Productivity Analysis of Horizontal Directional Drilling Muhammad Adel 1 and Tarek Zayed 2

Transcription

1 835 Productivity Analysis of Horizontal Directional Drilling Muhammad Adel 1 and Tarek Zayed 2 1 Graduate Student, Dept. of Building, Civil & Environmental Engineering, Concordia University, Montreal, Quebec, Canada H3G 1M7. mu_mahmo@encs.concordia.ca 2 Associate Professor, Dept. of Building, Civil & Environmental Engineering, Concordia University, Montreal, Quebec, Canada H3G 1M7. zayed@bcee.concordia.ca Abstract: With the rapid increase of new installations, replacement and repairs of pipe utilities, the demand for trenchless excavation methods with minimum disruption to the public such as horizontal directional drilling (HDD) has increased. Canadian National Research Council reports that rehabilitation of municipal water systems would cost $28 billion from year 1997 to 2012 (NRC 2004). Contractors, engineers, and decision makers are always facing a challenge of how to estimate the cost of new pipe installation using the HDD due to the presence of subjective factors. The HDD process involves a large number of factors to be considered for productivity prediction and cost estimation. Therefore, an emergent need for developing a dedicated HDD productivity model is currently undertaken to meet industrial needs. The presented research aims at identifying the main factors that affect productivity of HDD operations and designing a productivity model. A neurofuzzy approach is utilized to design the HDD productivity prediction model for underground pipe installations in clay soil. The neurofuzzy system is developed based on actual project data that are collected through interviews, phone calls and questionnaire surveys. Results show that crew and operator skills and pipe diameter greatly affect the HDD productivity and the project as a whole. Introduction An enormous task was added to the utility service companies (i.e., power, telecommunications, water mains, and sewer) in Canada and USA since the beginning of this century for maintaining, repairing the existing utilities and constructing the new installation due to the increase in demand. These operations had proven to be expensive, especially in the crowded urban areas. The total activity cost includes the costs of the ground surface repair (i.e., sidewalks, pavement, brick paving) and the social costs due to the disruption to traffic and unfavourable impact on nearby activities (Ariaratnam et al, 1999). To face the urgent demand of the aged utilities replacement or renovation and the environmental constrains, the municipalities and utility companies as well as contractors stared to seek other alternatives to install or repair their underground assets. Trenchless Technology (TT) proved to be what they were looking for according to its various ranges of methods, materials, and equipments. Canadian

2 836 National Research Council emphasized that only rehabilitation of the municipal water systems would cost $28 billion from year 1997 to 2012 (NRC 2004). Horizontal Directional Drilling (HDD) is a trenchless technique that proposes several benefits over traditional open-cut. The HDD was originally developed by the oil industry in the United States; this technique is now widely used for installing all pressure pipes under obstacles such as motorways, large rivers, airport runways, etc. A steerable drill bit of 90 mm diameter usually starts digging from the earth surface and generates a pilot hole. Upon completion, the pilot string is removed and a rotating reamer is attached to travel back along the pilot hole. Consequent reaming continues until the required diameter is achieved. Recently, in the last 15 years, the directional drilling industry in North America has grown from a small revolutionary group of contractors operating a few directional drilling rigs to a well established, extensive, multibillion dollar industry (Allouche et al., 2000; Ariaratnam and Allouche, 2000; Ariaratnam, 2005). Almost in 15 years time, horizontal directional drilling (HDD) industry in North America has grown from a bunch of contractors concerned with a few directional drilling units operations to a multibillion-dollar industry (Kirby et al., 1997). Canada and the USA today are facing a growing problem in rehabilitating its decaying underground utility systems (Ali et al., 2007). Currently, HDD has become the most favourite method for new underground conduits and pipelines installations (Lueke and Ariaratnam, 2005). The number of HDD contractors has increased as a result of the growth in size and difficulty of the HDD projects, thus there is a growing need for developing a dedicated model design to meet the special industrial needs due to the increase of complexity and size of the projects currently undertaken and implemented using HDD. The society is influenced by the great development of underground infrastructure, environmental concerns, and economic trends, which resulted in the technology advancement to achieve the most efficient and cost effective utility installation, reparation, renovation and maintenance (Allouche et al. 2003). Contractors, engineers, and decision makers are always facing a challenge of how to estimate the cost of new pipe installation using the HDD due to the presence of many qualitative factors. HDD is a process that involves a large number of factors that need to be considered for productivity prediction and cost estimation. Productivity of trenchless technology methods is usually predicted using heuristic techniques and experts opinions (common practices) without considering the subjective factors effect. Contractors always use their experience in estimating production rates. In addition, the industry lacks models that predict productivity of the trenchless techniques (Ali et al., 2007). Allouche et al. (2000) stated that the subsurface conditions and pipe diameter are the two main factors affecting productivity in utility projects. Allouche et al., 2003 stated that over 49 HDD contractors, engineers, and consultants showed the growing demand for the HDD productivity model(s). Importance of different HDD packages is shown in Table 1. Dubey et al. (2006) developed a simplified deterministic productivity assessment model for HDD. They implemented their application based on two projects only in

