Data estimation methods for predicting temperatures of fruit in refrigerated containers

Abstract

Improving the capability and resolution of monitoring perishable products during their transportation and storage is essential, but there is a key requirement it is not to increase costs or the number monitoring devices. Currently there lies a knowledge gap in studies on the spatial prediction and mapping of determinant parameters (e.g. temperature) for the shelf life of perishable products. Through the viewpoint of different refrigeration failure scenarios this paper investigates and compares three data estimation tools (artificial neural networks, Kriging and capacitive heat transfer) for improved food safety. Results indicate that using these techniques makes it possible to reduce the number of sensors (through estimation of temperature distribution) within an industrial scale fully loaded strawberry-shipping container, thus reducing the overall commercial cost. Using a set of eight source sensors, an average error of 0.1 °C was achieved, which represents an improvement of 97.14% in regards to the absolute error between the ambient and product temperatures. Even when using only a single container sensor as a source for prediction, with an average error of 1.49 °C there still was an improvement of 62% with regards to the same baseline. This paper demonstrates that the adoption of these technologies not only presents significant industrial value-added potential but also the data obtained can further improve cold chain strategies and reduce product losses through more accurate shelf life calculations.