Prediction of Cost of Quality Using Artificial Neural Network In Construction Projects

Transcription

1 Prediction of Cost of Quality Using Artificial Neural Network In Construction Projects Chinchu Mary Jose 1, Ambili S 2 1 M Tech scholar,department of Civil Engineering,MES College of Engineering, Kuttippuram, Malappuram, Kerala, India 2 Assistant Professor,Department of Civil Engineering,MES College of Engineering, Kuttippuram, Malappuram, Kerala, India Abstract: Quality is the fitness for use, conformance to requirements, predictable degree of uniformity and dependability, at low cost and suited to market. Cost of quality is an essential element of the total cost of any construction project. The main objective of this paper is to develop a neural network model that will enable the construction firms to access cost of quality for any future building project. The different sequences of the model development will be investigated. Moreover, the validity of the proposed model will be evaluated using case study applications. The main 29 factors affecting the expected cost of quality were identified. A questionnaire survey was carried out which was conducted among 60 expert in the construction domain to determine the importance of these twenty nine factors. By using Microsoft Excel, relative importance index of the factors are obtained and the factors with important index more than 70% was selected for a second stage questionnaire survey. Second stage questionnaire survey was conducted among construction experts from different construction industries and the result obtained was used as the input parameters of the proposed cost of quality model. Keywords: construction industry, cost of quality, questionnaire survey, artificial neural network I. INTRODUCTION Cost of quality is an essential element of the total cost of a construction project. Therefore, the accurate assessment of cost of quality can significantly affect the reliability of the estimated cost of any construction project. Cost of quality is the sum of conformance and non conformance cost, where cost of conformance is the price paid for prevention of poor quality and cost of non conformance is cost of poor quality caused by product and service failure. Cost of quality is generally affected by many factors. The COQ can be broken down into four categories. [1] Prevention cost: The cost of any action taken to investigate, prevent or reduce the risk of nonconformity Appraisal cost: The cost of evaluating the achievement of quality requirements Internal failure cost: The costs arising within an organization due to nonconformities or defects at any stage of the quality loop. External failure: The cost arising after delivery to a customer/user due to nonconformities or defects which may include the cost of claims against warranty, replacement and consequential losses and evaluation of penalties incurred. Cost of quality is affected by many factors like design errors, defected material, planned COQ for the project, accident, suppliers, contractors, labor skills. DOI : /IJRTER XVLPB 54

2 II. LITERATURE REVIEW Samadony et al. (2006) revealed that the mean expenditure on quality in the Egyptian construction firms is about 26% of total cost, and the internal failure cost is about 10% from total project cost. The key to continuing success in quality management is the ability to collect poor quality information to improve the performance of the construction process. This information should then be incorporated into the design and management of the new projects. This information can also be used to measure the performance of construction firms so that continuous improvement is based on measurement of performance can be effectively implemented [2]. Rosenfeld (2009) compared cost of quality versus cost of non-quality in construction. The methodology is based on quantifying the four types of quality-related costs in residential construction, and relates them to each other by expressing them all as percentages of the relevant total construction revenues [3]. Firuzan (2002) proposed a radical change in industry practice that will improve the quality of the construction process and the levels of customer satisfaction derived from it by evaluating the quality performance of the contractor. An alternative theory is developed of what constitutes quality, client satisfaction, performance, and their interrelationships in the context of the construction industry [4]. Mwamila et al. (1999) stated that construction speed is impacted by the number and productivity of workers and can be increased by reliable equipment and early planning and design that maximize use of limited available resources. Building quality is dependent on standardization, product suitability evaluation, defect identification, and thorough planning. Labor costs are generally a small portion of total construction costs; however, labor is a key cost factor because it affects both quality and speed [5]. Hany Shoukry Tawfek et al. (2012) identified planned COQ for the project, project duration, accident, external factors, equipment down time, design errors, labor skills, supervision team experience, class of contractor, project size, project type, client type, awareness of quality for project team, suppliers, execution errors, sub contractors nature, type of contract etc as the factors influencing cost of quality in a construction project[1]. M. Abas et.al. (2015) evaluated the factors affecting the quality of construction in Pakistan. They developed a questionnaire and conducted a survey amongst experts to take their opinion. Statistical analysis tools such as Chi square test and weighted mean method were used to rank the significance level of these factors [6]. III. OBJECTIVES OF THE STUDY To identify the factors affecting cost of quality of a construction project. To develop an Artificial Neural Network model that can help cost estimator to arrive at a more reliable assessment for the expected cost of quality of any building construction project. IV. MATERIALS AND METHODS In this study, to identify the significance factors affecting cost of quality in construction firm the research methodology adopted is given as follows: 4.1 Data Collection Method Data collection is divided into two stages; first stage is to perform a questionnaire survey using identified factors affecting COQ from the literature study. The second stage was to collect data for 60 projects from several construction companies that represent the several construction industries. On the basis of previous studies on factors influencing quality performance 29 factors are identified. These factors were identified through several literature surveys and suggestions from experienced persons. A questionnaire is prepared using these factors. The questionnaire is distributed among All Rights Reserved 55

