Structural reliability assessment is imperative to keep structural safety, durability and serviceability. One vital factor of such assessment is determination of dominant parameters of structure, called sensitivity analysis. There are many methods for determining dominant parameters, among them artificial neural networks are superior. Existing methods are generally based on a single neural network, but inadequate as a basis for parameter sensitivity analysis because of the instability of a single neural network. To address this deficiency, the paper describes a neural network ensemble-based parameter sensitivity analysis. The proposed method is applied to prestressed concrete bridge. Three dominant parameters were identified for limit state of decompression and six dominant parameters for limit state of crack initiation. The proposed method provides a common paradigm for analyzing the sensitivity of influential parameters, providing effective information to set up models and even to simplify reliability assessment.

           10.21495/91-8-637