Progress 05/01/22 to 04/30/23
Outputs
Target Audience:Researchers and educators, two of the target audiences, were reached individually and at the largest of the food engineering conferences in North America (an international one), Conference on Food Engineering. The database was presented, demonstrated, and tried out by the audience. Feedback was collected. Changes/Problems:Although there have been many hiccups in the past, the immediate past year has been productive and no major changes were needed. What opportunities for training and professional development has the project provided?A module for learning about thermal conductivity, using the database mentioned above, has been implemented in one formal classroom. The graduate student working on the project has been extensively trained in building data-driven models. How have the results been disseminated to communities of interest?Students in a formal classroom have used the database. What do you plan to do during the next reporting period to accomplish the goals?Complete the remaining development that will allow crowdsourcing, i.e., allow others to contribute to the database (so it can grow). Vastly increase the amount of data in the database by recruiting students who would enter data. Reach out to both researchers and industry personnel who are the eventual users.
Impacts
What was accomplished under these goals?
The database is functional now. Based on composition, it can predict density, specific heat, enthalpy,thermal conductivity, mass diffusivity, water activity, initial freezing point, latent heat, porosity, bound water, and dielectric properties of a large group of foods. These are available for use by anyone and useful in industrial product and process design, teaching, and research. We have also completed a rapid simulation platform that will allow instant "what if" in product and process design.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Ghosh, D., & Datta, A. (2023). Deep learning enabled surrogate model of complex food processes for rapid prediction. Chemical Engineering Science, 270, 118515. https://doi.org/https://doi.org/10.1016/j.ces.2023.118515
|
Progress 05/01/21 to 04/30/22
Outputs
Target Audience:Students in a scheduled food engineering/scienceclass at Washington State University used a teaching module that incorporatesthe food properties estimation database developed in this project. This is the education component of the project. Changes/Problems: There has been continuous difficulty in obtaining database-related help with the small amount of funds we had set aside in the proposal for this purpose. This has seriously delayed the project The surrogate model development, whichreduces simulation time from hours to seconds, is a cutting-edge computational technique that took longer to master than predicted. What opportunities for training and professional development has the project provided? Opportunity for the graduate student to be trained in absolutely the cutting-edgecomputational approaches that will make food process simulation use and computer-aided food manufacturing a reality. Opportunity for students in a food engineering class to use state-of-the-artsoftware for doing "what if" scenarios in a design context that they can later use in industry. Also, they are able to look deeper into the material science aspects of food. How have the results been disseminated to communities of interest? It has been disseminated to classrooms It is being disseminated to other researchers What do you plan to do during the next reporting period to accomplish the goals? Add crowdsourcing to the database so everyone can contribute, making it a community resource. This will ensure its usefulness as more data are added and also make it sustainable beyond the life of this project. Build a digital twin (predictive software) for the safety and quality of a drying process. Build the digital hub for tools in the prediction of food product and process development Build educational modules for use in classroom
Impacts
What was accomplished under these goals?
The software to estimate physical properties is now able to add discreet data (on top of estimation formulas) easily using an Excel spreadsheet. It is also able to show simultaneously measured and estimated data, providing more confidence in the estimated data and a deeper look into the data. Trying "what if" scenarios in process and product design is now enabled with a reduced order neural network simulation of the product-process combinations. Our approachis an enabler by bringing down simulation time from hours to seconds, while still keeping the mechanistic basis of the simulations. We have started to build learning modules that are based on the above software, allowing active learning in the classroom.
Publications
|
Progress 05/01/20 to 04/30/21
Outputs
Target Audience:
Nothing Reported
Changes/Problems:We had two major setbacks. One was the pandemic. The other was locating an IT person/company for the database development that we can afford in the small amount (now we realize) of funds we had budgeted. The database development is progressing at this time but has caused major delays completely beyond our control. In addition, we also learned the building of a surrogate model for fast simulation takes a lot longer than we had planned. What opportunities for training and professional development has the project provided?One student has been trained in his doctoral program in the emerging discipline of data science. Work is being presented at IFT and ASABE meetings. Another student has been trained in the understanding of food physical properties for prediction purposes. How have the results been disseminated to communities of interest?We have made a presentation at the annual meeting of the A1363 Advanced Manufacturing Technologies PD meeting of the USDA. What do you plan to do during the next reporting period to accomplish the goals? This year, we will make major progress in completing the web interface by including crowdsourcing and visualization capabilities. We will include the data and predictive equations for rheology, high pressure, and storage of fruits and vegetables. Afterward, we will develop the crowdsourcing protocol--how to get others interested in entering their research data into this database. We will invite several colleagues to have access in exchange for them entering data and equations. We will develop training modules for academia and industry for learning deeper aspects of food material properties using this database. On the simulation side, we will build an app that provides cloud-based simulation that is universally available and usable. We will invite users and obtain feedback. For the video database, we will reach out to more researchers for additional simulations to be included. As soon as the tools are built, we will be in touch with several food-industry personnel for their use and feedback, to help improve the tools.
Impacts
What was accomplished under these goals?
Important note: **Per guidance from our USDA Grants Management Specialist, the reporting below is the same as what has already been submitted for the year 2021-22, which was a cumulative report.** The web interface and, more importantly, the underlying data and knowledge base have been built. The entire system is functional barring small sporadic issues. This means interactively one can instantly obtain certain physical properties data for a food material from this database simply by choosing the food material. Food properties data for rheology, high-pressure processing, and storage of fruits and vegetables have been compiled. Conceptual frameworks for the prediction of such data have been partly built that will allow a coordinated and sustainable approach for prediction. To do "what if" scenarios through simulation more practical for their use in the product, process, and equipment design in food industry, we have successfully built a neural network surrogate model for drying-like processes. This surrogate model predicts reasonably well the same output as the complex full-physics model from which it is trained, but it does so at 1/100th the time, making "what if" predictions very practical. This work is novel, difficult, and has been computationally challenging. We have built the platform to store and retrieve annotated videos providing insight into complex food processes.
