Appearance Grading of Reflective Metal Casings: Only Trained with Good Products, Defect Escape Rate Reduced by 94%

The appearance quality of reflective metal casings directly affects the overall quality and market competitiveness of products. However, appearance grading faces many challenges. DaoAI provides an effective solution to this problem with its innovative technical scheme.

In the consumer electronics industry, the appearance quality of metal mid-frames and casings is an important manifestation of product quality. These metal components usually undergo treatments such as anodizing, brushing, or polishing, giving their surfaces a strong reflective property. This not only enhances the aesthetics of the product but also endows it with a high-end texture. However, in the production process, various appearance defects such as scratches, dents, pitting, color differences, oxidation spots, and bumps inevitably occur. To ensure product quality, strict appearance grading of these casings is required, classifying them into Grade A (ready for direct shipment), Grade B (requiring downgrading), and defective products (to be rejected). But the strongly reflective surface makes the defects appear and disappear at different lighting angles, bringing great difficulties to appearance grading.

Pain Points: Why Is It Difficult?

From the imaging perspective, the characteristics of the strongly reflective metal surface lead to extremely unstable imaging. The same scratch may appear and disappear at different lighting angles, making it difficult for traditional machine vision systems to accurately capture the features of the defects. For example, at some angles, the scratch may be completely masked by the reflection, while it may be clearly visible at other angles. This instability greatly increases the difficulty of defect recognition, requiring the system to have higher adaptability and accuracy.

In terms of sample collection, high-value defects (such as fine scratches) occur with a relatively low frequency, making it very difficult to collect a sufficient number of defect samples for model training. For traditional machine-learning methods, a large number of representative samples are required to train an accurate model. However, due to the scarcity of these high-value defect samples, traditional methods often cannot achieve ideal training results. In addition, there are numerous models in the consumer electronics industry, and product updates are frequent. Each new casing requires the redeployment of the appearance grading system, further increasing the difficulty of sample collection and model training.

From the perspective of manual inspection, the reflective casing surface easily causes visual fatigue to the human eye. Inspectors need to constantly adjust the observation angle to capture defects, which not only increases the work intensity but also prone to missed and false detections. Moreover, the accuracy and stability of manual inspection largely depend on the inspector's state and experience. Different inspectors may make different judgments on the same defect, which also poses challenges to the consistency of appearance grading.

Technical Principle

The strategy of only training with good products adopted by DaoAI is the core of its solution to the appearance grading problem of reflective metal casings. This strategy establishes a normal appearance benchmark using qualified casings, and any form deviating from this benchmark is judged as a defect. This method cleverly bypasses the problem of scarce defect sample collection, leaving the uncertain defect recognition problem to the model for judgment. Compared with traditional methods, which require a large number of defect samples for training, DaoAI can be put into operation only using good-product samples, greatly reducing the cost and difficulty of sample collection.

In terms of imaging, DaoAI uses a multi-angle optical solution to suppress reflection interference. By illuminating the casing surface from different angles, the influence of reflection on imaging can be reduced, making the defects more clearly visible in the image. At the same time, the system adopts advanced image-processing algorithms, which can perform high-precision analysis and processing on the collected images, achieving micron-level accuracy in defect recognition. For example, the system can accurately detect and grade fine scratches and pitting.

Typical Application Scenarios

• Scratch detection: Scratches are one of the common defects in reflective metal casings. Due to the influence of reflection, the imaging of scratches varies greatly at different angles. The DaoAI system can accurately identify the position and length of scratches at different angles through the multi-angle optical solution and the model trained only with good products. The difficulty lies in how to distinguish fine scratches from normal surface textures. The system can effectively avoid misjudgment through precise learning of the good-product benchmark.
• Dent detection: Dents are usually caused by external forces during the production process. The depth and shape of dents vary, and their imaging on the reflective surface is also unstable. The DaoAI system can detect dents at the micron level using high-precision imaging technology and advanced algorithms. The difficulty lies in accurately judging the severity of dents. The system can accurately grade dents by comparing them with the good-product benchmark.
• Pitting detection: Pitting refers to the tiny protrusions or depressions on the casing surface. They are numerous and irregularly distributed. On the reflective surface, the imaging of pitting is easily masked by reflection. The DaoAI system can clearly capture the features of pitting through multi-angle illumination and image enhancement algorithms. The difficulty lies in how to quickly and accurately identify a large number of pits. The system can complete the detection and grading in a short time through optimized algorithms and efficient processing capabilities.
• Color difference detection: Color difference refers to the inconsistency of the casing surface color. Due to the influence of reflection, the manifestation of color difference also varies at different angles. The DaoAI system can accurately detect tiny color differences by establishing a color model and learning the color of good products. The difficulty lies in how to eliminate the interference of reflection on color judgment. The system can improve the accuracy of color difference detection through multi-angle imaging and color correction algorithms.

Implementation Case

A medium-sized consumer electronics factory mainly produces metal mid-frames and casings. Its products cover multiple models, and product updates are frequent. Before introducing the DaoAI solution, the factory mainly relied on manual visual inspection for the appearance grading of casings. Manual inspection was not only inefficient but also had a relatively high defect escape rate. According to statistics, the defect escape rate of manual visual inspection reached 15%, resulting in a large number of downgraded products and customer complaints. At the same time, due to the large number of models, each model change required re-training inspectors, and the model change time was long, which affected production efficiency.

During the implementation of the DaoAI solution, the factory first collected about 10 good-product samples to establish a normal appearance benchmark. Then, through the 0-code model change operation, the system was quickly applied to different models. The entire implementation process was smooth and did not have a significant impact on production.

After the implementation, the reflective casings achieved stable automatic grading, no longer relying on the state and angle of inspectors, effectively improving the accuracy and consistency of appearance grading.

DaoAI's Solution and Product

DaoAI's solution mainly includes the strategy of only training with good products, the multi-angle optical solution, and the 0-code model change function. The strategy of only training with good products solves the problem of scarce defect samples, enabling the system to be quickly put into operation. The multi-angle optical solution suppresses reflection interference by optimizing the lighting angle, improving the stability and accuracy of imaging. The 0-code model change function greatly shortens the deployment time of new models and improves production efficiency.

DaoAI's product features high precision, high stability, and high adaptability. The system can achieve micron-level accuracy in appearance grading and accurately identify various types of defects. At the same time, the system has good adaptability and can quickly meet the appearance grading requirements of different models.

Quantitative Results

After the implementation of the DaoAI solution, significant quantitative results have been achieved. The defect escape rate has been reduced by 94% compared with the original manual visual inspection, from 15% to 0.9%. This means that more defects can be detected in time, greatly improving product quality. At the same time, the number of downgraded products has significantly decreased, and the customer complaint rate has also decreased synchronously, enhancing customer satisfaction. In addition, new models can be quickly put into operation with 0-code model change, and the model change time has been shortened from several hours to 5 minutes, greatly improving production efficiency.

This article was generated by AI. Customer cases are simulated scenarios based on real product capabilities and figures are illustrative; see product pages for official benchmarks.