Selected work

Representative Papers

Temperature Estimation in Induction Motors using Machine Learning

Mar 19, 2023Applied Power Electronics Conference

To address the challenge of real-time, accurate monitoring of stator winding and bearing temperatures in induction motors—critical for reliable thermal protection—this paper proposes a data-driven, multi-model comparative approach for temperature-rise prediction. Leveraging real-time, multi-source sensor data during operation, the method systematically evaluates the generalization performance of linear regression, support vector regression (SVR), random forest, and long short-term memory (LSTM) networks under transient operating conditions. Model robustness is enhanced via experimental calibration and Bayesian hyperparameter optimization. Crucially, this work first demonstrates LSTM’s superior capability in modeling non-stationary thermal dynamics: it achieves a mean absolute error ≤1.2°C—over 35% lower than conventional thermal models and shallow learning methods—while meeting millisecond-level response and thermal-protection accuracy requirements. The results establish a deployable, data-driven paradigm for online thermal-state awareness in electric drive systems.

2 citationsRead paper

Fault Detection in Electrical Distribution System using Autoencoders

Feb 16, 2026

This work addresses the challenges of fault detection in power distribution networks, including data scarcity, high randomness of fault parameters, and difficulties in deploying existing methods. The authors propose an unsupervised anomaly detection approach based on a convolutional autoencoder (CAE), leveraging a deep autoencoding architecture to achieve efficient dimensionality reduction and rapid training. Without requiring extensive labeled data, the method simultaneously accomplishes fault detection, classification, and localization. It significantly reduces model parameter count and training time while achieving detection accuracies of 97.62% on simulated data and 99.92% on a public dataset, outperforming current state-of-the-art techniques and demonstrating strong practicality and deployability.

1 citationsRead paper

Vessel Trajectory Prediction using COLREGs-aware Optimal Planning

Jul 17, 2026

This study addresses the challenge of simultaneously ensuring compliance with the International Regulations for Preventing Collisions at Sea (COLREGs) and achieving real-time performance in maritime vessel trajectory prediction. To this end, the authors propose an efficient rule-constrained prediction method that first employs A* search to generate a collision-free initial trajectory around static obstacles as a warm start, followed by numerical optimization that explicitly embeds COLREGs constraints. The approach requires only the current positions, velocities, and Automatic Identification System (AIS)-derived destination information of surrounding vessels, significantly reducing reliance on extensive historical trajectory data. Compared to conventional data-intensive methods, the proposed framework enhances both computational efficiency and regulatory compliance, demonstrating strong predictive accuracy and feasibility in realistic scenarios constructed from real-world AIS data.

0 citationsRead paper

ELMP: Efficient Learning for Motion Planning via Analytical Policy Gradients

Jun 30, 2026

Existing neural motion planners rely heavily on large amounts of expert demonstration data for adaptation to new environments, incurring substantial costs. This work proposes an efficient self-adaptation approach that directly optimizes the policy through a differentiable kinematics layer and self-supervised fine-tuning, thereby circumventing the need for expensive expert trajectory collection. The method explicitly encodes tool geometry using point-cloud inputs and replaces data-driven updates with analytical policy gradients. Evaluated in previously unseen environments, the approach improves task success rates from 57.3% to 89.8% (averaging 84.8%), reduces cold-start latency by two orders of magnitude compared to conventional methods, maintains millisecond-level inference times, and demonstrates empirical effectiveness on a Franka robotic platform.

0 citationsRead paper

Benchmarking Sensor-Fault Robustness in Forecasting

May 11, 2026

Existing predictive models lack robustness evaluation under sensor data faults such as noise, bias, missing values, or temporal misalignment, as standard benchmarks focus solely on nominal performance and fail to reflect real-world stability. This work proposes SensorFault-Bench, the first unified evaluation framework for sensor faults in cyber-physical systems, which enables fair comparison of zero-shot foundation models and diverse robustness methods through standardized fault injection, worst-case degradation metrics, and fault-timed mean squared error (MSE) that disentangles absolute error from relative robustness. Experiments reveal that models excelling on clean data can degrade significantly under faults—e.g., Chronos-2 performing worse than a naive predictor—and that different robustness approaches exhibit complementary strengths across value-type and availability-type fault scenarios, thereby validating the framework’s effectiveness and necessity.

