A groundbreaking advancement in nano-positioning technology has been unveiled by a team of researchers from Huazhong University of Science and Technology and the University of Victoria. Their study introduces a pioneering neural-network-based switching output regulation controller (NN-SORC) specifically designed to enhance the precision and responsiveness of high-speed nano-positioning stages. This innovation confronts one of the long-standing challenges in piezoelectric actuation systems—hysteresis nonlinearity—which has historically reduced the accuracy and stability of nanoscale motion control under dynamically changing reference signals.
The newly proposed NN-SORC marks a significant shift in how nonlinear hysteresis in piezoelectric materials is managed. Piezoelectric actuators, fundamental to the precision movement of nanoscale devices, exhibit complex nonlinear behaviors that are difficult to model and compensate with traditional control approaches. To address this, the researchers leveraged the adaptive capabilities of neural networks, enabling the control system to dynamically tune its parameters in real time, effectively counteracting hysteresis and ensuring superior tracking performance even when subjected to switching reference trajectories.
At the heart of this research lies an elegantly designed mechatronic platform that forms the experimental foundation for evaluating the controller’s performance. The nano-positioning stage developed features multiple high-fidelity thin piezoelectric ceramic layers bonded in parallel, symmetrically driven to promote uniform actuation and reduce mechanical asymmetries. Precision capacitive displacement sensors closely monitor stage position, while dedicated voltage amplifiers regulate the piezoelectric stack, allowing real-time feedback essential for high-bandwidth control loops. This setup replicates demanding operational environments typical in advanced manufacturing and precision detection, where nanometric accuracy is non-negotiable.
To tackle computational bottlenecks inherent in high-speed control systems, the team engineered an innovative dual-layer data processing architecture combining a field-programmable gate array (FPGA) with a central processing unit (CPU). This architecture delegates the FPGA to perform ultra-fast signal conversions and execute control algorithms at sampling frequencies nearing 10 MHz for the inner loop, a remarkable feat that facilitates rapid real-time responses. Meanwhile, the CPU oversees parameter optimization and system status, effectively orchestrating the entire control operation at a coarser sampling rate of 100 kHz. This FPGA–CPU symbiosis optimizes both speed and flexibility, overcoming limitations commonly associated with floating-point computations and compilation delays.
From a control theory perspective, the researchers employed a sophisticated feedback linearization technique to transform the inherent nonlinear hysteresis dynamics into a switched linear tracking error framework. This transformation allows the development of the NN-SORC as an adaptive closed-loop controller, capable of seamlessly regulating Nano-positioning behavior in response to switching reference signals. Importantly, the team applied rigorous stability analyses using Lyapunov theory and average dwell-time concepts to derive sufficient conditions guaranteeing the asymptotic stability of the entire closed-loop system when subjected to valid switching inputs.
Notably, their stability framework extends beyond ideal continuous references, accommodating piecewise continuous signals that may lack smooth second derivatives—situations commonly encountered in practical control scenarios with abrupt setpoint changes. By introducing a minimum dwell-time criterion that constrains the frequency of reference jumps, the controller ensures that transient oscillations are mitigated and stable tracking behavior is consistently maintained, contributing robustness to the system’s overall performance.
Experimental validation of the NN-SORC was conducted on the meticulously constructed test bench, which offers a stroke capability of approximately 10 micrometers and a frequency bandwidth enabling response up to 140 Hz. The NN-SORC’s performance was benchmarked against both a finely tuned traditional proportional–integral–derivative (PID) controller and a Prandtl–Ishlinskii inverse hysteresis compensation scheme. The tests employed both frequency-switching cosinusoidal and triangular reference signals, representative of operational conditions in micro- and nano-scale positioning tasks. Results vividly demonstrated that the NN-SORC consistently outperformed conventional methods, achieving notably lower tracking errors across the examined frequency spectrum.
The NN-SORC’s capability to maintain stable tracking during rapid and complex switching transients was particularly significant. While conventional PID controllers often exhibited oscillations or overshoot during reference setpoint changes, the neural network-based controller adapted swiftly, dynamically recalibrating its parameters to minimize switching-induced error. The experimental data corroborated the theoretical findings related to dwell-time criteria, establishing a robust connection between theoretical design and practical realizability.
This research synergistically combines a high-speed hardware platform, advanced machine learning control techniques, and switched-system stability theory into a cohesive, real-time implementation framework. The holistic approach not only addresses persistent nonlinear control challenges but also paves the way for practical adoption of intelligent control systems in precision engineering fields. Applications for such technology are far-reaching, encompassing scientific instrumentation, semiconductor manufacturing, and other domains requiring nanoscale positioning accuracy with high operational bandwidth.
Looking forward, the research team plans to extend their approach to more complex multi-axis nano-positioning stages, where coupled dynamics and cross-axis interference present additional challenges. By integrating dual-axis or multi-axis decoupling methodologies with the NN-SORC framework, future developments aim to further enhance control fidelity and robustness in spatially intricate positioning tasks. Such enhancements are expected to provide transformative improvements in next-generation micro- and nano-fabrication processes, bolstering productivity and product quality.
The significance of this work lies not only in its practical contributions but also in its theoretical rigor, blending control theory with artificial intelligence to manage nonlinearities and switching complexities effectively. The devised controller stands as a template for future research on integrating machine learning algorithms with traditional control methodologies, signaling a new era for precision mechatronic system design.
With its open-access publication in the journal Engineering, this study invites the broader scientific and engineering communities to engage with and build upon its findings. By sharing the integrated hardware–software platform and controller design, the researchers foster collaborative innovation geared toward overcoming scaling and complexity barriers in precision control systems.
In summary, this pioneering research heralds a transformative approach to high-speed nano-positioning control by melding neural networks, switching control theory, and advanced FPGA-based architectures. Its demonstrated efficacy in real-time hysteresis compensation and tracking accuracy enhancement positions it as a vital step forward for technologies demanding exceptional nanoscale motion control under challenging dynamic conditions.
Subject of Research: Nano-positioning stage control using neural network-based adaptive regulation and switching control theory.
Article Title: Neural Network-Based Switching Output Regulation Control for High-Speed Nano-Positioning Stages
News Publication Date: 17-Feb-2026
Web References:
Full article: https://doi.org/10.1016/j.eng.2025.07.023
Journal homepage: https://www.sciencedirect.com/journal/engineering
Image Credits: Hongwei Sun, Ning Xing et al.
Keywords
nano-positioning, neural network control, hysteresis compensation, piezoelectric actuators, FPGA–CPU architecture, switching systems, nonlinear control, feedback linearization, Lyapunov stability, dwell-time analysis, precision mechatronics, adaptive control
Tags: adaptive neural network controldynamic reference signal trackinghigh-fidelity piezoelectric ceramic layershigh-speed nano-positioning precisionmechatronic platform for nano-positioningnanoscale motion control stabilityneural-network-based switching output regulation controllernonlinear hysteresis in piezoelectric materialspiezoelectric actuation system advancementspiezoelectric actuator hysteresis compensationreal-time control parameter tuningswitching reference trajectory compensation
