One-Way Thermo-Mechanical Coupled System Identification Using Displacement and Temperature Measurements

TL;DR

This paper introduces an optimization-based, adjoint-driven framework for identifying structural weaknesses and temperature fields in thermo-mechanical systems using sparse displacement and temperature data, improving accuracy over traditional assumptions.

Contribution

It proposes a novel monolithic and partitioned approach for coupled system identification, effectively recovering temperature and material properties even with limited sensor data.

Findings

Both approaches accurately recover Young's modulus and temperature distributions.

The proposed methods outperform constant-temperature assumptions and interpolation.

Sensor placement critically impacts identification accuracy.

Abstract

Structural system identification in the presence of thermal loads is challenging, as unmeasured or poorly modeled thermal effects can mask or mimic damage, leading to unreliable conclusions. This work presents an optimization-driven, adjoint-based high-fidelity system identification framework for localizing structural weakness and recovering the temperature field in one-way thermo-mechanical coupled structures. The methodology builds upon a standard optimization formulation that minimizes weighted discrepancies between simulated responses and measured data from a sparse displacement and temperature sensor network. To account for thermal effects, two strategies are proposed: a monolithic approach, which simultaneously identifies Young's modulus and temperature distributions, and a partitioned approach, which iteratively couples two inexact sub-problems through a Gauss-Seidel type fixed-point scheme. The proposed approaches are evaluated using two numerical examples -- a Plate With a Hole and a Footbridge model -- under linearly varying and localized thermal fields, and for different sensor layouts. Both approaches successfully recover the Young's modulus and temperature distributions, even when sensor placement does not fully capture the underlying thermal trends. Compared with a constant-temperature assumption and interpolation of the temperature field from sensor data, the proposed approach achieves the most accurate damage localization and temperature reconstruction. The largest gains occur when localized thermal features are poorly sampled by sensors, where interpolation and constant-temperature assumptions underperform. Furthermore, results show that the location of the temperature sensors is as influential as the number of sensors: well-placed sensors substantially improve identification, while additional sensors that miss critical thermal features provide limited benefit.