Case Metsä: Discover Metsä with Data
Picture: Metsä Group
Metsä Group's Discover Metsä with Data summer internship program opened opportunities for students to delve directly into the core of the company's business development over the summer. The results have garnered praise.
In previous years, Metsä Group had offered individual advanced data analytics internships to students. The Head of Data Analytics Development and Services at Metsä Group, Heikki Sulonen, saw the caliber of applicants who were left without an internship spot.
"In recent years, we have seen on paper alone how high the quality of applicants for internships has been. Previously, only individual positions in advanced analytics were available. It was regrettable to leave so many competent young people unrecruited," says Sulonen, who initiated the program.
At Sulonen's initiative, Metsä Group launched a new summer internship program focused on advanced analytics this summer.
The program's goal was to provide students with the latest machine learning methods and technologies, as well as business-critical tasks. During the preparation phase of the program, the needs of Metsä Group's business units were carefully examined, and specific tasks needing solutions and supportive mentors were defined for each recruit.
Eventually, five data analytics students were recruited to the program. The preparation and training offered to the interns paid off. Several promising observations and applications were developed over the summer, with the potential to lower production energy consumption and save on logistics costs.
"I can gladly state that their starting level was excellent. They are ready to tackle business-critical challenges and bring new perspectives and thinking to the analysis. We hope that the program will be a new avenue for recruiting the best talent in the field for us," says Sulonen.
The Discover Metsä with Data program has so far resulted in one application moving into production and plans for three master's theses.
"Expectations seem to have been exceeded by a considerable margin. The feedback from the businesses has been very positive, even laudatory. The results achieved are genuinely such that we will continue to use them. Some are practically ready for use, while others are moving on to further development in master's theses," Sulonen rejoice.
Training Enabled Results
Metsä Group's data and technology partner Cloud1 Oy was responsible for training held every two weeks, in collaboration with Metsä Group's IT management. The training was provided by Cloud1's advanced data analytics team, which supported the interns throughout the summer. Cloud1's team trained the students to utilize the latest machine learning methods available in Microsoft Azure technology, run on the company's data platform.
"The mentoring and training of the interns were essential parts of the program. It is one thing to come into an industrial environment and another to practice work at school. We focused a lot on how to collect and process data from different locations for analysis. The students received up-to-date knowledge on working in an industrial scale and learning the latest Azure technology-enabled analysis methods," Sulonen says.
Cloud1 serves as Metsä Group's data partner and maintains the Microsoft Azure Databricks platform it provides.
"We have renewed Metsä Group's data architecture, which enables the development and rapid testing of modern machine learning and artificial intelligence applications. Participating in Metsä Group's data analytics summer internship program in a mentor role was a great opportunity to share our expertise with young talents aspiring to enter the field. The results indicate that the collaboration was even more successful than expected," says Tuomo Riihentupa, CEO of Cloud1.
The biggest challenge in the summer internship program was adapting to the demands of an industrial production environment. For example, students had good skills in Python and SQL coding.
"It is one thing to come into an industrial environment and another to practice work at school. We focused a lot on how to collect and process data from different locations for analysis. The students received up-to-date knowledge on working in an industrial scale and learning the latest Azure technology-enabled analysis methods," Sulonen says.
"We initially created work methods and frameworks in Metsä Group's cloud environment, where all data processing, coding, and analytics work were done. Systematicity is essential for cybersecurity and helps manage the different versions of developing applications as work progresses," says Jesse Turkia, Cloud1's data architect. Often, the problem with cybersecurity and version control arises from processing data and writing code in different places, such as on one's laptop.
"At the core of everything is our unique architectural expertise. When a systematic work model is built on top of that, results can be produced quickly. It was great to see what the students were able to achieve in such a short time," Turkia says.
Significant Savings from Forecasting
At Aalto University, Antti Honkanen, a master's student in production engineering, faced the challenge of automating the forecasting of Metsä Board's logistics needs and their updates, while also increasing the flexibility, manageability, and transparency of logistics planning.The work had to take into account road, rail, sea, and air transport, container reservations, and the purchasing behavior of customers in different markets and factors affecting demand.
"The crux of my job was to calculate how the volume of sales forecasts is distributed across the logistics network, taking into account that there are several different routes and they may have alternative transportation modes. In addition, logistics lead times, delivery models, and third-party carton processing plants as part of the logistics network presented their own challenges to the calculations," Honkanen says.
Honkanen developed a solution visualized with Microsoft Power BI, which optimizes the logistics required to meet the demand in different market areas. The application is estimated to have the potential for cost savings in the millions. The forecasts produced by the application enable Metsä Board to better anticipate logistics needs and avoid overcapacity in logistics reservations, which could previously occur when the results of manual forecasting work became outdated and more recent information was lacking, leading to precautionary reservations being added to ensure customer deliveries.
