PAPER – Construction Productivity

Find a current research article (published within the last 4 years) on the topic of Construction Productivity and cost estimation using artificial neural networks (Please refer to Chapter 11 Reference Word Document – Attached).

Save Time On Research and Writing
Hire a Pro to Write You a 100% Plagiarism-Free Paper.
Get My Paper

Write a two-page analysis of the article using at least two other peer-reviewed sources to support your analysis/discussion. You must have a title page, abstract page (minimum of 150 words), two full pages of text, and a reference page (3 References **Minimum) for this weekly assignment. References used must be in the U.S. APA format is required. No Plagiarism. Need this done by February 28, 2021, @ 2000 Central Time zone.

11 Construction productivity and cost estimation using artificial neural networks

Introduction

Because of the uncertainties and complexities involved in construction projects, expert system applications and artificial intelligence are helpful in the context of construction engineering and management. This chapter focuses on the applications of artificial neural networks (ANNs) for productivity and cost estimations, which are among the most crucial tasks of construction managers and general estimators.

Save Time On Research and Writing
Hire a Pro to Write You a 100% Plagiarism-Free Paper.
Get My Paper

The main objective of this chapter is to provide practical explanations of how to design, develop, analyse and validate ANNs as robust and reliable tools for productivity and cost estimations. An introduction to ANNs is provided, and several examples from the literature that have used ANNs in different areas of construction productivity and cost predictions are listed. As a result, a framework is presented to serve as a general guide on how to develop ANNs, and based on that, a detailed example is discussed to show the application in a real construction project setting. By the end of the chapter, readers should have some basic background about ANNs and should be able to develop a simple but efficient ANN for their own construction projects.

Artificial neural networks (ANNs)

An artificial neural network (ANN) can be defined as a massive parallel distributed processor composed of simple processing units (neurons) which are capable of storing experiential knowledge and retrieving it for future use (Haykin 1999). Neurons communicate by sending signals to each other over a large number of weighted connections. Thus, ANNs can be considered as an information processing technology that, by learning from different experiences and generalizing from previous examples, can simulate the human brain system. A simple schematic diagram of a neuron is shown in 

Figure 11.1

.

Figure 11.1 

shows that each neuron has two distinct segments: a summing junction that sums up the received inputs from neighbours and an activation function that computes the output signal, which is propagated to other neurons. The activation function can be theoretically in any form such as signum, linear or semilinear, hyperbolic tangent and sigmoid functions.

Figure 11.1  Schematic diagram of a neuron

To form a network, neurons are grouped into several layers, namely input, hidden and output layers. Two types of network topologies are shown in 

Figure 11.2

:

· Feed-forward networks: data flows strictly from input to output layers, and no feedback connections are allowed.

· Recurrent networks: feedback connections are allowed to provide data flow from the following layers to the preceding layers.

Figure 11.2 

 Different topologies of ANNs (Alavala 2006)

An ANN should be arranged in such a way that it can provide the desired outputs for a set of inputs presented to the network. To do so, either connection weights should be set using prior knowledge or the network should be trained by training samples (training sets). Connection weights are then updated based on ‘the learning rules’. There are two types of learning in ANNs:

· Supervised learning (associative learning): both input and output training pairs are presented to the network, and the network will learn based on the presented samples.

· Unsupervised learning (self-organization): there is no prior set of categories to be presented to the network, and the system develops its own representation of the input stimuli.

Almost all the learning rules are variants of the ‘Hebbian’ learning rule, which simply says that if two neurons are active, their interconnection must be strengthened. Based on this learning rule, the ‘delta rule’ uses the differences between actual and desired activation for modifying the connection weights.

Multilayer feed-forward neural networks (Figure 11.2) are the most commonly used architecture in the engineering fields, including construction engineering and management. The so-called back-propagation learning algorithm (Rumelhart, Hintont et al. 1986), which is a generalization of the delta rule, is generally used in these types of networks. A set of training samples is presented to the network, and based on the activation functions of the hidden and output layers, the network outputs are calculated, and are then compared with the actual (desired) outputs. The error is calculated (i.e. the difference between the actual and network output) and is propagated back to the network, the connection weights are modified and this procedure is repeated until the network reaches the stop criterion.

