Optical Character Recognition for Mother Plate Life Cycle Optimization

Affinerie CCR, based in the Montreal region, is a refinery specializing in copper and precious metals. CCR carries out the final refining stage using an electrolytic plating process, during which copper is deposited on stainless steel plates (known as master plates). After over 10 years of use, the 85,000 mother plates had started showing signs of surface corrosion, which affected the physical quality of the copper cathodes and slowed production. Our client therefore decided to restore them at a rate of about 300 mother plates a week. However, the client had to be able to identify the mother plates to track which ones had been repaired and aid process improvement.

Our task was to develop a custom optical system, because conventional character recognition methods could not be used on mother plates for the following reasons:

• The engraved numbers are copper on copper, resulting in poor contrast.
• The surface is uneven and presents numerous defects (scratches, stains, marks).
• The numbers are not always engraved in the same location, which means a broader field of vision is required, reducing image resolution.

The development project was conducted in three main phases.

During the initial phase, which ended in 2012, INO developed an optical system and a digital vision algorithm to segment and classify the various characters in the codes used to identify the master plates. The system and algorithm are used to monitor the path of each plate from one end of the production facility to the other. Thanks to this system, continuous monitoring is provided and any plates requiring restoration are taken out of production without interrupting the operations.

Later, CCR called on INO to modernize its system and boost the read rate. Thanks to their deep learning and artificial intelligence expertise, the members of the INO team developed a new character recognition algorithm using convolutional neural networks.

This type of algorithm is known to exceed human performance in various tasks, such as image recognition. The previous algorithm used hand-crafted features while the new algorithm learns the characteristics directly from the data with a view to minimizing the number of read errors. This makes the system much more adaptable since by adding images to the database, new learning is facilitated and performance is improved.

In addition, read accuracy performance was boosted from 94% to 99%, enabling CCR to better monitor the condition of master plates in the facility. Based on a total of 75,000 master plates read per week, the number of incorrect readings dropped from 4,500 to 750 per week.

This second phase of the collaboration with CCR showed that by combining optics and artificial intelligence, two of INO’s key areas of expertise, innovative solutions can be delivered to the manufacturing sector.

Phase three entailed completing development of the vision system and commissioning three systems. While the systems were being physically installed, we were able to optimize the character recognition algorithm using the images obtained with the first machine. We tested multiple versions of the software before arriving at a definitive version.

After restoring over 5,000 mother plates, our client noticed an improvement in the quality of electroplating on repaired mother plates, easier cathode stripping, and fewer machine blockages, which cut down on operator intervention and the risks of accidents.

The high level of cooperation between Xstrata Affinerie CCR, INO, and the other project partners was critical to its success. Moreover, the optical system we have delivered has paved the way for our client to implement other process improvement projects.