[Paper Review] High dimensional inference for the structural health monitoring of lock gates

Matthew Parno, Devin O’Connor+1 more|arXiv (Cornell University)|Dec 13, 2018

Water Systems and Optimization5 citations

TL;DR

This paper presents a scalable Bayesian inference framework that combines finite element models with high-dimensional Gaussian process priors to enable real-time structural health monitoring of lock gates using noisy strain data. By leveraging Karhunen-Loève decompositions, state-space representations of Gaussian processes, and Kalman smoothing, the method efficiently infers spatially distributed boundary conditions and thermal effects across nearly 1.4 million parameters in under an hour on a standard desktop, demonstrating near real-time applicability for critical infrastructure monitoring.

ABSTRACT

Locks and dams are critical pieces of inland waterways. However, many components of existing locks have been in operation past their designed lifetime. To ensure safe and cost effective operations, it is therefore important to monitor the structural health of locks. To support lock gate monitoring, this work considers a high dimensional Bayesian inference problem that combines noisy real time strain observations with a detailed finite element model. To solve this problem, we develop a new technique that combines Karhunen-Lo\\`eve decompositions, stochastic differential equation representations of Gaussian processes, and Kalman smoothing that scales linearly with the number of observations and could be used for near real-time monitoring. We use quasi-periodic Gaussian processes to model thermal influences on the strain and infer spatially distributed boundary conditions in the model, which are also characterized with Gaussian process prior distributions. The power of this approach is demonstrated on a small synthetic example and then with real observations of Mississippi River Lock 27, which is located near St. Louis, MO USA. The results show that our approach is able to probabilistically characterize the posterior distribution over nearly 1.4 million parameters in under an hour on a standard desktop computer.

Motivation & Objective

• Address the challenge of monitoring aging lock gates in the U.S. inland waterways, which are past their design life and pose high economic risks if failed.
• Overcome the computational intractability of high-dimensional Bayesian inference in structural health monitoring with noisy, temporally correlated strain observations and complex finite element models.
• Enable probabilistic characterization of boundary conditions, thermal effects, and sensor biases to support fatigue life prediction and maintenance planning.
• Develop a scalable, near real-time inference pipeline suitable for operational deployment on standard computing hardware.

Proposed method

• Apply Karhunen-Loève decomposition to reduce the dimensionality of spatially distributed boundary conditions and thermal strain fields.
• Represent Gaussian process priors for boundary loads, thermal strains, and sensor biases using time-dependent stochastic differential equations in state-space form.
• Use Kalman smoothing to perform efficient, linearly scalable inference over high-dimensional parameter spaces, with computational cost scaling linearly with the number of observations.
• Incorporate quasi-periodic Gaussian processes to model diurnal thermal strain effects observed in real strain gage data.
• Integrate a high-fidelity finite element model of Lock 27 with real-time strain observations from the USACE SMART Gate database.
• Apply static condensation to reduce the finite element model size, improving computational efficiency without sacrificing accuracy.

Experimental results

Research questions

• RQ1 Can a Bayesian inference framework efficiently handle high-dimensional structural health monitoring problems with over a million parameters using real-world strain data?
• RQ2 How can temporal correlations and environmental effects—such as diurnal thermal expansion—be effectively modeled and separated from elastic strain in strain gage data?
• RQ3 To what extent can state-space representations of Gaussian processes enable scalable, linear-time inference in complex structural models with noisy observations?
• RQ4 Can the method accurately infer unknown boundary conditions and sensor biases without direct measurement, using only strain gage data and a detailed finite element model?
• RQ5 Is the approach suitable for near real-time monitoring, allowing operators to detect structural anomalies promptly and respond to potential failures?

Key findings

• The method successfully performed Bayesian inference over nearly 1.4 million parameters in under one hour on a standard desktop computer, demonstrating high computational efficiency.
• Quasi-periodic Gaussian processes effectively captured diurnal thermal strain cycles in strain gage data, enabling accurate separation of thermal and elastic strain components.
• Kalman smoothing with state-space Gaussian process priors enabled linear scaling with respect to the number of observations, overcoming the cubic cost of standard Gaussian process inference.
• The posterior distribution revealed physically plausible boundary load patterns that were consistent with gate geometry and gage locations, with higher uncertainty in regions without strain gages.
• The approach identified significant discrepancies between prior and posterior loads during unsealed miter conditions, highlighting the model’s sensitivity to physical state and the importance of data-driven correction.
• An interactive Bokeh-based web interface was developed to explore the full joint posterior distribution, enabling deeper uncertainty quantification and visualization of model parameters.