0 citationsRead paper

Recent publications

Latest Papers

Vessel Trajectory Prediction using COLREGs-aware Optimal Planning

Jul 17, 2026

This study addresses the challenge of simultaneously ensuring compliance with the International Regulations for Preventing Collisions at Sea (COLREGs) and achieving real-time performance in maritime vessel trajectory prediction. To this end, the authors propose an efficient rule-constrained prediction method that first employs A* search to generate a collision-free initial trajectory around static obstacles as a warm start, followed by numerical optimization that explicitly embeds COLREGs constraints. The approach requires only the current positions, velocities, and Automatic Identification System (AIS)-derived destination information of surrounding vessels, significantly reducing reliance on extensive historical trajectory data. Compared to conventional data-intensive methods, the proposed framework enhances both computational efficiency and regulatory compliance, demonstrating strong predictive accuracy and feasibility in realistic scenarios constructed from real-world AIS data.

0 citationsRead paper

ELMP: Efficient Learning for Motion Planning via Analytical Policy Gradients

Jun 30, 2026

Existing neural motion planners rely heavily on large amounts of expert demonstration data for adaptation to new environments, incurring substantial costs. This work proposes an efficient self-adaptation approach that directly optimizes the policy through a differentiable kinematics layer and self-supervised fine-tuning, thereby circumventing the need for expensive expert trajectory collection. The method explicitly encodes tool geometry using point-cloud inputs and replaces data-driven updates with analytical policy gradients. Evaluated in previously unseen environments, the approach improves task success rates from 57.3% to 89.8% (averaging 84.8%), reduces cold-start latency by two orders of magnitude compared to conventional methods, maintains millisecond-level inference times, and demonstrates empirical effectiveness on a Franka robotic platform.

0 citationsRead paper

Benchmarking Sensor-Fault Robustness in Forecasting

May 11, 2026

Existing predictive models lack robustness evaluation under sensor data faults such as noise, bias, missing values, or temporal misalignment, as standard benchmarks focus solely on nominal performance and fail to reflect real-world stability. This work proposes SensorFault-Bench, the first unified evaluation framework for sensor faults in cyber-physical systems, which enables fair comparison of zero-shot foundation models and diverse robustness methods through standardized fault injection, worst-case degradation metrics, and fault-timed mean squared error (MSE) that disentangles absolute error from relative robustness. Experiments reveal that models excelling on clean data can degrade significantly under faults—e.g., Chronos-2 performing worse than a naive predictor—and that different robustness approaches exhibit complementary strengths across value-type and availability-type fault scenarios, thereby validating the framework’s effectiveness and necessity.

0 citationsRead paper

Safe Exploration for Nonlinear Processes Using Online Gaussian Process Learning

May 10, 2026

This work proposes a safe, data-driven control method for nonlinear systems with only partially known dynamics, ensuring both stability and satisfaction of state and input constraints during online learning. Built upon a stabilizable linear approximation, the approach employs an online Gaussian process to estimate the unmodeled nonlinear residuals in real time. By integrating Lyapunov theory, it constructs a probabilistic control-invariant set with finite-sample safety guarantees. A convex quadratic program is solved at each step to maximize information gain while preserving safety, and the safe region is adaptively expanded as uncertainty diminishes. Numerical experiments demonstrate that the method enlarges the feasible safe set by approximately 30% under safe exploration, while reducing the root mean square error of the Gaussian process predictions from 1.11 to 0.03.

0 citationsRead paper

Unsharp Measurement with Adaptive Gaussian POVMs for Quantum-Inspired Image Processing

Apr 06, 2026

This work proposes a novel quantum-inspired approach to image processing by introducing unsharp quantum measurements and adaptive Gaussian positive operator-valued measures (POVMs). Unlike conventional methods that rely on segmentation or thresholding and struggle to balance structural preservation with adaptive probabilistic transformation, the proposed framework embeds pixel intensities into a finite-dimensional Hilbert space. It constructs data-adaptive POVM operators based on a histogram Gaussian mixture model and incorporates a nonlinear sharpening parameter γ along with the number of Gaussian centers k to modulate the locality and resolution of the measurement process, thereby enabling a continuous transition from probabilistic smoothing to projective measurement. Experimental results on standard test images demonstrate that the method effectively achieves a balance between preserving structural information and performing adaptive grayscale transformation.

0 citationsRead paper