Generalization is the ability of neural networks to produce reasonable outputs for inputs for which they have not been trained. Learning capabilities and generalization are among the differences between ANNs and conventional models. ANNs are also useful where the processes to be modelled are too complex to be represented by analytical models. Based on the previous examples, neural networks can be built to make new decisions, classify new patterns and make forecasts and predictions. The most important advantage of ANNs over conventional models (mathematical and statistical) is their ability to adjust their weights to optimize their behaviour in different situations (Boussabaine 1996). Moreover, ANNs may include multiple outputs simultaneously (e.g. cost and productivity are both predictable using a single network providing the required inputs are used) when compared to multiple regression analysis (MRA), which can only estimate one output at a time.

Using ANNs for productivity and cost estimations

This section describes the procedure for the development of simple but efficient ANNs for productivity and cost predictions. Past studies are described and a framework established that shows the steps required for the development of ANNs for productivity and cost estimations. Based on that, an example is provided to show the accuracy of such networks in the context of real construction projects.

Suppose an estimator wants to predict the total construction cost of a new high-rise building project for a general contractor. The focus here is to describe the problem in such a way that it can be solved by the predictive ability of an ANN. To do so, the first step is to collect the factors (inputs) that affect the cost of the project, which are mainly project specifications such as total gross floor area, the type of building structure (steel or concrete), specific resources required and the like. Note that the main output is the total cost of the building. The estimator is then required to look at the previous projects (similar projects by the same contractor or other general contractors) to be used as the training examples for the ANN. The next step is to configure the network by choosing the number of layers, type of activation functions for hidden and output layers and number of neurons in each layer. These configurations are usually determined by trial and error. It has been shown that one layer of hidden units is enough to approximate any function to an arbitrary level of precision (Alavala 2006), but there is no specific rule regarding the number of neurons in the hidden layer.

There are several commercial packages available for design and analysis of ANNs. Most of them have graphical interfaces and can be used as add-ins for commonly used spreadsheet programs like Microsoft Excel. After the network is developed and trained, it is ready to be used for estimation purposes in the new projects with acceptable accuracy considering the uncertainties involved in the construction industry. A similar procedure can be used to estimate the productivity of different construction operations.

The following are some examples from the literature that have used the same method to develop ANNs for productivity and cost estimation of different construction projects. Some of the following examples have used other learning algorithms, but the procedure to develop the ANNs is more or less the same.

ANNs for total cost and duration estimations

One of the most important tasks for the construction management team is to provide cost and schedule estimations for different construction projects. Because of the predictive capabilities of ANNs, they have been widely used to provide more accurate cost and schedule predictions when compared to conventional estimation methods. This can result in fewer cases of cost overruns and fewer schedule discrepancies during the construction phase. Some of the examples in this area included predicting:

· highway and road tunnel construction costs (Hegazy and Ayed 1998; Wilmot and Mei 2005; Wichan et al. 2009; Petrousatou et al. 2011)

· duration of reconstruction projects and the related costs (Attalla and Hegazy 2003; Chen and Huang 2006)

· total cost and construction duration of residential buildings (Bhokha and Ogunlana 1999; Emsley et al. 2002; Kim et al. 2004)

· cost of structural systems of buildings (Günaydın and Doğan 2004)

· cost indices and unit price analysis (Baalousha and Çelik 2011).

The list shows that ANNs can be applied at any level for cost and duration estimations, from unit price analysis, which is the basis of cost predictions, to overall cost estimations of different construction projects including infrastructure and residential buildings.

ANNs for productivity modelling and estimation

Another vital task in the construction management field is to provide productivity rates and measurement, especially in areas such as resource allocation and management, scheduling, estimating, accounting, cost control and payroll (Herbsman and Ellis 1990). Contractors should have reliable estimates for different construction operations which are needed for planning and scheduling purposes. ANNs can be utilized in this area as a robust tool for productivity modelling and estimation. Some of the construction processes and operations for which ANNs have been used to model productivity are:

· concrete pouring, formwork and finishing tasks (Portas and AbouRizk 1997; Ezeldin and Sharara 2006; Dikmen and Sonmez 2011)

· hoisting times (hook times) of tower cranes (Tam et al. 2004)

· pile construction (Zayed and Halpin 2005)

· plastering activities (Oral et al. 2012)

· earth-moving equipment (Chanda and Gardiner 2010; Hola and Schabowicz 2010; Han et al. 2011).

