An Anti-Singularity Data-Driven-Control-Based Method for Soft Start and Multi-Machine Balance of Magnetic Couplers

Yuqin Zhu, Wei Liu, Hongju Zhao, Chuang Yang, Hao LiuMachines 2026 · DOI

TL;DR

This paper introduces a novel "anti-singularity data-driven control" method to tackle the challenges of soft starting and power balancing in magnetic couplers, common in heavy industrial applications like belt conveyors. By combining Lyapunov stability theory with advanced data processing techniques, the method achieves precise speed control, smooth starts, and efficient multi-motor coordination, outperforming traditional PI control.

What Problem Does This Paper Address?

Magnetic couplers are crucial for transmitting torque and regulating speed in various industrial systems, offering advantages like high efficiency and robustness. However, controlling them, especially for tasks like soft starting heavy loads or balancing power across multiple motors, is difficult. The core issues stem from:

  • Complex Dynamics: Magnetic couplers exhibit non-monotonic torque-speed characteristics, making it hard to model them accurately, especially under constant-torque loads.
  • Singular Noise: Real-world data from these systems often contains "singular noise" – outliers that can severely disrupt control algorithms.
  • Soft Start Challenges: Directly starting heavy loads can cause mechanical shock and motor burnout. Achieving a smooth, S-curve start-up is essential but difficult with traditional controllers.
  • Multi-Motor Imbalance: In systems with multiple motors driving a single load (like long belt conveyors), uneven power distribution leads to reduced efficiency and increased wear.

Existing control methods like PID, adaptive control, and fuzzy logic struggle with these combined challenges, necessitating a more robust and data-informed approach.

Methods

The paper proposes an "anti-singularity data-driven control" strategy that integrates three key innovations:

  1. Singularity Screening and Limiting (SVSL): This mechanism preprocesses raw data to identify and remove outliers (singular points) that deviate significantly from the local data trend. This enhances data quality and robustness.
  2. State-Space Memory Model with Generalized Distance Weighting: Instead of relying on a precise mechanistic model, this approach uses historical operational data. A "generalized distance weighting factor" is employed to extrapolate the estimated angular acceleration across the entire operating plane, effectively creating a data-driven model that captures nonlinear dynamics.
  3. Lyapunov Stability-Based Control Law: The control law for air-gap adjustment is derived using Lyapunov stability theory, theoretically guaranteeing exponential stability of the closed-loop system. This ensures predictable and reliable performance.

These components are integrated into a unified framework that allows for real-time estimation of system dynamics and computation of optimal air-gap control variables.

Datasets

The paper does not explicitly mention training datasets. Instead, it utilizes historical operational data from the magnetic coupler itself for constructing the data model. Experimental validation was performed on a dedicated test platform.

Key Insights and Ideas

  • Data Quality is Paramount: The SVSL mechanism is crucial for making data-driven control viable in noisy industrial environments. Simply using raw data would lead to inaccurate models and poor control.
  • Extrapolation is Key for Model Coverage: The generalized distance weighting factor is vital for extending the learned dynamics from observed sample points to the entire operating space, creating a comprehensive data model.
  • Lyapunov Guarantees Stability: Tying the data-driven model to a Lyapunov-based control law provides theoretical assurance of system stability, which is critical for industrial applications.
  • "Anti-Singularity" is a Holistic Approach: The method's strength lies in the synergistic combination of data preprocessing, robust data modeling, and theoretically grounded control design.

Key Results

  • Speed Control Accuracy: Experimental results demonstrated a speed control accuracy within 4%.
  • Soft Start Performance: For long-distance belt conveyors, the strategy achieved a soft-start acceleration of less than 0.15 m/s² with a start-up time of approximately 25 seconds.
  • Multi-Motor Power Balance: The system maintained power balance among motors within a 5% deviation.
  • Energy Efficiency: The method improved energy efficiency by 17.6%, translating to significant annual electricity savings.
  • Comparison to PI Control: The proposed method showed significantly better control accuracy and stability compared to conventional PI control, with a lower Mean Squared Error (MSE) and fewer instances of maximum tracking error.

Limitations and Future Work

The authors acknowledge that the system's performance relies on the accuracy of the data model. Future work will focus on:

  • Enhancing Online Adaptability: Developing mechanisms for the data model to dynamically incorporate new, high-quality data points during operation.
  • Safe and Efficient Online Learning: Investigating learning algorithms that can adapt to slow drifts in equipment characteristics or explore optimal operating points, enabling "lifelong learning."
  • Exploring More Optimal Operating Points: Continuously refining the model to achieve even better performance and efficiency.

Summary updated

Abstract

The speed-regulating magnetic coupler faces challenges in mechanism modeling for constant-torque-load soft start and multi-motor power balance. Moreover, system data contain singular noise values. To address these issues, an anti-singularity data-driven control strategy is proposed. This strategy enhances speed control accuracy and operational robustness. The method designs an exponentially stable air-gap regulation law using Lyapunov theory. A robust data model is constructed by estimating angular acceleration via piecewise least squares with singularity screening, followed by model extension using a generalized distance weighting factor, which enables the numerical solution of the control law. Experimental results demonstrate a speed control accuracy within 4%. In practical applications on long-distance belt conveyors, the strategy achieves a soft-start acceleration of ≤0.15 m/s2 with a 25 s start-up time, maintains power balance among motors within a 5% deviation, and improves energy efficiency by 17.6%. This work provides an effective data-driven solution for the high-performance control of magnetic couplers in complex industrial scenarios.
BibTeX
@inproceedings{zhu2026anti,
  author = {Yuqin Zhu and Wei Liu and Hongju Zhao and Chuang Yang and Hao Liu},
  title = {{An Anti-Singularity Data-Driven-Control-Based Method for Soft Start and Multi-Machine Balance of Magnetic Couplers}},
  booktitle = {Machines},
  year = {2026},
  doi = {10.3390/machines14030263},
}
Figure