Bayesian Damage Identification Using Strain Data from Lock Gates

Damage identification plays a significant role in the maintenance of navigation locks, which are part of the United States’ $500B replacement value in inland waterway infrastructure; maritime transport disruption for closed lock gates causes substantial economic and utility losses. Lock gates are normally instrumented with strain gauges, and one of the critical failure modes is the development of a gap between the supporting wall and gate, initiating from quoin and/or pintle part wear. This gap leads to undesirable load distributions that can induce gate failure by overload. The probability of damage exceedance from different values affects repair strategies. This work uses Bayesian inference to identify that damage. The input features are raw strain data, while the loading is assumed unknown. The inherent uncertainty in measurements and model assumptions result in a posterior distribution of parameters of damage such as gap size and location. The results show that the true parameter of damage, which is used to generate simulated data, could be predicted using the posterior.