Repository logo
 

Faculty of Engineering and Built Environment

Permanent URI for this communityhttp://ir-dev.dut.ac.za/handle/10321/9

Browse

Subject: search.filters.subject.Artificial neural network

Search Results

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Model based real time controller performance assessment for non-linear systems
    (Central University of Technology, 2016-12) Pillay, N.; Govender, P.
    The aim of this paper is to present a novel methodology for the performance assessment of proportional-integral-derivative (PID) controllers operating in the presence of process nonlinearities. The principle objective is to assess the quality of controller performance in real time when subjected to setpoint changes. Using prescribed operating regions, optimal PID controller settings are synthesized off-line by numerical optimisation from a trained artificial neural network (ANN) of the process. To demonstrate the effectiveness of the proposed controller benchmarking scheme, the procedure is applied to a simulation example, plus a real process control loop operating in a full scale pH neutralization pilot plant. Results obtained from the experiments indicate that the method is suitable for servo tracking in nonlinear control loops such as those found in the pulp and paper, and water purification industries.
  • Thumbnail Image
    Item
    Hybrid motion detection system using DSP and ANN ensembles
    (IMECS, 2013-03) Moorgas, Kevin Emanuel; Govender, Poobalan
    Hybrid systems are used in engineering systems and scientific applications to enhance and to improve their efficiency. This paper presents a hybrid system using digital signal processing (DSP) systems and artificial neural networks (ANN’s) for object motion detection, extraction and filtering. A summary of a DSP motion detection system is first presented and its performance is compared to that of an ANN ensemble system (AES). The two systems are then combined to form a hybrid motion detection system. Preprocessing of the hybrid system uses wavelet image compression to enhance its computational speed. During the testing of the system the efficiency of the (AES) is demonstrated as a powerful parallel processor in handling large amounts of image data.
  • Thumbnail Image
    Item
    ANN’s vs. SVM’s for image classification
    (International ASET, 2012-08) Moorgas, Kevin Emanuel; Pillay, Nelendran; Governder, Poobalan
    In this paper the dynamic performance of the artificial neural network is compared to the performance of a statistical method such as the support vector machine. This comparison is made with respect to an image classification application where the performance is compared with regards to generalization and robustness. Image vectors are compressed in order to reduce the dimensionality and the salient feature vectors are extracted with the principle component algorithm. The artificial neural network and the support vector machine are trained to classify images with feature vectors. A comparative analysis is made between the artificial neural network and the support vector machine with respect to robustness and generalization.