FAQ

Firstly, the customer needs to inform us of the products and their styles that need to be tested, as well as the types of defects that need to be tested. It is best to provide pictures so that we can evaluate the cost budget of the testing system. If the customer decides to purchase, they need to prepare all defective samples for us to confirm the final testing system. After the customer agrees to the final testing plan, we can sign the contract, pay the advance payment, and start production. The general delivery cycle is 4-8 weeks, and before shipment, we will conduct equipment operation testing and algorithm addition.

After the equipment is delivered to the customer's site, we will arrange for engineers to come and install it on site. At the same time, we will provide systematic training to the customer's engineers. In the later stage, we have a dedicated technical support team to solve all system software problems for the customer through the AI cloud platform. We can work 8 hours a day, 7 days a week, and reply within 15 minutes during working hours and 1 hour during holidays.

Compared to traditional manual detection, machine vision inspection has the characteristics of faster speed, better detection stability, and easier management. Our detection machine can work 24 hours a day, and the parameters can be set through software. The detection results will not be affected by various factors. The current detection algorithms have transitioned from full customization to a semi customized and semi standard state, and the cost of visual inspection equipment has been greatly reduced. The daily work efficiency of the equipment can be equivalent to 4-5 manual workers. From controlling product standards for enterprises to reducing production costs, to providing automation standards for companies, using visual inspection machine is the best choice.

Currently, the vast majority of visual inspection software in the world wide market are traditional algorithms, not AI algorithms, commonly known as forward comparison. By inputting good product data and comparing it with other products, those that are different are considered non-conforming products. Traditional algorithms are algorithm standards established based on the first generation of measurement data, which detect product matching and adapt to software. The software has pre-set various defect modules, and on-site operators circle different defect modules to different positions of the product for detection. As the detected product moves rapidly through the detection camera, the detection position may deviate to varying degrees. Therefore, the software needs to continuously adjust the detection position on the detected product through circle drawing to provide detection accuracy. The adjusted parameters are complex, the operation is locked, and the technical requirements for operators are high. It is also susceptible to interference, misjudgment, and missed detection due to the influence of the product's external structure.

Technical advantages:
Suitable for measurement and positioning needs, with high measurement accuracy;

Technical disadvantages:
1. The program parameters are numerous and complex, and there are often correlations between parameters, which heavily rely on the experience of professional debugging personnel;
2. Developing algorithms specifically for special needs requires a long development cycle and cannot quickly add algorithm models;
3. There is no fast algorithm for complex requirements, resulting in low computational efficiency and a single type of defect detection;
4. High requirements for operating conditions and poor environmental adaptability;

Keye AI algorithm is a vision system solution independently developed by the USTC doctoral team, which has the advantages of few training samples, low annotation volume, strong compatibility, fast operation and deployment speed, etc. At the same time, we have added forward training models according to different scenarios, which is suitable for use in scenarios with high product consistency and fixed detection requirements. That is what we call the forward and reverse AI dual training model.

Technical advantages:
1. Accuracy of results: Compared with traditional visual algorithms, deep learning algorithms have greatly improved stability and can adapt to general disturbances in the environment and background. At the same time, the accuracy of the algorithm is also higher than traditional algorithms;
2. Algorithm universality: For different defects, only their samples need to be collected, and after sufficient training, different defect samples can be automatically identified through deep learning, meeting the needs of large-scale training samples for deep learning. Theoretically, there is no upper limit requirement for the types of defects that can be detected;
3. Development timeliness: As there is no need to develop algorithms for different defects, the entire development cycle is greatly shortened, and creating a new model only takes about 15 minutes at the fastest;
4. Environmental reliability: Due to the guaranteed computing power of NPU, the entire system operates in a state with sufficient computing power margin, and can work for a long time in high temperature environments without easily crashing..

KeyeTech is the first company who applied AI algorithm into visual inspection system in plastic packaging products field since 2020, till now, we have two training modes, defect training mode and good product training mode, it is the most flexible and convenient AI system in the world, only 15min to create a new program and compatible for more different defect types.

KeyeTech is a branded manufacturer & designer of Industrial product visual defect detection, like plastic packaging products, plastic cap, closure, bottles, preform, cups etc, adopted with the latest self-developed AI algorithm system, small model AI training mode, high computing power, providing standard/customized AI vision inspection system for worldwide customers.