Most of these studies show the application of ANNs for productivity modelling at the task level. ANNs are, however, applicable to any level of productivity prediction just as they can be applied to various levels of cost estimation.

Based on the previous studies, 

Figure 11.3

 depicts a framework that shows all of the steps required to develop efficient ANNs for productivity and cost estimation.

Example: estimating on-site productivity of precast installation activities

Problem formulation

A contractor in charge of the installation of different precast elements wanted to have more accurate estimates of on-site erection activities in order to manage the required resources more efficiently. An estimator was hired to provide a tool which could predict the time needed to install different precast components such as walls, columns, beams and slabs. For better understanding of the problem requirements, several preliminary site visits were conducted, and the estimator determined that a typical erection process or cycle includes several activities and resources; these are shown graphically in 

Figure 11.4

.

Figure 11.3 

 A framework for the development of an ANN for productivity and cost estimation

Factors affecting productivity of precast installation

Based on Figure 11.4 and through site visits, expert interviews and a literature review, the following factors were identified as being important for productivity of precast erection processes:

· Weight: component weight (tonnes)

· Area: the largest surface area of the element (m2)

· Length: the longest dimension of the element (m)

· Height: component height (m)

· Storage type: the component is stored among others or isolated (0: isolated; 1: among others)

· Storage to crane: distance from the element to the centre of the crane at the storage area (m)

· Installation to crane: distance from the installation point to the centre of the crane (m)

· Crane angle: angular movement of crane (degrees)

· Installation type: the component is installed among others or isolated (0: isolated; 1: among others)

· Location type: the component is part of the exterior or interior (0: interior; 1: exterior)

· Elevation: elevation of the installation point (m).

Figure 11.4 

 Precast installation activities (Najafi and Tiong 2012)

Other factors such as weather conditions, management conditions, crew skill and the like could also be considered; however, for simplicity, they were not included in this example.

Data collection

The next step was to collect the required data needed for ANN training and validation. In this case, the estimator collected the data regarding the installation of 91 precast elements (walls, columns and slabs) through site observations. General specifications of the collected data are shown in 

Table 11.1

.

Table 11.1 

shows that the observed precast components are among the typical elements that are widely used in construction projects. The results of the study can therefore be generalized and applied to new projects as well.

Selecting the ANN architecture and training algorithm

The estimator then used a commercial package (e.g. NeuroSolutionsTM) to design the ANN. For simplicity, a typical feed-forward back-propagation network was chosen as the main architecture. The inputs of the network were the factors described in the previous sections, and the main output is the total installation time in minutes.

Table 11.1  General specifications of the collected data

From 91 observations, 60 data points were randomly selected for the training of the network, 16 cases for cross-validation, and the remaining 15 data points were used for validation purposes. Use of the cross-validation technique ensures that the network is not overtrained (a situation in which the network performs well on the training set but performs poorly on the test data). With this method, the stop criterion is chosen in such a way that as soon as the error in the cross-validation set starts to increase, the training will be stopped. Since there is no specific rule for the configuration of ANNs, the estimator should develop several ANNs and choose the one with minimum error.

Table 11.2

 shows different ANN architectures and the corresponding mean squared error (MSE) and mean absolute error (MAE), which are calculated using the following formulae:

YNN is the network output and YA is the actual (desired) output for each testing data point. The best-performing network, the one that contains the minimum error (min. MSE), is highlighted. Note that in Table 11.2, architecture 11–7–1 denotes a network with three layers (one hidden layer) containing 11, 7 and 1 neurons in the input, hidden, and output layers, respectively.

Validation and performance of the selected model

A set of 15 data points was used to test the performance of the model against the actual data collected from the construction sites, and the results are shown in 

Figure 11.5.

