Insta and the metal industry company Boliden Harjavalta are conducting a study at the Boliden sulfuric acid plant to determine if the gas distribution process of the sulfuric acid plant can be optimized with AI. In a joint project, the companies were able to model a part of the process using a neural network.
“Insta has the expertise to use machine learning algorithms in analyzing and improving adjustments. We also have a customer who understands the value in this. The possibility of piloting was very important for us,” states Mika Riikonen, Insta Automation’s Vice President, Regional Operations and Maintenance.
Identifying benefits with controller optimization
At Boliden Harjavalta, the controller and its steering was modelled again, and the model was used to see what the new results would be. As a result, a kind of digital twin was created. Boliden Harjavalta can use this digital twin to simulate production using different drive parameters, as well as to reduce emissions.
The results show that it is possible to reduce sulfur dioxide emissions by 5–10 percent. Because the benefit here is significant, the simulated model will be coded into the automation system.
“Controller optimization has many additional benefits as well; for example, improved environmental and occupational safety, reduced energy consumption, reduced raw material wastage and improved productivity,” says Riikonen.
Intelligent maintenance through machine learning
In the future, utilization of data will increase, and machine learning models will offer undeniable benefits also for maintenance applications. However, we cannot skip any important phases of the process, but instead, intelligent maintenance requires careful optimization and stabilization of processes.
“We can make use of machine learning only after the controllers have been tuned and equipped with the right data acquisition instruments. The Overall Control Efficiency (or OCE) value indicates process stability, and always falls between 0–100%. In production, a limit has to be set for OCE to determine how low the value can sink before action is taken,” Riikonen clarifies.
When the output data is valid, machine learning and simulated models can be used to predict the correct time for maintenance. If a device doesn't work as intended, its output data will change. Deterioration and failure can be predicted, and the device can be repaired at a suitable time according to plan.
“Even a small amount of data indicates if something strange happens in a process. When there’s enough data, the machine learning model can give suggestions as to the cause of the malfunction.”
Secure digitization enables continuous improvement
Machine learning for analytics is gaining in popularity. Mika Riikonen thinks that the competition will get tough in the next few years: If industrial companies won’t start using machine learning and AI early enough, their competitors may get really far ahead.
“Performance has to be constantly better. AI expertise alone won’t be enough. The experts will also have to have a genuine understanding of automation, which only very have. Another challenge is that there are still so few experts in this field around. Where can we find enough good experts when the demand rises?” Riikonen notes.
According to Riikonen, data analytics is the most cost-effective, fastest and most reasonable in a public cloud. However, without safe digitization there are risks involved. Insta is a pioneer in the fast-developing global market of safe digitization. The company also has long experience in the processing industry, instrumentation, as well as automation.
“With safe digitization, it is possible to make use of process data, realise full savings potential, and achieve truly genuine maintenance. We speak the same language as our customer, and we are able to handle the whole chain from the field all the way to the machine learning model,” Riikonen sums up.
Read more about our Industrial AI services by clicking