CN107516313A - Forging surface defect based on integrated study and Density Clustering is in position detecting method - Google Patents
Forging surface defect based on integrated study and Density Clustering is in position detecting method Download PDFInfo
- Publication number
- CN107516313A CN107516313A CN201710707884.8A CN201710707884A CN107516313A CN 107516313 A CN107516313 A CN 107516313A CN 201710707884 A CN201710707884 A CN 201710707884A CN 107516313 A CN107516313 A CN 107516313A
- Authority
- CN
- China
- Prior art keywords
- forging
- point cloud
- data set
- dimensional point
- finished product
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/001—Industrial image inspection using an image reference approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30116—Casting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Forging (AREA)
Abstract
The invention discloses the forging surface defect based on integrated study and Density Clustering in position detecting method, the three dimensional point cloud group of zero defect standard high temperature forging is obtained by setting Cahn-Ingold-Prelog sequence rule by three-dimensional laser scanner, and to two principal curvatures of each cloud computing, obtain standard forging five and tie up cloud data group；Contrast standard forging data set number n1, n2, n3, n4, n5 and finished product forging data set number N1, N2, N3, N4, N5 respectively, if there are at least 2 groups in n1=N1, n2=N2, n3=N3, n4=N4, n5=N5 while meet, then finished product forging is free of surface defects, and otherwise the finished product forging has surface defect.Using distance of the Euclidean distance of Weight as two five dimension point clouds, each five dimensions point cloud in each data set is divided based on density threshold d, integrate judged result the defects of under five density thresholds d1, d2, d3, d4, d5, realize and judge that finished product forging whether there is defect according to the integrated results of integrated study, realize the quick detection in place of high-temperature forging.
Description
Technical Field
The invention relates to the field of high-temperature forging detection, in particular to a forging surface defect in-situ detection method based on ensemble learning and density clustering.
Background
In industrial production, a free forging process is a relatively common production mode, and the production of forgings is also a very important position. The manufacturing process of free forging is implemented under extreme conditions of strong shock, high temperature, high pressure and the like, a large amount of manpower and material resources are required to be invested before the production of the forge piece, the manufacturing process is continuous and complex, the consumption of materials and energy is huge, and the manufacturing cost is high. In the forging process, the size of the large forging is an important index which needs to be detected in time in the production of the forging.
At present, the high-temperature forging piece is flawless through manual visual inspection, the detection accuracy is low, the size of the forging piece is generally larger than a specified value by a large margin, and the average loss amount is up to 15%; the characteristic dimension information of the key parts of the forge piece is not fed back to an upper computer system in time, so that the internal quality of the forge piece is not improved in time, the machining precision is improved, and the production efficiency is reduced. In addition, the die is easy to damage due to multiple times of extrusion under high temperature and high pressure in the free forging process, and the forged piece obtained by free forging with the damaged die is a waste piece.
Disclosure of Invention
In view of the background of the existing social demands and the current technical situation, the invention aims to provide an in-situ detection method for the surface defects of the forgings based on the integrated learning and density clustering, and the identification and the alarm of the defects of the forgings at the high temperature of more than 1000 ℃ are realized.
In order to achieve the purpose, the invention adopts the following technical scheme:
a forge piece surface defect in-situ detection method based on ensemble learning and density clustering comprises the following steps:
acquiring a large amount of three-dimensional point cloud data of a standard forge piece by a blue light three-dimensional laser scanner at a position 10 meters away from a standard high-temperature forge piece without defects according to a set sequence rule;
step two, uniformly collecting 3000 point clouds from a large amount of three-dimensional point cloud data of the standard forge piece obtained in the step one to form a standard forge piece three-dimensional point cloud data set, calculating two main curvatures corresponding to each standard forge piece point cloud in the standard forge piece three-dimensional point cloud data set according to a fitting method of an ellipsoid curved surface, and then selecting a three-dimensional coordinate value and the two main curvatures of each standard forge piece point cloud as five characteristics representing the standard forge piece point cloud, thereby obtaining a standard forge piece five-dimensional point cloud data set D;
step three, dividing the standard forging five-dimensional point cloud data group D into corresponding standard forging data set numbers n1, n2, n3, n4 and n5 according to density thresholds D1, D2, D3, D4 and D5 respectively;
and (3) detecting a finished product forging:
fourthly, obtaining a large amount of three-dimensional point cloud data of the finished product forge piece through a blue light three-dimensional laser scanner at a position 10 meters away from the finished product forge piece to be detected according to a set sequence rule;
step five, 3000 point clouds are uniformly collected from a large number of three-dimensional point cloud data of the finished product forged piece obtained in the step four to form a finished product forged piece three-dimensional point cloud data set, two main curvatures corresponding to each finished product forged piece point cloud in the finished product forged piece three-dimensional point cloud data set are calculated according to a fitting method of an ellipsoid curved surface, then a three-dimensional coordinate value and the two main curvatures of each finished product forged piece point cloud are selected as five characteristics representing the finished product forged piece point cloud, and a finished product forged piece five-dimensional point cloud data set E is obtained;
step six, dividing the finished product forging five-dimensional point cloud data group E into corresponding finished product forging data set numbers N1, N2, N3, N4 and N5 according to the density threshold values d1, d2, d3, d4 and d5 respectively;
seventhly, comparing the number N1, N2, N3, N4 and N5 of the standard forging data sets with the number N1, N2, N3, N4 and N5 of the finished product forging data sets respectively, wherein if at least 2 groups of N1= N1, N2= N2, N3= N3, N4= N4 and N5= N5 are simultaneously satisfied, the finished product forging has no surface defect, otherwise, the finished product forging has surface defect.
Preferably, the method further comprises the following monitoring alarm process:
and step eight, performing the detection process of the finished product forge pieces on each finished product forge piece on the production line, and giving an alarm and prompting that the forging and pressing die on the production line is damaged when the surface defects of three continuous finished product forge pieces are judged.