3 of construction industry to identify the most important factors among these to be used as the input parameters of the proposed cost of quality model. 4.2 Data Analysis Method In order to measure the importance of the factors Relative Importance Index (RII) was measured for each factor. RII calculation was used to determine relative significance and for ranking the factors affecting project performance. RII is given as; a R II *100 A* n a= weighting given to each factor by respondent and it ranges from 1 to 5 A= highest response N= total number of participants All factors are ranked in a descending order according to their importance index. Such factors represent the input parameters of the proposed cost of quality model. In Table I the investigated factors and their RII scores are represented. Table 1: RII scores of investigated factors Rank Factors RII Score 1 Suppliers Supervision team experience Planned COQ for the project 92 4 Awareness of quality for the project team Labor skills Design errors Defected material Execution errors Project duration Project type Class of contractors Weather conditions Equipment downtime Project size Client type Special construction engineering equipments Labor turnover Sub-contractors nature Wages of labors Auditing process periods Working shifts Plan of improving quality Project location New construction techniques Special site preparation techniques Percentage of rejected submittals accident External factors Type of contract All Rights Reserved 56

4 Based on the important index analysis, the COQ factors which has an importance index greater than 70% is considered for the second stage questionnaire. In this survey the respondents were given a checklist in which he/she was required to rate each factor according to the status of the organization in terms of project execution. The response was fed in as input data for artificial neural network modeling. V. MODEL DEVELOPMENT The main purpose of this study is to develop a neural network model to predict the percentage of cost of quality for building construction projects. 5.1 Neural network and overview In this paper, Artificial Neural Networks were used as a modeling tool that can enhance current automation efforts in the construction industry. The structure of the neural network model includes an input layer that receive input from the outside world, hidden layers that serve the purpose of creating an internal representation of the problem, and an output layer, or the solution of the problem. Before solving a problem, neural networks must be trained. Networks are trained as they examine a smaller portion of the dataset just as they would a normal-sized dataset. Through this training, a network learns the relationships between the variables and establishes the weights between the nodes. Once this learning occurs, a new case can be entered into the network resulting in solutions that offer more accurate prediction or classification of the case. The steps for the design of ANN model will be illustrated to predict the percentage of the expected cost of quality for building construction projects. All factors that have an effect on the expected cost of quality of the building construction projects in Egypt were identified. These factors were considered as the input variables for the proposed neural network model, while the expected cost of quality as a percentage from the total projects contract value is considered as the output variable of this model. Neural network models are generally developed through the following six basic steps: Identify the problem, decide what information to be used and what will the network do; come to a decision of how to gather the information and symbolize it; define the network, select network inputs and identify the expected outputs; structure the network; train the network; and analyze the trained network. This engages addressing novel inputs to the network and evaluates the network s results with the authentic life results. 5.2 Training the network All trial models experimented in this research was trained in supervised mode by a back propagation learning algorithm. Inputs were fed to the proposed network model and the outputs were calculated. The differences between the calculated outputs and the actual outputs (data taken from project documents) were then evaluated. The back propagation algorithm develops the input to output mapping by minimizing a root mean square error [RMS] which is expressed by the following equation [7]: RMS = (O i P i ) 2 n i=1 All Rights Reserved 57

