Engineers who design aero engines and gas turbines face a stubborn computational bottleneck: before a hot metal component can be certified for service, its stresses must be traced through every start-up, load change and shutdown of an operating cycle. Transient structural analysis of that kind is notoriously expensive, because the temperature field inside the part changes continuously, the material properties shift with temperature, and the resulting thermal strains drive stress fields that evolve second by second. A team of Japanese researchers now reports that a model reduction technique called proper generalized decomposition, or PGD, can slash the cost of these calculations on industrial-scale models while keeping the answers close to those of a conventional finite element analysis.
Tomohiro Ishida, Hiroshi Ito and Kenjiro Terada describe the work in a study published on 19 June 2026 in the journal Advanced Modeling and Simulation in Engineering Sciences. The authors applied PGD to transient thermoelasticity under the small-strain assumption, with material properties that depend on temperature, by decomposing the displacement field into spatial and temporal modes while accounting for thermal strain. In their most demanding test, a simplified axial-flow turbine vane model of roughly 1.5 million degrees of freedom, the in-house PGD analysis software they developed reduced the computation time to approximately one-fourth of that required by their in-house finite element software, while maintaining high accuracy.
The physical motivation is familiar to anyone who has watched a jet engine spool up. In high-temperature axial-flow rotating machinery, rapid start-stop cycles and load transitions generate transient temperature distributions, and those distributions produce significant thermal stresses. If such stresses exceed the yield stress or the fatigue limit, they can cause inelastic deformation and fatigue cracking, compromising structural reliability. Accurate evaluation of transient thermal stresses under realistic operating conditions is therefore essential to reliable design, yet transient and nonlinear analyses of full-scale industrial components remain computationally intensive, posing a persistent challenge to efficient design workflows.
PGD belongs to a family of techniques known as model order reduction, which simplify analytical models through physical or mathematical approaches to reduce computational complexity. Among the widely adopted methods are principal component analysis, reduced basis methods and proper orthogonal decomposition, or POD. POD, in particular, has been studied extensively in structural analysis: it derives low-dimensional bases through singular value decomposition of precomputed solution snapshots, and it has enabled applications such as virtual sensing, online fatigue prediction for structures under wind and wave loading, and residual stress estimation from surface sampling points. The catch, the authors note, is that POD-style approaches need those snapshots in advance.
That requirement becomes awkward in structural design, where deriving optimal solutions from a design space that includes boundary conditions and geometries often makes it difficult to prepare snapshots beforehand. PGD sidesteps the problem by generating reduced bases on the fly. The method iteratively computes reduced bases and solves reduced models until convergence, reducing computational time by deriving and solving individual governing equations through the separation of spatial, temporal or parametric variables in the solution of partial differential equations. Variants in the literature include LATIN-PGD, which combines the LATIN method proposed by Ladevèze, and approaches that incorporate the Newton-Raphson method into the PGD framework.
The technique has already been applied across a striking range of engineering problems, from advection-diffusion and Fokker-Planck equations to quantum mechanics, and from the Navier-Stokes equations, where Dumon and colleagues achieved significant computational time savings, to transient heat conduction with metal temperature as a parameter, unsteady thermal analyses, and temperature prediction in powder bed fusion additive manufacturing. In structural mechanics, PGD has been used for shell structures, NURBS-based geometry representations, geometrically nonlinear problems, elastoplasticity, viscoplasticity, crystal plasticity, frictional contact and damage mechanics. Notably, although PGD follows a different computational procedure from conventional finite element methods, it shares common operations such as stiffness matrix assembly and load vector construction, allowing partial reuse of existing FEM codes.
The formulation presented in the new study treats the quasi-static equilibrium of a three-dimensional elastic body under small-deformation theory with isotropic material behavior. Stress is evaluated from the mechanical, or elastic, strain, which excludes the thermal strain from the total strain, with thermal strain given by the product of the coefficient of linear thermal expansion and the temperature difference from a reference state. The key move is to write the displacement as the product of a spatial mode and a temporal mode, and then, for an arbitrary number of modes, as a superposition of such pairs. From the arbitrariness of the virtual displacement, the weak form of equilibrium splits into two equations: one solved for the spatial mode and one for the temporal mode.
The solution algorithm alternates between these two problems. For each new mode, the spatial mode is computed first assuming an initial temporal mode, then the temporal mode is updated pointwise so that the scalar residual vanishes at each discrete time step, and the cycle repeats until the change in the temporal mode falls below a tolerance. Modes are added sequentially until the magnitude of the newest temporal mode, relative to the first, drops below a residual tolerance. The spatial mode vector is normalized so that the temporal mode carries the physical magnitude of the displacement, which the authors say makes the contribution of each mode easier to assess and lets the temporal mode serve directly in convergence checks. Stiffness matrices and external force vectors are evaluated at each discrete time step using the temperature field from a transient heat conduction analysis, accounting for temperature-dependent material properties.
Three numerical examples test the method. The first, a strip plate under a pressure load ramping from 0 to 100 megapascals, is a linear problem in which displacement is proportional to load; convergence was achieved with a single mode, and the PGD results agreed well with the commercial code ABAQUS, Version 2022, including the stress concentration in a fillet region. The second, a perforated plate cooled by a hole and heated at its edges, required at least two modes because the transient thermal loading changes over time; PGD again matched ABAQUS for both displacement and stress and captured the stress concentration at the cooling hole edge.
The decisive test was the third case: a simplified axial-flow turbine vane with four blades, modeled with second-order tetrahedral elements comprising 497,074 nodes and 790,542 elements, made of Inconel 718 with temperature-dependent properties. Thermal loads transitioned through 40, 100 and 80 percent load over 2000 seconds, with the gas-side heat transfer coefficient varying from 1000 to 900 watts per square meter kelvin under full load. With a residual tolerance of 10 to the minus 3, the relative root-mean-square error in von Mises stress against ABAQUS was approximately less than 1.5 percent, and convergence was achieved in six modes, meaning the linear system solver ran only 12 times. Tightening the tolerance to 10 to the minus 4 lowered the error to less than 0.8 percent but required 20 modes, a trade the authors judge not worthwhile. The authors caution that the present scope is restricted to linear thermoelasticity with small strains and space-time separation, and they plan to extend the method to material and geometric nonlinearities, including viscoplastic effects; they note that low-cycle fatigue in aero engines and gas turbines is frequently driven by thermal stresses that do not exceed the elastic limit, which is where they see the approach’s practical value.
Subject of Research: Application of proper generalized decomposition to accelerate transient thermoelastic structural analysis of high-temperature rotating machinery components
Article Title: Transient structural analysis with proper generalized decomposition
Article References: Ishida, T., Ito, H., & Terada, K. (2026). Transient structural analysis with proper generalized decomposition. Advanced Modeling and Simulation in Engineering Sciences, 13(1), Article 10. https://doi.org/10.1186/s40323-026-00328-7
Image Credits: AI Generated
DOI: 10.1186/s40323-026-00328-7
Keywords: proper generalized decomposition, model order reduction, transient structural analysis, thermoelasticity, turbine vane, thermal stress, finite element method, computational mechanics, Inconel 718, aero engines, gas turbines, reduced order modeling
News Source: Denise Maddox. (October 29, 2026). Decomposed modes speed up transient stress analysis of hot turbine parts. Scienmag.