Preferably, the standard forging five-dimensional point cloud data set D is divided into corresponding standard forging data set numbers n1, n2, n3, n4 and n5 according to density thresholds D1, D2, D3, D4 and D5, and includes:
creating a standard forging point cloud distance two-dimensional table process:
calculating the distance between two points of all five-dimensional point clouds in the standard forging five-dimensional point cloud data set D according to the Euclidean distance with the weight, and making a corresponding standard forging point cloud distance two-dimensional table;
the integrated learning subprocess about the standard forging:
s11: randomly selecting a standard forging five-dimensional point cloud from a standard forging five-dimensional point cloud data set D, inquiring a standard forging point cloud distance two-dimensional table, finding out a data set C consisting of other standard forging five-dimensional point clouds of which the distances from the standard forging point cloud data set D to the standard forging point cloud are smaller than a density threshold value D1, and counting the number N of the standard forging point cloud in the data set C; if N is less than 5, marking the standard forging five-dimensional point cloud as 0, emptying the data set C, randomly selecting another standard forging five-dimensional point cloud again from the standard forging five-dimensional point cloud data set D, and executing the step S11 again; if N is more than or equal to 5, marking the standard forging five-dimensional point cloud as 1, creating a data set C1, uniformly distributing all the standard forging five-dimensional point clouds in the data set C into the data set C1, emptying the data set C, and executing the step S12;
s12: randomly selecting an unmarked standard forging five-dimensional point cloud from the data set C1 for accessing, inquiring the standard forging point cloud distance two-dimensional table, finding out all other standard forging five-dimensional point clouds of which the distance from the accessed standard forging five-dimensional point cloud is smaller than a density threshold value d1, putting the other standard forging five-dimensional point clouds into the data set C, and counting the number N of the standard forging five-dimensional point clouds in the data set C; if N is less than 5, marking the five-dimensional point cloud of the accessed standard forging as 0, emptying the data set C, detecting whether the five-dimensional point cloud of the unmarked standard forging exists in the data set C1, if so, executing the step S12 again, otherwise, executing the step S13; if N is more than or equal to 5, marking the standard forging five-dimensional point cloud as 1, uniformly distributing all the standard forging five-dimensional point clouds in the data set C into the data set C1, emptying the data set C, and executing the step S12 again;
s13: removing the standard forging five-dimensional point cloud of the data set c1 from the standard forging five-dimensional point cloud data set D to obtain a data set D1, randomly selecting one standard forging five-dimensional point cloud from the data set D1, and obtaining a data set c2 according to the operation of the steps S11 and S12 in the same way;
s14: removing the standard forging five-dimensional point clouds in the data set D1 to obtain a data set D2, obtaining a data set c3, a data set c4, a data set c5 and the like in the same way according to the operations of the steps S11, S12 and S13 until all the standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data group D are classified into corresponding data sets such as the data set c1, the data set c2, the data set c3 and the like, and obtaining the number of the data sets such as the data set c1, the data set c2, the data set c3 and the like at the moment, wherein n1 is the number of the data sets;
s15: from the operations of steps S11 to S14, the number of data sets n2, n3, n4, and n5 corresponding to the set density threshold values d2, d3, d4, and d5, respectively, is obtained in the same manner.
Preferably, the five-dimensional point cloud data set of the finished product forging is divided into corresponding finished product forging data set numbers N1, N2, N3, N4 and N5 according to the set density thresholds d1, d2, d3, d4 and d5, respectively, and includes:
establishing a point cloud distance two-dimensional table process of a finished product forging:
calculating the distance between two points of all the finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data set E according to the weighted Euclidean distance, and manufacturing a corresponding finished product forging point cloud distance two-dimensional table;
an integrated learning subprocess about finished forgings:
s21: randomly selecting a finished product forging five-dimensional point cloud from a finished product forging five-dimensional point cloud data set E, inquiring a finished product forging point cloud distance two-dimensional table, finding out a data set C consisting of all other finished product forging five-dimensional point clouds of which the distances from the finished product forging five-dimensional point clouds are smaller than a density threshold d1, and counting the number N of the finished product forging five-dimensional point clouds in the data set C; if N <5, marking the five-dimensional point cloud of the finished product forged piece as 0, emptying the data set C, randomly selecting another five-dimensional point cloud of the finished product forged piece from the data set E of the five-dimensional point cloud of the finished product forged piece, and executing the step S21 again; if N is more than or equal to 5, marking the five-dimensional point cloud of the finished product forged piece as 1, creating a data set C1, uniformly distributing all the five-dimensional point clouds of the finished product forged piece in the data set C into the data set C1, emptying the data set C, and executing the step S22;
s22: randomly selecting an unmarked finished product forging five-dimensional point cloud from the data set C1 for accessing, inquiring the distance two-dimensional table of the finished product forging point cloud, finding out all other finished product forging five-dimensional point clouds of which the distance from the accessed finished product forging five-dimensional point cloud is less than a density threshold value d1, putting the other finished product forging five-dimensional point clouds into the data set C, and counting the number N of the finished product forging five-dimensional point clouds in the data set C; if N is less than 5, marking the five-dimensional point cloud of the accessed finished product forged piece as 0, emptying the data set C, detecting whether the data set C1 contains the five-dimensional point cloud of the unmarked finished product forged piece, if so, executing the step S22 again, otherwise, executing the step S23; if N is more than or equal to 5, marking the five-dimensional point cloud of the finished product forged piece as 1, dividing all the five-dimensional point clouds of the finished product forged piece in the data set C into the data set C1, emptying the data set C, and executing the step S22 again;
s23: removing the finished product forged piece five-dimensional point cloud of the data set C1 from the finished product forged piece five-dimensional point cloud data set E to obtain a data set E1, randomly selecting one finished product forged piece five-dimensional point cloud from the data set E1, and obtaining a data set C2 according to the operation of the steps S21 and S22 in the same way;
s24: removing the finished product forging five-dimensional point cloud of the data set C2 from the data set E1 to obtain a data set E2, similarly obtaining a data set C3, a data set C4, a data set C5 and the like according to the operations of the steps S21, S22 and S23 until all finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data group E are classified into corresponding data sets such as the data set C1, the data set C2, the data set C3 and the like, and obtaining the number of the data sets such as the data set C1, the data set C2, the data set C3 and the like at the moment, wherein N1 is the number of the data sets;
s25: from the operations of steps S21 to S24, the number of data sets N2, N3, N4, and N5 corresponding to the set density thresholds d2, d3, d4, and d5, respectively, is obtained in the same manner.
Preferably, the standard forging five-dimensional point cloud defining method comprises the following steps:
defining the first characteristic of each standard forging point cloud as an x-axis coordinate value, the second characteristic as a y-axis coordinate value, the third characteristic as a z-axis coordinate value, and the fourth characteristic as the maximum curvature r of the two main curvatures 1 The fifth characteristic is the minimum curvature r of the two principal curvatures 2 。
Preferably, the method for creating the standard forging point cloud distance two-dimensional table comprises the following steps:
s31: all the standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data set D are compressed to be between-1 and 1 in a linear normalization mode, the absolute value of the maximum curvature r1 of all the standard forging five-dimensional point clouds is calculated, and the maximum value r1 is found out max Minimum value r1 min And calculates the mean value r1 m (ii) a Calculating the absolute value of the minimum curvature r2 of the five-dimensional point cloud of all standard forgings to find out the maximum value r2 max Minimum value r2 min And calculates the mean value r2 m (ii) a Calculating the absolute value mean xm of the x-axis coordinate value, the absolute value mean ym of the y-axis coordinate value and the absolute value mean zm of the z-axis coordinate value of the standard forging five-dimensional point cloud data set D, and finding the maximum value m of the mean xm, ym and zm max ；
S32: defining a standard forging five-dimensional point cloud (X) 1 ,Y 1 ,Z 1 ,r a1 ,r b1 ) And another standard forging five-dimensional point cloud (X) 2 ,Y 2 ,Z 2 ,r a2 ,r b2 ) Has a weight Euclidean distance L of
Wherein the content of the first and second substances,
and sequencing all the standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data set D from large to small according to the absolute value of the maximum curvature r1, calculating the weighted Euclidean distance L between every two standard forging five-dimensional point clouds, and manufacturing a corresponding standard forging point cloud distance two-dimensional table, wherein the distance numerical value of the ith row and the jth column in the standard forging point cloud distance two-dimensional table is the weighted Euclidean distance L of the ith standard forging five-dimensional point cloud and the jth standard forging five-dimensional point cloud after all the standard forging five-dimensional point clouds are sequenced according to the absolute value of the maximum curvature r 1.
