Kit media
We talk a lot about artificial intelligence, Machine Learning and especially Deep Learning. How and why does machine vision based on these different technologies arrive in the industry? What do they bring and what are their advantages compared to traditional solutions?
In the manufacturing industry, so-called classic machine vision solutions have existed for several decades, where quality controls are necessary and very present. They are used to make presence/absence checks, measurement or positioning when the objects are known in advance and their differences are perfectly identified.
These vision solutions are based on rules of distinction for a defined defect and therefore it is necessary to multiply these rules when defects multiply. This can become unmanageable when the product in manufacture has a very large number of possible variants without it being considered a quality defect.
The contribution of Deep Learning in machine vision
The advantage of deep learning is that we no longer need to provide these specific rules of distinction, but we will provide significant examples.
Using examples, i.e. images taken on the production line, we perform neural network training, which will then allow us to be able to analyze defects or products that we have never seen.
The approach is therefore to tell the algorithm that this or that image corresponds to this defect or perfect product, so that it then finds by itself how to process this task to provide correct results.
A neural network training is therefore carried out upstream of use in the product, in the Cloud or in the camera directly, to then allow classification or anomaly detection for example, which are two of the principles of Deep Learning.
Deep Learning as a relay of traditional machine vision methods
Classic machine vision already meets a large number of applications. Many technological bricks exist, such as shape searches or "matching" for localization, can be used upstream. Deep learning tools complement traditional machine vision tools when they become insufficient on their own.
Because the combination of these different methods makes it possible to solve applications that were practically impossible to do before.
For example, controlling products or parts made of "living" matter, such as wood, or controlling products in the field of food, was all a challenge that we can meet quite easily today!
In addition to solving complex applications, the configuration effort and skill level required are really very low and accessible to everyone. To detect anomalies on a product for example, a mechanical installation of a camera, an image adjustment (often automatic) and a sorting of the images received (in very large numbers <20 images) sont les trois uniques étapes pour réaliser des applications.
The fields of application and industries are very varied, and where traditional methods do not work, the addition of artificial intelligence gets results.
To learn more and how easy it is to implement a quality control solution through deep learning, watch the webinar conducted by SICK by clicking here.
