Efficient Signal Processing Techniques for Distillation Column Malfunctions Identification

Abstract

Listen

Translate article

This paper presents some advanced signal processing techniques for distillation column malfunctions identification as one of the most common industrial radioisotopes applications. In this industrial application, gamma scanning technique is used for examining the inner details of the column by recording the variations of the received gamma ray intensity resulting from variations of the material density inside the column. Gamma intensity variations can be analyzed for detection and identification of the distillation column malfunctions. In the proposed approach, the signals are divided into frames; each frame represents the signal of one column tray only to be able to accurately determine the position of the defected tray from the entire column. Discrete Cosine Transform, Discrete Sine Transform, matched filtering, the Higher Order Statistics (Bispectrum and Trispectrum) are the signal processing tools utilized for further cepstral feature extraction. The extracted features are fed to a Support Vector Machine (SVM) or an Artificial Neural Network (ANN) classifier for the identification purpose. The results prove thatthe proposed approach can be used efficiently for the distillation column malfunctions identification and that the Bispectrum achieves the highest identification rate. For the classifiers, the SVM classifier achieves a higher identification rate and takes a shorter time compared to the ANN classifier.