foxconn, NxVAE
Source: Shutterstock

Foxconn announced today that the unsupervised learning AI algorithm called the FOXCONN NxVAE has improved quality control on the factory floor. It is currently applied to some of their production lines and has resulted in reducing 50% of defect inspecting laborers.

It took the AI team 8 months to develop and it is worth it

According to Foxconn, the current defection rate is lower than 1%. The algorithm FOXCONN NxVAE is able to thoroughly inspect 13 most common unit defects, reaching clients’ demands for zero-miss results.

The unsupervised learning method is comparably more effective in spotting faults than supervised learning which would require hundreds to thousands of error images to train the model beforehand and can only achieve 90% precision.

This is where Foxconn NxVAE comes in. It takes in what a normal unit looks like to train the algorithm, the data comes in handier and minimizes the time pressure for quality inspection.

The accumulated data from the factory floor is colossal, it can continuously feed the training data ensuring the system to be truly efficient. By optimizing the inspection procedure with AI, the transforming period of introducing new products into their production line can also be eased off.

Terry Gou, Chairman of Foxconn, sought to transform the contract manufacturer into using AI, automation and big data solutions dating back to 2018. The AI team took 8 months to build AOI infrastructures and collect product images (data).

Now, three years later, the results can clearly be seen.

AI can improve quality control in not so subtle ways

Quality control has become a key differentiating aspect in the increasing competitive scene of manufacturing and AI has a strong role to play in it. Any defection can easily result in dissatisfied consumers and be time-consuming for error correction.

While typical computer vision needs each fault pre-programmed before catching errors, an algorithm can do the jobs of many in real-time that saves money and time. Fueled by data and machine learning, AI image processing in the production line allows quick quality inspection and is more reliable than fallible manual inspection.