Computer Vision based Quality Inspection in Manufacturing using Artificial Intelligence

Reliable quality inspection techniques play a significant role in ensuring higher productivity in manufacturing. Specifically in precision manufacturing current practice of quality assurance inspection is laborious as well as time consuming. The majority of the current practices are either partly or fully dependent on visual inspection by humans; resulting large number of defective products. The consequence is even more costly for high value manufacturing - not only for higher number of defective products, but also could result significant post-delivery penalties. 

Manual Testing in Product Quality Inspection

Human error in quality inspection is inconsistent due to varying experience and ability between individuals. In high precision manufacturing, the dimensions of defects are reasonably small for human eyes (at times within the range of fraction of a millimetre), leading towards the risk of missing one off and random defects.

Traditional Method for Defect -Detection

Ultrasonic technology is widely used to detect the defects in the internal structure of the sample. The main advantage of ultrasonic testing is the high rate of detection and simple operability. The defect-detection methods are based on filtering (Kalman, morphological and Gabor) has a strong ability to describe the disturbance signal and detection of the tool defect inside the machine.

The role of Artificial Intelligence

Artificial intelligence (AI) has come to play an increasingly integral role in real-time product quality inspection in general, particularly in the field of computer vision-based quality control. The applications of AI in product quality inspection are numerous, from accurate fault detection (using surface feature extraction and abnormality detection), to sorting abnormalities (using ranking based method), and their classification (using Deep learning).

Computer vision in Product Quality Inspection

Camera-based computer vision systems can significantly improve quality assurance. Some techniques that can improve inspection-time performance include Digital cameras, MMX-enabled vision software and the Image-acquisition hardware.

Digital cameras provide faster image-acquisition rates (up to thousands of frames per second) than analogue cameras (30 frames/s). MMX-enabled vision software can increase system processing for some functions by as much as 400%. The Image-acquisition hardware can support image-processing techniques such as on-board partial image scanning, programmable to image segmentation within a specific region of interest and on-board pixel decimation. This design strategy aims to reduce the number of processing pixels to decrease the time needed to process the image.

Machine Learning in Product Quality Inspection

The AI system is trained with thousands of images of parts collected by cameras. The system is trained to distinguish a bad part from a good part.  

1. The machine vision-based solutions can detect scratches, holes, scales, pitting, edge cracks, crusting and other surface defects.
2. Statistical analysis of machine learning techniques is then applied. Ranking algorithms, for example, ranks the defects from the highest to lowest probabilities.
3. If the defects can be classified, then the inspection process will generally be easy to automate. The images are labelled based on ranking. These images can be trained using different deep learning architectures, including convolutional neural networks. These architectures are then able to compare these images with actual images of non-faulty parts, identifying defective ones.

The defective parts are separated from the rest of the inventory and discarded or sent for correction.

SOLVD

AI-based solutions can not only automate the process of quality inspection in the manufacturing pipeline, but can also make the inspection process more reliable and efficient through a mix of Computer vision techniques. This includes real-time image processing for edge-detection, precision segmentation etc. The use of Deep Learning techniques based AI for segmentation is an example that can identify faulty products as well as being able to precisely detect the specific issue with the faulty product.

Our AI expert in SOLVD can help you to explore this potential for your businesses.

Include “call to action” i.e. for an informal discussion about how your company can become more efficient, please get in touch with our experts via:

solvd@wlv.ac.uk

www.wlv.ac.uk/solvd

Blog by Dr Lavanya Srinivasan. Research Fellow in Computer Vision and AI at The University of Wolverhampton.