Braskem - Smart Gels: Classifying Gels in Polymers Through AI
Team members:
- Luis Paulo Bernardi
- Mariele Kaipers Stocker
- Evelin Dalla Nora de Almeida
- Ana Paula Lobo
Country: Brazil
Organization: Braskem
Braskem is a global company with industrial units located in Brazil, the United States, Mexico, Europa and Asia. Founded in 2002 via the integration of six companies from the Odebrecht Organization and the Mariani Group, we are the sixth largest petrochemical company in the world in the production of thermoplastic resins. With customers in more than 71 countries, we are the market leader in the Americas and pioneers in the production of biopolymers (plastic made of renewable raw material) on an industrial scale.
Awards Categories:
- Best Acceleration Use Case
- Best Moonshot Use Case
Business Challenge:
Braskem is a chemical industry that produces different polymeric resins, such as PP, PE, PVC, and EVA. One of the most commercialized materials is resins for flexible plastic packing production. During the production of this kind of material, gels formation can occur. Gels can be defined as any imperfection visible to the naked eye or touch-sensitive that can assume a spherical shape, fibrous, striated, and sometimes stained. Gels cause imperfections that can impact the film production process, aesthetic imperfections (such as label printing flaws), or even cause mechanical failures, such as packing rupture during handling or transport.
The gel classification and identification are not trivial, it involves the sampling of these from the polymeric film and the posterior analysis by optical microscopy. Since this consists of, heating the film to a temperature above the material fusion, and after colling until the material solidification/crystallization. During these steps, images are taken that allow evaluation of the gel behavior in front of the heating and cooling process, which makes possible the classification of different source types, that can be linked to the resin or film production process, as well as external contamination. The indication of the gel origin guides the researchers or the process engineers to find a solution to this problem.
However, this analysis process is very costly for the Braskem laboratories because it comprises a great analyst time effort and the need to have experience in how to execute this analysis. Since the gels classification needs many comparisons with past analyses to better understand the gel behavior during the heating–cooling process. It results in extensive time to deliver the analysis results, in addition, to impact in other lab demands, which also can be delayed. Also, the training of new analysts is costly because it is needed many rounds of analysis until to get experience enough to correctly classify the gels.
So, the challenge was to find a way to accelerate/facilitate the gels classification, and also improve the navigation through the historical data, which helps the training of new people. For it, a project was elaborated that involves the development of artificial intelligence models for gel detection and classification, and a graphical interface to use the model and consult the past data. Thus, Dataiku was identified as the tool to meet the needs of this challenge.
Business Solution:
The development of artificial intelligence models that detect objects and classify them is an activity that demands a great amount of time and also requires a great number of scripts/codes. Dataiku helped the team to navigate through the historical database of images and to select the best images to train the models, develop the detection and classification models themselves, create an image processing flow for new analyses, and create the user interface that the users will use to interact with the models.
The project is in development for six months and consists in:
- Python scripts to process the images for models training;
- Recipes for data transformation and filtering;
- Gel detection model;
- Gel classification models;
- Visual recipes for the users to review the models' classification;
- Application development for the users to interact and use the models;
- Models’ results data storage.
After the development of all these steps in different projects in Dataiku, all of it was consolidated into only one project, that holds the application for users to engage.
Day-to-day Change:
Before the Dataiku usage, it was known that it will be possible to develop this project, however, it would demand a great time to develop, besides the need to use many different tools to reach the final goal. It would result in a considerable effort to maintain the scripts/codes, develop new features, and integrate new people into the project.
The platform sped up the project because it made it possible to aggregate all the developments into only one tool, which has facilitated considerably the management of the scripts/codes. The simple artificial intelligence development/implementation models interface for detection and classification models creation and training has made the work much easier. The integration of new developers into the projects was also greatly accelerated by the use of Dataiku as most of the functionalities are no-code, which has made it possible to onboard people with no previous knowledge of programming in the development, which sped up, even more, the project.
For the lab that executes the gel detection and classification analysis in polymers, it will also be a great change that will bring many benefits. Before it was a very manual activity and dependent on the analyst's experience. Now, with this tool, besides the automation of several steps of the data processing process, the analysts will count on its support of it to classify the gels, which will decrease the time to deliver the analysis results and speed up the training of new people.
Furthermore, this work with Dataiku increases the capacity of the team to deliver solutions to the polymer analyses laboratories, since simple data transformation (such as applying business rules) until advanced models for classifying objects in images.
Business Area Enhanced: Product & Service Development
Use Case Stage: In Progress
Value Generated:
The metrics to measure the success of the project were:
- Feasibility to develop models that are capable of classifying the gels;
- It was proved that is possible to develop models that are capable of correctly classifying the gels in polymeric films and that is possible to continuously improve the models' performance.
- Time-saving during the development;
- The time saved during the development was estimated at one FTE.
- How easy is it to maintain the application;
- It has been proven that the maintenance of the application and the development of new features/models is significantly improved by the use of the Dataiku;
- How easy is the use;
- By the use of the application designer a simple and intuitive interface can be developed, that can comprise all the steps to process one or more images.
- Time-saving in the execution of the analysis in the lab;
- The time that can be saved by the lab using the application in the data processing is estimated at 30%.
Value Brought by Dataiku:
The value brought by Dataiku is related mainly to time-saving during the development and maintenance of what was developed because due to the platform capabilities and the no-code/low-code features together with the capacity to add scripts created by the team in the flow, it greatly facilitated all the project developments. Due to it, the integration of new people into the project was also accelerated, since all things were in the same place that is the Dataiku.
Furthermore, this project and the usage experience in the platform brought by it increased the possibilities of what the team can deliver in solutions for the labs, this is in the most diverse types of projects, which vary from: small data automation, such as applying business rules; lab equipment results data processing; advanced historical data visualization; image data processing to extract analytical data. Therefore, it is possible to say that Dataiku is also accelerating the digital transformation in the Labs.
Value Type:
- Improve customer/employee satisfaction
- Reduce cost
- Save time
Value Range: Thousands of $