Fast and Precise Surface Inspection of Smart Devices Uses Deep Learning
Griffyn Robotech, headquartered in Pune, India, specializes in visual inspection, robotics and AI-based automation. With industrial manufacturing and quality control solutions, Griffyn serves the automotive, telecommunications, pharmaceutical, FMCG and machine tool industries. The classification of cell phones and tablets according to optical defects determines the necessary measures for reprocessing. In Griffyn Robotech’s inspection machine, for both new and used smart devices, the AI expert uses PC-based control from Beckhoff to precisely control product handling enabling micrometer-precise detection of surface defects in fractions of a second.
Due to easy upgrade options, warranty and insurance coverage, the reverse supply chain of smart display devices such as smartphones, tablet PCs and wearable electronics is increasing day by day. Visual inspection or evaluation is one of the most important steps in processing returned devices, as the defects discovered during this process determine further treatment. Until now, the search for scratches and other damage was usually carried out by manual visual inspection. But while nothing beats the versatility of the human eye, its repeatability and productivity are limited. Additionally, millions of devices must be inspected and evaluated each year before being reintroduced into the supply chain.
To meet the increasing demand in reverse logistics and eliminate the subjectivity of human inspectors, Griffyn Robotech developed the Deepsight Cosmetic Grading Machine (CGM). The patented vision system enables the quick and accurate detection, measurement and analysis of all surface defects with high repeatability. It identifies defects intelligently detects scratches while tolerating natural variations in complex patterns and surface textures, including shiny or rough surfaces. Significant differences in the tolerances that apply to the make and model of the devices are also taken into account.
Efficient Handling Reduces Cycle Times
To determine the ‘true’ quality of a handheld, the machine uses a high-resolution camera sensor. To do this, it takes multiple images from all six sides of the device. However, the basis for accurate image capture is appropriate product handling for turning, rotating and positioning the devices . For precise and fast motion control with precise object orientation, Griffyn uses the TwinCAT NC I motion control software on the ultra-compact industrial PC C6015 from Beckhoff. Combined with the AM81xx servo motors and the EL7211 servo motor terminals, this enables complex interpolated multi-axis movements The ultra-fast EtherCAT communication and sub-millisecond processing times reduce machine cycle times and increase throughput, according to Griffyn.
Deep Learning Algorithms
The captured images are processed using Griffyn’s proprietary deep learning algorithms to identify various surface defects such as scratches, cracks, dents and discoloration. After analysis, the machine produces a detailed report that includes raw images as well as an edited image with defects made clear to the human operator. While the human eye can detect scratches over 80 µm wide on the surface, Deepsight also identifies defects as small as 40 µm wide and 3 µm deep. The report includes information about the number of scratches on the device, the length of the largest scratch, and the depth of the deepest scratch.
The machine easily integrates into supply chain management systems, where the classification data controls downstream processes and the final treatment of the device. Smartphones with defects within acceptable ranges are sent for a polishing process. For example, devices with scratches that are less than 15 µm deep can usually be restored properly through grinding and polishing. The high accuracy of Deepsight’s data enables a reliable statement to be made as to whether simple reprocessing is promising or more intensive rework is necessary.
The Deepsight sorting machine is available as a standalone device with manual product feeding or as a combination of two to four machines with automatic product feeding via a robotic arm. With an inspection cycle of less than a minute, the machine in the 4-machine configuration delivers a total output of four inspected devices per minute. Precise motion control technology from Beckhoff contributes to a machine availability of 95% and a throughput in the range of 200 devices per hour.
For more information: www.griffyn.io