Preferably, the five-dimensional point cloud defining method for the finished product forging comprises the following steps:
defining the first characteristic of each finished product forging point cloud as an x-axis coordinate value, the second characteristic as a y-axis coordinate value, the third characteristic as a z-axis coordinate value, and the fourth characteristic as the maximum curvature r of the two main curvatures 1 The fifth characteristic is the minimum curvature r of the two principal curvatures 2 。
Preferably, the method for creating the finished product forging point cloud distance two-dimensional table comprises the following steps:
s41: all finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data set E are subjected to linear normalization compression to be between-1 and 1, the absolute value of the maximum curvature r1 of all finished product forging five-dimensional point clouds is calculated, and the maximum value r1 'is found out' max Minimum value r1' min And calculating a mean value r1' m (ii) a Calculating absolute values of minimum curvatures r2 of all finished product forging five-dimensional point clouds to find out maximum values r2' max Minimum value r2' min And calculating a mean value r2' m (ii) a Calculating the absolute value mean value xm ' of the x-axis coordinate value, the absolute value mean value ym ' of the y-axis coordinate value and the absolute value mean value zm ' of the z-axis coordinate value of the finished product forging five-dimensional point cloud data group E, and finding the maximum value m ' of the mean values xm ', ym ' and zm ' max ；
S42: defining five-dimensional point cloud (X ') of finished product forging' 1 ,Y' 1 ,Z' 1 ,r' a1 ,r' b1 ) And another finished forged piece five-dimensional point cloud (X' 2 ,Y' 2 ,Z' 2 ,r' a2 ,r' b2 ) Has a weight of Euclidean distance L' of
Wherein the content of the first and second substances,
and sequencing all the finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data set E from large to small according to the absolute value of the maximum curvature r1, calculating the weighted Euclidean distance L 'between every two finished product forging five-dimensional point clouds, and manufacturing a corresponding finished product forging point cloud distance two-dimensional table, wherein the distance numerical value of the ith row and jth column in the finished product forging point cloud distance two-dimensional table is the weighted Euclidean distance L' of the ith finished product forging five-dimensional point cloud and the jth finished product forging five-dimensional point cloud after all the finished product forging five-dimensional point clouds are sequenced according to the absolute value of the maximum curvature r 1.
Preferably, the density threshold values d1, d2, d3, d4 and d5 are set by:
firstly, searching the Euclidean distance L with the maximum weight from the point cloud distance two-dimensional table of the standard forge piece max And a minimum weighted Euclidean distance L min ；
Then, the density threshold d1= L is set min +0.01(L max -L min )，
Density threshold d2= L min +0.02(L max -L min )，
Density threshold d3= L min +0.03(L max -L min )，
Density threshold d4= L min +0.04(L max -L min )，
Density threshold d5= L min +0.05(L max -L min )。
The forge piece surface defect in-place detection method based on the integrated learning and density clustering mainly obtains three-dimensional point cloud data of a finished forge piece to be detected and calculates two principal curvatures of each point cloud to obtain a finished forge piece five-dimensional point cloud data set, linear normalization processing is carried out on the data, euclidean distance with weight is used as the distance between the two point clouds, each point cloud is divided based on a density threshold, defect judgment results under five density thresholds d1, d2, d3, d4 and d5 are integrated according to the integrated learning idea, judgment of the defects of the high-temperature finished forge piece to be detected is achieved, and detection precision is improved.
Drawings
The drawings are further illustrative of the invention and the content of the drawings does not constitute any limitation of the invention.
FIG. 1 is a flowchart of a high-temperature forging surface defect non-contact in-situ detection and alarm method based on ensemble learning and density clustering according to one embodiment of the invention.
Detailed Description
The technical scheme of the invention is further explained by the specific implementation mode in combination with the attached drawings.
In this embodiment, as shown in fig. 1, the method for in-situ detecting surface defects of a forged piece based on ensemble learning and density clustering includes:
the method comprises the following steps of firstly, obtaining a large amount of three-dimensional point cloud data of a standard forging by a blue light three-dimensional laser scanner at a position 10 meters away from a defect-free standard high-temperature forging according to a set sequence rule;
step two, uniformly collecting 3000 point clouds from a large amount of three-dimensional point cloud data of the standard forging obtained in the step one to form a standard forging three-dimensional point cloud data set, calculating two main curvatures corresponding to each standard forging point cloud in the standard forging three-dimensional point cloud data set according to a fitting method of an ellipsoid curved surface, and then selecting a three-dimensional coordinate value and the two main curvatures of each standard forging point cloud as five characteristics representing the standard forging point cloud so as to obtain a standard forging five-dimensional point cloud data set D;
thirdly, dividing the standard forging five-dimensional point cloud data group D into corresponding standard forging data set numbers n1, n2, n3, n4 and n5 according to density threshold values D1, D2, D3, D4 and D5 respectively;
and (3) detecting a finished product forging:
fourthly, obtaining a large amount of three-dimensional point cloud data of the finished product forge piece through a blue light three-dimensional laser scanner at a position 10 meters away from the finished product forge piece to be detected according to a set sequence rule;
step five, 3000 point clouds are uniformly collected from a large number of three-dimensional point cloud data of the finished product forged piece obtained in the step four to form a finished product forged piece three-dimensional point cloud data set, two main curvatures corresponding to each finished product forged piece point cloud in the finished product forged piece three-dimensional point cloud data set are calculated according to a fitting method of an ellipsoid curved surface, then a three-dimensional coordinate value and the two main curvatures of each finished product forged piece point cloud are selected as five characteristics representing the finished product forged piece point cloud, and a finished product forged piece five-dimensional point cloud data set E is obtained;
step six, the finished product forged piece five-dimensional point cloud data group E is divided into corresponding finished product forged piece data set numbers N1, N2, N3, N4 and N5 according to the density threshold values d1, d2, d3, d4 and d 5;
seventhly, comparing the number N1, N2, N3, N4 and N5 of the standard forging data sets with the number N1, N2, N3, N4 and N5 of the finished product forging data sets respectively, wherein if at least 2 groups of N1= N1, N2= N2, N3= N3, N4= N4 and N5= N5 are simultaneously satisfied, the finished product forging has no surface defect, otherwise, the finished product forging has surface defect.
