Wizer DP (Deep Learning)
The whole industry has been very excited over the 2 key terms “deep learning” & “artificial intelligence” and there have been many new applications the industry are using this technology for gaming, security & surveillance. Some of the applications include face recognition, number plate reading, behavior characterization, etc. Deep learning has been used for the industrial applications for reading OCR & defects inspection. It has been proven that deep learning will be able to recognize a defect even if the production parts has some different variation on the threshold or machining marks.
The whole idea of deep learning is to be able to train sufficient sample size of good parts, and “marking” the defects area on the NG parts. During the production, new group of NG parts can be marked & added to the existing database to improve the inspection capability.
We offer our deep learning software, Wizer DP and has successfully provided solution for the hard drive, semiconductor, display & PCB industry.
Wizer DP comes in 3 modules, namely Trainer, Detect & Classify. The Trainer module is needed to build up the recipe by training selected good images & different group of NG images. The Detect module is used to detect the defects on the production parts, and the Classify module is used to classify the defects type. However, it is important to know that the classification will never be 100% accurate as there will always be an overlap on the defect characteristics.
The images can first be trained offline to build up the recipe for the inspection database. This recipe can be used for the actual inspection to detect & classify the defects accordingly.