ST-One: where Industry and University comes together
Copyright ST-One
In the last year, ST-One worked together with UFMG and M. Dias Branco in research about the creation of machine learning models. These models were trained to identify patterns on the manufacturing of margarine, particularly on the deodorization process of vegetable oils.
In this process, it plays an indispensable role on removing undesirable substances that negatively affect the odor, taste, color, and stability of the final product. At first, the industry heats the oil to carry out its filtration. Then, Water vapor is injected into the heated oil to absorb the fatty acids, after which it is cooled to prevent oxidation. Finally, the oil goes into a vacuum distillation process, which removes volatile components that cause product instabilities.
Copyright Alfa Laval
This vacuum is essential in this process because it allows deodorization to happen at low temperatures. This action reduces the risk of thermal degradation of the oil and improves the efficiency of the process. To make sure that this process goes with no flaw is really important, as it ensures that the manufacture of margarine is secure and tasty to the customer.
To highlight the importance of this food, according to a survey by the Brazilian Association of Nutrology (Abran), in 2011 about 32.2% of Brazilians chose to consume margarine for breakfast.
Partnership structure
Since March 2023, the aim of the study was on vacuum breaking time, which happens when it exceeds an operational limit, and the process must be interrupted. The resulting information was used to improve the production line efficiency. The research was carried out with three institutions – ST-One, UFMG and M.Dias Branco – with each of them playing an essential role in the search for more technology and innovation.
M. Dias Branco is one of the largest food industries in Brazil, and it is always looking for more digitalization in order to enhance its production process. It has an industrial unit specialized in the production of special shortening and margarines. This industrial plant was used as a reference point and expertise for the project.
UFMG (Federal University of Minas Gerais), within its computer science department, has a renowned Artificial Intelligence Laboratory (LIA). In this laboratory, the faculty members encourage the development of projects, through scholarships, focused on machine learning and data processing. LIA is a reference in the field, even collaborating with international institutions, such as Stanford University, located in the United States. The surveys are applied in lots of areas, including industry.
ST-One contributed as the main sponsor, in addition to providing all the data necessary to carry out the study, based on the technology it develops. Holding weekly meetings throughout the process allowed for a rich exchange between ST-One’s R&D laboratory and the other parties.
The solutions developed aim to create predictive models, which assist the operations team to fix the system before the vacuum breakdown. The project resulted in more knowledge and ownership of the machine learning process for all parties involved. In addition, it was an opportunity to see the real-life application of the models created in an industrial environment, opening room for later improvements.
Copyright ST-One
Three steps to develop innovative technologies
As previously mentioned, vacuum plays a crucial role in the preparation of vegetable oil used in the production of margarine and shortenings. Throughout the project, 3 different kinds of machine learning models were created to test which one would bring the most assertive answers. This especial attention is essential because creating a vacuum is a delicate and expensive process, and any disruptions can have significant consequences.
Taking this into account, the core was to create a machine learning model that can predict when this break will occur. The project was carried out in three major stages: the investigative part, the mathematical modeling, and the results testing.
The first step focuses on the analysis and discovery of the issue to be worked on. This initial phase must be thought out in detail, mainly due to the great importance of the model in production that determine the good result. This investigation takes place by defining the tools that should be used, what questions should be asked to the model and what is its ideal answers.
The second phase involves applying several mathematical formulas to train these models using the created questions. The goal is to operationalize the process and enable visualization. During this stage, a specific quantity of data is taken, which then passes through the chosen machine learning model. The model processes the data and transforms it into a trainable format.
From this, LIA scientists developed questions used to create a model that explains the behavior of the data collected. Based on this, the M. Dias Branco team was able to make adaptations in the automation of the equipment.
Finally, the last step is testing the model, to ensure that after training the patterns of the data collected are recognized. This process is done using a test dataset that the model has not yet seen. From the test result, more questions appear, especially if the process cannot identify the correct answers with the necessary accuracy. If all goes well, the model is implemented.
These steps result in a cycle, which uses the “explain” method developed in the LIA (Artificial Intelligence Laboratory) itself, as shown below:
Copyright ST-One
Results of the Machine Learning application in industry
First of all, it is important to remind that the models developed are still in the application stage. Even so, the study has already showed several beneficial results.
Within the university walls, the data-academia initiative, in collaboration with industries, fosters innovation and creates opportunities. This good relationship occurs because postdoctoral students in the laboratories are devoted to research with a development-oriented focus. These studies, grounded in state-of-the-art theoretical foundations, are directly applied in the real industry routine, providing valuable practical experiences for all involved. Lastly, the partnership with renowned educational institutions results in additional resources for the industry.
For ST-One, in addition to the experience itself, having the opportunity to delve into domains of Artificial Intelligence was the key point. By engaging in the exploration, model development, and application stages, it was possible to improve the interpretation of the “world” through the data and visualize this within its own framework. ST-One improves itself through each new challenge, always aiming to bring more technology and productivity.
In addition, the factory was benefited by getting more knowledge, networking, and understanding of the possible ways to predict a vacuum breakdown considering its complexity. This is a path to further development and retraining, until the model reaches the ideal prediction time. Thus, the quality of the process is guaranteed, produced in an intelligent and assertive line.
Actions like this are good for all the parties involved. It assists in keeping students and professionals of data to always be updated and seeking for improvements. Learn more about ST-One.