According to the forge piece surface defect in-place detection method based on ensemble learning and density clustering, three-dimensional point cloud data of a finished forge piece to be detected are obtained, two principal curvatures of each point cloud are calculated to obtain a finished forge piece five-dimensional point cloud data set, euclidean distances with weights are used as distances of the two finished forge piece five-dimensional point clouds through linear normalization of the data, the finished forge piece five-dimensional point clouds are divided based on a density threshold value d, defect judgment results under the five density threshold values d1, d2, d3, d4 and d5 are integrated, whether defects exist in the finished forge piece or not is judged according to the integrated learning integration result, in-place detection of the high-temperature forge piece is achieved, and detection accuracy is improved.
The three-dimensional coordinate value and the two main curvatures of each standard forging point cloud in the standard forging five-dimensional point cloud data set are used as five characteristics representing the point cloud, and the description of each standard forging point cloud comprises more information and details. And dividing the standard forging five-dimensional point cloud data group D into corresponding standard forging data set numbers n1, n2, n3, n4 and n5 according to the set density threshold values D1, D2, D3, D4 and D5, and providing a check standard for judging whether the finished product forging has defects. The point cloud data obtained according to the set sequence rule is as follows: the starting point of the first scanning is the position of the upper left corner of the front view of the forge piece, the starting point of the second scanning is the position of the upper left corner of the back view of the forge piece, and the scanning sequence principle is from top to bottom and from left to right.
In the detection process of the finished product forged piece, the three-dimensional point cloud data set of the finished product forged piece is acquired in a non-contact mode through the three-dimensional laser scanner, and the characteristics of the high-temperature forged piece are acquired in situ. Calculating two main curvatures of each finished product forging point cloud according to the finished product forging three-dimensional point cloud data set, wherein a three-dimensional coordinate value and the two main curvatures corresponding to each finished product forging point cloud are used as five characteristics representing the finished product forging point cloud, and the description of each finished product forging point cloud comprises more information and details; then dividing a finished product forging five-dimensional point cloud data group E into corresponding finished product forging data set numbers N1, N2, N3, N4 and N5 according to the set density threshold values d1, d2, d3, d4 and d 5; if at least 2 groups of N1= N1, N2= N2, N3= N3, N4= N4, N5= N5 satisfy simultaneously, for example, N1= N1 and N2= N2 satisfy simultaneously, the finished forging has no surface defect; and for another example, only N1= N1 is satisfied, the finished forging has surface defects. And comparing the number of the standard forging data sets with the number of the finished product forging data sets one by one to achieve the purpose of judging whether the finished product forging has surface defects or not and improve the detection accuracy.
Preferably, the method further comprises the following monitoring alarm process:
and step eight, carrying out the detection process of the finished product forge pieces on each finished product forge piece on the production line, and giving an alarm and prompting that the forging and pressing die on the production line is damaged when the surface defects of three continuous finished product forge pieces are judged. Because the die is extruded for many times under high temperature and high pressure in the free forging process, the die is easy to damage, and the high-temperature forging piece obtained by using the damaged die for free forging is a waste piece. Therefore, the monitoring and alarming process is set, when the surface defects of three continuous finished product forgings are judged, the alarm is given out, the damage of the forging die is found in time, and the number of scrapped forgings can be effectively reduced.
Preferably, the standard forging five-dimensional point cloud data group D is divided into corresponding standard forging data set numbers n1, n2, n3, n4 and n5 according to density thresholds D1, D2, D3, D4 and D5, respectively, and includes:
creating a standard forging point cloud distance two-dimensional table:
calculating the distance between two points of all five-dimensional point clouds in the standard forging five-dimensional point cloud data set D according to the Euclidean distance with weight, and manufacturing a corresponding standard forging point cloud distance two-dimensional table;
the integrated learning subprocess about the standard forging:
s11: randomly selecting a standard forging five-dimensional point cloud from a standard forging five-dimensional point cloud data set D, inquiring a standard forging point cloud distance two-dimensional table, finding out other standard forging five-dimensional point clouds of which the distances from the standard forging point cloud two-dimensional point clouds are smaller than a density threshold value D1 to form a data set C, and counting the number N of the standard forging point cloud two-dimensional point clouds in the data set C; if N is less than 5, marking the standard forging five-dimensional point cloud as 0, emptying the data set C, randomly selecting another standard forging five-dimensional point cloud again from the standard forging five-dimensional point cloud data set D, and executing the step S11 again; if N is more than or equal to 5, marking the standard forging five-dimensional point cloud as 1, creating a data set C1, uniformly distributing all the standard forging five-dimensional point clouds in the data set C into the data set C1, emptying the data set C, and executing the step S12;
s12: randomly selecting an unmarked standard forging five-dimensional point cloud from the data set C1 for accessing, inquiring the standard forging point cloud distance two-dimensional table, finding out all other standard forging five-dimensional point clouds of which the distance from the accessed standard forging five-dimensional point cloud is less than a density threshold value d1, putting the other standard forging five-dimensional point clouds into the data set C, and counting the number N of the standard forging five-dimensional point clouds in the data set C; if N is less than 5, marking the five-dimensional point cloud of the accessed standard forging as 0, emptying the data set C, detecting whether the data set C1 contains the five-dimensional point cloud of the unmarked standard forging, if yes, executing the step S12 again, otherwise, executing the step S13; if N is more than or equal to 5, marking the standard forging five-dimensional point cloud as 1, uniformly distributing all the standard forging five-dimensional point clouds in the data set C into the data set C1, emptying the data set C, and executing the step S12 again;
s13: removing the standard forging five-dimensional point cloud of the data set c1 from the standard forging five-dimensional point cloud data set D to obtain a data set D1, randomly selecting one standard forging five-dimensional point cloud from the data set D1, and obtaining a data set c2 according to the operation of the steps S11 and S12 in the same way;
s14: removing the standard forging five-dimensional point cloud of the data set c2 from the data set D1 to obtain a data set D2, obtaining a data set c3, a data set c4, a data set c5 and the like in the same way according to the operations of the steps S11, S12 and S13 until all standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data group D are classified into corresponding data sets such as the data set c1, the data set c2, the data set c3 and the like, and obtaining the number of the data sets such as the data set c1, the data set c2, the data set c3 and the like at the moment, wherein the number of the data sets is n1;
s15: from the operations of steps S11 to S14, the number of data sets n2, n3, n4, and n5 corresponding to the set density threshold values d2, d3, d4, and d5, respectively, is obtained in the same manner.
