Digital Imaging Inspection System for Aluminum Case Grinding Quality Control of Solid-State Drive ^†

Abstract

The enterprise or data center does not use the M2 SATA because of the cooling problem. Therefore, SSDs employ metal cases similar to the traditional 2.5” or 3.5” hard disk. The metal case is made of aluminum, which must be ground after the metal plate forming process. Conventionally, quality control is conducted to check the ground quality of aluminum cases manually. This method is not accurate as the data are difficult to digitize. To improve the quality control, speed, and efficiency. We established a digital imaging-based inspection system for the aluminum case grinding quality control. The inspection system consists of a digital industrial camera, a closed-circuit TV lens, a light-emitting diode (LED) light source, and a personal computer. If the loading and unloading time is ignored, the test time is less than five seconds for one case. When the tested case is uploaded to the inspection system, the camera captures and sends images to the computer. The image was processed to evaluate the quality and record the tested results. Then, the tested case is classified by a robot or an operator.

1. Introduction

Automatic optical inspection (AOI) is a common automatic inspection method. The line scanning technology has the fastest inspection speed, but its system construction cost is relatively high [1,2,3]. In addition to the high price of the linear image sensor, the price of its matching lens is high due to its larger aperture. For high-speed measurement, a dedicated image capture card and workstation computer are required, which significantly increases the system construction cost. Therefore, most general inspections are based on area-scanning image sensors. There are many applications for AOI used in the electronics industry [4], textile industry [5], and production status monitoring [6].

Lim et al. [7] used convolutional neural networks (CNNs) to classify components on printed circuit boards (PCBs). Yang et al. [8] used a charge-coupled device (CCD) camera with a resolution of 2592 × 1944 and a field of view of 15 × 11.25 mm with light sources in six directions to detect surface scratches and flow of O-rings. Defects such as surface scratches, flow marks, and non-fills are detected, and elastomer O-rings with high surface reflection coefficients are identified.

Zhang et al. [9] applied the adaptive threshold gray transformation to image enhancement of defect images on the steel surface and processed subsequent Gabor filter and image segmentation. Li et al. [10] proposed an automatic defect pattern recognition system to search for hidden scratch/line defects in wafers. The proposed method achieved a higher recognition rate than 89% for scratch/line types and 94% for all common wafer defect types.

Șipoș et al. [11] proposed an automated optical inspection system for the detection of PCB mounting defects. The reference image was processed using hue-saturation-value (HSV) color space and image thresholding. Tahir et al. [12] developed a technique using machine vision and unsupervised machine learning for real-time defect detection and localization to support paper production industries in improving and monitoring product quality. The proposed system efficiently detects and localizes unwanted foreign objects in an image for a wide range of object sizes.

Sirvent et al. [13] determined an optimal subset of descriptors for obtaining the best possible classification of defects in semiconductor wafers using scanning electron microscope (SEM) images. A support vector machine (SVM) classifier was trained with all subset combinations of features, and the classification results were utilized to employ an Elasticsearch (ES) for providing the optimal subset of features. Li et al. [14] proposed a machine learning-based method for accurate wafer defect map classification. The proposed method is referred to as TestDNA-E, as it applies ensemble learning based on improved TestDNA features. The proposed method achieves a high hit rate for each defect type and overall accuracy. The TestDNA of a tested wafer consists of the TestDNA sequence (for electric information) and the wafer defect pattern (for spatial information).

2. System Structure

The inspection system includes a digital industrial camera, a closed-circuit TV lens, a light-emitting diode (LED) light source, and a personal computer. The camera and light source are mounted in a specified position by the aluminum fixture. The samples are picked up by a 6-axis robot arm and a vacuum suction cup. The robot arm moves the sample to the specified position and orientation. The sample must be orthogonal to the optical axis of the camera. The CCTV lens is mounted on the camera. The camera is connected to the personal computer by USB 3. When a sample reaches the specified position, the arm controller outputs the trigger signal to the computer. The computer captures one image and conducts an image processing process. After the image is processed, the inspection system outputs inspection results. The computer sends the results to the robot controller. The robot moves the approved and rejected samples to different locations. Figure 1 shows the presented inspection system without the personal computer and robot controller.

3. Experimental Results

Figure 2 shows the original image captured by the industrial camera of the presented inspection system. Figure 3 shows the appearance of the inspection system software. It shows the result after the imaging process and defect identification. Figure 4 shows five sample images. Polish defects appear on the side of the solid-state drive (SSD) case and on the inside of the SSD case. Figure 5 shows the final version of the inspection system software. In addition to displaying the recognition results, arm links are set up by transmission control protocol/internet protocol (TCP/IP).

4. Conclusions

We established an inspection system for SSD aluminum case inspection. It was used by the company in their production lines. The system saves time and costs for inspection. It enables the digitalized inspection result, and the test time is less than five seconds for capturing images, image processing, and recognition.