How to investigate the defects of the steel bars

Challenge

The steel bar rolling plants process the material at high temperature (900°C or more) and under very harsh environmental conditions. The bars quickly leave the plant still at high temperature, with no way for an operator to check for defects that could affect the quality of the final product.

In this way, it may happen that a lot of defective material continues to be rolled before the defect is noticed, and even sometimes the defective material is not intercepted and reaches the customer, resulting in economic and reputation losses.

The challenge of this application is to develop a system that is integrated into the rolling line and enables fast and affordable detection of defects.

Approach

An automated product inspection system has been developed, integrating hardware engineering solutions to obtain reliable product images under harsh environmental conditions, and AI software solutions that ensure fast and reliable classification of defects on product images.

The hardware of the system is designed to be optimized with respect to product speed and required resolution; lighting and imaging conditions are designed to ensure that defect images are properly visible in the collected images. The position of the imaging system along the rolling line is also chosen so that the defect is most visible.

These system design criteria made it possible to obtain a very sensitive device capable of collecting images in which the presence/absence of the defect is indisputable.
The system captures a large amount of images in real-time (hundreds of megapixels per second) and processes them at high speed. The system's detection capability ensures that the totality of defects on the material are detected and reported.

The use of AI techniques to process the images ensures to obtain reliable results even in the classification of the type of defect detected, thus being able to generate selective reports indicating the nature and cause of the defect. In this way, the end user has a basis for being able to selectively take corrective action on the processes to reduce the causes of defects.

Defect classification has been optimized for the identification of typical defects of this process. We were able to use AI and Machine Learning techniques because of our specific knowledge of this process, and the use of hundreds of thousands of examples of defect images from this process that we have available and that allowed us to instruct the system.

Conclusion

The automatic inspection system promptly alerts the customer of the presence and type of defects in the product. The use of the device led to two effects, the first one immediate (the ability to deviate material with most critical defects, and more objective management of customer complaints) and the second which is slower but equally important, which consists in the optimization of processes upstream of rolling (steel making, continuous casting) for which the information on the presence of defects, together with material tracking, allows to analyze what process conditions originated the defect in the first production steps.