3 837 which two regression linear models are designed between bore length and cycle time. The soil type, rig size, pipe material and diameter greatly affect the drilling, prereaming and pull back times. Other factors, such as weather condition, contractor s experience and job management conditions considered as efficiency factors. On the other hand, Allouche et al. (2003) developed two computerized applications tailored to the HDD industry. The first is an integrated data management system that combines asset management, cost control, estimation and project-tracking capabilities, to enable decision makers to intimately monitor field performance in terms of expenses and productivity. The second is a simulation model developed to optimize the utilization of drilling rigs and hydro-vacuum trucks on large-scale urban projects. Table 1: Current needs for devoted HDD software (Allouche et al., 2003) Software Package Yes % Somewhat % No% Rank Project Tracking Cost Control Project Management Asset Management Cost Estimation Accordingly, both qualitative and quantitative parameters - that affect horizontal directional drilling productivity - were investigated in order to identify the main factors that affect the productivity of horizontal directional drilling operations, thereby to refine the HDD process and obtain productivity. In addition, estimating the productivity of horizontal directional drilling projects under different soil conditions should be studied. Therefore, the presented research in this paper has the following objectives: (1) Identify the important factors that affect the productivity of horizontal directional drilling operation and (2) Develop a productivity model. Neurofuzzy Approach The integration of neural network and fuzzy logic are receiving the attention to develop real-world applications (Medsker L., 1996). Neurofuzzy approach refers to hybrids of artificial neural networks and fuzzy logic. The application of the neurofuzzy technique is based on the integration of the explicit knowledge representation of the fuzzy logic with the learning power of neural networks (Simon and Biro, 2005). The mechanism of neurofuzzy system can be simply explained as; having the input vector consisting of a set of fuzzy values, as well as having the connection weights of the nodes to the nodes in the previous layer in fuzzy values. Where, the weights and input values are each represented by membership function. A further summation process is implemented to find the product of the membership function of the fuzzy inputs and weights, and then add the resulting membership functions to get another single that represents the integration of weighted fuzzy inputs to the node. Lastly, a final operation takes place on the resultant finding out a crisp value for the node output (Medsker L., 1996). Crisp value is a numerical deterministic value of the factor effect. The process of converting crisp input value to fuzzy value and vice versa are called fuzzification and defuzzification, respectively.

4 838 Neural and fuzzy systems are model-free function estimators that can be adjusted or trained for improved performance, where they are by nature readily implemented with parallel processing techniques. Neural networks consist of connection among a distribution of nodes. On the other hand, fuzzy systems process rules that associate, in parallel, fuzzy inputs with fuzzy output sets (Medsker L., 1996). Georgy et al. (2005) presented a study utilizing neurofuzzy intelligent systems for predicting the engineering performance. The system employed statistical variable reduction techniques to develop linear regression models of similar engineering performance prediction scheme. Georgy et al. (2005) utilized neurofuzzy systems as credible approach for predicting engineering performance, based on their ability to model nonlinearity. The developed regression model that depict the whole set of input variables was not feasible due to the significant number of inputs. The neurofuzzy system showed fewer deviations from the actual outputs than regression models. The accuracy of the models is affected by the limited data set available during model development. Increasing the volume of actual project data could improve the models. Research Methodology The developed research methodology is shown in Figure 1. Data collection and analysis comes along with the intensive literature review on; factors affecting productivity of TT methods with a focus on HDD in addition to the previous models developed to calculate their productivity, and neurofuzzy modeling technique. A new automated HDD productivity prediction model is developed, followed by conclusion and future work needed for the present research. Model Development This paper utilized the neurofuzzy technique in developing the HDD Productivity Prediction (HDDPP) model in clay soil. The Fuzzy neural network model has three main functions fuzzifier, fuzzy rules, and evaluator/defuzzifier. The starting layers process crisp input data of the eight input modeling parameters and implement membership functions of each parameter. Therefore, crisp input data transforms into membership function values, which are the output of the first layer of nodes. The hidden middle layer(s) includes fuzzy rules operating on the fuzzified input. Finally, the last layer(s) collects the results of applying the rules, and defuzzify the results to get the crisp value representing the HDDPP. The presented study focuses on both quantitative and qualitative factors affecting productivity calculation in addition to the activity cycle time. The thorough literature review, as well as the industry experts input, comes out with identifying thirteen main factors to be the most significant factors affecting HDD productivity (Figure 2); operation/crew skills, safety regulations, rig size, machine condition, slurry flow rate, steering problems, soil types, unseen obstacles, site/ weather conditions, pipe diameter, pipe length, pipe depth and pipe type. These factors are sub-divided from four main categories; management, mechanical, environmental and pipe conditions. Due to data limitations, this research only focuses on analyzing the relationship

