INVESTIGATIONS ON FORCED-AIR PRECOOLING STRATEGIES FOR THE PRESERVATION OF PERISHABLE FRUITS

Abstract

Preserving horticultural produce (i.e., fruits and vegetables) is vital for quality products to be delivered to customers long after harvest. In this regard, temperature management strongly influences produce qualities and deterioration rates. The major causes of early deterioration generally include biological activities (e.g., respiration and transpiration) triggered by the field heat in harvested produce. Therefore, it is necessary to slow down these biological activities and remove field heat by lowering the temperature of the harvested produce. This requirement of lower temperatures immediately after harvest is fulfilled by rapidly cooling the produce (i.e., precooling process) to storage temperatures. For various perishable fruits and vegetables, forced-air cooling (FAC) is one of the most prevalent precooling techniques due to its low cost and adaptability. The present thesis aims to develop and generate efficient FAC strategies with experimental and numerical analysis and help mitigate existing problems by providing optimized and sustainable solutions. The work is divided into three major parts to meet the research objectives: experimental analysis of FAC of apples and investigations on flow reversal strategies, development of numerical modeling technique for flow reversal combined with velocity variations during the cooling process, and development of prediction models for temperature measurement of apples during cooling using artificial intelligence (AI) and machine learning (ML) techniques. The first part of the work focuses on developing reliable experimental facilities to accomplish the research objectives. After the successful development, experiments were used to analyze the effects of periodic airflow reversal on cooling characteristics (i.e., cooling rate, uniformity, and energy consumption). Firstly, an improved vent design was proposed for distributed flow and used for comparing different flow reversal strategies at three constant cooling velocities (i.e., 1 m/s, 2 m/s, and 3 m/s). Also, the overall heterogeneity index (OHI) based on surface heterogeneity maps and energy consumption based on fan power and seven-eighth cooling times (SECT) were used to quantify the whole process uniformity and conspicuously compare different strategies. The results show that the periodic flow reversal improves precooling time, energy consumption, and cooling uniformity within the package and fruit for all flow velocities. Especially for higher velocities (i.e., 3 m/s) that are generally preferred for faster cooling, a significant improvement (i.e., 43.89% reduction in OHI with three reversals) in cooling uniformity was observed with more flow reversals. Finally, some cooling strategies and guidelines for utilizing flow reversals are proposed to achieve more uniformities at different flow velocities. The second part focuses on the development of a numerical model and analysis of the effects of various parameters on the FAC process. The study analyzed the effects of flow velocity variations combined with flow reversals on primary cooling characteristics (i.e., cooling rate, cooling uniformity, and energy consumption) with both individual and multi-parameter optimization using validated numerical modeling. A simulation approach is proposed for combined velocity variations and flow reversals, providing consistent results with the experimental data. Fifteen cooling plans were formulated based on the number of flow reversals, reversal frequency, and velocity variation analysis. The analysis shows that the energy consumption based on the pressure drop increased significantly when the airflow velocity exceeded 2 m/s. The cooling rates were higher for higher velocity (i.e., 3 m/s); however, the energy consumption tripled when compared to 2 m/s. Therefore, some cooling strategies were suggested based on cooling rates and energy consumption, and appropriate compromises were made between both for an optimal system. Finally, some novel cooling strategies were proposed from cooling plans analyzed using multi-parameter optimization. The third part focuses on the development of prediction models for temperature using AI and ML algorithms. This study proposed a novel supervised learning technique to generate datasets and predict apple average and extreme temperatures inside a package during the FAC of apples for performance evaluation. The predictions using traditional ML, ensemble, and ANN models were analyzed to compare their accuracy and applicability for different FAC conditions. First, the prediction models were optimized using hyperparameter tuning, and then the optimum hyperparameters were used to train the models for predictions. The results were verified with the help of two datasets: one test dataset and the other unknown conditions dataset. The results on the test dataset show that ensemble models (i.e., XGBoost) made the most accurate predictions with the least mean square error (MSE) of 0.0093. For the unseen conditions, recurrent neural network (MultilayerRNN) predictions were more accurate than other models with an MSE of 1.211, followed by the ML support vector regression (SVR) model. It was revealed that MultilayerRNN and SVR have better adaptability and generalizability for temperature predictions than other models. A comparative study was also performed using trained prediction models for the effect of changing airflow velocity on FAC performance. A decrease of 24.73% was predicted in cooling time by increasing velocity from 1 m/s to 2 m/s, comparable to the actual decrease, which was 25% during simulations, proving the applicability of prediction models for comparative studies. Finally, the challenges and further improvements needed for the practical applications of prediction models in the FAC of fresh produce were highlighted.