Using Tensor Factorization to Predict Network-Level Performance of Bridges
Journal of Infrastructure Systems
The nation’s bridges are critical assets that facilitate travel and commerce, making them important components of economic growth. However, bridge repair and rehabilitation needs far exceed the resources available to maintain them. Prediction of future bridge network conditions is essential to bridge asset management because it provides valuable information for budgeting and medium- to long-term planning. This paper focuses on the use of tensor factorization, an advanced data analysis approach, as a tool that can be used for both exploratory analysis and prediction of network-level performance of bridges in states across the United States. The network-level performance indicators in this work are defined as a percentage of bridges that are structurally deficient, functionally obsolete, or both. These indicators are calculated based on bridge areas and counts. The tensor decomposition approach decomposes multidimensional data into lower-order forms while preserving variation over time. Prediction of future bridge network conditions was done using the compressed time factor matrix, which yielded mean absolute error (MAE) values indicating promise for this new approach.
Locate the Document
Adarkwa, O., Attoh-Okine, N., & Schumacher, T. 2016. Using Tensor Factorization to Predict Network-Level Performance of Bridges, Journal of Infrastructure Systems, 23(3).