BAÜP

Quality control with artificial vision

The occurrence of defective products in industrial manufacturing processes increases companies’ costs, decreases the average life of products and results in wasted resources.

Defect detection is a key competence that companies must possess in order to improve the quality of their products without affecting production.

We know that automatic defect detection technology has advantages over manual detection, as it adapts to any environment and works over the long term with high accuracy and efficiency. If we think about the repetitive and monotonous visual inspection tasks performed by operators, we realise that inspection performance and reliability can decline rapidly. The inherent human subjectivity that causes two different operators to provide different results for the same situation must also be taken into account.

SOLUTION

A manufacturing company is faced with the challenge of detecting defects in certain parts automatically, with little operator intervention.

Visual inspection using computer vision is faster and more consistent over time, addressing the drawbacks of manual inspection in another way. However, computer vision inspection techniques still require a great deal of manual support to work properly, especially when a number of adjustments and settings are necessary before testing. These traditional inspection techniques allow the evaluation of parameters such as orientation, presence or absence of parts, measurement, etc., which require parameterisation by a user and do not allow the detection of surface level defects such as deformations, scratches, dents, etc.

Innovative machine vision techniques, such as the use of Deep Learning, are more versatile and autonomous compared to traditional computer vision inspection methods. Through the use of algorithms, no human intervention is required to parameterise all the steps (shape, orientation, etc.), but the ‘work’ of finding (or in this case, learning) the features and thresholds is done by the machine learning model itself.

BAÜP has developed a Deep Learning algorithm for this defect detection task, which classifies a given image as being ‘ok’ or ‘not ok’, depending on the type of defect found.