Thermal conductivity reconstruction method with application in a face milling operation

• Published in: International journal of numerical methods for heat & fluid flow Vol. 33; no. 8; pp. 3025 - 3055
• Main Authors: Boos, Everton, Bazán, Fermín S.V., Luchesi, Vanda M.
• Format: Journal Article
• Language: English
• Published: Bradford Emerald Publishing Limited 22.06.2023; Emerald Group Publishing Limited
• Subjects: Boundary conditions; Composite materials; Conduction heating; Conductive heat transfer; Conductivity; Dependent variables; Discretization; Face milling; Finite element analysis; Heat conduction; Heat conductivity; Heat transfer; Inverse problems; Mathematical models; Methods; Optimization; Ordinary differential equations; Partial differential equations; Reconstruction; Regularization; Scaling; Smoothness; Temperature; Temperature measurement; Thermal conductivity; Thermal conductivity; Levenberg–Marquardt method; Chebyshev pseudospectral method; Milling operations
• ISSN: 0961-5539; 1758-6585; 0961-5539
• DOI: 10.1108/HFF-12-2022-0720

Purpose This paper aims to reconstruct the spatially varying orthotropic conductivity based on a two-dimensional inverse heat conduction problem described by a partial differential equation (PDE) model with mixed boundary conditions. The proposed discretization uses a highly accurate technique and allows simple implementations. Also, the authors solve the related inverse problem in such a way that smoothness is enforced on the iterations, showing promising results in synthetic examples and real problems with moving heat source. Design/methodology/approach The discretization procedure applied to the model for the direct problem uses a pseudospectral collocation strategy in the spatial variables and Crank–Nicolson method for the time-dependent variable. Then, the related inverse problem of recovering the conductivity from temperature measurements is solved by a modified version of Levenberg–Marquardt method (LMM) which uses singular scaling matrices. Problems where data availability is limited are also considered, motivated by a face milling operation problem. Numerical examples are presented to indicate the accuracy and efficiency of the proposed method. Findings The paper presents a discretization for the PDEs model aiming on simple implementations and numerical performance. The modified version of LMM introduced using singular scaling matrices shows the capabilities on recovering quantities with precision at a low number of iterations. Numerical results showed good fit between exact and approximate solutions for synthetic noisy data and quite acceptable inverse solutions when experimental data are inverted. Originality/value The paper is significant because of the pseudospectral approach, known for its high precision and easy implementation, and usage of singular regularization matrices on LMM iterations, unlike classic implementations of the method, impacting positively on the reconstruction process.