About

Lei Zuo is a Professor in Mechanical Engineering at the University of Michigan, with a joint appointment in Naval Architecture and Marine Engineering. His research interests encompass marine renewable energy sources such as ocean waves, tidal currents, and offshore wind, with a focus on the blue economy, energy harvesting, vibration and dynamics, control, mechatronics design, vehicles and transportation, and advanced manufacturing. He is involved in projects that aim to advance AI-powered materials discovery and human-centered smart manufacturing, as well as improving testing, durability, and noise levels of wave energy devices and offshore wind technologies to enhance their reliability and environmental compatibility. His work contributes to the development of sustainable energy solutions and innovative manufacturing processes, with a particular emphasis on marine renewable energy and its integration into the broader energy landscape.

Research topics

• Materials science
• Computer science
• Engineering
• Acoustics
• Electrical engineering

Selected publications

• Surrogate Boundary Element Modeling for Frequency-Domain Wave–Structure Interaction

  SSRN Electronic Journal · 2026-01-01

  preprintOpen accessSenior author
  PublisherDOI

• System Modeling and Power Optimization of a Point Absorber Wave Energy Converter

  Proceedings of the ... European Wave and Tidal Energy Conference · 2025-09-08

  articleSenior author

  A conventional approach to evaluating the performance of a wave energy converter (WEC), including point absorbers, is to represent the power dissipated in the electrical domain using an equivalent electrical damping coefficient. However, realistic WECs are complex systems comprised of various constituent components: (i) a wave energy capture structure, such as a single- or two-body buoy, (ii) a mechanical energy transmission mechanism, usually referred to as a gear system, and (iii) an electromagnetic generator for converting kinetic energy into electricity. Here, we show that utilizing an electrical damping constant is not sufficient to fully capture the dynamics of these systems, and optimizing output power based on this metric as an independent variable does not yield the true maximum. In general, the design of a WEC involves sophisticated processes, which have traditionally been iterative and sequential, focusing separately on individual objectives. This approach often leads to suboptimal solutions because it neglects the coupling effects and interdependencies among subsystems. In contrast, an engineering (control) co-design method simultaneously considers the dynamics of the entire system to maximize overall energy conversion, ensuring that design rules for one subsystem align with the objectives of others, thereby enhancing system performance. While several studies have explored various aspects of co-design in WECs, comprehensive methodologies, and generalized principles remain limited in the literature. This limitation comes from the focus of these studies on specific numerical examples of wave converters, which might have been selected differently and could potentially lead to different conclusions. In light of these challenges, this paper aims to develop a comprehensive model that captures the physical properties (e.g., equivalent mass, mechanical damping coefficient, generator inductance and resistance) and the coupled interactions of all subsystems, paving the way for analytically investigating and maximizing overall system performance. Addressing this challenge is critical for ensuring efficient energy harnessing in real-world applications and is the primary objective of the paper. In addition, the findings can provide a general and robust framework for optimizing the generated power of WECs, significantly aiding the co-design process. In this paper, we focus on the dynamics of a point absorber WEC driven by regular wave excitation. We analytically determine the maximum possible power that can be harvested for a specific geometry of a single-body buoy structure. Furthermore, we demonstrate that these geometric dimensions and other system components can be optimized independently to maximize output power under given ocean wave conditions. The upper bound of power is expressed as a function of an effective figure of merit for the WEC, combining electromagnetic transducer coupling and parasitic losses. Our findings indicate that for complex systems with multiple energy conversion stages, such as WECs, the gradient descent method is more suitable for optimizing objective variables compared to the impedance matching principle. In such cases, although the latter is widely used in the literature, it does not yield the global maximum power delivered to the load. We derive analytical solutions for system parameters, including the complex load, mechanical transmission, and generator moment of inertia.

  PublisherDOI

• Analysis of feed fluctuations on Ocean Wave-Powered Reverse Osmosis (WPRO) desalination using transient model

  Desalination · 2025-12-10

  articleSenior authorCorresponding
  PublisherDOI

• High-resolution Fourier single-pixel SWIR imaging at extremely low sampling rate via learning-based approach

  2025-07-29

  article

  Fourier Single-pixel Imaging (FSI) reconstructs target scenes by directly measuring Fourier coefficients using a single-pixel detector, offering practical physical significance as well as superior sparsity and energy concentration capabilities. However, the reconstruction quality of existing methods remains unsatisfactory under extremely low sampling rates. This paper proposes a deep learning-based Fourier single-pixel imaging method for extremely low sampling rates, termed Fourier Deep Residual Attention Network (FDRANet). The method employs an autoencoder network constrained by initial masks that conform to the Fourier spectrum distribution of grayscale images, and integrates residual neural networks (ResNet) with channel attention mechanisms in parallel. It simultaneously optimizes the sampling masks and the neural network, thereby achieving high-quality image reconstruction at extremely low sampling rates. Additionally, we constructed a passive short-wave near-infrared single-pixel imaging system to validate the effectiveness of the proposed method (FDRANet). Extensive experiments on both simulated and real-world data demonstrate that our method achieves state-of-the-art performance.