Table 11.2 

 Selection of the best performing network

Figure 11.5 

 Predicted vs. actual installation times

The mean absolute percentage error (MAPE) calculated by 

Equation 11.3

 was equal to 19.86%, which means that the installation times predicted by the ANN are about 20% higher or lower than the actual installation times, which can be considered to be an acceptable performance considering the uncertainties involved in construction estimations.

different construction projects. The main differences are in the factors affecting productivity or cost estimations and the procedure followed to collect the required data. It should be noted that the aforementioned errors can be further minimized by increasing the sample size of the collected data (training set). Additionally, the use of other network types with different learning algorithms may result in better estimates. Statistical tests and analysis are also recommended for further comparison between the characteristics of the actual data and the estimated data provided by ANNs.

Hybrid use of artificial intelligence

The performance of ANNs can be further enhanced when they are used together with other artificial intelligence techniques such as fuzzy logic and genetic algorithms. Basically, fuzzy and neural systems are structurally different; however, they share a complementary nature and can be integrated to improve the overall flexibility and expressiveness of neural networks (Tsoukalas and Uhrig 1997). ANNs can be applied in function approximation, pattern recognition and automatic learning. On the other hand, fuzzy logic is efficient in modelling uncertainties. Their combination is therefore useful in dealing with certain construction management research problems that require accurate approximation while including different types of uncertainties. The performance of ‘neurofuzzy’ systems can be further improved by using genetic algorithms (Holland 1992) to optimize ANN parameters such as connection weights or network topologies (number of neurons in the hidden layer).

Additionally, with the increasing adoption of building information modelling (BIM) packages, there is a need to develop ANN analysis as a separate module that can be integrated with BIM systems. Architects, engineers, contractors and owners use BIM programs to plan, collaborate and predict performance before breaking ground on buildings. This chapter shows that ANNs are able to provide accurate productivity modelling of any construction operation, and the integration of ANNs with BIM systems is beneficial for both tools. This can be shown by the example of precast installation. Note that to use ANNs to estimate the installation time of each precast element, an estimator should manually feed all of the required input data such as element weight, surface area, the longest length and so on, as these are the factors affecting productivity in precast installation. Since most of this data is already available in the BIM model of the building, a computer program can be developed to automatically extract the required input for ANNs from the BIM system and perform the analysis. The results from ANNs can be then linked back to the BIM system to predict the performance more accurately.

Conclusions

Artificial neural networks are an information-processing technology that simulates the human brain system through learning from different experiences and generalizing from previous examples. Since the 1990s, this technique has been widely used in various areas of construction engineering and management, including cost and project duration estimations, bidding models and mark-up estimations, productivity modelling, prequalification of contractors and risk management.

This chapter summarized how ANNs can be easily applied to the two most important construction management tasks, cost estimation and productivity modelling. The procedures for using ANNs in other areas are similar to those depicted in the framework described here.

Some construction management problems deal with different levels of uncertainty. In these cases, the use of ‘neurofuzzy’ systems is highly recommended to add the flexibility required to solve these problems with the integration of ANNs with fuzzy systems. Additionally, by increasing adoption of BIM packages, the integration of BIM systems with ANNs is suggested in order to automate the data entry process as well as to improve the accuracy of performance prediction in the BIM system.

Finally, it is recommended that construction management research scholars and industry players further investigate the applications of ANNs. There are several commercial and open-source packages that provide user-friendly interfaces which can be easily used to develop efficient ANNs with only basic knowledge of this technique. Although ANNs have been used for more than two decades in the construction industry, further research gaps can be identified through extensive literature review, especially in those areas where ANNs have been utilized in other industries.

Calculate your order
Pages (275 words)
Standard price: $0.00
Client Reviews
4.9
Sitejabber
4.6
Trustpilot
4.8
Our Guarantees
100% Confidentiality
Information about customers is confidential and never disclosed to third parties.
Original Writing
We complete all papers from scratch. You can get a plagiarism report.
Timely Delivery
No missed deadlines – 97% of assignments are completed in time.
Money Back
If you're confident that a writer didn't follow your order details, ask for a refund.

Calculate the price of your order

You will get a personal manager and a discount.
We'll send you the first draft for approval by at
Total price:
$0.00
Power up Your Academic Success with the
Team of Professionals. We’ve Got Your Back.
Power up Your Study Success with Experts We’ve Got Your Back.

Order your essay today and save 30% with the discount code ESSAYHELP