An Advanced Framework to Manage Uncertainty and Buffers in Construction

Abstract

Variation is an inherent attribute of a production process. It represents one of the largest problems in construction projects by inducing uncertainty in plans. Among the strategies adopted to deal with variation in production, the use of buffers has been a typical part of the solution. Buffers, in the form of extra time, inventory, or capacity, are used to shield the production process against the likely variation. From a practical standpoint, buffers are instrumental in production systems. Their absence will make the systems vulnerable to the inevitable process variations. Therefore, systems with no buffer are a rare practice in ordinary construction projects. Meanwhile from a theoretical standpoint, in lean production, buffers represent waste. Accordingly, they are suggested to be minimized in the system. The tension between theory and practice necessitates finding a balanced state in allocating buffers to the production systems. The balanced state should provide a compromise between an excessive use of buffer, which results in undesirable waste, and a scarce-buffer scenario, which causes poor project performance. A wide range of concepts and techniques have been developed to cope with the buffer allocation problem in construction; however, the developed strategies have not been well accepted by construction practitioners. The construction industry has traditionally employed informal approaches to allocate buffers. It often undermines the required balance in buffer allocation resulting in a flawed construction plan. In order to determine the required balanced state, this research aims to model the extent to which the production project can diverge from its planned value. It uses time as an indicator of the production performance and terms the possible extent of deviation from the planned value as variability. The expected variability to be treated in a project can be reduced using planning and control strategies such as the Last Planner System. Accordingly, the expected variability that is discussed in this research might have been reduced by the commitment control strategies prior to be modelled. This study converts the expected variability into a set of random variables at activity level, and analyses its propagation to the project level using a statistical approach. This method can result in a comprehensive statistical characterization of the system performance. To do so two primary impediments should be tackled: a) Although the beta density function has achieved extensive acceptance as a suitable statistical model to represent variability in performance of construction activities, its application faces serious limitations in certain production situations. The limitations are apparent when a coefficient of variation associated with the process variability is higher than 100 percent. This research has explored a reliable probability density function to model the situations where the beta distribution is not suitable. As a case study, it collected data from 73 construction samples corresponding to 25 projects from three different countries. The collected data were analysed using statistical inference techniques and stochastic simulation. The main finding from this part of the study shows that the Burr density function can give a more accurate statistical model of the variability of the construction activities with variability levels between 100 and 150 percent. b) In order to establish a reasonable model of variability propagation from activities to the project level, an accurate and efficient calculation method is required to combine the identified variability models at the activity level. Probabilistic methodologies based on sampling methods (such as Monte Carlo simulation) are readily available for this purpose. However, they are subjected to the sampling theory problems including sampling errors that can reduce the accuracy of the results. Also, they suffer from a slow convergence ratio. These problems have restricted their application in real projects. This study discusses an Approximate Combinatorial Method, which uses a mathematical approach to combine the estimated functions at the activity level along the project activities network. The method satisfies the practical requirements while provides a suitable level of accuracy. With the two primary impediments of characterizing the variability in a construction project removed, the balance problem can be resolved through finding the optimum size and location of buffers in the system. Accordingly, this study has developed two probabilistic-based buffer allocation methods that can account for the specific features of the underlying scheduling policy adopted by the construction planners. They also address multiple conflicting objectives normally exist for a construction project. The first Probabilistic-based Buffer Allocation method (PBAL) allows a flexible start time for activities and supposes they will start as soon as possible. The computational advantages of PBAL have been shown through a set of computational experiments. The experiments indicated the high level of computational accuracy supported by PBAL. It can incorporate the impact of variability from any activity into the overall distribution at the project level and preserves the quality of results independent of the number of activities on the network. The second method applies a Robust Probabilistic-Based Buffer Allocation framework (RPBAL) that helps to find the optimum buffer allocation in construction schedules where activities will not start before their planned time. RPBAL has been tested using data from a road construction project with 16 activities. The results were compared to a discrete event simulation experiment. The simulation experiment included 50,000 replications to minimize the potential sampling error. This comparison demonstrated the computational accuracy and robustness of the calculations by RPBAL. These two developed methods have been integrated into an inclusive probabilistic-based buffer allocation method (IPBAL). Accordingly, an implementation pathway has been introduced that enables the IPBAL framework to bridge the gap between the concerns over the usability aspects, and the computational accuracy required in buffer allocation to the systems. At the final stage, the usability aspects of IPBAL have been evaluated based on eight interviews completed with experts from Construction Industries in New Zealand and Australia. Thematic analysis of the responses acknowledged the major practical capabilities embedded in the implementation pathway of IPBAL. The experts corroborated the structured approach, graphical presentation of results, and the employed Approximate Combinatorial method as the features that promote usability of IPBAL. Also provided are a number of recommendations to pave the way for future developments of the IPBAL framework; these include data pre-processing for probability encoding, the effects of human behaviour such as Student Syndrome and Parkinson law, and the value degree of products.