5 Where n is the number of samples to be evaluated in the training phase. Oi is the actual output related to the sample. Pi is the predicted output. The training process stopped when the value mean square error is the minimum. 5.3 Development of ANN model using MATLAB The data analysis of the study includes the development of an ANN model for predicting the contingent cost of construction due to lack of quality using Matlab. The Matlab R 2015 b version is used for the same. The inputs were already prepared in the excel spread sheet. 1. First step in the model analysis is to import the data. MATLAB has got the command for calling the data. While clicking the command Import Data a window will appear asking to select the file where the input data, which has been previously been prepared, is selected. Then a window as in Fig.1, will appear. From the window we can upload the required data and save as matrix data. The input data and target data are imported to the MATLAB. Fig. 1 Importing data in MatLAB 2. The next step is to train the data. Here we are using Neural Fitting Application. Open the Neural Network Fitting App which can be found under Apps tab. GUI with the command nftool will be opened as shown in All Rights Reserved 58

6 Fig. 2 Neural Fitting App Window 3. Then click Next to proceed. Window will appear as in Fig.3 below the datasets which were imported earlier can be loaded. Fig.3 loading data set 4. Click Next to display the Validation and Test Data window, shown in the Fig. 4. From the window fix the number of training, testing and validating dataset. Here 70% data is used for training. The validation and test data sets are each set to 15% of the original data. Fig.4 Selecting the number of samples for All Rights Reserved 59

7 5. Then click Next to view the Network Architecture as in Fig.5. The standard network that is used for function fitting is a two-layer feed forward network, with a sigmoid transfer function in the hidden layer and a linear transfer function in the output layer. The default number of hidden neurons is set to 10. If the network training performance is poor, the number of hidden neurons can be increased or decreased. Fig.5 Fixing the number of neurons 6. Once the numbers of neurons are fixed then click Next. The window appeared is as shown in Fig. 6. Select a training algorithm, and then click Train. Levenberg- Marquardt (trainlm) is recommended here to obtain a better solution. The training continued until the validation error failed to decrease for six iterations (validation stop). Click the Plot Error Histogram button to view the error in training data as in Fig. 7. Adjust the number of neurons and repeat the iteration until minimum error is obtained. Fig.6 Training the All Rights Reserved 60

8 Fig.7 Error Histogram The blue bars represent training data, the green bars represent validation data, and the red bars represent testing data. The histogram can give you an indication of outliers, which are data points where the fit is significantly worse than the majority of data. 7. Click Plot Regression. This is used to validate the network performance. The following regression plots display the network outputs with respect to targets for training, validation, and test sets. For a perfect fit, the data should fall along a 45 degree line, where the network outputs are equal to the targets. If even more accurate results were required, we could retrain the network by clicking Retrain in nftool. The obtained plot is shown in the Fig.8. Fig.8 Regression plot 8. Click Next in the Neural Network Fitting Tool to evaluate the network as shown in the Fig.9. At this point, test the network against new data. If dissatisfied with the network's performance on the original or new data, do one of the All Rights Reserved 61

9 Train it again. Increase the number of neurons. Get a larger training data set. International Journal of Recent Trends in Engineering & Research (IJRTER) If the performance on the training set is good, but the test set performance is significantly worse, which could indicate over-fitting, then reducing the number of neurons can improve the results. If training performance is poor, increase the number of neurons. Fig 9. Evaluating the network 9. If satisfying network performance is obtained, click Next and the screen shown in Fig.10 will be obtained. Fig. 10 Deploy the network 10. Use this panel to generate a MATLAB function or Simulink diagram for simulating your neural network. The generated code or diagram can be used to better understand how the neural network computes outputs from inputs, or deploy the network with MATLAB Compiler tools and other MATLAB code generation All Rights Reserved 62

