Skip Navigation Links.





Aluminium extrusion profile defect detection



1 Aluminium extrusion profile defect detection

1-1System features

Aluminum extrusion workers set extrusion rates as high as possible to increase productivity. At that time, look at the surface of the aluminum profile and check for any abnormalities. This can only be done by skilled workers. Therefore, we created a computerized extruded material defect detection system using deep learning so that anyone could increase their productivity. The system has the following features: 1 Collection of profile images during automatic / manual extrusion. 2 Import prediction model file by deep learning. 3 Automatic / manual extrusion profile defect detection. Deep learning application areas include voice, images, and languages. For images, Google in 2015 and Microsoft in 2016 declared that Deep Learning has exceeded human discrimination. The area where deep learning is best at image recognition. Therefore, it is reasonable to use deep learning to detect anomalies in images at the factory. As an application to the control of aluminum extruded material defect detection, it is conceivable to incline the set extrusion speed when detecting the signs of defects.
The platforms for creating predictive models include the following in the cloud: Azure Machine Learning, Amazon Machine Learning, Bluemix IBM Watson, Google Cloud Platform. The framework for creating a program is as follows. CNTK, TensorFlow, Caffe, Chainer, Keras, Theano Predictive models can be read as files. By accumulating image data, the new model has a better defect detection rate. Replacing it with a new model will improve your productivity.

1-2 System configuration

A surveillance camera compatible with IP network is required. The surveillance camera performance is below. 1 lens type: Optical 20-30x zoom autofocus lens. 2 Two million pixels. 3 Remote control available (smartphone / PC). 4 PTZ mode: preset / scan / patrol / tour. SUNBA 406-D20X-PoE was used to test the program. Because it zooms and takes an image, it is necessary to pay attention to the shutter speed (frame rate) of the camera at high product speeds. For 405-D20X-PoE, the shutter speed is indicated in the specifications. Shutter control: 1/1 ~ 1 / 30000s. This product is safer.

Figure 1-1 A surveillance camera has been added to the RcdWin, EM, EDA program system. Since the surveillance camera has a large amount of communication, we added a USB LAN adapter exclusively for the surveillance camera.

↑TOP of Page

2 Aluminum extruded part defect detection screen

Usually, I think that the program will automatically collect good images. If you get a defective product, I think that you will collect images manually. In the case of deep learning, it is not during extrusion, and storing an image of a defective product after extrusion with a camera should be fine as data for predictive model creation. This is because extrusion workers can recognize defects as defects during and after extrusion. The screen has the following functions.

  • Display of extruded profile (the photograph is not an actual factory extrusion site).
  • Automatic / manual defect detection setup.
  • Automatic / manual image acquisition settings.
  • Selection of prediction model.

Figure 2-1 Aluminum extruded part defect detection screen (This is not an image during extrusion)

↑TOP of Page

3 Image acquisition with iSpy plug-in

iSpy is free surveillance camera software. By creating an iSpy plug-in, you can extend the functionality of the software. With a self-made plug-in, only the 400 x 400 pixel square area is transferred to the extrusion management program. The plug-in can reduce the reflection of light in the image and can process the image so that there is no problem as deep learning data.

Figure 3-1 Display video of surveillance camera with iSpy. Use the plug-in to send a square image to the extrusion management program. (This is not an image during extrusion)

↑TOP of Page

4 Adjusting the surveillance camera

You can zoom and change the direction of surveillance camera images by accessing the IP address of the surveillance camera in IE. Camera control is easier to use on the camera web than on iSpy.

Figure 4-1 Access the IP address of the surveillance camera with IE. You can adjust the camera. (This is not an image during extrusion)

↑TOP of Page