Neural networks as a diagnosing tool for industrial level measurement through non-contacting radar type and support to the decision for its better application

Abstract

Abstract: The aim of this study was to develop an analysis tool based on artificial neural networks (ANN) to detect level measurement problems with free wave propagation radars. The trend of using this type of radar has been growing in the last ten years mainly because of its easy installation on the top of tanks and reservoirs, and for its low rate maintenance comparing to other level measurement technologies. For the experiments, a Rosemount radar was used and the training of the neural network was based on the data from the software Radar Master. Therefore, some network topologies in different scenarios were tested and it was possible to demonstrate the efficiency of the ANN with accuracy rate between 94.44 to 100% for the first experiment with networks using 10, 20 or 50 neurons in the hidden layer. This technique was applied in a real industrial application, a sugar and ethanol mill, and accuracy rate was about 87,0 to 96,1%. This methodology can be applied to asset management software for diagnosis report or troubleshooting which would increase the level measurement reliability and plant safety.