Non-Invasive Milk Quality Monitoring using mm-wave Radar System

Abstract

The Internet of Things (IoT) technologies is becoming increasingly valuable in industries for real-time monitoring of goods during manufacturing. This thesis explores the implementation of a low-cost electromagnetic radar solution to detect the quality of goods as they progress throughout the production, enabling companies to respond swiftly to potential quality issues that may arise. A specific application is demonstrated in the dairy industry, where milk can be monitored by tracking changes in the dielectric properties as deviations from a defined standard.

The dielectric properties of various concentrations of salt and sugar dissolved in distilled water are measured using an open-ended coaxial probe technique over the frequency range 200 MHz to 20 GHz. The conducted literature survey indicates a decrease in the value of the dielectric constant and an increase of the dielectric loss with an increase in concentration over this frequency range. Since the dielectric properties vary, the survey indicates that it is possible to develop a radar that can accurately detect these concentrations and the slight variances in a food product.

To investigate the relationship between milk fat content and dielectric properties, milk and cream samples with fat content ranging from 0.1-35% were tested using an open-ended coaxial probe connected to a vector network analyzer across frequency range 200 MHz to 20 GHz. The results revealed a linear decrease in both the dielectric constant and dielectric loss with increasing milk fat content, while other milk constituents also influenced these properties. Two regression models were developed to predict the fat content based on dielectric measurements, yielding promising accuracy.

Further measurements were extended up to 67 GHz to evaluate dielectric property variations at 60 GHz, the targeted frequency for radar implementation. Using this data, simulations were conducted in Ansys HFSS to validate the ability of a simple radar system to distinguish between different products based on their scattering parameter responses. Additionally, two electromagnetic millimetre-wave frequency-modulated continuous wave (FMCW) radars were tested on milk cartons, both stationary and moving on a conveyor belt. Data collected from the moving cartons were used to train and test a support vector classifier algorithm in Python, achieving 87.5% accuracy. The performance was compared with two alternative classification algorithms. This proof-of-concept demonstrates the potential for extending the proposed approach to other applications and industries, offering a versatile solution for real-time quality monitoring.