From traditional techniques to the integration of deep learning and advanced sensors. Artificial intelligence is revolutionising energy efficiency and the preservation of organoleptic properties in drying processes.
Drying is a fundamental process in food processing, playing a significant role in extending shelf life and preserving product quality. Compared to traditional sun drying, mechanical drying solves the problem of prolonged drying times. Conventional methods such as hot air drying, vacuum drying and freeze-drying have made it difficult to meet current food preservation requirements due to long operating cycles, lower sensory quality, poor retention of bioactive substances and high energy consumption.
However, relying solely on mechanical drying often results in significant degradation of product quality after processing, failing to meet consumer expectations. This is because mechanical drying typically has inherent h l limitations, including uneven heat distribution and irreversible quality deterioration. These drawbacks stem from an over-reliance on fixed process parameters, which fail to adapt to the dynamic interactions between material characteristics and the process environment.
Although some advanced drying technologies, such as pulsed vacuum drying or radiofrequency drying, have proven effective in improving quality, the causes of deterioration in the characteristics of the treated food are due to inaccurate process control, ineffective or absent real-time monitoring, lack of consideration of the dynamic evolution of the physical and chemical property parameters of the materials, and poor applicability of processing technologies. For this reason, the concept of ‘intelligent drying’ has gained widespread attention in recent years. Smart drying allows real-time adjustment of drying parameters based on changes in the state of the food during the process, thereby reducing drying time and minimising deterioration in quality characteristics.
However, the application of smart drying is accompanied by certain challenges. In the challenging drying environment, which involves high temperatures and significant humidity variations, accurate, real-time monitoring of food quality is difficult, making it challenging for researchers to obtain accurate and continuous databases of the different variables. The drying process is a complex system characterised by non-linearity, temporal variability, uncertainty and a degree of randomness, making it difficult to describe the states of the system through precise modelling. Furthermore, the drying process is irreversible; once quality deterioration occurs, it cannot be restored through subsequent processing and production steps, resulting in high trial and error costs for control decisions.
Deep learning for drying control
The development of deep learning technology offers a new way to solve the above-mentioned problems. Deep learning is a branch of machine learning that can adaptively acquire key features during data processing. This end-to-end modelling approach can achieve better generalisation performance in data processing. Before applying the deep learning model, it is necessary to precisely construct specific datasets for each food item, which requires time. However, the use of deep learning-assisted online monitoring is useful for improving the accuracy of continuous system control in environments with strong interference.
For example, by combining deep learning with computer vision to monitor the appearance quality of food, real-time semantic segmentation of food images during the drying process can be achieved. This solves the problem of difficult segmentation of food regions caused by significant changes in appearance or water vapour during drying. As a result, a comprehensive and accurate database can be constructed.
Based on this database, deep learning can be used to build non-linear adaptation models in the food drying process under the influence of multiple factors. This is because an accurate and continuous dataset can improve the adaptive adjustment capability of the model, which will be able to make accurate and advanced predictions before quality deteriorates, thereby reducing the costs of drying device control decisions and providing a reference for intelligent control of parameters in the process.On this basis, the use of deep learning to support intelligent decision-making control of the drying process significantly improves the accuracy of process system control, solving problems such as exceeding temperature and humidity limits and air velocity fluctuations.
Application challenges
However, deep learning-assisted intelligent drying still poses challenges in the application process. The ‘black box’ nature of the deep learning model limits its interpretability and hinders understanding of the quality evolution mechanism. Furthermore, deep learning models are trained on specific data sets, which limits their applicability to different types of food and production environments. Emerging solutions, such as physical information-based neural networks and digital twins, i.e. dynamic virtual replicas of foods and processes used to simulate, monitor and optimise performance, allowing AI to experiment, can bridge this gap.
By using digital models that mimic reality, AI becomes more accurate, reliable and capable of dealing with new situations it has never seen before. At the same time, advances in multiscale data fusion are expected to overcome the limitations of low-dimensional data in capturing collaborative characteristics between qualities and provide an overview of multiscale data . In the field of drying, the application of these new technologies will allow the integration of physical mechanisms and data-based methods, achieving interpretable monitoring of the quality evolution mechanism in the drying process and improved generalisation capabilities between food types and environments.
Although progress has been made in improving efficiency and quality, persistent challenges remain in generalising the model to different types of food and drying conditions, computational complexity and mechanistic interpretability. Future research will focus on developing hybrid physics-data frameworks, combining data collected from machines with the laws of physics, systems capable of controlling the process at every level, towards an intelligent drying system capable of regulating itself to save energy, time and improve food quality.
Integration of advanced sensors
An intelligent drying system is also based on the integration of various types of advanced sensors into the drying system, as well as algorithms to achieve visualisation, automation and intelligence of the drying process. For example, electronic noses and artificial vision can non-destructively capture data such as visual, chemical/biochemical and morphological changes in the drying process, which in turn can achieve intelligent analysis and automated control of drying parameters. These sensor technologies are fast, efficient, non-destructive, environmentally friendly and sustainable, and are widely used in the food processing and quality control industries.
Artificial intelligence and sensor technology will be at the heart of future research on intelligence in the food processing industry. Their integration into the food industry can help visualise highly non-linear, complex and constantly evolving drying processes, to fully understand their kinetics and guide their optimisation. Artificial intelligence can be combined with 5G and IoT technologies for real-time online monitoring of food drying. Together with sensor technologies, they represent the future of real-time online monitoring of food drying.
Artificial vision for real-time shrinkage monitoring
Shrinkage is an important aspect to consider when developing a model for describing food drying. It changes the shape and size of products and is directly related to water loss during drying, which in turn influences the mass transport and cementation phenomena that occur in some drying processes. One of the innovative technologies used for online measurement of shrinkage during the drying process is machine vision.
Researchers have estimated the level of dehydration of prawns by analysing colour during the drying process based on machine vision. Others have presented a computer vision-based method to analyse the effect of drying conditions on the shrinkage, colour and texture of apple samples. A linear relationship between shrinkage and moisture content was found. Moisture content is a key factor in the quality of the dried product, therefore, the development of a drying model for the food to be processed is essential for predicting, controlling and optimising the process.
