Application of machine vision in optical fiber end face defect detection

Traditional fiber end face defect detection uses manual detection, which is inefficient and the test results are highly subjective. Using machine vision to detect fiber end face defects can greatly improve detection efficiency and accuracy. First, the collected image is binarized through image processing, then the fiber core center is located, and then the fiber end face is divided into different circular detection areas with the fiber core center as the center. Since the defects on the fiber end face may be dark or light, different binary processing is required for each area detection in order to distinguish between the two. If any area fails the test, the fiber end face does not meet the requirements. The results show that the use of machine vision for fiber end face detection can quickly and accurately detect the location and size of the defects.

0 Introduction

As the carrier of the information highway and an important part of the optical fiber communication system, optical fiber embodies very superior information transmission characteristics and is an indispensable element of the information society in the 21st century. In optical fiber communication, the active connection of optical fiber is realized through optical fiber connectors, and the cleanliness of the optical fiber end face has a decisive influence on the performance of the connector. In addition to the permanent damage (such as scratches and cracks) to the optical fiber end face caused during the polishing process or during online business operations such as optical fiber plugging and unplugging, the optical fiber end face may also be subject to various temporary contaminations (such as dirt, oil stains, water or detergent residues) during normal use due to contact with unclean hands, the edge of the optical fiber cap, the metal edge of the flange, the dirty optical fiber end face and particles in the air, which affects its performance. This will not only increase the connection loss and reduce the communication performance, but also cause the fiber core to be blocked and unable to transmit light in severe cases, resulting in the core being burned by high-power lasers. In order to ensure the stability and efficiency of optical fiber communication, the end face of each optical fiber must be kept clean to a certain extent.

The traditional method of optical fiber end face defect detection is to use manual detection. This method first collects the image of the optical fiber end face, then observes the dirt with the naked eye, and then makes a manual judgment. Since this method requires the inspector to continuously observe the end face of the optical fiber with the naked eye, it is easy to cause visual fatigue, so the efficiency is relatively low. Moreover, everyone's experience and sense of responsibility are different, and the quality of the obtained products will also vary greatly. In order to improve the reliability and production efficiency of the product, this paper proposes a method of abandoning manual naked eye detection and using machine vision for detection. Machine vision mainly uses computers to simulate human visual functions, extract information from the image of objective things, perform image processing and understand it, and then use it for defect detection of optical fiber end faces. Compared with manual detection, machine vision detection methods improve detection accuracy, reduce testing costs, enhance testing capabilities, reduce the difficulty of training operators on the production line, and can obtain more production line monitoring data information.

This article introduces how to use machine vision to detect defects on the end face of optical fiber, and uses the VBAI visual automatic inspection development environment of National Instruments (NI) to complete the development of a machine vision system. VBAI (Vision Builder for Automated Inspection) is a visual inspection software launched by NI, which serves as a visual builder for automatic inspection. This tool is an ideal tool for rapid visual verification in the laboratory, and is also a good simple test platform for production lines. The results show that the system can detect the location and size of defects more accurately, at a faster speed, and meets the requirements for optical fiber end face defect detection.

1. Detection system

The fiber end face defect machine vision inspection system proposed in this paper consists of a fiber end face detector and a PC system. During the inspection, the fiber connector is inserted into the fixed test platform of the fiber end face detector, and the fiber end face detector is connected to the PC system via a USB cable, so that the image can be collected in the computer. Adjust the focal length of the microscope, and once a satisfactory image is obtained, start the software to analyze the fiber end face and compare it with the standard indicators preset by the software, so as to quantitatively determine the information of each area and judge whether the fiber end face is qualified or not.

The system's inspection results are related to the software's capabilities, the microscope's performance, and the operator's skills in focusing the image. It has been proven to be significantly superior to manual inspection in terms of accuracy, repeatability, reproducibility, and inspection efficiency. The solution can also provide detailed records of inspection results, including end-face images and damage detection data, to improve the system's automation level.

2. Testing process

The detection software used in this system is developed based on the VBAI visual automatic detection development environment and has the ability to process and analyze optical fiber end face images. The processing flow is shown in Figure 1.

Figure 1 System software flow chart

Threshold method It is a simple and effective image segmentation method. This method uses one or several thresholds to divide the grayscale of the image pixels into several levels. Pixels belonging to the same level are considered to be the same type of objects. It should be noted that during the inspection of the optical fiber end face, since the cladding may belong to the same grayscale as the dirt outside the cladding, the dirt on the cladding cannot be inspected at the same time as the dirt outside the cladding. The cladding needs to be shielded before inspection.

