Mar 4.0: Marine Process Industry 4.0

The integration of optical spectroscopy with multivariate modeling has become a widely adopted technology for online monitoring and control of chemical processes. In this BIA-funded project, partly supported by the Norwegian Research Council, the project partners have focused on developing predictive models for online concentrations of EPA and DHA during the distillation of fish oils. The project involves the installation and validation of in-line optical sensors (NIR and Raman) in a distillation plant. Databases with reference samples and corresponding spectra have been established for modeling, prediction, and control to achieve the desired product quality. The project partners are Pattern Recognition Systems, the Chemical Institute at the University of Bergen, and the Seattle-based Raman supplier Marqmetrix Ltd. The project aligns with the principles of Industry 4.0, emphasizing digitalization and automation as key drivers for the next phase of industrial evolution. Industry 4.0 encompasses a range of technologies, including Big Data, analytics, Internet of Things (IoT), cloud services, robotics, and sensors. However, the realization of Industry 4.0's potential depends on addressing challenges such as competency gaps and issues related to data availability, traceability, and quality. GC Rieber VivoMega, located in Kristiansund, plays a central role in this project, with a focus on the distillation process for manufacturing high-concentration omega-3 products. The distillation process involves multiple steps to achieve the desired composition of EPA and DHA. Optical spectroscopy, specifically NIR and Raman spectroscopy, has been utilized in this project for online measurements during the distillation process. A significant aspect of the project is the development of adaptive modeling software aimed at reducing the reliance on in-house expertise for multivariate modeling. The software incorporates elements from local weighted linear regression, assigning importance to samples based on their similarity to samples in a reference database. The emphasis is on local adaptive models, which, based on experience, tend to outperform global static models by approximately 30%. This approach allows companies to focus on maintaining reference databases with less demand for advanced modeling expertise. The concentration information in the spectra is crucial for online predictions. The project builds databases of spectra and reference values for components like EPA and DHA, covering feed, intermediate products, and distillation products. Predictive models for concentrates are constructed without temperature correction, and the entire modeling process, including validation and prediction, is automated. Temperature information is also extracted from the spectra. Both NIR and Raman spectra exhibit changes in band width and intensity with temperature variations. The developed software leverages this temperature information to automatically predict optimal temperature settings for distillation, considering target concentrations for EPA and DHA. The effectiveness of the temperature models has been demonstrated through tests on various productions, with promising results. The predicted temperatures closely follow the set temperatures. The correlation between predicted and set temperatures is strong, indicating the model's reliability. In conclusion, the combination of optical spectroscopy and multivariate modeling offers a powerful tool for swiftly optimizing temperature settings in a distillation processes. The adaptive model development and on-line monitoring contribute to the efficient control of the fish oil distillation process, making it feasible for production operators to achieve and maintain desired product quality efficiently.

The developed software solution will be further tested and validated at GC Rieber VivoMega. As is the case for all machine learning algorithms, the amount and quality of data is crucial for success and more data will improve the prediction quality. The project has developed a software suite (off-line and real time prediction, optimization and calibration transfer) for a consistent handling of spectroscopic data. This may be complementary to other machine learning solutions in various chemical/food industries. The adaptive modelling approach will be particularly important in areas with little spectroscopic modeling competence. The solution is expected to give increased efficiency, product quality and reduced cost. This in a time where we have seen historical high levels of raw material and energy cost. Making the proposed solution highly relevant for economical productivity and resource efficiency.

The underlying idea is to enhance the competitiveness of processing industries relying on biological raw materials by creating a novel technology for automation of production processes based on a multi-disciplinary approach. This approach leans on key concepts from Industry 4.0 and includes flexible data management, advanced online process instrumentation, self-modelling data analysis, and multi-objective optimization. The complexity of this approach mirrors the complexity of the inherent and unique challenges that the processing industry depending on biological raw materials is currently facing. These challenges include large variations in the raw material, complex processing lines, and manual production set-ups with low level of digitisation. This way of processing affects both production capacity and production control negatively, leading to reduced earnings, lower long-term sustainability and poorer international competitiveness. Specific R&D challenges must be solved to automate process control. Modelling and optimizing several variables simultaneously is challenging when the variables are dependent on each other. It is also challenging to obtain a robust, informative signal from online sensors during production of complex raw materials. This project aims to solve these issues by developing a novel, generic, and digitised solution for process control and automation, starting off with one of the key processes in production of omega-3 concentrates from fish oil. Developing this technology in a representative company for the marine processing industry in Norway will pave the way for similar industries in the future to digitise and automate their production. This will increase their competitiveness by taking them into the fourth industrial revolution.