在3CAppearance detection scenario中，There are many kinds of products, a large number, fast update and iteration, and the demand for accuracy, flexibility, intelligence, cost control and other aspects of appearance testing is strong, but there are many pain points in traditional testing methods:1.The accuracy of traditional visual detection is poor。Manual detection is greatly affected by subjective factors and visual fatigue, and wrong detection and leakage detection occur from time to time, which seriously damages the brand efficiency and economic interests of enterprises。2. Low degree of flexibility, efficiency improvement is difficult。The traditional detection method relies on the crowd tactics and part of the traditional machine vision, can not quickly adapt to the market's high requirements for production efficiency, can not flexibly adapt to the enterprise order production plan。3. Data value is not released, and the degree of intelligence is low。Traditional detection data has not been effectively preserved and utilized, resulting in the repeated construction of some big data systems, but did not play their due value。4. Labor costs continue to rise。With the difficulty of hiring and recruiting workers, labor costs continue to rise, and many enterprises have been overwhelmed。

针对3CThe pain points and difficulties in the detection of appearance defects of industrial products, based on multi-axis linkage control and artificial intelligence technology, launched a general solution of soft and hard integrated intelligent inspection。ACOI Automatic Cleaning and Optical Inspection (Automatic Cleaning and Optical Inspection), developed by Huiyan Technology, consists of feeding, cleaning, testing, cutting and other links.Includes model training and prediction services,OK/NG automatic identification of incoming materials is carried out by AI deep learning algorithm，The product is aimed at the problems of poor accuracy and low efficiency of the appearance detection of 3C parts，Creative use of one-stop intelligent solutions，Break through the bottleneck of manufacturing appearance inspection automation，Greatly improve the detection efficiency，Liberating quality inspection manpower and optimizing cost，Solve the bottleneck problem of manufacturing appearance inspection automation。

In addition, through the Insight technology self-developedKai's DashboardAnalysis and detection ability and effect, which mainly involves the following parameters:

TermNeighbor Greyscale Difference (NGD), NDS (NG Defect Sample), ONS (OK Noise Sample)。

4 Actions 4-Actions：

Optical Capability: Raise NGD Optical capability: Raise NGD

Algorithm Capability: Reduce AE%AO% Algorithm capability: Reduce AE%AO%

Cleaning: Reduce ONS. Cleaning: reduce ONS

Defect Threshold: Define a larger defect threshold that reduces the inspection requirement

Indicator KPI：

OE% (Optical Escape)=NGD[0,39]/NDS

Algorithm Escape (AE%)=NGD[40,80]/NDS Optical Overkill (OO%)=NGD[81,255]/ONS

AO% (Algorithm Overkill)=NGD[40,80]/ONS。

Through Kai's Dashboard, you can quickly analyze the inspection and testing capabilities of the equipment, improve the accuracy of the appearance inspection of parts and components, and provide factory production efficiency。

Performance advantage:

1. High speed fly shot Support fast moving uninterrupted shooting, can flexibly and quickly obtain the detailed image of any surface of the three-dimensional workpiece at any position, each point only0.2秒

2. Flexible control design Highly flexible combined light source configuration, with coaxial, circular, low Angle, spherical integration, backlight and photometric three-dimensional light source functions, surface appearance defects are not hidden

3. Intelligent algorithms based on deep learning Artificial intelligence algorithms based on deep learning adapt to complex products and conditions and accurately distinguish between real defects and inherent features of products, as well as interference features caused by the environment。At the same time, by sharing highly consistent data from multiple sites and machines, the small sample size and model migration problems of deep learning industrial applications can be overcome。

4. The replacement only needs to change the fixture to avoid repeated investment Small body, large content, can be compatible with the same field of vision of a number of products。Product replacement only need to design the corresponding fixture, without equipment transformation, greatly saving hardware input cost。

Industry application:

Product Specifications: