What we do
AI-based defect detection or classification is a further development of machine vision inspection solutions. It uses an expect of machine learning, called deep learning. Just like a human eye, visual inspection AI solutions capture an image and process it. The system learns from examples and understands differentiations of characteristics and anomalies. Due to its repetitive nature, human visual inspection is error-prone. The advantage of an AI-based system is the reliability and high accuracy. Furthermore, these systems can support humans in their decision making and thereby reduce time, costs and security.
AI-based visual inspection for industrial parts
Our client is an international automotive supplier producing different car components. Manufacturing companies, especially in the automotive suppliers, must meet very high-quality standards. In this use case we improved the efficiency of the automated visual inspection system of metal components at the end of the production line.
The client uses a state-of-the-art quality control visual inspection devices for the final inspection of the product at its high volume production line. The system is able to detect quality issues such as missing glue or misalignments. Although the current inspection solution detects issues very well, it also has a high pseudo-scrap rate. Thus, quality inspectors still have to inspect a large number of product manually, which lowers the full efficiency that the system potentially offers.
Our AI-based visual inspection software uses images from industrial cameras and detects defects on components using a semi-supervised approach, meaning a minimal number of labeled images were required to train the model. It processes images in real-time and classifies between good parts and bad parts. Furthermore, the system can also learn different types of defects to classify even more accurately between types.
The system supports automated quality control and could significantly improve the pseudo-scrap rate to a minimum. This does not only save the quality inspectors time but also makes the production line more efficient and cost-effective.
Together with craftworks, the automotive supplier defined the specific use case as well as software and hardware requirements. Subsequently, the solution was developed in several phases. First, the feasibility of the project was evaluated by creating an image preprocessing pipeline and we trained a simple model that constituted a baseline for further optimization and development. In the second phase, the model was further developed. The solution uses a semi-supervised learning approach, meaning that in its core it uses an anomaly detection approach to identify abnormalities. Moreover, it uses some human-labeled images to validate the anomalies. After several tests in the production, the model was ready for further optimization and continuous increase of performance and accuracy.
How you can work with us
A selection of our Industrial AI clients
With craftworks it was simple and uncomplicated. After an initial conversation, we could start a project within a short amount of time. While the data scientists at craftworks supported us in drawing valuable conclusions from our data, the close collaboration together with their domain experts made it possible to have fast results within a month.
Wien Energie, Head of Innovation and Strategic Projects