"Previously, this work was mainly done in Excel, and now, thanks to the application, this labor-intensive manual work has been eliminated, and forecasts are generated automatically based on sales data," Honkanen says.
Honkanen praises both the data experts at Metsä Group and Cloud1's training as enablers of the results.
"I only knew Azure by name before my summer job. The training provided by Cloud1 was an excellent help, and they were always available to answer questions related to coding or data processing. Additionally, Metsä Group's data analytics experts played a central role in guiding my work," Honkanen says.
"About 70 percent of my time was spent collecting data and thinking about how to achieve the desired final result. Power BI allows, for example, the integration of data from an enterprise resource planning system with other data sources in a new way. This, in turn, enables data to be utilized in a value-creating manner. Also, the possibilities for visualizing data analysis are excellent," Honkanen explains.
At Metsä Board, Honkanen's work has been enthusiastically received.
"We get a longer-term view of demand across the world, its impact on transportation needs, and, for example, container reservations. The potential for improved work efficiency and effectiveness is significant," says Ville Virtanen, the business process expert responsible for the program at Metsä Board.
Rapid mastery of Azure The Azure environment and its machine learning tools were not previously familiar to the summer interns. In the training, students were taught to utilize Databricks tools, such as Automated ML machine learning models, the Experiment tool, and delta tables useful in data versioning.
"For example, Automated ML tools make it possible to quickly test the usefulness of a dataset and the model used. Versioning Delta tables and the ML Flow Framework Experiments ensure that experiments and the datasets that change in them are kept safe and are reproducible," says Jesse Turkia, Cloud1's data architect.
During the summer, workshops addressed typical challenges in machine learning and data analysis. Common challenges in advanced analytics and machine learning often arose. Especially, the amount of data may often be too small, prompting a need to reformulate the original question.
"It is crucial to evaluate whether we are asking the right questions or aiming for a result that is achievable with the given dataset and ML model used. We spent a lot of time going through different solution options and how the available data could be used in various ways. Typically, a solution is found by fine-tuning the model, dataset, code, or the problem formulation itself," Turkia says.
Participation in Metsä Group's Discover Metsä With Data program provided Cloud1 with an opportunity to deepen the expertise of aspiring students with the best practices in the field.
"Our team put themselves on the line. Mentoring the students was a really inspiring experience for all of us. It felt like we enabled something for five young people that is not usually available to summer interns or at the university. Metsä Group had created excellent conditions for success, and we were able to share knowledge from our core expertise," Turkia says.
Application to Predict Energy Consumption
Henna Roinisto, a fifth-year master's student in data science at the University of Helsinki, was tasked by Metsä Tissue with finding solutions to optimize energy consumption in the production of soft paper at the Mänttä plant. Roinisto had access to a large amount of data from various sources, including data from production line IoT sensors, energy consumption data, end-product quality data, raw material data, and historical data on the impact of local weather on energy consumption.
"During the summer, it became clear how much time ultimately goes into data processing and compilation. For me, the ratio was probably closer to 80-20, meaning 80 percent of the summer was spent searching for data and compiling the dataset," Roinisto says.
With training from Cloud1, Roinisto learned the secrets of utilizing Microsoft Azure tools. Roinisto had not previously used Azure technology for data processing and model building.
"Most of the Azure-related expertise came from Cloud1's training. Help was always available, whether it was about data processing or writing code."
In Azure Databricks, Roinisto built models for energy optimization based on production and production cost data. The main model predicted parameters of sensors measuring end-product quality and energy consumption on the production line. Another model assessed the production line operating parameters predicted by the main model and determined how the line should be run to minimize energy consumption. Roinisto's analysis of the data revealed several opportunities to change the production line settings in such a way that energy and raw material consumption would decrease, but without affecting the quality of the final product.
"We have extensive experience in running and optimizing the production line. Now, we harnessed the latest machine learning methods and data collected from various sources. In a relatively short time, we found initial results of new ways to run the soft paper production line so that energy consumption would still be as low as possible, and the final product quality would still be excellent," says Joonas Kukkonen, leader of the Data team at Metsä Tissue.
"The models can be further developed and supplemented with additional data. We demonstrated that such a model can be built in Azure's Databricks environment," Roinisto says.
"Every day was absolutely great to go to work, and I couldn't always bring myself to end the workday. Thanks to Metsä Group and Cloud1 for that. Both are work communities where the atmosphere was really encouraging. I didn't feel any stress during the summer at all. This also enabled the results."
The Discover Metsä with Data program left Roinisto with good memories and showed how practical work in an industrial environment ultimately teaches a data scientist's professional skills. Roinisto is currently planning to work on an advanced data analytics-related thesis in collaboration with Metsä Group.