Preferably, the five-dimensional point cloud data set of the finished product forging is divided into corresponding finished product forging data set numbers N1, N2, N3, N4 and N5 according to the set density thresholds d1, d2, d3, d4 and d5, respectively, and includes:
establishing a point cloud distance two-dimensional table process of a finished product forged piece:
calculating the distance between two points of all the finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data set E according to the weighted Euclidean distance, and manufacturing a corresponding finished product forging point cloud distance two-dimensional table;
the integrated learning subprocess about the finished product forging:
s21: randomly selecting a finished product forging five-dimensional point cloud from a finished product forging five-dimensional point cloud data set E, inquiring a finished product forging point cloud distance two-dimensional table, finding out a data set C consisting of all other finished product forging five-dimensional point clouds of which the distances from the finished product forging five-dimensional point clouds are smaller than a density threshold d1, and counting the number N of the finished product forging five-dimensional point clouds in the data set C; if N <5, marking the five-dimensional point cloud of the finished product forged piece as 0, emptying the data set C, randomly selecting another five-dimensional point cloud of the finished product forged piece from the data set E of the five-dimensional point cloud of the finished product forged piece, and executing the step S21 again; if N is more than or equal to 5, marking the five-dimensional point cloud of the finished product forged piece as 1, creating a data set C1, uniformly distributing all the five-dimensional point clouds of the finished product forged piece in the data set C into the data set C1, emptying the data set C, and executing the step S22;
s22: randomly selecting an unmarked finished product forging five-dimensional point cloud from the data set C1 for accessing, inquiring the distance two-dimensional table of the finished product forging point cloud, finding out all other finished product forging five-dimensional point clouds of which the distance from the accessed finished product forging five-dimensional point cloud is less than a density threshold value d1, putting the other finished product forging five-dimensional point clouds into the data set C, and counting the number N of the finished product forging five-dimensional point clouds in the data set C; if N is less than 5, marking the five-dimensional point cloud of the accessed finished product forged piece as 0, emptying the data set C, detecting whether the data set C1 contains the five-dimensional point cloud of the unmarked finished product forged piece, if so, executing the step S22 again, otherwise, executing the step S23; if N is more than or equal to 5, marking the five-dimensional point cloud of the finished product forged piece as 1, dividing all the five-dimensional point clouds of the finished product forged piece in the data set C into the data set C1, emptying the data set C, and executing the step S22 again;
s23: removing the finished product forging five-dimensional point cloud of the data set C1 from the finished product forging five-dimensional point cloud data set E to obtain a data set E1, randomly selecting one finished product forging five-dimensional point cloud from the data set E1, and obtaining a data set C2 according to the operation of the steps S21 and S22 in the same way;
s24: removing the finished product forging five-dimensional point cloud of the data set C2 from the data set E1 to obtain a data set E2, similarly obtaining a data set C3, a data set C4, a data set C5 and the like according to the operations of the steps S21, S22 and S23 until all finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data group E are classified into corresponding data sets such as the data set C1, the data set C2, the data set C3 and the like, and obtaining the number of the data sets such as the data set C1, the data set C2, the data set C3 and the like at the moment, wherein N1 is the number of the data sets;
s25: according to the operations of steps S21 to S24, the number N2, N3, N4, and N5 of data sets corresponding one-to-one to the set density threshold values d2, d3, d4, and d5, respectively, is obtained in the same manner.
Preferably, the standard forging five-dimensional point cloud defining method comprises the following steps:
defining the first characteristic of each standard forging point cloud as an x-axis coordinate value, the second characteristic as a y-axis coordinate value, the third characteristic as a z-axis coordinate value, and the fourth characteristic as the maximum curvature r of the two main curvatures 1 The fifth characteristic is the minimum curvature r of the two principal curvatures 2 . For a certain point cloud in a group of point clouds, the method for acquiring the two main curvatures comprises the steps of taking the point cloud needing to solve the two main curvatures as a center, selecting 19 point clouds closest to the point cloud according to the three-dimensional Euclidean distance, adding the point cloud needing to solve the two main curvatures, totaling 20 point clouds, fitting an ellipsoid curved surface equation by using an ellipsoid curved surface fitting method, and solving the two main curvatures of the point cloud according to the ellipsoid curved surface equation.
Preferably, the method for creating the standard forging point cloud distance two-dimensional table comprises the following steps:
s31: all the standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data set D are subjected to linear normalization and compression to be between-1 and 1, the absolute values of the maximum curvatures r1 of all the standard forging five-dimensional point clouds are calculated, and the maximum r1 is found out max Minimum value r1 min And calculates the mean value r1 m (ii) a Calculating the absolute value of the minimum curvature r2 of the five-dimensional point cloud of all standard forgings to find out the maximum value r2 max Minimum value r2 min And calculates the mean value r2 m (ii) a Calculating the absolute value mean xm of the x-axis coordinate value, the absolute value mean ym of the y-axis coordinate value and the absolute value mean zm of the z-axis coordinate value of the standard forging five-dimensional point cloud data set D, and finding the maximum value m of the mean xm, ym and zm max ；
S32: defining a standard forging five-dimensional point cloud (X) 1 ,Y 1 ,Z 1 ,r a1 ,r b1 ) And another standard forging five-dimensional point cloud (X) 2 ,Y 2 ,Z 2 ,r a2 ,r b2 ) Has a weight Euclidean distance L of
Wherein the content of the first and second substances,
and sequencing all the standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data set D from large to small according to the absolute value of the maximum curvature r1, calculating the weighted Euclidean distance L between every two standard forging five-dimensional point clouds, and manufacturing a corresponding standard forging point cloud distance two-dimensional table, wherein the distance numerical value of the ith row and the jth column in the standard forging point cloud distance two-dimensional table is the weighted Euclidean distance L of the ith standard forging five-dimensional point cloud and the jth standard forging five-dimensional point cloud after all the standard forging five-dimensional point clouds are sequenced according to the absolute value of the maximum curvature r 1.
The method has the advantages that the standard forging point cloud distance two-dimensional table is created in advance, so that various subsequent calculations are greatly simplified, the standard forging point cloud distance two-dimensional table can be obtained only by inquiring the standard forging point cloud distance two-dimensional table in all calculations with the weight Euclidean distance between two point clouds, and recalculation is not needed, so that the identification speed of the finished product forging to be detected by the method is increased.
Preferably, the five-dimensional point cloud defining method for the finished product forging comprises the following steps:
defining the first characteristic of each finished product forging point cloud as an x-axis coordinate value, the second characteristic as a y-axis coordinate value, the third characteristic as a z-axis coordinate value, and the fourth characteristic as the maximum curvature r of the two main curvatures 1 The fifth characteristic is the minimum curvature r of the two principal curvatures 2 。
Preferably, the method for creating the finished product forging point cloud distance two-dimensional table comprises the following steps:
s41: all finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data set E are subjected to linear normalization compression to be between-1 and 1, the absolute value of the maximum curvature r1 of all finished product forging five-dimensional point clouds is calculated, and the maximum value r1 'is found out' max Minimum value r1' min And calculating a mean value r1' m (ii) a Calculating the absolute value of the minimum curvature r2 of all five-dimensional point clouds of the finished forgings to find out the maximum value r2' max Minimum value r2' min And calculating a mean value r2' m (ii) a Calculating the absolute value mean value xm ' of the x-axis coordinate value, the absolute value mean value ym ' of the y-axis coordinate value and the absolute value mean value zm ' of the z-axis coordinate value of the finished product forging five-dimensional point cloud data group E, and finding the maximum value m ' of the mean values xm ', ym ' and zm ' max ；
S42: defining five-dimensional point cloud (X ') of finished product forged piece' 1 ,Y' 1 ,Z' 1 ,r' a1 ,r' b1 ) And another five-dimensional point cloud (X ') of finished forged piece' 2 ,Y' 2 ,Z' 2 ,r' a2 ,r' b2 ) Has a weight Euclidean distance L' of
Wherein the content of the first and second substances,
and sequencing all the finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data set E from large to small according to the absolute value of the maximum curvature r1, calculating the weighted Euclidean distance L 'between every two finished product forging five-dimensional point clouds, and manufacturing a corresponding finished product forging point cloud distance two-dimensional table, wherein the distance numerical value of the ith row and jth column in the finished product forging point cloud distance two-dimensional table is the weighted Euclidean distance L' of the ith finished product forging five-dimensional point cloud and the jth finished product forging five-dimensional point cloud after all the finished product forging five-dimensional point clouds are sequenced according to the absolute value of the maximum curvature r 1.
