Advanced deep learning algorithms accurately identify and classify defective products. They deliver more consistent and reliable results than traditional visual inspections.
Advanced deep learning algorithms accurately identify and classify defective products. They deliver more consistent and reliable results than traditional visual inspections.
All AI hardware and software are provided turnkey at a low cost, allowing you to conduct quality inspection projects on your own without the expense of hiring experts. The camera can be any low-cost, high-quality HDMI camera, maximizing cost-effectiveness.
All AI hardware and software are provided turnkey at a low cost, allowing you to conduct quality inspection projects on your own without the expense of hiring experts. The camera can be any low-cost, high-quality HDMI camera, maximizing cost-effectiveness.
Deep learning training is possible with just a mouse click, without any specialized knowledge. The intuitive interface makes it easy for anyone to implement AI quality inspection.
Deep learning training is possible with just a mouse click, without any specialized knowledge. The intuitive interface makes it easy for anyone to implement AI quality inspection.
No separate computer is required for the quality inspection process. Simply connect the camera to a monitor and you can immediately integrate it into your production line.
No separate computer is required for the quality inspection process. Simply connect the camera to a monitor and you can immediately integrate it into your production line.
Image Capture Steps
Access the Deep Runner control panel via an internet browser on your management laptop. Click the capture icon at the top of the screen to enter image capture mode. There are two capture modes: automatic and manual. Manual capture allows the operator to specify the timing for capturing images, while automatic capture automatically captures images when the production line is stopped.
Click the hat icon on the left to open the options window. Selecting the manual capture option will display the capture button in image capture mode. Each press of the capture button will capture and save the image. If automatic capture is selected, the camera measures changes in input and captures the image 0.5 seconds after no change is detected.
Manual Classification
Run WinSCP on your PC, log in to Deep Runner using the admin account, and then move the product image files stored in the capture folder to your PC. The moved files are then (1) created in a user-defined data home directory, (2) subfolders are created under that directory for the categories to be categorized, and (3) manually distributed within the subfolders. For example, if the subfolders are Good and Bad, images of normal products are moved to Good, and images of defective products are moved to Bad. As another example, if the subfolders are Good/Crack/Stain, images of normal products are placed in Good, images of cracked products are placed in Crack, and images of rusted products are placed in Stain. The classification type here is determined by the user and is the same as the categories that the AI automatically classifies when performing the AI automatic quality inspection later. The number of image data should consist of independent product images for each category, and a minimum of 1,000 images is required, but the more the better.
Deep Learning Training
Install and run the Deep Trainer software included with the device on the PC where the data is stored. Click the "Tensorflow Training for Deep Runner" button in the upper left corner of the main screen, then click the Set button in the upper left corner of the new window. Then, select the home folder for the manually classified data. Then, click the Start Training button in the lower left corner to automatically begin deep learning training. Training can take several hours or even tens of hours, depending on the amount of training data and whether a GPU is installed. Once training is complete, it can be used on multiple devices indefinitely, so it's important to invest sufficient time. Once training is complete, the message "Training successful" will appear on the screen. Clicking the "Open Folder" button in the "Training Process" section of the window will reveal the trained drmodel.par file. Change the file name to suit your project.
In the Deep Runner control panel, click the brain icon on the left, then the "Add PAR" button in the upper right corner. Select the renamed par file to register it on the device. Next, click the hat icon on the left side of the control panel and, in the "Deep Learning Model" section, specify the model you just registered. This completes the quality inspection process.
Automatic AI Quality Inspection
Once the model setup is complete, quality inspection is performed based on what's learned from the camera images. The camera must be positioned in the same position as the image capture stage. Two recognition methods are available: real-time and timing recognition. Real-time recognition is a mode where the deep runner continuously performs defective product recognition and outputs the results in real time. Timing recognition mode performs recognition only at specified times. In this mode, two modes are available: manual and automatic. Manual recognition allows the operator to select the timing for recognition, while automatic recognition automatically captures and recognizes when the production line is stopped.
