Visual detection method of chamfer R on the root bottom of textile workpiece rol

HU Ling-hao, LIU Chang-jie, SHI Chun-min, FU Lu-hua, LU Gang

(1. State Key Laboratory of Precision Measuring Technology and Instruments, Tianjin University, Tianjin 300072, China；2. China Railway Test & Certification Center Locomotive & Car Inspection Lab, China Academy of Railway Sciences, Beijing 100015, China)

Abstract： Roller is an important workpiece of automatic spinning machine. Only if the chamfer R on the root bottom of roller’s guide pillar meets processing accuracy requirement can the end face of one roller’s guide pillar match the correct position of another roller’s end face on the guide hole. Therefore, the size of chamfer R on the root bottom of the roller’s guide pillar has an important influence on the operating state of automatic spinning machine. In order to achieve the rapid, automatic and precise measurement of chamfer R on the root bottom of roller, an auto-detection system for roller’s chamfer based on computer vision technology is proposed. Firstly, the principle of measurement based on computer vision technology is introduced. And then the extraction method of chamfer’s characteristic parameters is presented, which uses image processing technique to obtain these characteristic parameters by means of collected images of roller contour, including extraction of region of interest, extraction of subpixel-precise edge, segmentation of arc and line, fitting of geometric primitives, etc. Finally, after experimental verification, the measurement error is within ± 5 μm and repeated accuracy is 0.1 μm. The results show that this measurement method is applicable to not only the chamfer on the textile workpiece, but also the workpieces of shaft type with various of sizes.

Key words： textile workpiece, roller, chamfer, image measurement, auto-detection

0　Introduction

As people’s demand for textile products increases and automation and precision of modern spinning machine improves, accurate detection of textile workpieces is indispensible. Different types of rollers are important shaft workpieces in automatic spinning machine. Only if the chamferRon the root bottom of the roller’s guide pillar meets the processing accuracy requirement can it ensure that one roller’s guide pillar matches the correct position of another roller’s guide hole. Therefore, the size of the chamferRon the root bottom of roller’s guide pillar has an important influence on the operating state of automatic spinning machine. At present, the production of roller mainly depends on cutting technology, but the size and shape of chamfer on roller’s root bottom cannot be processed with high precision. Thus the detection of geometry dimensioning of chamfer on roller’s root bottom is particularly essential for the safety and reliability of roller’s operating. As shown in Fig.1, the structure of roller is entirely cylindrical and the features of chamfer are tiny. Currently, in the textile industry, the method for inspecting roller’s chamfer is mainly based on manual sampling detection, by comparing the roller by means of standard model loop to judge whether the roller is qualified or not, which means that this method cannot obtain the precise geometry features of the chamfer on roller’s root bottom. Besides, this process is tardy with low efficiency and it also brings inevitable artificial errors. The standard model is easily abraded, which cannot guarante the accuracy of verification. In conclusion, it is necessary to have a rapid, precise and automatic measuring system for the geometry feature parameters of the roller’s chamfer on the root bottom.

Fig.1　Workpiece roller and roller’s chamfer on the root bottom

The existing methods for measuring the chamfer are divided into two types, contact measurement and non-contact measurement.

The contact measurement is represented by the coordinate measuring system, in which the probe contacts with multiple samples on the workpiece’s surface to acquire the coordinates of the points, and then by calculating the coordinate position of the points set, the coordinate data are converted into geometry size, shape and related position of the measured pieces. For example, the Germany company Hommel developed T8000 and C8000, and they have been used to detect the groove of crankshaft in automotive engine and the detection accuracy can reach micrometer[1-2]. The ultraprecise coordinate measuring machine produced by Japanese company Panasonic and the inductance contact equipment developed by Italian company Marposs are also widely used in the industrial engineering[3-4]. The precision of contact measurement is able to attain micron level and thus it is applied to complex shape artifact. However, it has disadvantages such as slow speed, low efficiency, high cost, limited measured objects due to the size of the probe, and so on. Therefore, it is difficult to measure the chamfer of columnar spinning object like roller.

