Machine learning for antenna array failure analysis
| dc.contributor.advisor | Ludick, D. J. | en_ZA |
| dc.contributor.advisor | Grobler, Trienko Lups | en_ZA |
| dc.contributor.author | De Lange, Lydia | en_ZA |
| dc.contributor.other | Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. | en_ZA |
| dc.date.accessioned | 2020-02-24T08:41:13Z | |
| dc.date.accessioned | 2020-04-28T12:07:18Z | |
| dc.date.available | 2020-02-24T08:41:13Z | |
| dc.date.available | 2020-04-28T12:07:18Z | |
| dc.date.issued | 2020-03 | |
| dc.description | Thesis (MEng)--Stellenbosch University, 2020. | en_ZA |
| dc.description.abstract | ENGLISH ABSTRACT: This work investigated the use of machine learning to detect failed elements in an antenna array. The aim was to identify a trustworthy means of early detection and isolation of faulty elements to improve the reliability of measured data. Previous work has shown that it is theoretically possible to identify failed elements from the far-field radiation pattern, using machine-learning algorithms such as artificial neural networks and support vector models. However, literature seems void of studies that test howthe input data affects the accuracy of the machine-learning algorithm. It is possible to measure the far-field radiation pattern of earth-based antenna arrays, but very few researchers have validated their proposed techniques on a manufactured array. We therefore investigated the effects of various far-field sampling methods on the accuracy and training time of a feedforward neural network, and on the accuracies of different out-of-the-box classification algorithms, and the effect of the antenna array configuration on the accuracy of a support vector model. We simulated, manufactured and measured a 16-element circular patch antenna array to determine the feasibility of using the simulated far-field pattern as training data for a machine-learning algorithm designed to identify failures in a measured far-field pattern. We found it would not currently be feasible to employ machine learning to detect single element failures by measuring distortions in the far-field radiation patterns generated by a very large array of antennas in an irregular sparse configuration, such as those planned for the Square Kilometer Array (SKA) radio astronomy project. | en_ZA |
| dc.description.abstract | AFRIKAANSE OPSOMMING: Die gebruik van masjienleerword ondersoek vir tydige opsporing en uitsluiting van foutiewe elemente antenna samestellings. Die doel is om ’n betroubare manier te vind om foutiewe elemente vroegtydig op te spoor en uit te sluit, sodat die gemete data meer betroubaar sal kan wees, soos byvoorbeeld in groot antenna samestellings, soos die SKA radio astronomie projek. Vorige studies het bevind dat dit teoreties moontlik is om foutiewe elemente vanaf die ver-veld patroon te identifiseer deur die gebruik van ’n masjienleer algoritme soos ’n neurale netwerk(NN)of ondersteunings-vektor masjien ("SVM"). Daar is ’n tekort aan studies in die literatuur wat die invloed van die leerdata op die akkuraatheid van die masjienleer algoritme toets. Die ver-veld van ’n aardvaste antenna samestelling kan gemeet word, maar min navorsers het al hul voorgestelde metodes op ’n vervaardigde antenna samestelling getoets. In hierdie studie is daar ondersoek ingestel na die invloed van verskeie ver-veld steekproefmetodes op die akkuraatheid en opleidingstyd van ‘n FNN; en op die akkuraathede van verskeie standaard klassifiseringsalgoritmes; asook die invloed van die uitleg van die saamgestelde antenna op die akkuraatheid van ‘n SVM. ‘n 16-element sirkelvormige plak antenna samestelling is gesimuleer, vervaardig en gemeet, om vas te stel of die gesimuleerde ver-veldpatroon suksesvol gebruik kan word as leerdata vir ‘n masjienleer-algoritme vir foutsporing op ‘n gemete ver-veldpatroon. Ons sou nie tans aanbeveel om ’n enkele foutiewe element te probeer identifiseer deur versteurings te meet in die ver-veldpatroon van ‘n baie groot saamgestelde antenna met ‘n yl verspreide, ongeordende uitleg, soos beplan vir die SKA radio astronomie projek nie. | af_ZA |
| dc.description.version | Masters | en_ZA |
| dc.format.extent | xiii, 92 leaves : illustrations (some color) | |
| dc.identifier.uri | http://hdl.handle.net/10019.1/107874 | |
| dc.language.iso | en | en_ZA |
| dc.publisher | Stellenbosch : Stellenbosch University | en_ZA |
| dc.rights.holder | Stellenbosch University | en_ZA |
| dc.subject | Antenna arrays | en_ZA |
| dc.subject | Machine learning | en_ZA |
| dc.subject | UCTD | en_ZA |
| dc.subject | Failure analysis (Engineering) | en_ZA |
| dc.subject | Neural networks (Computer science) | en_ZA |
| dc.subject | Support vector machines | en_ZA |
| dc.subject | Antenna radiation patterns | en_ZA |
| dc.title | Machine learning for antenna array failure analysis | en_ZA |
| dc.type | Thesis | en_ZA |