Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Parameter-varying Data-enabled Predictive Control of Cold Atmospheric Pressure Plasma Jets

Pegah GhafGhanbari, Mircea Lazar, Javad Mohammadpour Velni

TL;DR

This work tackles real-time control of nonlinear, parameter-varying cold atmospheric pressure plasma jets (APPJs) by introducing Neural Parameter-Varying DeePC (NPV-DeePC). The approach combines a HyperDNN, which adaptively modulates a neural predictor conditioned on the tip-to-surface distance, with the data-enabled predictive control (DeePC) framework to achieve accurate multi-step predictions and constrained optimization across varying operating conditions. Offline training optimizes neural features and an output layer, while online control computes actions within a predictive horizon, balancing tracking accuracy and control effort. Simulations on surface temperature tracking and thermal dose delivery show that NPV-DeePC outperforms neural DeePC, DeePC, and MPC in accuracy, robustness to noise, and adaptability, with real-time computational feasibility. This framework offers a generalizable pathway for controlling other nonlinear, parameter-varying systems beyond APPJs.

Abstract

Cold Atmospheric Pressure Plasma Jets (APPJs) show significant potential for biomedical applications, but their inherent complexity, characterized by nonlinear dynamics and strong sensitivity to operating conditions like tip-to-surface distance, presents challenges for real-time control. This paper introduces the Neural Parameter-varying Data-enabled Predictive Control (NPV-DeePC) framework to address these issues. By integrating hypernetworks into the neural DeePC paradigm, NPV-DeePC adaptively captures system nonlinearities and parameter variations, dynamically adjusts the neural network's learned representation of the system, enabling accurate multi-step trajectory prediction and control. Simulation studies on surface temperature tracking and thermal dose delivery demonstrate that NPV-DeePC achieves higher accuracy and adaptability than existing controllers. Moreover, its computational efficiency supports real-time implementation, making it a practical approach for precise APPJ control and a generalizable solution for other nonlinear, parameter-varying systems.

Neural Parameter-varying Data-enabled Predictive Control of Cold Atmospheric Pressure Plasma Jets

TL;DR

This work tackles real-time control of nonlinear, parameter-varying cold atmospheric pressure plasma jets (APPJs) by introducing Neural Parameter-Varying DeePC (NPV-DeePC). The approach combines a HyperDNN, which adaptively modulates a neural predictor conditioned on the tip-to-surface distance, with the data-enabled predictive control (DeePC) framework to achieve accurate multi-step predictions and constrained optimization across varying operating conditions. Offline training optimizes neural features and an output layer, while online control computes actions within a predictive horizon, balancing tracking accuracy and control effort. Simulations on surface temperature tracking and thermal dose delivery show that NPV-DeePC outperforms neural DeePC, DeePC, and MPC in accuracy, robustness to noise, and adaptability, with real-time computational feasibility. This framework offers a generalizable pathway for controlling other nonlinear, parameter-varying systems beyond APPJs.

Abstract

Cold Atmospheric Pressure Plasma Jets (APPJs) show significant potential for biomedical applications, but their inherent complexity, characterized by nonlinear dynamics and strong sensitivity to operating conditions like tip-to-surface distance, presents challenges for real-time control. This paper introduces the Neural Parameter-varying Data-enabled Predictive Control (NPV-DeePC) framework to address these issues. By integrating hypernetworks into the neural DeePC paradigm, NPV-DeePC adaptively captures system nonlinearities and parameter variations, dynamically adjusts the neural network's learned representation of the system, enabling accurate multi-step trajectory prediction and control. Simulation studies on surface temperature tracking and thermal dose delivery demonstrate that NPV-DeePC achieves higher accuracy and adaptability than existing controllers. Moreover, its computational efficiency supports real-time implementation, making it a practical approach for precise APPJ control and a generalizable solution for other nonlinear, parameter-varying systems.

Paper Structure

This paper contains 15 sections, 2 theorems, 33 equations, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Notations
  3. Problem Description
  4. Structure of Atmospheric Pressure Plasma Jets
  5. Control Challenges for APPJs
  6. Control Objectives for APPJs
  7. Hybrid Data-driven Control Approach
  8. Preliminaries
  9. Hyper Deep Neural Network Architecture
  10. Neural Parameter-Varying DeePC
  11. Simulation Results and Validation
  12. Data Generation and Preprocessing
  13. HyperDNN Training and Neural Space Construction
  14. Control Performance
  15. Conclusion

Key Result

Lemma 1

(Willems' Fundamental Lemma willems2005note) Consider an LTI system of order $n$. Suppose that following conditions hold: Then, any $D$-samples long trajectory $\{(\tilde{u}(k), \tilde{y}(k))\}_{k=1}^{D}$ of the system can be represented as a linear combination of the columns of the Hankel matrices constructed from the original trajectory. That is, it is also a trajectory of the system if and onl

Figures (9)

  • Figure 1: Schematic of atmospheric pressure plasma jets.
  • Figure 2: Overview of the proposed HyperDNN architecture. The target network $\mathfrak{T}_\theta$ predicts the output $\hat{\mathbf{y}}_{NN}$ from the input $\mathbf{u}_{NN}$, constructing the neural space $\upphi_{HL}$. Its parameters include fixed weights and biases for the first $m$ hidden layers, capturing general features, and dynamic parameters for the subsequent layers, computed by the hypernetwork $\mathfrak{H}_\psi$. The hypernetwork consists of subnetworks $\mathfrak{h}_j$, each generating weights and biases for the corresponding $j$-th dynamic hidden layer of $\mathfrak{T}_\theta$ based on the contextual input $\mathbf{p}$, enabling adaptation to varying APPJ operating conditions.
  • Figure 3: Dataset used for HyperDNN training, showing a representative 500-sample subset.
  • Figure 4: Tip-to-surface distance variation for the reference tracking experiment.
  • Figure 5: Tracking performance of the proposed controller compared to the benchmarks for noise-free and noisy measurements.
  • ...and 4 more figures

Theorems & Definitions (4)

  • Definition 1
  • Lemma 1
  • Lemma 2
  • proof