Publications
|
Progress 05/01/19 to 04/30/20
Outputs
Target Audience:
Nothing Reported
Changes/Problems:We had two major setbacks. One was the pandemic. The other was locating an IT person/company for the database development that we can afford in the small amount (now we realize) of fundswe had budgeted. The database development is progressing at this time but has caused major delays completely beyond our control. In addition, we also learned the building of a surrogate model for fast simulation takes a lot longer than we had planned. What opportunities for training and professional development has the project provided?One student has been trained in his doctoral program in the emerging discipline of data science. Work is being presented at IFT and ASABE meetings. Another student has been trained in the understanding of food physical properties for prediction purposes. How have the results been disseminated to communities of interest?We have made a presentation at the annual meeting of the A1363 Advanced Manufacturing Technologies PD meeting of the USDA. What do you plan to do during the next reporting period to accomplish the goals? This year, we will make major progress in completing the web interface by including crowdsourcing and visualization capabilities. We will include the data and predictive equations for rheology, high pressure, and storage of fruits and vegetables. Afterward, we will develop the crowdsourcing protocol--how to get others interested in entering their research data into this database. We will invite several colleagues to have access in exchange for them entering data and equations. We will develop training modules for academia and industry for learning deeper aspects of food material properties using this database. On the simulation side, we will build an app that provides cloud-based simulation that is universally available and usable. We will invite users and obtain feedback. For the video database, we will reach out to more researchers for additional simulations to be included. As soon as the tools are built, we will be in touch with several food-industry personnel for their use and feedback, to help improve the tools.
Impacts
What was accomplished under these goals?
The web interface and, more importantly, the underlying data and knowledge base have been built. The entire system is functional barring small sporadic issues. This means interactively one can instantly obtain certain physical properties data for a food material from this database simply by choosing the food material. Food properties data for rheology, high pressure processing, and storage of fruits and vegetables have been compiled. Conceptual frameworks for prediction of such data have been partly built that will allow a coordinated and sustainable approach for prediction. To do "what if" scenarios through simulation more practical for their use in product, process, and equipment design in food industry, we have successfully built a neural network surrogate model for drying-like processes. This surrogate model predicts reasonably well the same output as the complex full-physics model from which it is trained, but it does so at 1/1000 the time, making "what if" predictions very practical. This work is novel, difficult, and has been computationally challenging. We have built the platform to store and retrieve annotated videos providing insight to complex food processes.
Publications
|
Progress 05/01/18 to 04/30/19
Outputs
Target Audience:The annual report was shared with an academic audience showing the potential of the tool being built in education and research. Changes/Problems:Building a database of material properties with an engaging web interface and various functionalities has the inherent challenge of finding a good computational framework for sustainability. After much searching, we could locate a database professional to not only advise but also tobuild the database from scratch with the latest technological applications. This took longer as the project has only a small amount of funds for such activity. For the process simulation app to be successfully used in food manufacturing, it is imperative to come up with a tool which can be operated on most computing systems without any requirement of specialized and resource intensive license. The app should also generate the desired results relatively quickly without surrendering the robustness of physics-based model. The first bottleneck is addressed by the compiler technology of the commercial software COMSOL that only recently became available. The research challenge associated with the reduction in computing time of a complex model has been a bigger challenge than we predicted. The main idea behind presenting a collection of videos in a database is to offer clear visualization of different phases of a food process and offer scientific insights behind them. The challenge behind the visualization partwas to come up with an appropriate video format and and provide synchronization of text with video. Our greatest challenge is to populate the database where we are not getting many takers. We are working on incentivization to make the process go faster, with a Cornell faculty member who specializes in information science to come up with innovative solutions. What opportunities for training and professional development has the project provided?Three graduate students got trained in food properties prediction and the application of computer science tools to effective food process visualization. Of these, onegraduate student with food engineering backgroundworked to produce fast simulation of food processes. This graduate student also provided overall supervision of the project involving the three tools, thus learning about multiple aspects of digital food. Two graduate students with computer science background helped develop the video database. How have the results been disseminated to communities of interest?The tools have been presented to some of the academic community. What do you plan to do during the next reporting period to accomplish the goals?We plan to enable different functionalities and convenient visualization in the web interface for food properties. This will help the user interact with the database actively to draw physical insights into properties data from the database rather than just using the database as a standalone collection of property data. For the second tool, we plan to implement physics-based reduction of the complex porous media models and also make some progress in towards a parallel neural network solution. For the third tool we plan to populate our video repository with some more good quality videos and annotated insights. We also plan to start crowdsourcing of the video database.
Impacts
What was accomplished under these goals?
1) We have developed a knowledge base of properties with a queryable web interface with regard to the first tool of the project. The backend of the database as well as the web interface is built by a group of professional and competent personnel who understands the importance of a sound data infrastructure for the long-term success of the database. Collaborating with of our Co-PDs we have come up with a framework for building the knowledge base for various sets of properties. 2) We have made some headway in terms of solving the research challenges involved in making a process simulation app (the second tool). We have chosen to explore three methods of reducing the computational needs of the mechanistic model and completed some preliminary research. These are operating parameter dependent, physics-based reduction of governing equation, an artificial neural network trained with the model solutions, and numerical reduction techniques of the solution procedure. 3) The web-based repository of videos describing food processes is built with careful addition of timed annotation containing scientific insights behind the visual changes in the food seen in the video. The tool will help specialized food process developers by providing valuable scientific information regarding various aspect of these processes.
Publications
|
|