5 839 between activities cycle time and the eight most comprehensive factors (i.e. operation/crew skills, pipe diameter, rig size, machine condition, unseen obstacles, pipe length and the site/ weather conditions) as well as identifying their relevant impact on productivity. The collected data are clustered according to the eight identified input parameters. These eight input parameters are clustered into four levels as shown in (Figure 2). The first level identifies the working soil type, which are limited to clay soil. The second input cluster level defines the pipe condition and the operational conditions, which leads to the third level of the pipe length and diameter representing the pipe conditions and the rig size, machine condition, site/weather conditions and operator/crew skills representing the operational conditions. Data Collection The main set of collected data includes, real HDD project data, representing the project input variables, cycle time and productivity measures, which is identified in the problem formulation stage. This set of data is used to train the structure connections of the neurofuzzy system to properly relate the defined HDD input variables with the corresponding productivity measures. Figure 1: Schematic Methodology of Studying HDD Productivity Prediction

6 840 Figure 2: Modeling Data Clustering A questionnaire was designed based on literature review and interviewing construction industry professionals to investigate the most effective factors on HDD productivity. Questionnaire was sent to professionals, consultants, contractors and equipment operators in HDD technique. The first part was designed to collect the participant s and project s information. The second part collects the effect of various factors on productivity using a unified fuzzy performance scale. The participants were given the possibility to add more factors and evaluate their impact. The last part collects the HDD cycle time duration. Approximately, 220 questionnaires were sent to professionals in USA, Canada, UK and South Africa; however only 12% replied (28 individuals replied with multiple soil types). Almost twenty percent of this data (Two projects from each soil type) were excluded from the modeling phase and used only for validation purposes. Collected data was organized and analyzed according to the various soil types and to be applicable by the modeling neurofuzzy tool. Based on the received questionnaires, the HDD productivity prediction model was developed. Model Implementation The model mechanism is basically divided into training and validation phases. The training starts by both qualitative and quantitative data entry in terms of parameter weights and cycle time/production rate, respectively. The model was developed using MATLAB ver. 7.0 using the neurofuzzy tool box ANFIS. Data entry layer takes both crisp and fuzzy factors and process all factors to the training process, which starts by data fuzzification where the degree of each parameter is determined. The data are