Preferably, the density threshold values d1, d2, d3, d4 and d5 are set by:
firstly, searching the Euclidean distance L of the maximum belt weight from the point cloud distance two-dimensional table of the standard forge piece max And a minimum weighted Euclidean distance L min ；
Then, the density threshold d1= L is set min +0.01(L max -L min )，
Density threshold d2= L min +0.02(L max -L min )，
Density threshold d3= L min +0.03(L max -L min )，
Density threshold d4= L min +0.04(L max -L min )，
Density threshold d5= L min +0.05(L max -L min )。
The technical principle of the present invention is described above in connection with specific embodiments. The description is made for the purpose of illustrating the principles of the invention and should not be taken in any way as limiting the scope of the invention. Based on the explanations herein, those skilled in the art will be able to conceive of other embodiments of the present invention without inventive effort, which would fall within the scope of the present invention.
Claims (9)
1. The on-site detection method for the surface defects of the forged piece based on ensemble learning and density clustering is characterized by comprising the following steps of:
acquiring a large amount of three-dimensional point cloud data of a standard forge piece by a blue light three-dimensional laser scanner at a position 10 meters away from a standard high-temperature forge piece without defects according to a set sequence rule;
step two, uniformly collecting 3000 point clouds from a large amount of three-dimensional point cloud data of the standard forging obtained in the step one to form a standard forging three-dimensional point cloud data set, calculating two main curvatures corresponding to each standard forging point cloud in the standard forging three-dimensional point cloud data set according to a fitting method of an ellipsoid curved surface, and then selecting a three-dimensional coordinate value and the two main curvatures of each standard forging point cloud as five characteristics representing the standard forging point cloud so as to obtain a standard forging five-dimensional point cloud data set D;
step three, dividing the standard forging five-dimensional point cloud data group D into corresponding standard forging data set numbers n1, n2, n3, n4 and n5 according to density thresholds D1, D2, D3, D4 and D5 respectively;
and (3) detecting a finished product forging:
fourthly, obtaining a large amount of three-dimensional point cloud data of the finished product forge piece through a blue light three-dimensional laser scanner at a position 10 meters away from the finished product forge piece to be detected according to a set sequence rule;
step five, 3000 point clouds are uniformly collected from a large number of three-dimensional point cloud data of the finished product forged piece obtained in the step four to form a finished product forged piece three-dimensional point cloud data set, two main curvatures corresponding to each finished product forged piece point cloud in the finished product forged piece three-dimensional point cloud data set are calculated according to a fitting method of an ellipsoid curved surface, then a three-dimensional coordinate value and the two main curvatures of each finished product forged piece point cloud are selected as five characteristics representing the finished product forged piece point cloud, and a finished product forged piece five-dimensional point cloud data set E is obtained;
step six, dividing the finished product forging five-dimensional point cloud data group E into corresponding finished product forging data set numbers N1, N2, N3, N4 and N5 according to the density threshold values d1, d2, d3, d4 and d5 respectively;
seventhly, comparing the number N1, N2, N3, N4 and N5 of the standard forging data sets with the number N1, N2, N3, N4 and N5 of the finished product forging data sets respectively, wherein if at least 2 groups of N1= N1, N2= N2, N3= N3, N4= N4 and N5= N5 are simultaneously satisfied, the finished product forging has no surface defect, otherwise, the finished product forging has surface defect.
2. The in-place detection method for the surface defects of the forgings based on the ensemble learning and density clustering as claimed in claim 1, further comprising the following steps:
and step eight, performing the detection process of the finished product forge pieces on each finished product forge piece on the production line, and giving an alarm and prompting that the forging and pressing die on the production line is damaged when the surface defects of three continuous finished product forge pieces are judged.
3. The forge piece surface defect in-situ detection method based on ensemble learning and density clustering according to claim 1, wherein the standard forge piece five-dimensional point cloud data set D is divided into corresponding standard forge piece data set numbers n1, n2, n3, n4 and n5 according to density thresholds D1, D2, D3, D4 and D5 respectively, and comprises the following steps:
creating a standard forging point cloud distance two-dimensional table process:
calculating the distance between two points of all five-dimensional point clouds in the standard forging five-dimensional point cloud data set D according to the Euclidean distance with the weight, and making a corresponding standard forging point cloud distance two-dimensional table;
the integrated learning sub-process for standard forgings:
s11: randomly selecting a standard forging five-dimensional point cloud from a standard forging five-dimensional point cloud data set D, inquiring a standard forging point cloud distance two-dimensional table, finding out other standard forging five-dimensional point clouds of which the distances from the standard forging point cloud two-dimensional point clouds are smaller than a density threshold value D1 to form a data set C, and counting the number N of the standard forging point cloud two-dimensional point clouds in the data set C; if N is less than 5, marking the standard forging five-dimensional point cloud as 0, emptying the data set C, randomly selecting another standard forging five-dimensional point cloud again from the standard forging five-dimensional point cloud data set D, and executing the step S11 again; if N is more than or equal to 5, marking the standard forging five-dimensional point cloud as 1, creating a data set C1, uniformly distributing all the standard forging five-dimensional point clouds in the data set C into the data set C1, emptying the data set C, and executing the step S12;
s12: randomly selecting an unmarked standard forging five-dimensional point cloud from the data set C1 for accessing, inquiring the standard forging point cloud distance two-dimensional table, finding out all other standard forging five-dimensional point clouds of which the distance from the accessed standard forging five-dimensional point cloud is less than a density threshold value d1, putting the other standard forging five-dimensional point clouds into the data set C, and counting the number N of the standard forging five-dimensional point clouds in the data set C; if N is less than 5, marking the five-dimensional point cloud of the accessed standard forging as 0, emptying the data set C, detecting whether the data set C1 contains the five-dimensional point cloud of the unmarked standard forging, if yes, executing the step S12 again, otherwise, executing the step S13; if N is more than or equal to 5, marking the standard forging five-dimensional point cloud as 1, dividing all the standard forging five-dimensional point clouds in the data set C into the data set C1, emptying the data set C, and executing the step S12 again;
s13: removing the standard forging five-dimensional point cloud of the data set c1 from the standard forging five-dimensional point cloud data set D to obtain a data set D1, randomly selecting one standard forging five-dimensional point cloud from the data set D1, and obtaining a data set c2 according to the operation of the steps S11 and S12 in the same way;
s14: removing the standard forging five-dimensional point clouds in the data set D1 to obtain a data set D2, obtaining a data set c3, a data set c4, a data set c5 and the like in the same way according to the operations of the steps S11, S12 and S13 until all the standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data group D are classified into corresponding data sets such as the data set c1, the data set c2, the data set c3 and the like, and obtaining the number of the data sets such as the data set c1, the data set c2, the data set c3 and the like at the moment, wherein n1 is the number of the data sets;
s15: from the operations of steps S11 to S14, the number of data sets n2, n3, n4, and n5 corresponding to the set density threshold values d2, d3, d4, and d5, respectively, is obtained in the same manner.