10 11. Use the buttons on the as in Fig. 11 to generate scripts or to save results. Fig. 11 Saving the results Simple Script or Advanced Script can be clicked to create MATLAB code that can be used to reproduce all of the previous steps from the command line. Creating MATLAB code can be helpful to learn how to use the command-line functionality of the toolbox to customize the training process. In Using Command-Line Functions, the generated scripts will be investigated in more detail. The network can be saved as net in the workspace. Additional tests can be performed on it or put it to work on new inputs. When we have created the MATLAB code and saved the results, click Finish. 12. After the network is trained and validated, the network object can be used to calculate the network response to any input. 5.4 Testing the validity of the model To assess the prognostic recital of the network, the five projects that were previously arbitrarily chosen and reticent for testing from the total collected projects are introduced to the best model. Percentage of the model will forecast the expected cost of quality. The calculated percentage will be evaluated to the real life projects percentage (stored outside the program) and the disparity between them will be premeditated if it is equal or under the value of the designed model s Absolute Difference. Then it is considered to be a correct calculation. If it exceeds the value of the designed model s Absolute Difference then it is considered to be an incorrect prediction attempt. Table 2 presents the actual and predicted percentages for the test sample. The model correctly predicted five (5) of the five (5) testing projects samples (100% of the test sample). The test sample outcome correct percentage is considered to be hundred percent (100%) which is very good and the model can be accepted. Table 2: Actual and predicted percentage of COQ for the test sample Project Actual real life Predicted Percentage Comments No percentage percentage Deviation % Correct % Correct % Correct % Correct % All Rights Reserved 63

11 VI. CONCLUSIONS The survey results illustrated that cost of quality are greatly affected by many aspects. Among these aspects come Suppliers, Supervision team experience, Planned COQ for the project, Awareness of quality for the project team, Labor skills, Design errors, Defected material. All of these factors make the detailed estimation of such cost of quality amore difficult task. Hence, it is expected that an ANN s model would be a suitable tool for assessment of cost of quality in construction projects. The following conclusions may be deduced from this study: All the way through the literature review, potential factors that control the percentage of cost of quality for building construction projects were recognized. Twenty nine factors were identified. The analysis of the composed data gathered from a questionnaire survey among the construction experts illustrated that Suppliers, Supervision team experience, Planned COQ for the project, Awareness of quality for the project team, Labor skills, Design errors, Defected material, Execution errors, Project duration, Project type, Class of contractors, Weather conditions, Equipment downtime, Project size, Client type are the top 15 factors affecting the percentage of cost of quality for building construction projects. A satisfactory neural network model was obtained for predicting the percentage of cost of quality for building construction projects for the future projects. This model consists of one input layer with 15 neurons (nodes), one hidden layer having twenty eight hidden nodes with a tangent transfer function and one output layer. The learning rate of this model is set automatically by the N-Connection while the training and testing tolerance are set to 0.1. Testing the validity of the proposed model was carried out on five (5) facts that were still unseen by the network. The results of the testing indicated an accuracy of 100%. ACKNOWLEDGMENTS Authors like to express sincere gratitude to our principal Dr. V H Abdul Salam for providing such quality facilities in our esteemed institution for our paper. Authors extend sincere gratitude to all teaching and non-teaching staffs for their valuable advice and help to move forward with the paper. REFERENCES 1. Hany Shoukry Tawfek et.al Assessment of the expected cost of quality (COQ) in construction projects in Egypt using artificial neural network model HBRC Journal (2012) 8, A. Samadony et al., The cost of quality in the Egyptian Construction industry, HBRC Journal 2 (3) (2006). 3. Y. Rosenfeld, Cost of quality versus cost of non-quality in Construction, Construction Management and Economics 27 (2) (2009) Y. Firuzan et al., Assessing contractor quality performance,construction Management and Economics 20 (3) (2002) B. Mwamila et al., Semi-prefabrication concrete techniques in developing countries, Building Research and Information Journal 27 (3) (1999) M.Abas, S.B. Khattak, I.Hussain, S.Maqsood, I. Ahmad (2015)- UET Technical paper,vol.20 (SI) No.II (S), pp.: I. Dikmen et al., Strategic use of quality function deployment (QFD) in construction industry, Building and Environment Journal 40 (2) (2005) 245 All Rights Reserved 64