May 14, 2021
As with all tools based on Deep Learning, the performance of solutions for automated visual quality inspection depends directly on the quality of the model training. In order to achieve accurate results, you have to provide relevant data for your system to learn. To be able to recognize defects in the manufacturing process an AI-based visual quality assurance solution requires images of the objects for inspection. The images are used to train the Deep Learning model. When training begins, the object’s features are analyzed and categorized. After the training is complete, the Deep Learning system is capable of recognizing and evaluating the relevant objects on the image.
With this approach, the visual quality assurance process takes place automatically. The system analyzes existing data and recognizes patterns, based on which it identifies the properties of the objects. For example, the quality assurance system learns that there must be a hole at a certain location on the part. So if during production there is a hole located in the other place of the part, it will be detected as a fault.Such a solution is much more precise than the human eye and is a relief in the manufacturing process.
However, there are a couple of disadvantages that make the classic visual quality check solutions less flexible. To achieve accurate results, classic Deep Learning-based vision technologies need an extensive database. For quality assurance solutions, hundreds of images of both correct and defective products are needed. The problem here is that manufacturers often do not have an extensive database of pictures of faulty parts. Furthermore, such a solution requires weeks to months of training.
This was the challenge for which FotoNow was looking for a solution. We wanted to develop a technology for AI-based surface inspection that could be trained using a small amount of data, be ready to use quickly, and deliver accurate results.
After years of research and development, we created the solution that allows visual quality check automation fulfilling the criteria mentioned above. We launched the quality automation tool that needs only five images to train, is ready to use within minutes, and delivers over 99% accuracy.
How does FotoNow manage to perform such accurate visual quality inspections with only five images for training?
Our product development team created a strong base of proprietary algorithms and libraries generating the most power from Convolutional Neural Networks. The combination of different pre-training and training strategies allow us to develop an accurate and flexible visual quality management solution. Our team developed its own data augmentation techniques and called it “Industrial Data Augmentation for Quality Assurance”. As a result, we generate a robust model that is trained within a day. We also refrain from using the images of the defective parts by applying our anomaly detection model based on learning of real manufacturing contexts accumulated over the years. Our proprietary algorithms are able to use the data learned from different datasets to train the model for all possible visual defects, such as cracks, scratches, holes, etc.
What also makes FotoNow able to perform high-precision quality inspection using only five images for training is its labeling and segmentation technology. Image processing and analysis is performed pixel by pixel, allowing our solution to classify the objects on the image with the highest precision.
Using these techniques based on proprietary algorithms and libraries, FotoNow can perform manufacturing quality control quickly and accurately, requiring very little data. It precisely identifies such defects as scratches, blowholes and dents, which are a major challenge for vision solutions. Furthermore, due to our proprietary Deep Learning model training techniques, FotoNow precisely detects faults on metal and glass surfaces, distinguishing surface reflections from scratches.
With FotoNow, manufacturers benefit from quick and easy implementation of a quality inspection technology. Our vision solution does not require much data and input from the customer’s side to be implemented and thus is available at the moment when the manufacturer needs it.
AI technologies are changing our way of living and working. Enterprises are actively looking for innovation through AI. The benefits of...
Defects that occur during manufacturing are not all equally relevant for the quality of the final product. For example, a small scratch on an...
One of the main strengths of the visual quality automation solution FotoNow is its compatibility with any camera device. This allows the...