virebox - case BOLIDEN HARJAVALTA
Improving industrial processes with AI based machine vision
In basic material industry, a growing number of companies are starting to use artificial intelligence to improve their processes. Boliden Harjavalta in Finland is using an intelligent machine vision solution provided by Vire Labs to monitor the quality of copper cathodes and improve the production processes.
As raw material for copper production Boliden uses both copper concentrate and recycled metals. The raw material contains approximately 20 to 30 percent of copper. During the smelting process in Harjavalta, the purity of the copper rises and reaches the level of 98 to 99 percent. The copper anode plates are then sent to the nearby city of Pori for electrolysis, after which the purity of the copper cathodes is 99,998 percent or more.
The system helps us to find out the causal connections related to quality of the products. This way we can develop our production processes.
Juuso Karjala, Boliden
An intelligent solution for the quality control
Previously the quality of the copper cathodes has been inspected mainly by human eye, and from time to time by also photographing the plates coming out from the process.
– We have photographed and rated thousands of plates during the years, but no automatic quality control system has been in use. For this task we need intelligent image analysis, and such a solution was not available before Vire Labs stepped in to help us with the AI solution, explains Jussi Lehtinen, the Production Engineer for Copper Electrolysis at Boliden Harjavalta..
The AI based machine vision system that Vire Labs delivered to Boliden consists of two cameras, and both sides of every copper cathode plate are photographed at the end of the production process. The Virebox.AI system then analyses each image, rates the quality of the cathode plate, and passes the information on to the production system.
Copper plates appear for analysis every eight seconds. Three quality parameters are inspected in every plate: the flatness (or unevenness) of the surface, the growth in the edges and lacing in the fringes.
– Regarding the flatness of the surface we have four quality classes, S stands for superior quality and A for good, B for normal and C for the plates that have noticeable unevenness, explains Juuso Karjala, the Manager at Copper Refinery at Boliden Harjavalta.
An intelligent analysis is needed because the phenomena on the cathode plates vary a lot. In addition to the parameters that affect the quality of the product, there can be many kinds of visible things like stripes, colour dyes, lighter and darker patches that should not be taken into account when rating the quality. A traditional machine vision system, which is based on comparing the images, cannot cope with a difficult challenge like this.
An AI based system is able to detect the factors affecting the quality of the cathode and it can make a reliable analysis even if the copper cathode analyzed differs from all previous plates. The neural network model has been taught by showing several images of copper cathodes of different quality classes, and it is able to make conclusions of the quality the plates so that it does not take into account the parameters which do not affect the quality of the product.
The system works like a human eye, but unlike a human, the system is automatic and objective and it works tirelessly.
– We now have access to the data history for all the copper cathode plates we produce. The system helps us to find out the causal connections related to quality of the products, and this way we can develop our production processes, tells Juuso Karjala.
– It is also easier to analyze customer feedback as we can afterwards check what the quality of the product delivered to the customer was at the end of the production process. The purity of the copper is always very high after the electrolysis, but to make the process better, we want to be able to recognize even the smallest deviations in quality and the reasons behind them, explains Karjala.
Now we get a constant feed of data of all the things that might affect the quality, and a blind spot in the process has been removed.
Jussi Lehtinen, Boliden Harjavalta
The project started in April 2021. Development was fast and by the end of year everything was tested and in operation.
– Now we get a constant feed of data of all the things that might affect the quality, and a blind spot in the process has been removed. The longer the system is used, the better understanding we will have of the factors affecting the quality of our product, explains Jussi Lehtinen.
The Vire Labs operating model consists of three phases. In the PoC (Proof of Concept) the feasibility of the solution is tested, and this usually takes a few weeks. After the PoC a pilot is carried out with the customer. In the case of Boliden the pilot consisted of several iterations during a six month period. In the end the solution is integrated fully into to the production process of the customer.
According to the Systems Engineer Kimmo Viitala, the model suited well for Boliden Harjavalta.
– Both the PoC and the pilot were extremely important for us, and without them the end result would not have been as good. We had several rounds of iteration, and the solution we now have in our use fulfils all the objectives and delivers everything agreed.
– We will continue the development work, and at least one round of iteration will be completed this year, explains Kimmo Viitala.
Does your company have similar challenges?
Please contact us to see if we can help you!