Show simple item record

dc.contributor.advisorVillacampa Esteve, Yolanda
dc.contributor.advisorMontoyo Guijarro, Juan Andrés
dc.contributor.authorGuerrero Lázaro, Miguel Ángel
dc.contributor.otherUniversidad de Alicante. Departamento de Edificación y Urbanismoes
dc.date.accessioned2018-05-11T09:18:36Z
dc.date.available2018-05-11T09:18:36Z
dc.date.issued2016
dc.identifier.urihttp://hdl.handle.net/20.500.12251/153
dc.description.abstractThe estimation of the time required to construct building projects has been a topic of great interest to many researchers and practitioners. Delays are a common problem in the construction industry and may be motivated by different factors. In this context, prediction of the construction time of building projects at early project phases has been considered a key element for project success. Initially, the construction time of a building is affected by several factors related to project features, although some factors are more crucial than others. Based on these factors, for the purpose of providing proper tools to estimate construction time and minimise the subjectivity in such estimation, to date, most research works have presented parametric models which were built using linear regression analysis (LRA). Nevertheless, there is an increasing trend for using artificial neural networks (ANNs) to develop better predictive models. In order to produce the best possible predictive models and provide a clearer explanation regarding the relationships that exist between different project scope factors and the construction time of new builds, the research work presented in this thesis used two data sets and three different modelling techniques: LRA, ANNs and a new numerical methodology based on the finite element method (FEM). In particular, this thesis addressed the general assumption that nonlinear modelling techniques are likely to better represent the previously mentioned relationships than LRA. According to available data, predictor variables related to construction costs, gross floor area (GFA), number of floors, and the type of facility were selected to analyse their influence on the duration of the construction process. Additionally, and since that there is no general agreement in the literature regarding which is the most appropriate dependent variable for predicting construction time, in this thesis both time and speed of construction were analysed to determine which of these offer better predictive models. In this regard, construction speed can be used as a useful and robust benchmark for comparison of contractor performance. In the case of ANNs, two different types of network architectures were tested: the multilayer perceptron (MLP) and the radial basis function (RFB). The modelling process of MLP networks was divided into five stages: (i) selection of the training methodology, (ii) data division, (iii) design of the initial network structure, (iv) network optimisation, and (v) validation of the optimised models. MLP networks were used in conjunction with two different training algorithms and five options for calibration data division. In addition, a methodology was defined to obtain optimised MLP networks with an adequate predictive performance. This methodology develops a stepwise trial and error procedure in which a basic MLP network structure, with enough consistency, is first established and subsequently this initial structure is modified at each step of the proposed optimisation process in order to achieve the best possible network configuration.. This thesis also proposes a framework to evaluate the performance of predictive models which includes five different assessment criteria: (i) verification of compliance with the underlying assumptions regarding the statistical procedure used to obtain the models, (ii) checking the goodness of fit of the models to the data set used for generating them, (iii) validation of models in terms of ability to generalise, (iv) assessment of the balance existing between the ability of a model to generalise and the accuracy obtained with the calibration data, and (v) development of a sensitivity analysis to verify model stability. Finally, a sensitivity analysis was also proposed to evaluate the impact of the construction cost variability, caused by the uncertainty in its estimation, on the performance of predictive models. The results obtained with this thesis showed that construction speed is a more appropriate dependent variable than construction time to develop predictive models to estimate the construction process duration of building projects, and that such construction speed is affected more by GFA than by construction cost. Furthermore, the FEM-based numerical methodology provided better predictive models than those generated by MLP networks and LRA. In this regard, the findings of this research work support the idea that linear regression models can provide a good starting point from which to search for better predictive models using nonlinear modelling techniques. The knowledge gained from this thesis will allow for new approaches to be explored in order to better determine the relationships existing between project scope factors and the construction speed of new builds, but always taking into account that the results provided by the models proposed herein are only initial construction speed estimates at early stages of project development, when only basic information is available, and are not intended to replace detailed schedules undertaken by builders.es
dc.language.isoenges
dc.publisherUniversidad de Alicantees
dc.rightsAttribution-NonCommercial-NoDerivatives 4.0 Internacional
dc.rights.urihttp://creativecommons.org/licenses/by-nc-nd/4.0/
dc.titleModelling building construction speed by using linear regression analysis, artificial neural networks and n-dimensional finite elementses
dc.typedoctoralThesises
dc.identifier.urlhttp://hdl.handle.net/10045/66350es
dc.rights.accessRightsopenAccesses
dc.subject.keywordAnálisis de regresión lineales
dc.subject.keywordRedes neuronales artificialeses
dc.subject.keywordLinear Regression Analysis (LRA)es
dc.subject.keywordArtificial Neural Networks (ANN)es
dc.subject.keywordRadial basis functiones
dc.subject.keywordMultilayer perceptrones
dc.subject.unesco1209.14 Técnicas de Predicción Estadísticaes
dc.subject.unesco1203.26 Simulaciónes
dc.subject.unesco1209.03 Análisis de Datoses
dc.subject.unesco3310.07 Estudio de Tiempos y Movimientoses


Files in this item

FilesSizeFormatView

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record