Modeling of Portland Cement Concrete Pavement Longitudinal Joint Distress

摘要

The neural network technique was found to be more accurate in predicting theextent of spalling for the longitudinal joints from the knowledge of the parameters related to both the joint and pavement. In the neural network method, the inclusion of the contractor (reflecting the construction quality) in the neural model with 18 inputs resulted in improvement of the prediction accuracy, when compared to another network with 11 inputs which assumed that all pavements were constructed with similar construction quality. The developed neural spalling model was used to perform the following additional tasks: (1) ranking the construction contractors according to the quality of construction, (2) studying the sensitivity of the model parameters vs spalling which can serve as a guideline for revising the current practice for constructing concrete pavements, and (3) determining the level of service of the surveyed pavements to be interpreted in terms of maintenance required to prevent further deterioration of the joints.