Classification of Range-Doppler Radar Echoes for Condition Monitoring in Industrial Processes

In process industry, for instance in the chemical, pharmaceutical and food sectors, a wide range of processes are applied. These include the storage and processing of liquids inside tanks, but also in chemical and biological reactors, among others. Extending our previous work, the goal of this contribution is to present and evaluate an approach for condition monitoring (CM) in such applications. Since radar systems are already successfully used to determine filling levels, their measurement capabilities are augmented, allowing to distinguish different process conditions (PCs). Our proposed approach is to capture the movements of the surface by range-Doppler radar measurements and to extract distinguishing quantitative features therefrom. A subsequent classification is performed employing techniques from machine learning. The potential of this approach has been experimentally evaluated in terms of typical PCs, such as heating, boiling, stirring, filling and draining. Measurements have been performed using a frequency modulated continuous wave (FMCW) Doppler radar system operating in the range of 68 to 92 GHz. It is shown that the chosen PCs can be distinguished with almost 96% accuracy using Gaussian Processes (GPs) and quantitative features, such as the velocity at the peak amplitude and the peak-to-average power ratio, extracted from the amplitude of range-Doppler maps (RDMs).