Dissertation (Ph.D)--University of Stellenbosch, 1999. / ENGLISH ABSTRACT: System identification has been sufficiently formalized for linear systems, but not for empirical
identification of non-linear, multivariate dynamic systems. Therefore this dissertation
formalizes and extends non-linear empirical system identification for the broad class of nonlinear
multivariate systems that can be parameterized as state space systems. The established,
but rather ad hoc methods of time series embedding and nonlinear modeling, using multilayer
perceptron network and radial basis function network model structures, are interpreted
in context with the established linear system identification framework.
First, the methodological framework was formulated for the identification of non-linear state
space systems from one-dimensional time series using a surrogate data method. It was clearly
demonstrated on an autocatalytic process in a continuously stirred tank reactor, that validation
of dynamic models by one-step predictions is insufficient proof of model quality. In addition,
the classification of data as either dynamic or random was performed, using the same
surrogate data technique. The classification technique proved to be robust in the presence of
up to at least 10% measurement and dynamic noise.
Next, the formulation of a nearly real-time algorithm for detection and removal of radial
outliers in multidimensional data was pursued. A convex hull technique was proposed and
demonstrated on random data, as well as real test data recorded from an internal combustion
engine. The results showed the convex hull technique to be effective at a computational cost
two orders of magnitude lower than the more proficient Rocke and Woodruff technique, used
as a benchmark, and incurred low cost (0.9%) in terms of falsely identifying outliers.
Following the identification of systems from one-dimensional time series, the methodological
framework was expanded to accommodate the identification of nonlinear state space systems
from multivariate time series. System parameterization was accomplished by combining
individual embeddings of each variable in the multivariate time series, and then separating
this combined space into independent components, using independent component analysis.
This method of parameterization was successfully applied in the simulation of the abovementioned
autocatalytic process. In addition, the parameterization method was implemented
in the one-step prediction of atmospheric N02 concentrations, which could become part of an
environmental control system for Cape Town. Furthermore, the combination of the embedding strategy and separation by independent component analysis was able to isolate
some of the noise components from the embedded data.
Finally the foregoing system identification methodology was applied to the online diagnosis
of temporal trends in critical system states. The methodology was supplemented by the
formulation of a statistical likelihood criterion for simultaneous interpretation of multivariate
system states. This technology was successfully applied to the diagnosis of the temporal
deterioration of the piston rings in a compression ignition engine under test conditions. The
diagnostic results indicated the beginning of significant piston ring wear, which was
confirmed by physical inspection of the engine after conclusion of the test. The technology
will be further developed and commercialized. / AFRIKAANSE OPSOMMING: Stelselidentifikasie is weI genoegsaam ten opsigte van lineere stelsels geformaliseer, maar nie
ten opsigte van die identifikasie van nie-lineere, multiveranderlike stelsels nie. In hierdie tesis
word nie-lineere, empiriese stelselidentifikasie gevolglik ten opsigte van die wye klas van nielineere,
multiveranderlike stelsels, wat geparameteriseer kan word as toestandveranderlike
stelsels, geformaliseer en uitgebrei. Die gevestigde, maar betreklik ad hoc metodes vir
tydreeksontvouing en nie-lineere modellering (met behulp van multilaag-perseptron- en
radiaalbasisfunksie-modelstrukture) word in konteks met die gevestigde line ere
stelselidentifikasieraamwerk vertolk.
Eerstens is die metodologiese raamwerk vir die identifikasie van nie-lineere,
toestandsveranderlike stelsels uit eendimensionele tydreekse met behulp van In surrogaatdatametode
geformuleer. Daar is duidelik by wyse van 'n outokatalitiese proses in 'n deurlopend
geroerde tenkreaktor getoon dat die bevestiging van dinamiese modelle deur middel van
enkelstapvoorspellings onvoldoende bewys van die kwaliteit van die modelle is. Bykomend is
die klassifikasie van tydreekse as 6f dinamies Of willekeurig, met behulp van dieselfde
surrogaattegniek gedoen. Die klassifikasietegniek het in die teenwoordigheid van tot minstens
10% meetgeraas en dinamiese geraas robuust vertoon. /
Vervolgens is die formulering van In bykans intydse algoritme vir die opspoor en verwydering
van radiale uitskieters in multiveranderlike data aangepak. 'n Konvekse hulstegniek is
V:oorgestel en op ewekansige data, sowel as op werklike toetsdata wat van 'n binnebrandenjin
opgeneem is, gedemonstreer. Volgens die resultate was die konvekse hulstegniek effektief
teen 'n rekenkoste twee grootte-ordes kleiner as die meer vermoende Rocke en Woodrufftegniek,
wat as meetstandaard beskou is. Die konvekse hulstegniek het ook 'n lae loopkoste
(0.9%) betreffende die valse identifisering van uitskieters behaal.
Na aanleiding van die identifisering van stelsels uit eendimensionele tydreekse, is die
metodologiese raamwerk uitgebiei om die identifikasie van nie-lineere, toestandsveranderlike
stelsels uit multiveranderlike data te omvat. Stelselparameterisering is bereik deur individuele
ontvouings van elke veranderlike in die multidimensionele tydreeks met die skeiding van die
gesamenlike ontvouingsruimte tot onafhanklike komponente saam te span. Sodanige skeiding
is deur middel van onafhanklike komponentanalise behaal. Hierdie metode van parameterisering is suksesvc1 op die simulering van bogenoemde outokatalitiese proses
toegepas. Die parameteriseringsmetode is bykomend in die enkelstapvoorspelling van
atmosferiese N02-konsentrasies ingespan en sal moontlik deel van 'n voorgestelde
omgewingsbestuurstelsel vir Kaapstad uitmaak. Die kombinasie van die ontvouingstrategie en
skeiding deur onafhanklike komponentanalise was verder ook in staat om van die
geraaskomponente in die data uit te lig.
Ten slotte is die voorafgaande tegnologie vir stelselidentifikasie op die lopende diagnose van
tydsgebonde neigings in kritiese stelseltoestande toegepas. Die metodologie is met die
formulering van 'n statistiese waarskynlikheidsmaatstaf vir die gelyktydige vertolking van
multiveranderlike stelseltoestande aangevul. Hierdie tegnologie is suksesvol op die diagnose
van die tydsgebonde verswakking van die suierringe in 'n kompressieontstekingenj in tydens
toetstoestande toegepas. Die diagnostiese resultate het die aanvang van beduidende slytasie in
die suierringe aangedui, wat later tydens fisiese inspeksie van die enjin met afloop van die
toets, bevestig is. Die tegnologie sal verder ontwikkel en markgereed gemaak word.
| Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:sun/oai:scholar.sun.ac.za:10019.1/51258 |
| Date | 12 1900 |
| Creators | Barnard, Jakobus Petrus |
| Contributors | Aldrich, C., Gerber, M., University of Stellenbosch. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. . |
| Publisher | Stellenbosch : Stellenbosch University |
| Source Sets | South African National ETD Portal |
| Language | en_ZA |
| Detected Language | Unknown |
| Type | Thesis |
| Format | 1 v. (various pagings) : ill. |
| Rights | Stellenbosch University |
Page generated in 0.0099 seconds