A spark for machine learning led to dissertation and to the invention of the first Finnish AI beer
Sometimes one can find a way into unexpected life paths by following one’s interests and passions. DSII, Metsäteho, and Ponsse were looking for a doctoral student who would find the means to collect new forest data efficiently. Lari Melander saw the opportunity to improve himself and his expertise in the field of machine learning and applied for the job. Now he uses the knowledge brought by his dissertation work for developing the first artificial-intelligence-generated beer in Finland.
Lari Melander, Doctor of Science (Engineering), is 33 years old, father of three and an AI-beer inventor who accomplished his doctoral dissertation 05/3/2021 in DSII.
Melander did his master’s thesis in engineering — while working in Fastems — thinking that he would not return to university. Even less he had ever thought of doing his dissertation. But then an interesting job offer occurred.
“I happened to notice this very interesting DSII job opportunity to work with two different companies, Ponsse and Metsäteho, machine learning and data. At that time, around 2016 there was a lot of buzzing around those subjects”, Melander recalls.
The companies involved in this job ad were the ones that gave Melander the push to apply for the job.
“If it were only the university, I wouldn't have been so interested. But with those companies and with all their professional support and ideas involved, it was a match made for me.”
After graduating from university one can feel like not getting back to school. Melander verbalizes this feeling back then with the words “It’s time to get to work and make a career.”
But when Melander took the chance, applied for the job, and got it, he did not want to turn back.
“Of course I was thinking, do I have what it takes? Will I ever finish this project? Now I feel good that I took the risk. I would definitely do it again!”
Overcoming the challenges
According to Melander, at first, the idea for the dissertation work was quite universal and a little too extensive. There was just a general thought of generating something new in the field of machine learning to improve forest operations.
“I really had to think and ponder where and how to find the main core for my research. What would be new and interesting also for the machine manufacturers and not just for science.”
Balancing between the academic world and the companies had its difficulties, but somehow the equilibrium was found.
Metsäteho and Ponsse both had slightly different angles for the dissertation work. Metsäteho was interested in evolving the ways to collect and combine forest data and Ponsse wanted to develop novel and more intelligent work machines for the forestry work.
Making use of existing resources
In his dissertation, Melander suggests two novel methods for perceiving data from the forest while the forest machines work in the forest environment. It is not very well known yet how changing forest environment affects the performance of the machines.
“The forest machines are working almost 24 hours a day in the forest, so why not utilize the time they spend there? We can collect a lot of data from the environment with different perception methods.”
Melander came up with measuring the severity of wheel ruts made by logging machines on the ground. The measurements were done by time-of-flight depth camera. The idea wasn’t new, but the technology is. Forest owners want to know the amount of forest damage caused by the machines. With more environmental knowledge the forest work can be planned more efficiently.
The other method is an automated soil stoniness classification system done by continuous vibration measurements of the excavator boom while the machine works in the mounding operation as the basis for detecting the soil stoniness level.
“In Finland, we have very good public forest resource data. But some data is missing, like information about the stone content in the topsoil of the forest. It’s important because it affects the quality of the tree planting operations, the growth of the trees, and the hydrological system of the soil.”
More precise forest data for the new platform
Melander remarks that forests are regenerated by soil preparation and planting. When the mounding is made in the forest, it is good to know in advance where the stone content is thick.
“In some maps, we have recommendations on in which season the forest can be harvested. For example, only in winter when the ground is frozen and not soft, so that the machines don’t sink in and get stuck.”
As the other half of the dissertation deals with the novel methods of perceiving information from the forest during forestry work, the other half is about combining the fieldbus data, collected by the forestry machines, into an existing forest data platform.
Melander narrates that Metsäteho has devised a platform that compounds all the forest data in Finland under one service. He used the data gathered from Metsäteho’s forest data platform and combined it with the fieldbus data collected by the forest work machines. With this information Finnish forests can be grouped and compared, and also the forest machines and their driver’s efficiency could be compared with each other taking into account the environmental factors.
According to the Metsäteho website, the service platform has been proposed to be commercialized in connection with the Finnish Forest Centre forest resource information system to support producing, updating, and distributing up-to-date forest resource information to industry actors and stakeholders.
In his own words, it took five years for Melander to accomplish his doctoral work. He points out that the support of family and friends is crucial. Nights can be prolonged and close ones need patience. Even so, Melander states that the DSII studies were just for him.
“The meetings and roundups at the university were always nice. Even at times when the meetings didn’t have that much content, it was nice to exchange thoughts, problems, and experiences and get peer support from other students. You get to know a lot of different new people, which is nice.”
Melander had monthly check-up meetings with Ponsse and Metsäteho where he introduced his accomplishments and got comments on his work from the contact persons.
The first Finnish AI Beer?
What does that mean? He has utilized the expertise he gained from his dissertation and his 10-year history of home-brewed beer – a combination of hobby and expertise.
“I have developed a system which improves the beer according to the feedback and the evaluations it gets.”
The beer is American IPA variety, and the first version is to be out in the shops by the 1st of May. The beer is brewed at United Gypsies Brewery in Lohja, southern Finland, but the brand and the AI recipe system are maintained by Melander and his friends, known as Kopla AI Brew -team.
“There is just a huge QR-code on the beer can which you can read with your phone and you get into the feedback system.”
The idea is to create different taste profiles of beers and produce beers for different taste preferences.
“At first there is going to be only one variety of beer that is evolving according to the customer feedback, but later we plan to increase the amount of different AI beer varieties.”
You can read Lari Melander's doctoral dissertation here.