![]() | Only 14 pages are availabe for public view |
Abstract Radioisotope scan be used to obtain signals or images of the industrial systems in order to recognize the information inside it. The main problems of using these techniques are the difficulty of identification of the obtained signals or images and the requirement of skilled experts for the interpretation process of the output data of these applications. Now, the interpretation of the output data from these applications is performed mainly manually, depending heavily on the skills and the experience of trained operators. This process is time consuming and the results typically suffer from inconsistency and errors. The objective of the thesis is to propose different approaches depending on different advanced digital signal processing techniques for improving the treatment and the interpretation of the output data from the different Industrial Radioisotopes Applications (IRA). Thesis is focused on gamma scanning of a distillation column in petroleum industry as one of IRA. Column scanning is an on-line technology without interference and interruption to the process. When scanning a column, a suitably gamma-ray emitting radioactive sealed source in a panoramic source container and a scintillation detector are selected. The data then collected and analyzed. The thesis proposes two approaches for the identification of distillation column malfunctions in the presence of noise. In all the proposed approaches, the signals are firstly divided into frames; each frame contains only the signal of one column tray to be able to identify not only the type of column malfunction, but also the position of the malfunction tray to facilitate the maintenance process. Then different signal processing techniques have been applied on the segmented signal (frames) as Discrete Cosine Transform (DCT) and Discrete Sine Transform (DST) in the first approach or as Higher order Statistics (HOS) with its different estimators and orders in the second approach. Mel Frequency Cepstral Coefficients (MFCCs) and polynomial coefficients are used as feature extractor techniques to extract the discriminating features from the processed signal. Artificial neural networks and support vector machines have been used as different classifier for the classification process. The simulation results show that all approaches can be used efficiently at low noise levels, but at high noise levels, HOS based techniques give better identification rate than other approaches especially when using the indirect estimator of the bispectrum. |