In this system, let F(x, y) represent the output of the binary image, and its pixel grayscale range is [a, b]. When detecting the same type of defects, it is only necessary to set a threshold TH between a and b to divide the pixels of the image into two parts: the pixel group greater than TH (defects) and the pixel group less than TH (background). That is:

The image binarization sets the gray value of defective pixels to 0 and the gray value of background pixels to 1. In the visual assistant function module of VBAI, there is a function submodule for setting the threshold. When calling it, you only need to find the peaks and valleys on the threshold histogram according to the bimodal method, and manually adjust the threshold value so that it can distinguish defects from the background, as shown in Figure 2.

Figure 2 Original image and its threshold histogram

2.1 Positioning of the fiber core

The method used in this paper to locate the fiber core is to first find the entire cladding of the fiber. Since the shape of the fiber cladding is a circle, the center of this circle is the center of the fiber core. If there is a large degree of pollution on the fiber end face, if you only set a certain threshold to binarize the image, the binary image obtained may have a lot of dirt in addition to the fiber cladding, which will have a great impact on the positioning of the fiber cladding. Therefore, after the image is binarized, it is necessary to use some sub-function modules in the visual assistant to perform some morphological processing on the image. Use the two function sub-modules of Remove Small Objects and Remove Large Objects to adjust the number of iterations, filter out particles smaller and larger than the fiber cladding, thereby removing interference with the positioning of the fiber cladding, so that the binary image obtained only contains the image of the fiber cladding, as shown in Figure 3.

[page]

After obtaining the binary image of the optical fiber cladding, the position of the optical fiber core can be accurately located by using the function modules of Find Circular Edge and Set Coordinate System in VBAI. The purpose of finding the circular edge is to find the edge of the optical fiber cladding circle, so as to find the center of the optical fiber cladding circle, which is also the center of the core circle, and then establish a coordinate system with this center as the origin of the coordinate system. In VBAI, the function of the function of establishing a coordinate system is a positioning feature. It can automatically locate the center of the cladding circle according to the found cladding circle. Even if the position of the cladding in the image changes, the origin of the coordinate system can be accurately located at the center of the cladding circle, and the center of the cladding circle is the center of the core circle. As shown in Figure 4, the origin of the coordinate system can be accurately located at the center of the core circle at different positions, even on an irregular cladding surface.

2.2 Division of detection area

After locating the fiber core, since the detection standards used in different circular rings on the fiber end face are also different, circular rings of different areas should be made with the center of the fiber core as the center, and then the detection should be carried out in each circular ring according to the specified detection standards. If the detection in any circular ring fails, the fiber is defective and cannot pass. When dividing the area, since the image collected by the fiber end face detector is measured in pixels (pix) in the VBAI environment, and the fiber end face detection requirements given in the general detection standard are measured in microns (μm), it is necessary to convert microns (μm) into pixels (pix) through formula (2). When converting, you need to know a parameter: dpi (dots per inch). Knowing dpi, you can get the conversion relationship between pixels and microns from formula (2). Let P be pixels, D be dpi, I be feet, and M be microns, then:

The dpi of an image can be obtained through some commonly used image viewing software (such as Acdsee, Photoshop, etc.). The dpi of the standard resolution of 640×480 or 800×600 is a constant: 96. In this way, when dividing the fiber end face into regions, the diameter of each ring can be accurately calculated, so that the entire fiber end face can be carefully and accurately detected.

2.3 Defect Detection

2.3.1 Different Thresholds

Fiber end face defects include white spots (chips), black spots (dirt), shadows (internal cracks) and scratches. Chips and scratches are brighter than the fiber end face, while dirt and shadows are darker than the fiber end face. To detect these defects, the original image must be reprocessed before each bright part of each area is detected, and the original image must be reprocessed before each dark part of each area is detected, so as to set different thresholds to distinguish between parts brighter than the fiber end face and parts darker than the fiber end face. In this way, the dark part and the bright part are tested successively during the detection process. If any of the two tests fails, the detection of this area will fail.

The area of ​​the fiber end face that needs to be inspected includes the fiber cladding and the ceramic part outside the fiber cladding, so light defects and dark defects may be distributed on the ceramic surface in addition to the fiber cladding. Since the fiber cladding is dark in the collected image, the color is close to the dark defect, while the color of the ceramic part outside the cladding is closer to the light defect. Therefore, when detecting defects on the fiber cladding and on the ceramic surface outside the cladding, different thresholds need to be set for dark defects and light defects respectively to accurately detect defects on the entire end face. Therefore, when using VBAI to detect the ceramic surface area outside the fiber cladding, it is necessary to re-set the threshold according to the double peak method, as shown in Figure 5.

