Performing Department
BIOSYSTEMS AG EGR
Non Technical Summary
Thermal food processing, which includes drying and cooking, is essential for preserving and sterilizing our food. However, it's not very energy-efficient, often using more energy than necessary.Improving the efficiency of this process could significantly reduce carbon emissions, benefiting our environment.However, the challenge is that this food processing method is subject to many variables. For example, changing the type of food being processed or the equipment used can disrupt the entire system, requiring it to be recalibrated or even rebuilt. Furthermore, certain essential data, like the exact moisture content in food during processing, is tricky to measure in real-time, which complicates optimization efforts.This project aims to revolutionize the way we manage these challenges. By creating a Cyber-Physical System (CPS), we should be ableto manage and account for both the variables we can measure and those we can't. This is done by building flexible models that can be quickly adjusted or combined with others, making the process more adaptable.Moreover, while there are detailed simulations available that can help predict the outcomes of certain inputs, these simulations take a lot of computational power and time. This project proposes an innovative solution: creating a simplified version of these simulations (Reduced-Order Modeling, ROM) which is faster but still reasonably accurate.The ultimate goal is to make thermal food processing smarter and more energy efficient. By using these new models and systems, webelieves they can optimize the processing conditions in real-time. The models and systems will be validatedusing specialized equipment at Michigan State University, focusing on how well the systemscan handle drying food with hot air.In summary, this project is about making our food processing greener and more efficient. It combines knowledge from food science, engineering, and technology to make a difference in the industry and, ultimately, our environment.
Animal Health Component
30%
Research Effort Categories
Basic
30%
Applied
30%
Developmental
40%
Goals / Objectives
This project focuses on optimizing thermal food processing operations, including drying, cooking, and pasteurization. These operations play a crucial role in food preservation and have a significant impact on the food supply chain. Unfortunately, they are known for their low energy efficiency, with energy losses ranging from 10% to 60%, primarily due to the use of fossil fuels. A 10% improvement in energy efficiency could lead to a substantial reduction in carbon emissions.Research Goal: The main goal of this research is to develop and implement a novel cyber-physical system (CPS) for real-time management of process variability in thermal food processing. This system aims to optimize food processing operations by addressing both measurable and unmeasurable sources of variation.Objectives:?UMD will take the lead in Tasks 1 and 4 while MSU will be responsible for Tasks 2 and 3. A detailed description and milestones of each task are given below. Table 2 lists the project schedule and milestones.Task 1: Reduced-Order Modeling (M1-M24)Task Summary: A CFD model of continuous hot-air drying involving both equipment and food will be built using nominal food model parameters. The CFD model will be used as a baseline model, and supply necessary information for construction of the ROM. The CFD model will be validated by drying experiments (Milestone 2). Equipment ROM with limited input will be built by implementing the Krylov subspace method based on 'alter-and-excite' scheme. Food ROM will be built based on the spectral method. The two ROMs will be combined to simulate hot-air drying of a stationary food product. In the second phase of this task, the equipment ROM will be modified to take multiple food locations and the coupling for moving products will be implemented. Optimal placement of interpolation points in the complex plane will be implemented to improve the accuracy of ROM. Validation experiments will be conducted with the actual drying process with two different types of food products to demonstrate the capability of plug-and-play assembly of ROMs.Milestone 1.1 (M12): Demonstrate an initial ROM having less than 1% error from the CFD model. Milestone 1.2 (M24): Demonstrate the capability of plug-and-play ROM assembly using two types of food products and achieve less than 10% overall root mean square error (RMSE) with both types after validation experiments.Task 2: Initial Food Model (M1-M12)Task Summary: Disc-shaped food products, such as apple and potato slices, will be modeled using primary heat and mass transfer PDEs with first order kinetic models for microbial safety and quality retainment. Off-line parameter estimation will be done by a) Plotting scaled sensitivity coefficients to determine parameter identifiability; b) Fixing the value of parameters to which the model is not sensitive; and c) Estimating the remaining parameters using sequential estimation, which updates the parameters as data are added and the experiment proceeds. The primary food model will be inserted into the CFD model of Task 1 and validated by drying experiments for Milestone 2.Milestone 2 (M12): Estimate drying/safety/quality parameters to achieve the accuracy of the model within expected biological variability of the food material and model bias, which may range up to 20% RMSE.Task 3: Enhanced Food Model and Validation Experiments (M13-M36)Task Summary: The initial model will be further calibrated with additional quality or safety data, variables, and more accurate parameters for enhanced prediction of the overall goals. To improve the models, parameter values from initial experiments will be used to create a more efficient set of the next experiments using optimal experimental design. Residual analysis will be used to determine whether the models need additional terms or modifications. The final target accuracy to achieve will be approximately 10% error in terms of RMSE. We will enhance the food model (hot-air drying) for possible improvement of overall performance by introducing humidity control in the system. For this, a previous MSU modified moist hot-air model will be incorporated into the developed primary model to increase drying efficiency and microbial inactivation efficacy by 10% overall. The MSU team will participate in the experiments for validation of soft-sensing and on-line control (Milestones 4.1 and 4.2).Milestone 3 (M24): Demonstrate the accuracy of enhanced food model reaches less than 10% overall RMSE after validation experiments, by reducing model bias.Task 4: Soft-Sensing and On-Line Control (M13-M36)Task Summary: Robust on-line estimation of product quality and important process variables will be achieved by implementing a multi-rate identification scheme and by carrying out the identifiability study based on Fisher information theory. The proposed on-line control framework will be developed by setting up the optimization scheme first and then by implementing adaptation framework using feedback. Both soft-sensing and on-line control methods will be tested and tuned using the CFD model of hot-air drying. Validation experiments will then be conducted with the actual drying process by applying unmeasurable disturbances on purpose (e.g., a sudden change in moisture contents of incoming food).Milestone 4.1 (M30): Demonstrate achieving less than 5% RMSE in soft-sensing of product qualities with the actual drying process by accounting for the unmeasurable food variations and demonstrate maintaining the accuracy when disturbances are applied.Milestone 4.2 (M36): Demonstrate maintaining the optimal performance in terms of energy efficiency and yield when disturbances are applied to the actual drying process.