Non-contact measurement is mainly based on machine visual technology, which is high-precision and efficient[5-6]. The measuring media can be divided into laser and structure light[7]. By combining a multi charge coupled device (CCD) camera, the measuring system scans the workpiece wholly to obtain the overall contour. In order to extract the feature parameters of the measured object, it then processes the contour image and integrates the data. Various of 3D and 2D inspection systems have been widely used for measuring different kinds of objects[8-10]. For example, in 2000, Seoul National University developed a precise detection system for gear chamfer contour with the accuracy of sub-micron[11], which made use of optical microscope and combined image processing technology. Guo Y C from Chongqing University utilized optical grating amesdial to modify the microscope and develop the chamfer measuring system with double screens[12]. Zhang H from Shanghai Jiaotong University took advantage of tele-centric lens and CCD camera to measure the crankshaft groove[13]. In conclusion, the chamfer measuring system based on computer visual system mainly relies on the accuracy of object’s position, which means the plane of the chamfer is parallel with the plane of the camera. As for a cylindrical object, it is difficult to define the surface of the chamfer. Accordingly, this method is hard to do well in the process of camera focusing and cannot adapt to the rapid, automatic and precise requirements of industrial inspection.

In this paper, taking a spinning workpiece roller as an example, a method for measuring the chamfer on the roller’s root bottom is presented. This method is based on the backlight image and it also illustrates the chamfer feature extraction technology for the captured image. It collects the contour of the roller’s chamfer on the root bottom and acquires its dimension parameters instantly, automatically and precisely.

1　Measurement principle

In this paper, measuring the measurement of roller’s chamfer is on the basis of computer visual technology, as shown in Fig.2.

Fig.2　Structure of measuring system

Backlighted by a parallel light source, the contour of roller’s chamfer on the root bottom is imaged on the CCD of the camera. To ensure that the contour image of the roller is undistorted and reserves the complete information of measured parameters, it is necessary to select suitable parallel light. Besides, the shadow needs to be imaged on the same plane of CCD camera with matching luminance accurately, and the plane is also perpendicular to the optical axis of the lens. If the light is not parallel, the image will be unsatisfactory, as shown in Fig.3. The contour will not be projected on the same plane and the focusing of part of the image is not accuracy, which leads to being obscure. What’s more, the distribution of brightness is uneven, causing the distortion of the image, which makes it difficult to implement the subsequent image processing. Due to the particular position of the chamfer on the spinning workpiece, backlighting the object by parallel light also solves the difficulty of focusing the camera.

Fig.3　Image of roller’s chamfer backlighted by non-parallel light

CCD sensor converts optical signals to electrical signal to obtain the image of the chamfer on the root bottom of the roller and then sends it to the computer. Next, by a series of image processing algorithms based on the shadow image of the roller’s chamfer, the information of roller’s chamfer on the root bottom of the roller is extracted. In the end, according to transformation relationship between image coordinate and world coordinate, the final life-sizes of measured parameters are obtained.

The photo of roller’s chamfer on the root bottom captured by CCD camera (50 mm focus length) is shown in Fig.4.

Fig.4　Image of roller’s chamfer backlighted by parallel light

By positioning the optical platform, it is ensured that not only the plane of the roller’s chamfer, the illuminated plane of the parallel light source and the plane of the CCD camera are parallel, but also they are all perpendicular to the main optical axis.

2　Chamfer detection and feature parameters auto-extraction

On the basis of captured image, the steps of processing the image and calculating the data are as follows. At first, preprocessing the image, which includes region of interest extraction and subpixel-precise edge extraction. Next, implementing the arc extraction from the contour and fitting the arc to the circle. After that, obtaining the pixel size of the roller’s chamfer on the root bottom based on the feature of fitting circle. At last, by transforming the picture coordinate to the world coordinate, the pixel size is converted to the real geometric dimension and the final detection result of roller’s chamfer is obtained.

2.1　Subpixel-precise edge extraction

The region of interest is shown in Fig.5.

Fig.5　Region of interest

In this paper, Canny edge detection algorithm equipped with double threshold is used to locate the subpixel edge from the regions of interest. This algorithm involves Gaussian smoothing, gradient calculation and non-maxima suppression[14]. Thereinto choosing the threshold is extremely important in this process. As shown in Fig.6, if the settings of low threshold and high threshold are inappropriate, it will lead to losing of the edge, mixing with plenty of noises, increasing of the fake edge, etc.