7 841 then proceeded to be trained via ANN. The fuzzy outcome is then diffuzified, where the crisp input parameter variable is determined given its degree of membership. In other words, the functions performed by the Fuzzy neural networks are fuzzify system inputs, defuzzify model outputs and develop a structure weight that properly represent the nonlinear relationships across the model inputs and outputs (Figure 3). This system is able to build relationships between the factors affecting the HDD process and the overall productivity. Therefore, the model can predict HDD process productivity given a specified set of project input variables (parameters), with an adequate error percent. The Clay FNN model considered seven input parameters, i.e. operation and crew skills, pipe diameter, rig size, machine condition, site and weather conditions, pipe length and unseen obstacles; where all factors had three membership functions except for the operator/crew skills and rig size, which had only two membership functions. The clay model was trained and tested via 40 data points; of which 32 were used in training and 8 in testing phase. Furthermore, clay FNN model was developed based on 1992 nodes, 7776 linear parameters, 57 nonlinear parameters, 32 training data pairs, and 972 fuzzy rules. The developed model is tested after the modeling phase, where the neurofuzzy system, splits the modeling data into training and testing data. Eight of the total collected data points are unexposed to the neurofuzzy system during the training phase to be used for testing purposes. Afterwards, the testing data set is used to predict productivity and compare results with the real time productivity in which it showed robust results with an average validation of 96%. Conclusion Horizontal Directional Drilling (HDD) has proven itself in the underground construction market as one of the most effective TT methods for new underground pipe/cable installations. Due to HDD competitive market conditions, client expectations, and technological advancements, an emergent need for HDD contractors has risen to identify the major factors that affect the HDD productivity. There is insufficient information in the literature regarding models and software for the HDD productivity analysis. Therefore, a productivity prediction model for HDD projects was developed using the neurofuzzy system. The model helped experts to estimate and predict the project duration. It is found that crew and operator skills, soil type and pipe diameter were the most significant factors affecting HDD productivity, while pipe length and weather conditions had the least effect. This model was a tool for experts and professionals to help them justify their productivity calculation by quantifying some of the subjective factors effect. This will better impact their schedule and cost estimation of HDD works. Moreover, it provides researchers and experts with the most significant factors that contribute to HDD installations. The developed neurofuzzy methodology and model can be used in similar research applications.

8 842 Figure 3: Basic Architecture of Clay FNN Model

9 843 REFRENCES Ali, S., Zayed, T. and Hegab, M. (2007). Modeling the Effect of Subjective Factors on Productivity of Trenchless Technology Application to Buried Infrastructure Systems, Journal of Construction Engineering and Management, ASCE, October, Vol. 133, pp Allouche, E., Ariaratnam, S. and Lueke, J. (2000). Horizontal Directional Drilling: Profile of an Emerging Industry, Journal of Construction Engineering and Management, ASCE, January/February, Vol. 126, No. 1, pp Allouche, E., Ariaratnam, S. and Macleod, C. (2003). Software for Planning and Cost Control in Directional Drilling Projects, Journal of Construction Engineering and Management, ASCE, July/August, Vol. 129, No. 4 pp Ariaratnam, S. (2005). Invited Presentation Development of Good Practices for Horizontal Directional Drilling, Toward 21st Century for Trenchless Technology, Guangzhou, China. Ariaratnam, S., Lueke, J. and Allouche, E. (1999). Utilization of Trenchless Construction Methods by Canadian Municipalities, Journal of Construction Engineering and Management, ASCE, March/April, Vol. 125, No. 2, pp Ariaratnam, S. and Allouche, E. (2000). Suggested practices for installations using horizontal directional drilling, Practice Periodical on Structural Design and Construction, ASCE, November, Vol. 5, No. 2, pp Dubey, B., Gupta, M. and Zayed, T. (2006). Deterministic Productivity Model for Horizontal Directional Drilling, INFRA 2006, Quebec City, QC, Canada, November Georgy, M., Chang, L. and Zhang L. (2005). Prediction of Engineering Performance: A Neurofuzzy Approach, Journal of Construction Engineering and Management, ASCE, May, 131 (5), pp Kirby, M., Kramer, S., Pittard, G. and Mamoun, M. (1997). Design Guidelines and Procedures for Guided Horizontal Drilling, Part II, No-Dig Engineering, 3(4), pp Lueke, J. and Ariaratnam, S. (2005). Surface Heave Mechanisms in Horizontal Directional Drilling, Journal of Construction Engineering and Management, ASCE, May, Vol. 131, No. 5, pp Lawson, G. (2003). Water and Sewer Construction with Horizontal Directional Drilling Equipment, American Society of Civil Engineers, May 8, pp Medsker, L. (1996). Microcomputer Applications of Hybrid Intelligent Systems, Journal of Network and Computer Applications, Vol. 19, pp National research council of Canada (NRC), Assessing Canada s Infrastructure Needs: A Review of Key Studies September 2004 < surfed on Dec Simon, A. and Biro, D. (2005). About Neurofuzzy Module of the FuzzyTECH 5.5 Software. Online Edition at College of Nyíregyháza, Proc. of 6th International Symposium of Hungarian Researchers on Computational Intelligence, Budapest,