Under the fluorescent light in the laboratory, dense codes are reflected on the lenses of engineer Zhao Lei's glasses. His fingers slide quickly on the touch screen, debugging the parameters of the latest generation of visual recognition system. "If the difference is 0.1 mm, the seal of the packaging box will leak." He has been repeating this sentence for three years, and now he has finally stabilized the error within 0.3 mm.
This breakthrough stems from the fusion innovation of multispectral imaging technology and deep learning models. The visual inspection of traditional packaging machines relies on a single visible light camera, and the misjudgment rate is as high as 15% when encountering reflective materials or transparent films. The third-generation system developed by ply-pack can penetrate the interference of the composite film surface and accurately capture the three-dimensional contour of the packaged object through synchronous scanning of infrared, ultraviolet and visible light bands.
On the production line of a nut company in Hangzhou, special-shaped products such as cashews and macadamia nuts have caused the packaging machine to frequently "stuck". Traditional mechanical positioning requires manual repeated debugging of the fixture, and each change of shape requires at least 2 hours of downtime. Ply-pack's visual recognition system is equipped with an adaptive positioning algorithm. Just take a photo of the product and the corresponding grasping path can be generated within 30 seconds. "The system automatically identifies the center of gravity and force point of the material, which is more accurate than the experience of the master." When the on-site technical director demonstrated, the manipulator was grasping irregularly shaped chocolate gift boxes at a frequency of 4 times per second, and the packaging qualification rate reached 99.7%. This dynamic calibration capability enables the same production line to be compatible with 12 packaging forms, and the changeover efficiency is increased by 6 times. Vibration, temperature fluctuations, and mechanical wear in the packaging workshop may cause precision attenuation. The R&D team designed a real-time error compensation model for this purpose: when the visual system detects that the robot arm is offset, it will synchronously send compensation instructions to the drive motor, and the whole process takes only 0.08 seconds. After a daily chemical company in Guangdong used this technology, the continuous working time of the filling-sealing linkage production line was extended from 8 hours to 72 hours. "In the past, the machine had to be stopped for calibration every half a day. Now the system can "check for deficiencies and make up for them" by itself, and the overall efficiency of the equipment has increased by 23%." The production supervisor pointed to the green curve on the monitoring screen and said. In the continuously flashing data stream, more than 2,000 sensors are silently building a precision defense line.
In the late night pharmaceutical packaging workshop, the blue scanning beam of the visual recognition system flows like stars. The fully automatic blister packaging machine is running at a speed of 120 plates per minute, and the position deviation of each tablet is controlled within a safe range of 0.5 mm. This GMP-certified pharmaceutical company just reduced its packaging loss rate from 1.2% to 0.3% last month.
"Precision is the lifeline." The quality director sighed as he stroked the laser anti-counterfeiting mark on the aluminum-plastic plate. With the penetration of visual recognition systems in food, electronics, medical and other fields, the packaging machine industry is moving from "extensive packaging" to the "micron-level control" era.