By comparing the gray values of the background and the chamfer’s contour, high threshold of 400 and low threshold of 200 are suitable. The extraction of the contour’s subpixel edge by Canny edge detection algorithm is shown in Fig.7.

Fig.6　Images of unsatisfied contour extraction

Fig.7　Subpixel-precise edge extraction

2.2　Arc extraction

To achieve the automatic detection of roller’s chamfer, the system has to automatically fit the circle of the contour at the location of chamfer. However, because the roller has different types and sizes, it is unable to ensure that every image of roller’s chamfer on the root bottom is completely identical. In addition, because of location error, the measured object may not be at the same position each time. On the basis of above reasons, the key of this design is to determine the connection of the chamfer arc and two straight lines (the generatrix of roller’s guide pillar and the straight line of bottom plane). By comparing different algorithms of arc region extraction, an automatic arc region extraction method which is suitable for this condition is proposed.

The first method is based on the Hough line detection algorithm. By means of the mapping relationship between the lines and points in the space, the coordinates of feature points are converted from the picture space into the parameter space. After a process called “voting”, it is impossible to find out the peak in the parameter space and it corresponds to the line object[15]. By using Hough straight line detection algorithm, the two straight lines (the generatrix of roller’s guide pillar and the straight line of bottom plane) connected with the chamfer arc are captured. As shown in Fig.8, the terminuses of straight lines are regarded as diagonal points to determine rectangle box and then as the fitting region.

Fig.8　Fitting region determined by Hough line detection algorithm

However, during the actual measurement, due to different sizes and characteristics of various rollers and the influence of working place, it is difficult to ascertain that the constant parameters of Hough algorithm satisfy the whole detection conditions. As shown in Fig.9, it is apparent that on the one hand, if the detected straight line is too long, the information of arc region will be incomplete, which will affect the accuracy of the subsequent fitting process. On the other hand, if the detected line is too short, it will automatically sense redundant information, which will introduce certain interference from the straight lines and they are difficult to be removed. Consequently, as it contains unavoidable system error and has difficulty in realizing the automatic detection process of the workpiece, it is not reliable and stable to extract arc contour by this method alone.

Fig.9　Fitting region with redundant straight-line information

The second method is based on the multi-primitive segmentation of planar curves. As it is arduous to obtain the characteristic data of roller’s chamfer from the whole contour and to seek the connection between the arc and straight lines, the chamfer’s contour should be segmented into two characters, the linear primitives and arc primitives. This system adopts a two-level breakpoint classification and tuning approach[16].

The first step of this algorithm is the two-level classification of breakpoints. In level 1, an adaptivek-curvature (AKC) function is defined to classify the break points into corners and smooth joints[17]. The function is defined based on thek-curvature values[18]. As shown in Fig.10, three breakpointsPi-1,PiandPi+1separate two consecutive segmentsSiandSi+1of the arc. Defininguas the minimum value of the lengthens ofSiandSi+1,k=u/2, thek-vectors ofPjareajkandbjk, wherePjis a point which belongs to [Pi-k,Pi+k].

Fig.10　Region of support of AKC function

Defining AKC function of the pointPito becjk, it can be obtained by

ajk=(Pj+(x)-Pj(x),Pj+(y)-Pj(y)),

(1)

bjk=(Pj-(x)-Pj(x),Pj-(y)-Pj(y)),

(2)

Pj+=Pj+k,

(3)

Pj-=Pj-k,

(4)

(5)

Next, a projective height function (PHF) is employed to determine the type of the segments on both sides of a breakpoint, which is line or arc. The projective heighth(Pj) of pointPjis obtained by

(6)

(7)

(8)

Subsequently, by calculatingh(Pj) of every point on the regions of segments, a line accumulator and an arc accumulator are establisheed. Ifh(Pj)>0.5, the arc accumulator increases by one, otherwise, the line accumulator increases by one. After that, the two accumulators are compared until the type of segments is determined. If the value of the arc accumulator is greater than that of the line accumulator, the segment is arc; otherwise, it is line. Consequently, by means of joint tuning, the line and arc segments are merged and the contour is recovered with the breakpoints, as shown in Fig.11.

