Imputing Missing Strain Monitoring Data in Structural Health Monitoring

Abstract

Data missing often occurs in structural health monitoring, this article deals with imputing missing strain monitoring data. Raw data collected by strain sensors are divided into stochastic response and temperature-induced response. For stochastic response, the copula is used to construct the conditional distribution of data from the intermittent fault sensor given measurements of a covariate sensor, missing data are imputed by post-processed samples generated from the conditional distribution. To overcome the drawback of conventional copula-based imputation method with fixed marginal distributions estimated from non-missing data, the distribution to distribution regression method is applied to impute missing marginal distributions adaptively. For temperature-induced response, missing data are imputed by the functional regression method. A case study based on field test data from a real bridge is conducted to illustrate the proposed approach.