As an essential step in the manufacturing process, surface defect detection is widely used in various industrial fields, including 3C, semiconductor and electronics, automobile, chemical industry, medicine, light industry, military industry and other industries, which has spawned many upstream and downstream enterprises. Since the beginning of the 20th century, surface defect detection has roughly experienced three stages, which are manual visual detection, single electromechanical or optical technology detection and machine vision detection.

Manual visual method is the earliest and the most widely used. Although advanced detection technologies such as artificial intelligence and machine vision are gradually mature, relying on naked eyes for defect detection still accounts for a large proportion and is widely used in small and medium-sized enterprises. According to statistics, more than 80% of the current industrial surface defect detection still relies on manual testing, and the number of manual testing workers on the product line every day exceeds 3.5 million. Manufacturing enterprises represented by Foxconn, Bonn Optics, etc., recruit a large number of quality inspection workers to carry out testing in the form of assembly line. However, with the disappearance of demographic dividend, as well as boring jobs, low freedom and low salary, fewer and fewer people are willing to engage in quality inspection, and the problem of labor difficulties becomes more and more prominent.

From the perspective of the current development trend, machine vision and other advanced detection systems will gradually replace manual detection, which is mainly because manual detection has the following shortcomings:

High labor intensity, poor stability and consistency of testing

Manual testing requires workers at fixed stations to observe products with naked eyes to determine whether there are defects. Long time detection work is easy to cause damage to the human eye, especially in the detection of glass, metal and other strong reflective object surface. The scene of defect detection in metallurgy, rail transit, machinery manufacturing and other industries is noisy, heavy smoke, high risk, long-term poor working environment will have a bad impact on the physical and mental health of workers. Affected by the emotional fluctuation of workers, technical level, judgment standards, individual differences and other factors, the stability and consistency of the same batch of products can not even be guaranteed, making the product quality fluctuate and uneven.

Low automation, low production efficiency

Since the upper limit of quality inspection efficiency of workers is low, and labor costs are increasingly high, enterprises generally adopt sampling detection strategy to ensure production efficiency. Assessing the quality of a batch of products by randomly selecting a small number of products is far less rigorous than comprehensive testing. Therefore, it is easy to fall into a contradiction when conducting defect detection manually: quality control and production efficiency cannot be both. This contradiction is particularly prominent in the surface defect detection of fabric, strip, film and other products. This kind of products are mostly high-speed continuous production, when the production speed is higher than 3m/s, it is difficult for the human eye to distinguish the defects.

It is difficult to form lean production

Quality is made, not detected. Testing is only an afterthought, which is costly and does not guarantee that there will be no errors. Many tests require not only judging the appearance quality of the product, but also recording and counting the location, size and quantity of defects. Traditional manual testing uses paper and pen to record quality inspection results, and the testing data is incomplete and scattered, which cannot form valuable feedback information to guide lean production.

Recruitment is difficult, labor is difficult, training is difficult, cost is high

Low pay, long working hours, labor intensity and so on directly affect the stability of recruitment. More and more young people would rather deliver food than work in factories, making it difficult to recruit workers for the traditionally labor-intensive job of defect detection and a serious brain drain among trained skilled workers. The trend of the disappearance of demographic dividend is irreversible, the cost of employing people keeps rising, and the online automatic defect detection system has changed from "optional" to "required".

In order to take advantage in the ever-changing and increasingly competitive market, enterprises should not only continuously improve product quality standards to meet customer needs, but also continuously improve the efficiency of production lines to adapt to the fast pace of the market. The use of automatic and intelligent surface defect detection method is an important means for both quality and efficiency. The implementation of automated surface defect detection system has the following technical difficulties:

The technical difficulties caused by the defect itself

The complex types of different defects and the large differences of similar defects bring great difficulty to the detection.

The complexity of different defects is mainly reflected in three aspects. First of all, there are large differences between categories, and the appearance defects of industrial products are complex and diverse. The morphological characteristics of defects of different categories may vary greatly, which leads to the low universality of detection algorithms, and many defects need to be independently developed with high development complexity. Secondly, the ambiguity between classes is large, and the ambiguity between classes is the other extreme of the large difference between classes. That is, the apparent characteristics of different classes of defects are similar to a certain extent, so it is difficult to distinguish the types of defects, which makes it impossible to accurately judge the causes of defects and accurately grade products. Third, the background is complex, so it is difficult to completely separate the defect from the background in the production scene, and the defect characteristics are not obvious.

The large difference of defects in the same category indicates that for defects in the same category, the size, contrast, gray value and other apparent characteristics of defects in the same category show great changes due to the influence of lighting conditions, different production batches, equipment status and other factors in the production process, and the defect characteristics do not follow the same distribution.

The technical difficulties brought by the detection system

The surface defect detection system is generally composed of various sensors such as mechanical motion, electrical control and vision, and the system tends to be customized, which is difficult to replicate in batches across fields. When designing the testing system, the designer should not only understand the testing system itself, but also be fully familiar with the characteristics of the tested products and the processing technology of the products. Only on this premise can the feasible and reliable implementation scheme be put forward.

The final detection effect of the system may be affected by every link of the system. Taking the detection system based on machine vision technology as an example, the consistency of workpiece position, the stability of lighting, the matching degree of camera and lens, and the effectiveness of detection algorithm will directly affect the quality of image acquisition and the application performance of the detection system, which needs the cooperation of machine, electrical, vision, sensing and other systems. There are many difficulties in the most basic lighting, such as which scenes need diffused light, scattered light, direct irradiation, low-angle irradiation or backlight irradiation, and how to light undeveloped surfaces such as spherical surfaces, cambered surfaces and inner cavities, etc.

At the same time, the complex and harsh external environment such as high temperature, high humidity and dust also put forward higher technical requirements for the integration and protection ability of the detection system. System development enterprises need to study the system integration technology of surface defect detection equipment, explore the data acquisition system to overcome the high temperature and humidity in the field environment and the system protection technology against external interference.