  PublisherDOI

• Evaluating adversarial robustness of single-pixel imaging models

  2025-07-29 · 1 citations

  article

  Single-pixel imaging (SPI), an emerging computational imaging technique, uses a single-pixel detector to obtain low-dimensional measurements and solve ill-posed inverse problems for image reconstruction. As this technology is widely used in security-sensitive fields like remote sensing and biomedical imaging, its adversarial robustness has become crucial for system reliability. Yet, there is a lack of systematic research on this issue. This study comprehensively evaluates the vulnerability of deep learning-based SPI reconstruction models under adversarial attacks. By analyzing the optimization objectives of SPI reconstruction, three novel first-order gradient-based attack methods are proposed: full-image perturbation, center-mask attack, and universal adversarial attack. Comparative experiments on the Set11 and Urban100 datasets show that: 1) Existing SPI methods are highly sensitive to ι_∞ adversarial perturbations; 2) Patch-based reconstruction models are more vulnerable than full-image ones; and 3) There is a nonlinear relationship between model reconstruction performance and robustness. This research reveals potential security risks in current SPI models through a systematic attack evaluation framework and offers new ideas for developing more robust reconstruction algorithms.

  PublisherDOI

• Positionally restricted masked knowledge graph completion via multi-head mutual attention

  Journal of Information and Intelligence · 2025-03-23

  articleOpen accessSenior authorCorresponding

  Knowledge graph completion aims to enhance the completeness of knowledge graphs by predicting missing links. Link prediction is a common approach for this task, but existing methods, particularly those based on similarity computation, are often computationally expensive, especially for large models. To address this, we propose a novel method, positionally restricted masked knowledge graph completion (PR-MKGC), which reduces inference time by leveraging masked prediction and relying solely on structural information from the knowledge graph, without using textual data. We introduce a multi-head mutual attention mechanism that aggregates neighbor information more effectively, improving the model's ability to predict missing links. Experimental results demonstrate that PR-MKGC outperforms existing models in terms of both predictive performance and inference time on the FB15K-237 dataset.

  PublisherDOI

• Suppression mechanism of vortex-induced vibrations using non-linear energy sink with inerter based mechanical networks

  Chaos Solitons & Fractals · 2025-09-30 · 3 citations

  articleSenior author
  PublisherDOI

• Melt pool dynamics and microstructural growth prediction of additively manufactured thermoelectric material

  CIRP journal of manufacturing science and technology · 2025-04-04

  articleSenior authorCorresponding
  PublisherDOI

• Understanding the adversarial robustness of deep learning-based single-pixel imaging

  Pattern Recognition · 2025-10-10 · 2 citations

  article
  PublisherDOI

• Extending the statistical linearization method to multi-variate non-differentiable nonlinearities in floating renewable energy devices

  Renewable Energy · 2025-07-16 · 2 citations

  articleOpen access

  This article investigates the methodology and applicability of the statistical linearization (SL) method to incorporating multi-variate non-differentiable nonlinearities, with a focus on floating renewable energy devices. The SL method serves as a highly competitive approach for analyzing floating renewable energy structures, such as wave energy converters (WECs) and floating wind energy turbines, because it inherently combines adequate accuracy and high computational efficiency. The origin of high accuracy comes from its incorporation of nonlinear effects through statistically linearized representations. Yet, the statistically linearized solutions have only been derived and verified for a limited number of nonlinearities of floating renewable energy devices, mostly simply-formed and differentiable in their mathematical expressions. However, floating renewable energy devices usually exhibit a complex dynamic mechanism, in which the relevant nonlinear effects could appear to be highly complex for linearization process to describe. These nonlinear effects could make a significant impact on the system dynamics, exemplified by external machinery force saturation and nonlinear hydrostatics of floaters with a non-uniform geometry. To push forward the boundary of the SL method, it is crucial to demonstrate how it applies to nonlinearities of different features. In this paper, the existing SL method is extended to address the nonlinear effects expressed as multi-variate non-differentiable functions. Several case studies are carried out to exemplify the application of the extended SL approach to the concerned nonlinearities in floating renewable energy devices. The accuracy and computational efficiency of the extended SL approach are evaluated by verifying against the corresponding nonlinear time-domain (TD) and linear frequency-domain (FD) models. Despite the complexity of the given nonlinearities, the relative errors of the SL approach are no more than 6 % while its computational time is comparable to the FD model, being thousands of times faster than the TD model. Comparatively, the FD model leads to a relative error of over 70% in some cases.

  PublisherDOI

Recent grants

• Vibration Control of Tall Buildings Using Electricity Generating Tuned Mass Dampers

  NSF · $195k · 2014–2015

• GOALI: Energy Efficient and Reliable Motion Mechanism for Ocean Wave Energy Harvesting

  NSF · $314k · 2014–2018

• Collaborative Research: GOALI: Bio-inspired bistable energy harvesting for fish telemetry tags

  NSF · $231k · 2022–2025

• LEAP-HI: US-Ireland R&D Partnership: Control Co-Design for Ocean Wave Energy Conversion

  NSF · $1.5M · 2022–2026

• Collaborative Research: Net-Shape and Scalable Additive Manufacturing for Thermoelectric Waste Heat Recovery Materials and Devices using Selective Laser Melting

  NSF · $264k · 2019–2022

Frequent coauthors

• Shuyu Wang

  Northeastern University

  57 shared

• Shifeng Yu

  Columbia University

  53 shared

• Ming Lu

  Nanchang University

  48 shared

• Michael Siedler

  AbbVie (Germany)

  27 shared

• Peter M. Ihnat

  Regeneron (United States)

  27 shared

• Dana I. Filoti

  AbbVie (United States)

  27 shared

• Jia Mi

  University of Michigan–Ann Arbor

  26 shared

• Xiudong Tang

  Stony Brook University

  25 shared

Education

• PhD, Mechanical Engineering

  Massachusetts Institute of Technology

  2004

• BS, Automotive Engineering

  Tsinghua University

  1997