativity with the rank of center segregation, can be used as and characterize other characteristic quantity of center segregation level, and the rank of area and underbead crack has stronger correlativity, can be used as to characterize other characteristic quantity of underbead crack level.But many characteristic quantities might cause defective to declare the inconsistent of level.For example, its deciding grade and level can be the 1.5C level, but, can cause the difficulty of defining the level thus with its deciding grade and level for the 1.0C level again according to features such as its continuity, length, areas according to shape and gray feature for the center segregation on certain the sufur printing photo.Solution to this problem is to obtain the corresponding relation of graded features amount and Level by self-learning method, writes in the knowledge base as knowledge and uses when declaring grade.And the result who at every turn correctly declares level can be used as the foundation of self study next time again.
Below by Fig. 4 self-learning method based on the characteristic quantity weights is described.As shown in Figure 4, with several correctly declared the level sufur printing photos as learning object, utilizing above-described method that they are made image handles, treatment step comprises photo scanning formed to have the digitized image of certain gray scale and carry out the pre-service filtering noise, extracts the primitive pattern in the image and the center segregation defective discerned.Shape, continuity, length, area and the gray scale etc. of extracting center segregation in each photo subsequently are as characteristic quantity, and statistics is determined each characteristic quantity and other degree of correlation of center segregation level in all are learnt the photo scope.Then the degree of correlation that obtains according to statistics gives each characteristic quantity different weights, it is evident that degree of correlation is big more, and the weights of then giving are also big more.Determine the relation of center segregation rank and characteristic quantity weights then.At last, the relation of center segregation rank and characteristic quantity weighted sum and the corresponding weights of characteristic quantity are stored in the knowledge base when declaring grade for center segregation use, thereby finish whole self study process based on the characteristic quantity weights.
Below by Fig. 5 self-learning method based on classification thresholds is described.As shown in Figure 5, with several correctly declared the level sufur printing photos as learning object, utilize and they are done the image processing based on the identical method of characteristic quantity weights self study process, it is to have the digitized image of certain gray scale and carry out pre-service that treatment step comprises photo scanning, extracts the primitive pattern in the image and defectives such as center segregation and underbead crack are discerned.For center segregation, then with the same based on characteristic quantity weights self study process, extract its shape, continuity, length, area and gray scale etc. as characteristic quantity, and for underbead crack, the pixel count sum of then extracting all underbead crack pattern primitives in every photo is as characteristic quantity.Subsequently, learnt to add up in the photo scope relation of determining center segregation or underbead crack rank and individual features amount at all, obtain the classification thresholds of each characteristic quantity, and classification thresholds is stored in the knowledge base when declaring grade for defective uses, thereby finish whole self study process based on classification thresholds.
Below by Fig. 6 grade process of declaring of center segregation is described.As shown in Figure 6, identify one in the sufur printing photo center segregation and extract characteristic quantities such as its shape, continuity, length, area and gray scale after, they and the knowledge that obtains by self study can be compared.Because the hierarchical knowledge of center segregation can obtain by two kinds of self study approach, therefore two kinds of comparative approach are arranged also correspondingly.First method utilization each characteristic quantity weights by obtaining based on characteristic quantity weights self study process, to declaring each characteristic quantity weighted sum of center segregation in grade photo, and determine the rank of this photo center segregation according to the relation of center segregation rank in the knowledge base and characteristic quantity weighted sum.Second method is earlier according to the rank of determining to declare each characteristic quantity of center segregation in grade photo by the classification thresholds of each characteristic quantity of obtaining based on classification thresholds self study process, then that occurrence number is maximum ranks is defined as the rank of center segregation in this sufur printing photo, this method is called " nose count method " again, each characteristic quantity is represented one " ticket ", and the rank that " gained vote " is the highest promptly is considered as declaring a grade result.
Below by Fig. 7 grade process of declaring of underbead crack is described.As shown in Figure 7, identify one in the sufur printing photo the underbead crack pattern primitive and the pixel count of this quasi-mode primitive added up after, can with it with compare by the knowledge that obtains based on the classification thresholds self study.Owing to have only characteristic quantity of pixel count accumulated value, only need according to the rank that can determine to declare underbead crack in grade photo by the single classification thresholds that obtains based on classification thresholds self study process.
As shown in Figure 1, if it is correct to declare grade result, then this can be declared a grade photo and include the self study scope in, thereby utilize more sufur printing photo to carry out above-described self study process, improve and declare a grade knowledge accuracy, rationality and applicability.
It is worthy of note, it will be apparent to one skilled in the art that, above by reading by the embodiment description of this invention, can make various modifications or change to the present invention at an easy rate under the prerequisite that does not depart from the present invention's spirit and essence, therefore spirit of the present invention and essence are limited by following claims.