Studying an improved recursive rough method for three-dimensional visual examination of steel plate butt welding surface quality

An improved recursive rough algorithm is proposed to extract feature points in order to address the slow speed and poor accuracy of feature point extraction in 3D visual inspection, as well as the difficulty in meeting the requirements of efficient and quantitative analysis of welded surface quality detection. The preprocessed point cloud is sliced by section. The feature points are obtained according to the improved recursive rough algorithm. The defect evaluation is carried out to obtain the inspection conclusion of the weld appearance defect. Finally, according to the formulated weld defect evaluation process and standards, the typical weld template is selected for weld width, weld misedge and weld straightness test. Weld detection accuracy can reach 3 decimal places, the speed is 4 times the current manual detection speed. The detection results show that the system has the characteristics of high accuracy and fast speed which can replace manual detection. It also has a good application prospect.


Introduction
Welding has been widely used in many production industries such as aerospace, shipbuilding, and nuclear industry [1][2][3] as a significant aluminum alloy material processing technology.Welding is a non-completely controlled physicochemical reaction process that is influenced by a variety of elements [4].Traditional methods for detecting welding appearance defects include ultrasonic testing [5][6], infrared heat wave detection [7][8], X-ray examination [9] and human inspection [10].At the moment, the majority of surface defect inspection in the construction process is done manually.The inspection method is very rough and easy to miss.It has a sluggish detection speed and lacks quantitative analysis.
With the rapid advancement of computer technology, 3D non-contact measurement technology based on laser scanning is now widely employed in the industrial sphere.3D visual measurement technology has a high external ambient light adaptability to match any working environment.However, the use of 3D vision detection has the problem of slow extraction of feature points and poor accuracy, which affects subsequent measurements.This paper proposes a recursive coarse improvement algorithm to extract feature points for this problem.

Weld point feature extraction and defect evaluation
The point cloud is preprocessed (point cloud cutting, point cloud reduction) by the 3D scanner using a Compman K6 series scanner, as shown in Figures 1 and 2.

Extraction of weld features
2.1.1Sliced section generation.The sliced section is utilized to create the weld section profile, which is then converted to two-dimensional space for the following extraction of weld feature points in order to acquire the weld feature information.The precise procedure is as follows: 1.Which way the weld line has to be separated is decided.2. Taking the cutting y direction as an example, the basic location in the contour cutting is chosen as the x coordinate of a point in the weld line after cutting.Due to some welding flaws, the curve is not smooth during sectioning.Therefore, in order to "best" portray the overall change trend of the weld profile data, the interpolation method is used to suit the weld's actual contour curve, as illustrated in Figure 3.  Assuming that the known points are ( 0 X , 0 Y ) and that there are two points ( 1 between the lines, the distance is found between the two.
Consequently, the formula for the distance is used between a point and a line.The feature points are shown in Figure 6, according to the definition of the forming size of the weld surface, the specific parameters of the geometric features of the weld can be obtained and the calculation process is as follows:   The collected feature points from the preprocessed point cloud indicated above serve as data support for the future evaluation of welding surface flaws.The forming size of the weld is one of the key factors in determining the quality of a weld.It is possible to assess the forming quality by measuring the forming size of the weld.The relevant evaluation standards for various welding techniques vary as well.The specific requirements are displayed in Table 1.
Tables 1.The allowable range of welded steel structure melting weld width and misalignment.

Sample production
Five 120x60x10 steel plates were welded together with a 10 mm gap between them by using a welding robot to test the viability of the idea.The sample is shown in Figure 7.

Comparison of system inspection and manual inspection results
The weld width results that the system detected and the weld width results that were manually examined are shown in Tables 2 and 3, respectively.
The findings of the system's weld misalignment detection and manual inspection are displayed in Tables 4 and 5, respectively.The weld width detection is shown in Figure 8.The weld misalignment detection is shown in Figure 9.

Conclusion
The above shows that the 3D detection system's detection result can be accurate to three decimal places whereas manual detection is only accurate to one and a half decimal places.In addition, 3D visual inspection's detection speed is around 1/4 of manual inspection, which is an improvement over the latter.The findings of the system inspection are consistent with those of the manual testing, which further demonstrates the detection accuracy and system accuracy.This paper only studies two defects and other defect detection can be added later to improve universality and promote automation.The system detects the wrong edge of the weld Manually detected mistaken weld edges

Figure 3 .
Figure 3. Contour curve of weld after fitting.2.1.2Improved recursive coarse algorithm extracts feature points.In order to measure the forming size of the weld, the characteristic points of the contour line of the weld section need to be further extracted.The extraction method uses an improved recursive coarse algorithm to extract feature points.The principle of the improved recursive rough algorithm is: as shown in Figure4, the initial point of the contour is 0 P ; The endpoints n P , 2 P and 3 P are the left end of the weld, the highest point of the weld and the right end of the weld; The point 1 P is the point with the largest distance from the
, the maximum distance can be chosen as the feature point; Figure5displays the outcome of the feature point selection.

Figure 5 .
Figure 5. Feature point selection results.The feature points are shown in Figure6, according to the definition of the forming size of the weld surface, the specific parameters of the geometric features of the weld can be obtained and the calculation process is as follows:

Figure 6 .
Figure 6.Schematic diagram of the characteristic point of the welding profile curve.The weld width (D) is the difference between the X value of 3 P and 2 P .
The wrong edge (M) is the difference between the Y value of 3 P and 2 P .
of the weld / m m

Table 2
Weld width tested by the system.

Table 3
Weld width for manual inspection.

Table 4
Mistaken welds detected by the system.

Table 5
Manually detected wrong edge of weld.