One of the areas where AI has a big impact is visual inspection of products and materials. Many factories and businesses still rely on human operators for visual control of quality and classification of material parts. These tasks were very hard, or even impossible, to automatize with “classical” approaches. But now, with the power of deep neural networks, we can finally make huge progress in automatizing even these hard cases and save work for human operators.
AI Visual Inspection Application
AI Machine vision algorithms designed by Foxconn 4Tech can automate many
visual industrial activities - even under complex conditions like
variable position and orientation of units, hard lighting conditions and
Artificial Intelligence Visual Inspection
Selected features and customer benefits
Full-fledged, integrated solutions including hardware, software and deep-learning models
Stable classification and detection during full work processes
Faster and highly accurate than human implementation
You will achieve outstanding accuracy for classification
Replaces human visual controls with an automated system
Saves operating time when implementing the solution and re-training components or processes change often
Cuts costs for an optical equipment because it can work with cheaper cameras and in worsened lighting conditions
Boosts process efficiency because it's faster and more thorough than humans are.
Examples of usage
Material classification, parts or the entire product
Deep neural networks are able to recognize materials and products (or their parts) and classify them faster than humans.
Components can be recognized and classified even in positions and orientations where it is impossible to have a clear field of view (i.e. when parts are randomly scattered) or when view thereof is partially blocked. This allows our models to function fully not only on production lines (with fixed anchoring), but also in other environments such as warehouses where lighting conditions vary and image material is captured using a mobile device (smartphones, tablets).
Visual quality control
Visual quality control is a key step in the process for ensuring quality. In many industries, it is still done using trained operators where production throughput and agreement in recognition is limited by the human factor. Of particular importance are checks for missing components, correctness of fittings or connections and cosmetic flaws (scratches, damage, color deviations, etc.).
Given that many of the aforementioned factors are random (i.e. each cosmetic defect is unique), classic machine visualization techniques can lead to inconsistent results (even if the conditions are only slightly changed, the system's reliability is disrupted).
Deep learning offers a unique way to generalize way beyond the boundaries of training data exponents (image material that we provide at the start of implementation). Such a solution is, as a result, much more resistant to new conditions and deals well with previously unrecognizable defects/flaws.
What makes our solution different
We use the best algorithms developed by experts on artificial
intelligence and we draw on their many years' experience with
integrating machine visualization solutions in industrial processes. The
models we design can work reliably in difficult and non-standard
conditions and can, over time and with more collected data,
improve. Part of our solutions is also design and integration of optical
components (cameras, lighting, assembly) in production processes
together with the development of the most modern neural networks for
visual inspection and the accompanying classification software.