In reality, due to the difference between various rollers and the influence of residual scraps of iron and metal burrs on the roller’s surface after cutting, as shown in Fig.12, during this process, redundant breakpoints are easily detected, resulting in the generation of interference information. Since this interference is hard to be eliminated, it exerts a negative influence on the result and efficiency of detection.

Fig.11　Segmentation of roller’s contour

Fig.12　Redundant breakpoints in segmentation of roller’s contour

In this paper, the system combines these two methods for circle extraction.

At first, it adopts Hough algorithm to detect two straight lines in the roller’s image. To ensure the rectangle region determined by two endpoints contains complete information of roller’s chamfer, threshold value is selected to be properly small.

Then, on the basis of the two-level breakpoint classification and tuning approach, the contour in the rectangle is segmented and the points of the highest vertical coordinate and horizontal coordinate are defined as the connections of straight lines and arc so as to filter redundant line information. As shown in Fig.13, the arc is extracted for the next fitting process.

The combination of two methods not only solves the problems of setting constant value, low efficiency and reliability in straight line detection, but also filters redundant connecting points in the contour segmentation because of unclean matter and roller’s difference. Moreover, it reduces the region of contour segmentation, which enhances the speed of calculating and testing.

Fig.13　Arc extraction

3　Roller’s chamfer circle fitting

To obtain the characteristic parameters of the roller’s chamfer on the root bottom, it is indispensable to fit the captured contour into circle. This article adopts weighted least squares method which accumulates the squared distance of all the points on the arc and then minimizes the sum to determine the fitting circle.

As the central angle of chamfer is large and it consists of plenty of fitting points, it is appropriate to employ the Kasa approach to improve the fitting speed[19-20]. The objective function can be rewritten as

(9)

(10)

The center coordinate and radius of the fitting circle can be transformed into solutions of parametersA,BandCregardless of iteration, there are

(11)

C=R2-A2-B2.

(12)

The fitting circle obtained by Kasa algorithm is shown in Fig.14.

The radius’ size of fitting circle is regarded as the pixel size of roller’s chamfer on the root bottom. According to mapping relationship between the picture coordinate and world coordinate, the pixel size can be converted into the real geometric quantity of roller’s chamfer.

Fig.14　Figure of circle fitting

4　Experimental test

To evaluate the measuring accuracy and repeated accuracy of the proposed detection method, it is feasible to measure the standard photolithographic template circles of 0.6, 0.5 and 0.4 mm, respectively. As these surface accuracies are up to 0.2 μm, much higher than the requirement of roller detection. The results are shown in Table 1.

Table 1　Results of measuring standard photolithographic template circle

Compared with the standard value, the accuracy of this method is within ±5 μm and the repeated accuracy is within 0.1 μm, which satisfies the conditions of roller measurement.

On the basis of the above test, Fig.15 shows the detecting of the roller from Jingwei textile machinery Co., Ltd. The measuring results are shown in Table 2.

Fig.15　Roller’s chamfer measurement system

Table 2　Results of measuring roller from Jingwei textile machinery Co., Ltd

The repeated accuracy is shown in Fig.16.

Fig.16　Figure of measuring repeated accuracy

5　Conclusions

This paper proposes an automatic detecting method for textile workpiece roller chamfer based on computer vision technology. It aims to solve the problems in contemporary roller detection such as difficulties in measuring, low efficiency of manual sampling, being unable to obtain the real size of roller’s chamfer, etc. This method means a lot to the processing of textile workpiece and working of spinning machine.

1) Measurement accuracy. Compared with the existing approaches that compare the roller’s chamfer with standard module, this method can acquire the real parameter of roller’s chamfer on the root bottom with measuring error within ±5 μm and repeated accuracy within 0.1 μm. It also lays the foundation for the future manufacturing and disposing of the defective products.

2) Measurement safety and popularity. As the measuring process is non-connect, it will not abrade the measuring object, and the operation is safe and reliable. Moreover, the method is not limited to the size of the probe, which means it is feasible to adjust the positions of camera, measuring object and illumination according to the size range of the chamfer. Therefore, it is suitable for chamfer measurement of shaft workpieces with various sizes.

3) High efficiency. The measuring process can be realized automatically. Besides, compared with the connect measurement like multi-sampling, it is more convenient to operate.