Eighty Years of the Finite Element Method: Birth, Evolution, and Future

Wing Kam Liu, Shaofan Li, Harold ParkarXiv 2021 · arXiv

Abstract

This paper provides a comprehensive 80-year retrospective on the Finite Element Method (FEM), tracing its journey from early structural "lattice" models to its current transformation into a data-driven powerhouse. We categorize the evolution of FEM into four distinct eras, highlighting the shift from rigorous physics-based modeling to the integration of machine learning and reduced-order modeling. This "New Era" aims to solve high-dimensional physics problems that were previously computationally untouchable, paving the way for real-time structural health monitoring and predictive medicine.

Introduction

The Finite Element Method (FEM) serves as the "computational workhorse" of modern engineering, enabling the simulation of complex phenomena ranging from automotive crashworthiness to cardiovascular mechanics. Despite its success, traditional FEM faces significant bottlenecks as engineering moves toward nanotechnology and real-time decision-making. The computational expense of multi-scale systems often precludes real-time application. This paper documents the historical evolution of FEM to provide context for its current challenge: integrating 80 years of rigorous physics-based modeling with the explosive growth of data science and artificial intelligence.

The Evolution of FEM: Four Eras

The development of FEM can be divided into four transformative periods:

1. The Birth (1941–1965)

The foundations of FEM were laid through a transition from physical "lattice" frameworks, championed by Alexander Hrennikoff, to mathematical variational principles. Richard Courant's work was instrumental in establishing these mathematical underpinnings, bridging the gap between pure mathematics and engineering application.

Richard Courant's 1943 paper established the mathematical variational principles that underpin modern FEM.

The era culminated in the coining of the term "Finite Element Method" by Ray Clough. Clough's practical applications, such as the 1962 analysis of the Norfork Dam, provided the method with the engineering credibility required for global adoption.

Ray William Clough (1920-2016) coined the term 'Finite Element Method' and applied it to landmark structures like the Norfork Dam.

2. The Golden Age (1966–1991)

This period saw the development of rigorous mathematical foundations, such as the LBB condition, and the introduction of fluid-structure interaction (ALE). Explicit time integration revolutionized automotive safety simulations and crashworthiness during this time.

3. Industrial & Materials Modeling (1992–2017)

The rise of "smart" elements characterized this era. Key innovations included X-FEM for modeling cracks without remeshing and Isogeometric Analysis (IGA), which directly linked Computer-Aided Design (CAD) with FEM solvers.

4. The New Era (2018–Present)

We are currently witnessing the integration of Deep Learning and Mechanistic Data Science (MDS). Novel approaches like Hierarchical Deep-learning Neural Networks (HiDeNN) and Physics-Informed Neural Networks (PINNs) embed physical constraints directly into neural network architectures, treating NNs as the ultimate interpolants for physical systems.

Key Insights and Methodological Shifts

The paper credits the "Musketeers" of FEM—Ray Clough, J.H. Argyris, and O.C. Zienkiewicz—for standardizing the method globally. A notable historical curiosity is the era of "Variational Crimes," where engineers used non-conforming elements that violated mathematical principles but yielded superior results, eventually leading to the development of the "patch test" for convergence.

Today, the focus has shifted toward efficiency. Methods like FE-SCA (Self-consistent Clustering Analysis) have reduced microscale analysis costs by multiple orders of magnitude. Furthermore, Reduced Order Modeling (ROM) techniques now allow for simulation response times in the millisecond range, enabling active control of complex engineering systems.

Future Frontiers and Limitations

While the merger of FEM and AI offers unprecedented speed, challenges remain regarding the "black box" nature of machine learning. Ensuring that models remain physically consistent and explainable is a primary concern. Additionally, the dependency on high-quality "offline" data remains a hurdle for novel materials or extreme environments.

The next frontiers for FEM-ML hybrids include Predictive Medicine, providing patient-specific diagnoses, and Digital Twins, which will allow for the real-time monitoring and management of critical infrastructure.

Summary updated

BibTeX
@misc{liu2021eighty,
  author = {Wing Kam Liu and Shaofan Li and Harold Park},
  title = {{Eighty Years of the Finite Element Method: Birth, Evolution, and Future}},
  year = {2021},
  eprint = {2107.04960},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2107.04960},
}
Figure