Learning viscoplastic constitutive behavior from experiments: II. Dynamic indentation

This paper extends a previously developed inverse problem framework for identifying viscoplastic constitutive behavior from full-field observations to dynamic indentation scenarios by incorporating contact constraints via Lagrange multipliers and slack variables, and validates the method using both synthetic data and experiments on rolled homogeneous armor steel and polycrystalline aluminum alloy.

The Gist

Imagine you have a mysterious, super-strong piece of metal. You want to know exactly how it behaves when you push, pull, or smash it. In the engineering world, this behavior is described by something called a constitutive relation—a fancy math formula that acts like the material's "personality profile."

Usually, to find this profile, scientists have to guess the numbers in the formula, run a simulation, compare it to a real experiment, and then tweak the numbers again. It's like trying to tune a radio by turning the dial blindly until the static clears up.

This paper, the second in a series, introduces a much smarter way to do this tuning. Here is the story of how they did it, explained simply.

The Problem: The "Black Box" of Indentation

Think of a dynamic indentation test like a high-speed game of "poke the bear."

• The Setup: A hard, rigid ball (the indenter) is slammed into a piece of metal at high speed.
• The Measurement: We can easily measure two things: how deep the ball goes and how hard the metal pushes back (the force).
• The Mystery: We cannot see inside the metal. We can't directly measure the stress or strain happening deep inside the material. The metal is a "black box." We only see the input (the poke) and the output (the pushback).

The goal is to reverse-engineer the metal's internal personality (its constitutive law) just by looking at the poke and the pushback.

The Solution: The "Shadow Puppet" Detective

The authors treat this like a detective solving a crime. They have a suspect (the experimental data) and a theory (a computer model). They need to find the specific settings for their theory so that the "shadow puppet" it casts matches the real crime scene perfectly.

Here is how their method works, step-by-step:

1. The Forward Problem: The Simulation

First, they build a digital twin of the experiment. They guess a set of numbers for the metal's personality (how stiff it is, how hard it gets when squeezed, etc.) and run a simulation.

• Analogy: Imagine you are trying to guess the recipe for a cake. You guess the amount of sugar and flour, bake a virtual cake, and see how it looks.

2. The Adjoint Method: The "Time-Traveling" Feedback

This is the paper's superpower. Usually, if your virtual cake tastes wrong, you have to guess again and bake another one. That takes forever.
Instead, the authors use a mathematical trick called the Adjoint Method.

• Analogy: Imagine you have a time machine. You bake the cake, taste it, and then travel backwards in time to the moment you mixed the ingredients. The time machine instantly tells you exactly how much more sugar or less flour you needed to get the perfect taste.
• In physics terms, this "time-traveling" calculation tells the computer exactly how to tweak the material's personality parameters to reduce the error between the simulation and the real experiment, all in one go.

3. The Contact Problem: The "Bouncy Wall"

A major hurdle in this specific experiment is contact. When the indenter hits the metal, they can't pass through each other. In math, this is a "non-holonomic constraint" (a fancy way of saying "you can't go through the wall").

• The Fix: The authors introduced a "slack variable" and a "Lagrange multiplier."
• Analogy: Think of the metal surface as a trampoline. If you try to push through it, the trampoline pushes back. The math they developed ensures the indenter bounces off perfectly without sinking into the metal, even while the computer is doing its time-traveling calculations.

The Experiments: Synthetic vs. Real

The team tested their method in two ways:

1. The "Fake" Data (Synthetic)
They created a computer simulation where they knew the "true" personality of the metal. They then tried to recover those numbers using only the force and depth data.

• Result: It worked perfectly! Even when they started with terrible guesses, the "time-traveling" math found the right answer.
• Key Insight: They discovered that the tiny, rapid wiggles (fluctuations) in the force curve contain a treasure trove of information. If you smooth those wiggles out (like blurring a photo), you lose the ability to figure out the material's true nature.

2. The Real Data (Experiments)
They took the method to the real world, testing:

• RHA Steel: Used in military armor.
• Aluminum Alloy: A common, strong metal.
• Result: They successfully identified the stiffness, yield strength, and hardening properties of these real metals using just a few indentation tests. The results matched well with known data from other scientists.

Why This Matters

This paper is a big deal because it turns a slow, guess-and-check process into a fast, precise science.

• Efficiency: You don't need to run hundreds of physical tests. A few smart tests are enough.
• Versatility: It works for complex, fast-moving events (dynamic indentation) where things get messy and chaotic.
• Future Proof: The authors hint that in the next part of their series, they will use this same "time-traveling" math to train Artificial Intelligence (Neural Networks) to learn material laws automatically, without needing a human to guess the formula structure at all.

The Bottom Line

The authors built a mathematical "time machine" that lets engineers look at how a material reacts to a high-speed poke and instantly reverse-engineer its internal rules. It's like being able to look at a car's tire tracks and instantly know the exact engine specs and tire pressure of the car that made them. This allows for faster, more accurate design of everything from armor plating to airplane parts.

Technical Summary

1. Problem Statement

The paper addresses the challenge of identifying complex material constitutive relations (specifically elasto-viscoplastic behavior) from experimental observations. Unlike direct measurement, quantities like stress, strain, and strain rate cannot be measured directly in complex loading scenarios; they must be inferred from macroscopic data such as displacement and reaction forces.

The specific focus of this work (Part II) is dynamic indentation, a test where a rigid indenter strikes a material at high velocity. This presents a challenging indirect inverse problem because:

• The stress state is highly heterogeneous and time-dependent.
• The problem involves unilateral contact (a nonholonomic constraint) between the indenter and the material, which complicates the mathematical formulation.
• Traditional methods often struggle to extract quantitative constitutive parameters from such complex, dynamic fields without extensive prior assumptions.