4. The method for detecting the surface defects of the forgings on site based on the ensemble learning and the density clustering as claimed in claim 1, wherein the five-dimensional point cloud data groups of the finished forgings are divided into corresponding finished forging data set numbers N1, N2, N3, N4 and N5 according to the set density thresholds d1, d2, d3, d4 and d5 respectively, and the method comprises the following steps:
establishing a point cloud distance two-dimensional table process of a finished product forging:
calculating the distance between two points of all the finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data set E according to the weighted Euclidean distance, and manufacturing a corresponding finished product forging point cloud distance two-dimensional table;
the integrated learning subprocess about the finished product forging:
s21: randomly selecting a finished product forging five-dimensional point cloud from the finished product forging five-dimensional point cloud data set E, inquiring the distance two-dimensional table of the finished product forging point cloud, finding out a data set C consisting of all other finished product forging five-dimensional point clouds of which the distance from the finished product forging five-dimensional point cloud is smaller than a density threshold value d1, and counting the number N of the finished product forging five-dimensional point clouds in the data set C; if N <5, marking the five-dimensional point cloud of the finished product forged piece as 0, emptying the data set C, randomly selecting another five-dimensional point cloud of the finished product forged piece from the data set E of the five-dimensional point cloud of the finished product forged piece, and executing the step S21 again; if N is more than or equal to 5, marking the five-dimensional point cloud of the finished product forged piece as 1, creating a data set C1, uniformly distributing all the five-dimensional point clouds of the finished product forged piece in the data set C into the data set C1, emptying the data set C, and executing the step S22;
s22: randomly selecting an unmarked finished product forging five-dimensional point cloud from the data set C1 for accessing, inquiring the distance two-dimensional table of the finished product forging point cloud, finding out all other finished product forging five-dimensional point clouds of which the distance from the accessed finished product forging five-dimensional point cloud is less than a density threshold value d1, putting the other finished product forging five-dimensional point clouds into the data set C, and counting the number N of the finished product forging five-dimensional point clouds in the data set C; if N is less than 5, marking the five-dimensional point cloud of the accessed finished product forged piece as 0, emptying the data set C, detecting whether the data set C1 contains the five-dimensional point cloud of the unmarked finished product forged piece, if so, executing the step S22 again, otherwise, executing the step S23; if N is more than or equal to 5, marking the five-dimensional point cloud of the finished product forged piece as 1, dividing all the five-dimensional point clouds of the finished product forged piece in the data set C into the data set C1, emptying the data set C, and executing the step S22 again;
s23: removing the finished product forging five-dimensional point cloud of the data set C1 from the finished product forging five-dimensional point cloud data set E to obtain a data set E1, randomly selecting one finished product forging five-dimensional point cloud from the data set E1, and obtaining a data set C2 according to the operation of the steps S21 and S22 in the same way;
s24: removing the finished product forging five-dimensional point cloud of the data set C2 from the data set E1 to obtain a data set E2, similarly obtaining a data set C3, a data set C4, a data set C5 and the like according to the operations of the steps S21, S22 and S23 until all finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data group E are classified into corresponding data sets such as the data set C1, the data set C2, the data set C3 and the like, and obtaining the number of the data sets such as the data set C1, the data set C2, the data set C3 and the like at the moment, wherein N1 is the number of the data sets;
s25: according to the operations of steps S21 to S24, the number N2, N3, N4, and N5 of data sets corresponding one-to-one to the set density threshold values d2, d3, d4, and d5, respectively, is obtained in the same manner.
5. The in-situ detection method for the surface defects of the forgings based on the ensemble learning and density clustering as claimed in claim 3, wherein the standard forging five-dimensional point cloud defining method comprises the following steps:
defining the first characteristic of each standard forging point cloud as an x-axis coordinate value, the second characteristic as a y-axis coordinate value, the third characteristic as a z-axis coordinate value, and the fourth characteristic as the maximum curvature r of the two main curvatures 1 The fifth characteristic is the minimum curvature r of the two principal curvatures 2 。
6. The integrated learning and density clustering-based forging surface defect in-situ detection method according to claim 5, wherein the method for creating the standard forging point cloud distance two-dimensional table comprises the following steps:
s31: all the standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data set D are compressed to be between-1 and 1 in a linear normalization mode, the absolute value of the maximum curvature r1 of all the standard forging five-dimensional point clouds is calculated, and the maximum value r1 is found out max Minimum value r1 min And calculates the mean value r1 m (ii) a Calculating the absolute value of the minimum curvature r2 of the five-dimensional point clouds of all the standard forgings to find out the maximum value r2 of the five-dimensional point clouds max Minimum value r2 min And calculates the mean value r2 m (ii) a Calculating an absolute value mean xm of an x-axis coordinate value, an absolute value mean ym of a y-axis coordinate value and an absolute value mean zm of a z-axis coordinate value of the standard forging five-dimensional point cloud data set D, and finding a maximum value m of the mean xm, ym and zm max ；
S32: defining a standard forging five-dimensional point cloud (X) 1 ,Y 1 ,Z 1 ,r a1 ,r b1 ) And another standard forging five-dimensional point cloud (X) 2 ,Y 2 ,Z 2 ,r a2 ,r b2 ) Has a weight Euclidean distance L of
Wherein the content of the first and second substances,
sorting all the standard forging five-dimensional point clouds in the standard forging five-dimensional point cloud data set D from large to small according to the absolute value of the maximum curvature r1, calculating the weighted Euclidean distance L between every two standard forging five-dimensional point clouds, and manufacturing a corresponding standard forging point cloud distance two-dimensional table, wherein the distance numerical value of the ith row and the jth column in the standard forging point cloud distance two-dimensional table is the weighted Euclidean distance L of the ith standard forging five-dimensional point cloud and the jth standard forging five-dimensional point cloud after all the standard forging five-dimensional point clouds are sorted according to the absolute value of the maximum curvature r 1.