Figure 5 Different binary treatments for dark defects and bright defects inside and outside the cladding
[page]

As shown in Figure 5, during the detection process, different thresholds must be set for the different distributions of dark defects and bright defects inside and outside the fiber end coating, otherwise the detection accuracy will be greatly affected. It should be noted that the fiber core itself is bright, so the fiber core needs to be ignored during the bright defect detection process.

2.3.2 Defect determination

In the defect detection of optical fiber end faces, there are both unacceptable defects and acceptable defects. For defect particles such as chipping, dirt, internal cracks and scratches, whether they are acceptable depends on their size and length. Generally, their size and length are evaluated mainly based on the size of their Feret diameter. Feret diameter is a commonly used method to express particle diameter. For regular spherical particles, "diameter" can be used to accurately describe their size. However, in most cases, the shape of particles, especially scratches, is not spherical. Using diameter to express is obviously inaccurate and easy to cause misunderstanding. Therefore, the concept of "particle diameter" is used to express particle size. The so-called particle diameter is the "one-dimensional" size that represents the particle size. "Dimension" is also called dimension, which is a unit of measurement for basic physical quantities, such as length, volume, mass, time, etc. For the same particle, due to different application scenarios, the measurement methods are often different, and the obtained particle diameter values ​​are of course different. For example, what is observed under a microscope is the size of the particle on the plane perpendicular to the line of sight, the particle size obtained by screening is the sieve hole size, and the diameter obtained by sedimentation is the diameter of spherical particles with the same sedimentation characteristics, etc.

In the machine vision inspection of fiber end face defects in this article, the Feret diameter of the defect to be measured after the binary image is the size on the plane perpendicular to the line of sight under the microscope. The Feret diameter of any irregular object is large or small. Usually, the maximum Feret diameter is needed, and then it is compared with the inspection standard. If the maximum Feret diameter is larger than the acceptable defect particle diameter, the inspection cannot pass. VBAI is very powerful. It provides a function that can directly measure the maximum Feret diameter (Max Feret Diameter), so that the maximum Feret diameter of various defect particles can be measured quickly and easily, including scratches with linear features. The maximum Feret diameter of the scratch is its length. In the visual assistant function module of VBAI, there is a particle filter (Particle Filter) sub-function module, which can set a certain range of maximum Feret diameter values, and then filter out the defect particles with the maximum Feret diameter within this range, and then make a judgment. For example: in a detection area, it is required that the number of defect particles with a maximum Feret diameter less than or equal to 5μm cannot exceed 5, and there are no defect particles larger than 5μm. Using formula (2), we can calculate that after 400 times magnification, 5 μm is converted into a pixel value of approximately 7.559 pix. Then, using the particle filter function module, first filter out defective particles with a maximum Feret diameter less than 7.559 pix, and use the particle analysis (Detect Objects) function module to detect the number of particles. If particles are detected, it is determined that the test is not passed; then use the particle filter function module again to filter out particles with a maximum Feret diameter greater than 7.559 pix, and still use the particle analysis function module to detect the number of particles. If more than 5 particles are detected, it is determined that the test is not passed.

After inspecting all areas, call the Set Inspection Status function module in VBAI, which has an option "Fail if Any Previous Step Fails" to fail the inspection. If this option is selected, if the inspection of any previous area fails, the inspection of the fiber end face will be judged as failed, so that any inspection area that does not meet the inspection requirements will not be missed.

2.4 Report Generation

As the most intuitive and important evidence for testing test results, reports are an indispensable part of the test system. After each fiber end face inspection is completed, a lot of data will be generated, including the number and size of defects in each inspection area. If these data are automatically imported into Excel or Word files after specifying the file path, it can not only improve the automation of the entire inspection system, but also greatly reduce the workload of testers. Therefore, a data export (Data Logging) function module is added to the end of the program to save the data in Excel format on the local computer hard disk or upload it to the FTP server and save it, so as to improve the security and reliability of the data and facilitate viewing at any time.

3 Conclusion

This paper combines image processing technology and develops a fiber end face defect detection system based on machine vision according to the collected fiber end face images. Experiments have proved that this system can detect and judge the defects of the fiber end face with high efficiency and high quality, avoiding operational errors caused by manual inspection and greatly improving the reliability of detection.

In addition to the application of optical fiber end face detection in this article, with the help of high-tech detection technologies such as infrared, ultraviolet, X-ray, and ultrasonic, machine vision has more outstanding advantages in detecting non-visual objects and high-risk scenes. Therefore, machine vision detection will become an increasingly popular solution.