2. Methodology

The authors formulate the identification of material parameters as a PDE-constrained optimization problem. The goal is to find a set of constitutive parameters that minimizes the difference between experimental observations and model predictions while strictly satisfying the laws of physics (balance laws).

A. Forward Problem Formulation

• Governing Equations: The material is modeled using small-strain $J_2$ plasticity with power-law isotropic hardening and power-law rate sensitivity. The equations include balance of linear momentum, yield conditions, and flow rules.
• Contact Handling: To address the frictionless contact between the rigid indenter and the deformable body, the authors introduce a Lagrange multiplier ($\lambda$) and a slack variable ($\ell$).
  • The contact condition $C \geq 0$ is enforced strictly by setting $C = \ell^2$.
  • This transforms the contact constraint into a differentiable form suitable for gradient-based optimization.
• Numerical Scheme: The forward problem is solved using a staggered predictor-corrector algorithm. Displacements are updated explicitly (central difference), while plastic variables are updated implicitly (backward Euler). Contact forces are corrected if penetration is detected.

B. Inverse Problem & Adjoint Method

• Objective Function: The objective is to minimize the $L_2$ norm of the difference between the experimental reaction force history and the simulated reaction force.
• Sensitivity Analysis: To efficiently compute the gradient of the objective function with respect to the material parameters, the authors employ the Adjoint Method.
  • They derive adjoint equations for displacement ($v$), accumulated plastic strain ($\gamma$), plastic strain tensor ($\zeta$), and the contact Lagrange multiplier ($\tau$).
  • These adjoint equations are solved backward in time from final conditions (zero) to the start of the simulation.
  • The sensitivity of the objective with respect to parameters is computed using the forward and adjoint solutions.
• Optimization: The parameters are updated using the Method of Moving Asymptotes (MMA), a gradient-based optimization algorithm.

3. Key Contributions

1. Rigorous Contact Formulation: The paper introduces a novel approach to handle dynamic contact in inverse problems by strictly enforcing the non-penetration constraint via Lagrange multipliers and slack variables, rather than using penalty methods which can introduce numerical stiffness or softening.
2. Adjoint Derivation for Dynamic Contact: The authors derive the specific adjoint equations required for dynamic indentation problems involving unilateral contact, extending the framework established in their previous work (Part I).
3. Robustness to Initial Guesses and Geometry: The method is demonstrated to be robust against poor initial parameter guesses and variations in indenter shape (spherical vs. ellipsoidal).
4. Data Efficiency: The study highlights that macroscopic fluctuations in the force-displacement curves (often ignored or smoothed out) contain sufficient information to accurately recover complex viscoplastic parameters, even without measuring local strain fields.

4. Results

A. Synthetic Data Validation

• Parameter Recovery: Using synthetic data generated from known "ground truth" parameters, the method successfully recovered the constitutive parameters ($\sigma_y, \varepsilon^p_0, n, \dot{\varepsilon}^p_0, m$) for a $J_2$ viscoplastic material.
• Convergence: The objective function decreased by over 4–6 orders of magnitude across 500 iterations.
• Uniqueness vs. Equivalence: While the recovered parameters did not always match the ground truth exactly (indicating non-uniqueness in the parameter space), the constitutive response (stress-strain curves) of the recovered model was nearly indistinguishable from the ground truth in independent uniaxial tension tests.
• Importance of Fluctuations: When synthetic data was linearized (smoothing out dynamic fluctuations), the parameter recovery failed to capture the true material behavior, proving that high-frequency dynamic responses are critical for identification.
• Shape Robustness: The method worked equally well for spherical and ellipsoidal indenters.

B. Experimental Data Application

The method was applied to dynamic indentation experiments on two materials:

1. Rolled Homogeneous Armor (RHA) Steel:
   • Recovered elastic modulus ($E \approx 140$ GPa) and plastic parameters.
   • The recovered model matched experimental force-displacement curves closely.
   • In independent uniaxial tension tests, the recovered model's response fell within the bounds of two different Johnson-Cook models found in literature.
   • Note: The recovered elastic modulus was lower than the standard literature value (200 GPa). The authors attribute this to the small elastic regime in indentation and the assumption of a perfectly rigid indenter (the actual tungsten carbide indenter deforms slightly).
2. Aluminum Alloy (Al 6061-T6):
   • Successfully recovered yield stress and hardening parameters.
   • The recovered model showed excellent agreement with literature Johnson-Cook models in independent uniaxial tension tests.

5. Significance and Future Work

• Generalizability: This approach provides a general framework for characterizing complex materials without needing to assume a specific functional form for the constitutive law (though this paper uses a parametrized form).
• Efficiency: By using the adjoint method, the computational cost of sensitivity analysis is independent of the number of parameters, making it scalable.
• Future Directions (Part III): The authors plan to extend this methodology to identify constitutive relations using neural networks (specifically recurrent neural operators) as the material model. This would allow for the discovery of the functional form of the constitutive law directly from data, rather than just fitting parameters to a pre-defined equation.
• Broader Impact: The method is applicable to a wide range of materials, including anisotropic HCP metals and potentially brittle materials (with extensions for fracture mechanics).

In summary, this paper demonstrates a powerful, gradient-based inverse framework that leverages dynamic indentation data and rigorous contact mechanics to accurately identify complex material constitutive behaviors, bridging the gap between macroscopic experimental observations and microscopic material properties.