Explainable Artificial Intelligence for Mechanics: Physics-Explaining Neural Networks for Constitutive Models

Abstract

(Artificial) neural networks have become increasingly popular in mechanics and materials sciences to accelerate computations with model order reduction techniques and as universal models for a wide variety of materials. However, the major disadvantage of neural networks remains: their numerous parameters are challenging to interpret and explain. Thus, neural networks are often labeled as black boxes, and their results often elude human interpretation. The new and active field of physics-informed neural networks attempts to mitigate this disadvantage by designing deep neural networks on the basis of mechanical knowledge. By using this a priori knowledge, deeper and more complex neural networks became feasible, since the mechanical assumptions can be explained. However, the internal reasoning and explanation of neural network parameters remain mysterious. Complementary to the physics-informed approach, we propose a first step towards a physics-explaining approach, which interprets neural networks trained on mechanical data a posteriori. This proof-of-concept explainable artificial intelligence approach aims at elucidating the black box of neural networks and their high-dimensional representations. Therein, the principal component analysis decorrelates the distributed representations in cell states of RNNs and allows the comparison to known and fundamental functions. The novel approach is supported by a systematic hyperparameter search strategy that identifies the best neural network architectures and training parameters. The findings of three case studies on fundamental constitutive models (hyperelasticity, elastoplasticity, and viscoelasticity) imply that the proposed strategy can help identify numerical and analytical closed-form solutions to characterize new materials.

Data‐Driven Virtual Material Analysis and Synthesis for Solid Electrolyte Interphases

“…The training data and model architecture are stored, handled, and processed within the Kadi4Mat ecosystem. KadiStudio workflows [42] and KadiAI's machine learning workflow [44,45] achieve reproducibility and track data provenance.…”

A stepwise physics‐informed neural network for solving large deformation problems of hypoelastic materials

“…The loss function includes three terms: the first term represents the residual between exact solution ũ(x) and network prediction u(x); the last two terms represent the norm values of the PDE and boundary condition in Equation (13). When the loss function value approaches to zero, the prediction of the network will be close enough to the solution of PDE.…”

“…PINN is a typical physics‐guided data‐driven method 12 and inherits the advantages of successful training on small or noisy data sets 10 . With the introduction of PINN, we can gradually use the well‐known physical laws to partially explain and understand the internal working mechanism of ANN 13 and turn this “black box” into a “gray box.”…”

Identifying dominant industrial sectors in market states of the S&P 500 financial data

“…Most importantly, this methodology can allow us to perform a dimensionality reduction by identifying the most important sector correlations without any qualitative ambiguity. XAI is nowadays considered for applications in physics [14,22,37] and finance alike [49] and in particular, financial regulators have already embraced the methods of XAI to increase…”