Project Methods
A real-time Cyber-Physical System (CPS) with distributed-parameter components needs efficient, high-fidelity models for optimal performance. Reduced Order Models (ROMs) have gained interest, but a suitable model for on-line control of thermal food processing remains elusive. Existing ROM techniques often depend on the discretization operator of the Computational Fluid Dynamics (CFD) model, which is not always accessible. The variability in ROM performance based on training data selection also poses challenges. To overcome this, a robust industry-compatible ROM technique will alter the system's digital twin for better identification in real-time CPS.Quick model assembly is common for lumped-parameter systems, but is rare for distributed-parameter ones due to the intensive computational demands of simulation and a lack of prior experience. This project aims to bridge this gap by introducing a model-building process that partitions the model into modular, replaceable components. This includes developing separate model components with ROMs and ensuring their smooth interplay.The research focuses on the hot-air drying process where food moves continuously on a belt through a drying equipment. Specifically, a pilot-scale oven at MSU is under investigation. It comprises a heat exchanger for heat provision and a blower for airflow regulation. The process's dynamics are influenced by multiple controls, such as heat input and blower speed, and unpredictable factors like food moisture content. The goal is to develop a model linking control inputs with outputs determining food quality and safety.The drying model is modular, mirroring the physical events during drying. It integrates heat and moisture transport equations, predicting food safety and quality outcomes. Key to the research is the Food Model, using a two-dimensional disc, like an apple or potato slice, to accurately represent the subject. This model employs Partial Differential Equations (PDEs) to predicttemperature and moisture changes, abiding by mass conservation principles. During drying, food undergoes temperature rise and moisture decrease, altering its properties. Determining these changing parameters is intricate. We willuse methods to measure these variables, utilizing model sensitivity analysis and experimental designs. The drying process affects food safety and quality, needing precise models to guide it. The model will encompasshow drying influences food attributes, such as color and vitamin content. The outlined approach ensures the drying retains food quality and nutritional essence.The CFD model will bedesigned to replicate the physics of the hot-air drying process, capturing the intricate evolution of temperature and humidity. Despite its complexity, encompassing multi-physics and time-dependent elements, the task is achievable through current CFD tools and simulation advancements. This model will dynamically simulate turbulent two-phase flow (air and moisture), and changes in temperature and moisture in moving food items. Each simulation offers a comprehensive view of flow and food states over time. The domain represents the pilot-scale oven's interior, accounting for variables like velocity, temperature, and turbulence at inlet points. Variability in temperature and humidity necessitates a model with adaptable gas properties. Techniques like Unsteady Reynolds-Averaged Navier-Stokes (URANS) and Detached Eddy Simulation (DES) will be applied for turbulence effects. The model will also consider temperature and moisture changes within food using discretized PDEs based on ideal 2D shapes. The food properties and parameters will be derived from experimental data and literature. Interaction between air flow and food models is ensured by applying instantaneous air property values as boundary conditions to PDEs. Simulations will be done using STAR-CCM+ software, but their computational intensity poses challenges. Although direct CFD simulations can analyze specific cases, they aren't ideal for quick optimization or real-time control due to long processing times. Reduced-order modeling offers a solution to this limitation.Reduced-Order Modeling (ROM) is a technique developed to simplify the complex computations inherent in Computational Fluid Dynamics (CFD) models. The CFD model of air/vapor mixtures is composed of extensive nonlinear differential equations, which can be represented in a polynomial system. This project will use a physics-based ROM approach, where the high-dimensional solution vector is constrained on a low-dimensional subspace. This reduces the computational complexity significantly. The challenge arises as many industrial CFD tools, like ANSYS or CCM+, are closed source, making it difficult to derive the necessary equations. The project team has devised the 'alter-and-excite' method, which allows for generating ROMs without the need to access the CFD model's source code. This technique has proven efficient in previous applications, like simulating the spread of aerosol particles, where it achieved massive reductions in computational cost with minimal error. For this project, the method will be extended to nonlinear systems. In the context of food modeling, due to the food's ideal 2-D shape, a simpler spectral method will be employed, which uses analytical functions to represent spatial modes, eliminating the need for spatial discretization. This method has previously been successful in modeling Lithium-ion battery cells. The primary goal is to produce accurate yet less computationally intensive models.Models will be tested using a pilot-scale oven at MSU, measuring food slice moisture and temperatures under various conditions. A datalogger and infrared camera will record temperature data, and results will be compared to model predictions using RMSE. The CPS's real-time adaptability will be assessed through three milestone tests, focusing on model adaptability during food switches and optimal conditions, measuring yields and energy. Energy savings from optimization will be gauged against benchmarks from literature and industry advice.