Prediction of design error rework cost in EPC industrial projects

Design errors constitute the major part of rework in construction engineering projects. Due to the significant overlap between engineering, design and construction works industrial Engineering, Procurement and Construction (EPC) projects have high propensity of encountering design errors. The reviewed literature revealed that design error rework on industrial projects can reach 12. 4% of projects value. Given the high investment costs of industrial projects in the USA, design error rework cost could reach $4. 6Bn per annum. Design error rework costs and the failure to agree on their magnitude have been reported as one of the major disputes causes in construction engineering projects. The goal of this study is to examine the design error rework costs encountered in industrial EPC projects in Texas and to develop a quantitative predictive model to assist EPC contractors to obtain accurate estimates of design error rework cost in industrial EPC projects. To accomplish this, the research objectives are to (1) identify the main contributors of design error rework cost through analyzing change orders in industrial EPC projects identifying the causes of the change orders, and pinpointing the percentage and cost of design errors, (2) develop a predictive model to quantify the design error rework cost, (3) use the developed predictive model to provide accurate estimates of design error rework that will be used by EPC contractors to negotiate down the proposed subcontractor’s design error price, thus reducing the cost of design errors. Accordingly, the predictive model will also reduce the disputes related to design error cost through providing reliable estimates that satisfy the expectations of both subcontractors and EPC contractor. The analysis of fifty (50) contracts revealed that mean design error rework cost accounted for around 15. 5% of the projects value and for 62% of the rework cost. Utilizing actual historical data, three (3) machine learning algorithms were deployed to predict the design error rework cost during the detailed engineering and construction phase of industrial EPC projects. The results were compared with those of multiple linear regression models. The machine learning algorithms results outperformed the multiple linear regression models results. Support Vector Regression (SVR) algorithm utilizing the important parameters revealed by Random Forest (RF) algorithm outperformed ll other machine learning models and was the most accurate algorithm in predicting design error rework cost. The SVR algorithm had coefficient of determination (R2) of 98. 3%, Root Mean Square Error (RMSE) of 104,876. 70 and Mean Absolute Error (MAE) of 7,195. 11 on the testing dataset. The SVR algorithm capability to assist the EPC contractor in negotiating down the subcontractors’ design error rework price was demonstrated through a case study. Using the SVR algorithm, the EPC contractor successfully negotiated down the design error rework price of subcontractors by 19%.