From lean production perspective, the physics of production flow can be thought of as comprising value adding and non-value adding (or waste) activities. Moreover, one of its core principles for work improvement is the elimination or mitigation of the latter component. This should be translated into increased productivity at the work site. The aims of this paper are to identify the relationship between productivity at the work site and the waste or non-value-adding activities, and to find out the root causes of the wastes. For this purpose, the waste activities are categorized into 20 sources according to their causes. Productivity data of formwork crews on multiple projects are collected together with the associated wastes. A neural network is then developed to model the influence of the wastes on measured productivity. The model is incrementally pruned so that, eventually, only eight significant wastes are identified and remain. The final model shows very good conformance when compared with observed data. After that the eight significant wastes have been correlated to the project level factors to find out their root reasons.
productivity, waste, neural networks, artificial intelligence