Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.583536
Title: Artificial Neural Networks for loudspeaker modelling and fault detection
Author: Fox, Caroline
Awarding Body: Cardiff University
Current Institution: Cardiff University
Date of Award: 2005
Availability of Full Text:
Access through EThOS:
Access through Institution:
Abstract:
This thesis is the result of a collaborative project between Cardiff University and Harman/Becker Automotive Systems. It investigates the application of Artificial Neural Networks to loudspeaker fault detection and modelling of the loudspeaker transfer function. The aim was to utilise the ability of artificial neural networks to model high order nonlinear systems to generate a model of the loudspeaker transfer function which could be used in a linearisation scheme to reduce distortion in loudspeaker output. This thesis investigates a practical approach to transfer function modelling through the use of musical excitation signals. This would allow data to be collected during normal operation of the loudspeaker and, as the transfer function changes over time due to time dependent nonlinearities, would facilitate regular updating of the neural network model to incorporate these nonlinearities. It was determined that although very accurate models could be produced over long training periods, a significant compromise in ANN training set size and number of training epochs were required to reduce the ANN training duration to the required time period, which ultimately resulted in a decline in performance. The aim in the case of fault detection was to improve on current end of production line testing for loudspeaker distortion. Neural networks were trained with harmonic distortion data in order to emulate the end of line test result. Excellent classification accuracy was achieved when neural network classification results were compared to the end of line test results. An investigation was also conducted to determine if neural networks could be trained to recognise specific loudspeaker faults. In a development of the end of line test, a system of neural networks were trained to produce an output vector that described which of five frequency regions the loudspeaker distortion levels were above the limits, thus giving an indication of the possible fault.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.583536  DOI: Not available
Share: