Stipendiatstilling 1 Nofima (2021-2024)

Industrial food quality measurements open new possibilities for improved sustainability in food production processes, by optimizing the use of raw materials and reducing food loss and waste. There are usually many quality attributes and process parameters representing different stages of the production process, from raw materials and ingredients to end products. By combining all relevant data sources, we can develop digital systems for monitoring, optimisation, or decision support. The road from sensor data to well-functioning systems is however not straightforward. The main challenges lie in pre-processing and combining data from different sensors, developing reliable models that relate quality attributes to controllable production parameters, and finally designing and implementing user-friendly systems based on such models. The aim of this project is to develop data analytical methods and strategies addressing all these challenges. The work is based on use cases and data from the two Norwegian companies Bioco and Biomega, who process rest raw materials from poultry and salmon through enzymatic protein hydrolysis. The first part of the work addresses the challenge of combining data from different sensors along a continuous production process. A given raw material need time to travel through the process, and sensor data therefore need to be shifted in time to represent the same raw material at different points in the process. We have made an overview of a wide selection of methods for estimating time lag between sensors and compare their performance in different scenarios. Based on these results we have developed guidelines for choice of method, and these are applied in modelling applications at Bioco and Biomega. When analysing industrial time-series data, dealing with time dynamics is crucial. These dynamics can arise from feedback control systems or be inherent to the process and ignoring them can lead to inaccurate models. Therefore, the second part of the work focuses on methods for modelling such dynamics. We have proposed and compared three alternative methods for modelling dynamics in situations where there are several blocks of variables, for instance corresponding to different steps of the production process. Our results show that all perform well in terms of predictive ability, but they are different when it comes to interpretation of the effects. Based on this, we have developed recommendations for choice of method. In the last part of the project, we combine the previous work into a comprehensive workflow for developing prediction models from industrial time series. The workflow includes pre-processing of the raw sensor signals, synchronization, machine learning and validation. We also suggest a systematic way of optimizing the entire workflow, all the way from raw data to model validation. In conclusion, this project has developed methods and guidelines for modelling industrial time series data. We have used data from enzymatic protein hydrolysis as case studies, but the methodology can be applied to any type of continuous production processes.

For Norilia, Bioco og Biomega har prosjektet ført til bedre forståelse av variasjonen i råvarer og sluttprodukter, og sammenhengen mellom dem. Denne innsikten brukes nå til å planlegge videre forskningsaktiviteter for å undersøke hvordan prosessen kan optimaliseres, med et langsiktig mål om å innføre systemer for kontinuerlig måling og styring. Forskningspartnerne i prosjektet har fått en utvidet verktøykasse for å analysere industridata. Disse verktøyene vil brukes videre i andre prosjekter, slik at lignende prosjekter med andre industripartnere blir mer effektive. Metodene er dessuten delt åpent i forskningsartikler og som kjørbar kode, og kan brukes fritt av andre aktører innenfor forskning og utvikling. Prosjektet har etablert et nytt forskningssamarbeid mellom Nofima og det polytekniske universitetet i Valencia (UPV), som fortsetter videre i SFI DigiFoods. Resultatene i prosjektet vil gagne næringsmiddelindustrien og teknologileverandører ved at vi har demonstrert hvordan måleteknologi og datamodellering kan gi ny innsikt og potensielt brukes til overvåking, optimalisering og styring av prosesser. På samfunnsnivå kan prosjektet bidra til mer effektiv bruk av råvarer, redusert matsvinn og dermed økt bærekraft i matindustrien.