7. The in-place detection method for the surface defects of the forgings based on the ensemble learning and density clustering as claimed in claim 4, wherein the definition method for the five-dimensional point cloud of the finished forgings comprises the following steps:
defining the first characteristic of each finished product forging point cloud as an x-axis coordinate value, the second characteristic as a y-axis coordinate value, the third characteristic as a z-axis coordinate value, and the fourth characteristic as the maximum curvature r in the two main curvatures 1 The fifth characteristic is the minimum curvature r of the two principal curvatures 2 。
8. The integrated learning and density clustering-based forging surface defect in-situ detection method according to claim 7, wherein the method for creating the point cloud distance two-dimensional table of the finished forging is as follows:
s41: all finished forgings in the five-dimensional point cloud data group E of the finished forgingsThe five-dimensional point clouds are all compressed to be between-1 and 1 in a linear normalization mode, the absolute value of the maximum curvature r1 of all finished product forging five-dimensional point clouds is calculated, and the maximum value r1 'is found out' max Minimum value r1' min And calculating a mean value r1' m (ii) a Calculating the absolute value of the minimum curvature r2 of all five-dimensional point clouds of the finished forgings to find out the maximum value r2' max Minimum value r2' min And calculating a mean value r2' m (ii) a Calculating an absolute value mean value xm ' of the x-axis coordinate value, an absolute value mean value ym ' of the y-axis coordinate value and an absolute value mean value zm ' of the z-axis coordinate value of the finished product forging five-dimensional point cloud data group E, and finding a maximum value m ' of the mean values xm ', ym ' and zm ' max ；
S42: defining five-dimensional point cloud (X ') of finished product forged piece' 1 ,Y' 1 ,Z' 1 ,r' a1 ,r' b1 ) And another five-dimensional point cloud (X ') of finished forged piece' 2 ,Y' 2 ,Z' 2 ,r' a2 ,r' b2 ) Has a weight Euclidean distance L' of
Wherein the content of the first and second substances,
and sequencing all the finished product forging five-dimensional point clouds in the finished product forging five-dimensional point cloud data set E from large to small according to the absolute value of the maximum curvature r1, calculating the weighted Euclidean distance L 'between every two finished product forging five-dimensional point clouds, and manufacturing a corresponding finished product forging point cloud distance two-dimensional table, wherein the distance numerical value of the ith row and jth column in the finished product forging point cloud distance two-dimensional table is the weighted Euclidean distance L' of the ith finished product forging five-dimensional point cloud and the jth finished product forging five-dimensional point cloud after all the finished product forging five-dimensional point clouds are sequenced according to the absolute value of the maximum curvature r 1.
9. The in-place detection method for the surface defects of the forgings based on the ensemble learning and density clustering as claimed in claim 6, wherein the density threshold values d1, d2, d3, d4 and d5 are set by a method comprising the following steps:
firstly, searching the Euclidean distance L of the maximum belt weight from the point cloud distance two-dimensional table of the standard forge piece max And a minimum weighted Euclidean distance L min ；
Then, the density threshold d1= L is set min +0.01(L max -L min )，
Density threshold d2= L min +0.02(L max -L min )，
Density threshold d3= L min +0.03(L max -L min )，
Density threshold d4= L min +0.04(L max -L min )，
Density threshold d5= L min +0.05(L max -L min )。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710707884.8A CN107516313B (en) | 2017-08-17 | 2017-08-17 | Forging surface defect based on integrated study and Density Clustering is in position detecting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710707884.8A CN107516313B (en) | 2017-08-17 | 2017-08-17 | Forging surface defect based on integrated study and Density Clustering is in position detecting method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107516313A true CN107516313A (en) | 2017-12-26 |
| CN107516313B CN107516313B (en) | 2018-06-05 |
Family
ID=60723376
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710707884.8A Expired - Fee Related CN107516313B (en) | 2017-08-17 | 2017-08-17 | Forging surface defect based on integrated study and Density Clustering is in position detecting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107516313B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108320329A (en) * | 2018-02-02 | 2018-07-24 | 维坤智能科技（上海）有限公司 | A kind of 3D map creating methods based on 3D laser |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101788532A (en) * | 2010-02-11 | 2010-07-28 | 中国航空工业集团公司北京航空材料研究院 | Ultrasonic testing method used for large-sized complex forging |
-
2017
- 2017-08-17 CN CN201710707884.8A patent/CN107516313B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101788532A (en) * | 2010-02-11 | 2010-07-28 | 中国航空工业集团公司北京航空材料研究院 | Ultrasonic testing method used for large-sized complex forging |
Non-Patent Citations (2)
| Title |
|---|
| 姜炜 等: "锻件近表面缺陷的超声检测", 《无损检测》 * |
| 窦华梅: "钛合金锻件夹杂和偏析缺陷状况对超声波检测回波幅度的影响", 《陕西省第十一届无损检测年会》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108320329A (en) * | 2018-02-02 | 2018-07-24 | 维坤智能科技（上海）有限公司 | A kind of 3D map creating methods based on 3D laser |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107516313B (en) | 2018-06-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110135468B (en) | Coal gangue identification method | |
| CN107492098B (en) | It is a kind of based on PCA and CNN high-temperature forging surface defect in position detecting method | |
| CN109272500B (en) | Fabric classification method based on adaptive convolutional neural network | |
| CN109190752A (en) | The image, semantic dividing method of global characteristics and local feature based on deep learning | |
| CN105787486B (en) | A kind of girder steel crack detecting method based on image procossing | |
| CN110135486B (en) | Chopstick image classification method based on adaptive convolutional neural network | |
| CN111242202B (en) | Method for monitoring wear state of turning tool based on metric learning | |
| CN106599915B (en) | A kind of vehicle-mounted laser point cloud classifications method | |
| CN114581782B (en) | Fine defect detection method based on coarse-to-fine detection strategy | |
| CN112465809B (en) | Mold defect detection method based on image recognition, computer readable storage medium and device | |
| CN106651882A (en) | Method and device for identifying and detecting cubilose impurities based on machine vision | |
| CN104680157A (en) | Bundled bar material identification and counting method based on support vector machine | |
| CN111798447B (en) | Deep learning plasticized material defect detection method based on fast RCNN | |
| CN103679207A (en) | Handwriting number identification method and system | |
| CN113222982A (en) | Wafer surface defect detection method and system based on improved YOLO network | |
| CN105976390A (en) | Steel tube counting method by combining support vector machine threshold statistics and spot detection | |
| CN104200215A (en) | Method for identifying dust and pocking marks on surface of big-caliber optical element | |
| CN111914902A (en) | Traditional Chinese medicine identification and surface defect detection method based on deep neural network | |
| CN110443303B (en) | Image segmentation and classification-based intelligent identification method for coal-rock microcomponents | |
| CN115375672A (en) | Coal gangue detection method based on improved YOLOv4 algorithm | |
| Zeng et al. | Steel sheet defect detection based on deep learning method | |
| CN107516313B (en) | Forging surface defect based on integrated study and Density Clustering is in position detecting method | |
| CN114092478B (en) | Anomaly detection method | |
| CN115082422A (en) | Image-based building steel pipe cross section size measurement method | |
| Meria | Development of automatic industrial waste detection system for leather products using artificial intelligence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180605 Termination date: 20200817 |