PhD Thesis Colloquium: Mr. Suman Sadhu (16/03/26)
Thesis title:
Multicomponent Diffusion Analysis: A New Method, Novel Experimental Strategy, and Physics-Informed Neural Network-Based Numerical Inverse Modelling
Faculty advisor(s):
Prof. Aloke Paul
When?
16th March, 2026 (Monday), 11:00 AM (India Standard Time)
Where
KPA Auditorium, Department of Materials Engineering
Abstract:
Historically, quantitative multicomponent diffusion analysis was considered impossible due to the complex, serpentine nature of diffusion paths, which could not be intersected in multicomponent space to estimate diffusion coefficients. The use of radioactive tracers, although capable of providing component-specific diffusion data, is also limited by the restricted availability of suitable isotopes and the requirement for specialized laboratory facilities to ensure safe handling. Recent developments, including the pseudo-binary and pseudo-ternary approaches, together with the concept of intersecting dissimilar diffusion paths, enable the estimation of diffusion coefficients for all components in a multicomponent system using only two diffusion paths. However, the successful formation of the required constrained diffusion couples (pseudo-binary or pseudo-ternary) in complex systems can be challenging, requiring strict discipline to produce such couples that fulfil thermodynamic constraints. Moreover, the effect of non-ideal behaviour in such diffusion couples has not yet been systematically standardized. Furthermore, existing numerical inverse methods for continuous diffusivity mapping rely on composition profile fitting, which often yield ill-posed, non-unique solutions.
This thesis addresses these challenges by developing a systematic experimental–theoretical–computational framework for multicomponent diffusion analysis. First, the pseudo-binary (PB) diffusion couple approach was critically examined. A re-adjustment factor (RAF) is introduced to account for the effect of the non-ideality range on estimated diffusion coefficients. This is demonstrated through several diffusion couple experiments across multiple systems of technological importance, producing a range of ideal, near-ideal, minor, and major non-ideal pseudo-binary diffusion couples. RAF values provide a guideline for pseudo-binary diffusion analysis in multicomponent systems, depending on the extent of non-ideality and composition range of the main diffusing elements.
Following, a strategic diffusion-couple design methodology is further developed and demonstrated in the Fe–Ni–Co–Cr quaternary system for estimating diffusion coefficients at multiple compositions using different types of diffusion profiles from a limited number of alloy end members. Using only eight carefully selected alloy compositions, this approach enables the generation of binary, ternary, pseudo-binary, pseudo-ternary, and fully multicomponent diffusion couples within a single, systematically designed experimental framework. The proposed framework enables systematic mapping of diffusion coefficients across the multidimensional composition space.
A central advancement of this thesis is the development of a novel experimental strategy that overcomes the limitations of intersection-based diffusion analysis. For the first time, it is demonstrated that all types of diffusion coefficients—tracer, intrinsic, and interdiffusion—can be estimated from a single diffusion profile in multicomponent systems. This significantly reduces experimental burden and complexity of preparing multiple diffusion couples. The thesis further shows that interdiffusion coefficients alone are often insufficient and can be misleading in complex alloys and establishes tracer diffusion coefficients as the most fundamental descriptors of atomic mobility.
To extend diffusion analysis beyond discrete experimental compositions, the thesis integrates constraint-enhanced numerical inverse modelling using physics-informed neural networks (PINNs) to extract composition-dependent tracer diffusivities over high-dimensional composition spaces. This integrated approach effectively addresses the challenges of parameter dimensionality and non-uniqueness, providing a scalable, physically consistent methodology for constructing reliable mobility databases in high-entropy, complex engineering alloys. These are demonstrated in Ni-Co-Fe and Ni-Co-Fe-Cr-Mn as model systems before extending to the other technologically important material systems.
Overall, the thesis offers a unified and reliable pathway for advancing quantitative diffusion modelling and mobility database development in technologically relevant multicomponent alloy systems.