The convergence of drying technology and Digital Twins is enabling a transition from traditional drying methods to a new phase where the process can be continuously monitored. A crucial aspect is the development of digital model of kinetic reactions. In this context, hyperspectral imaging has been widely used, however, it has limitations due to its high cost and computational intensity, which restrict its application to lab-scale settings. To address this limitation, this study introduces technically simple and cost-effective imaging techniques of Light-emitting diode (LED) and Band-pass filter (BPF). These techniques were implemented at 980 nm and 1450 nm to measure moisture content, Soluble Solids Content (SSC), and shrinkage of apple slices undergoing drying at 60 °C. Images of apple slices during the drying process were captured to train Gaussian Process Regression (GPR) models. Moisture content achieved the highest accuracy, with GPR models yielding R-squared values of 0.996 and 0.992, and RMSE values of 1.74% and 2.21% for LED and BPF, respectively. Similarly, SSC (R-squared ≥0.874 and RMSE ≤7.85%) and shrinkage (R-squared ≥0.967 and RMSE ≤5.23%) were well predicted. Furthermore, accurate prediction results for external apples demonstrated the models reliability and robustness. In line with the concept of digital twins-based smart drying, the 980 nm LEDs embedded into an experimental dryer exhibited high accuracy (R-squared ≥0.968 and RMSE ≤4.6%) in inline prediction of moisture content. This represents a significant step towards the development of digital twins-based smart dryers using cost-effective and technically simple imaging sensors.
Read full abstract