Pith. sign in

REVIEW 3 major objections 5 minor 22 references

Online Learning Control Strategies for Industrial Processes with Application for Loosening and Conditioning

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read HPC-AK-MPC, an online-learning Koopman MPC with historically derived safety corridors, claims significantly better Cpk on all 16 test batches of a real tobacco loosening line.

desk verdict The framework is a clean integration of online Koopman identification with a historical safety corridor, but the headline Cpk gains come from a circular advisor-mode validation and don't survive contact with the process's transport delay. read the letter →

arxiv 2506.08983 v1 pith:HNR2HXE4 submitted 2025-06-10 eess.SY cs.SYmath.OA

Yue Wu , Jianfu Cao , Ye Cao This is my paper
classification eess.SYcs.SYmath.OA
keywords KoopmanoperatorModelpredictivecontrolRecursiveleastsquaresOnlinelearningHistoricalprocessconstraintsSafeTobaccolooseningandconditioningcapabilityindex
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that an industrial controller can learn a nonlinear, time-varying process online and still operate safely by constraining its actions near historically successful operations. It combines recursive EDMDc—a Koopman-based linear model updated by recursive least squares—with a Historical Process Constraint that tightens around past safe control inputs when model confidence is low. The proposed HPC-AK-MPC is tested in an 'advisor mode' that replays real production batches and predicts what would have happened had its control advice been used. Across 16 held-out batches, the predicted trajectories yield higher Process Capability Index values for furnace temperature, outlet moisture, and outlet temperature than the historical trajectories. If this is true, it offers a practical route from static data-driven models to safe adaptive control in process industries.

What carries the argument

The load-bearing piece is the rEDMDc estimator: the lifted state dictionary built from second-order polynomials of the state, with parameters Θ_k=[A_L,k | B_L,k] updated sample-by-sample by recursive least squares with a forgetting factor and covariance reset, giving an always-current linear model of the lifted dynamics. Around it sits the Historical Process Constraint: for the current lifted state, K-nearest neighbors in the historical database yield a reference control u_ref^hist, and the first control move is restricted to the corridor u_ref ± max((α_base+β_adapt·Conf_k)|u_ref|, δ_abs), where Conf_k=max(0,1−tr(P_k)/tr(P_max)) measures model confidence. The MPC then solves a quadratic program that combines physical limits with this adaptive corridor. The experimental evaluation uses advisor mode: the actual historical state is fetched, MPC proposes an action, and the rEDMDc model predicts the next state that would have resulted; those predicted states are scored by the Process Capability Index.

What would settle it

Replay the same 16 batches while recording the rEDMDc model's one-step prediction error, comparing the predicted next state against the actual next historical state under the historical action. If the mean absolute error on the primary variable, outlet moisture, exceeds the historical fluctuation band used in the Cpk calculation, then the claimed Cpk improvement is an artifact of the model rather than a property of the controller. A direct closed-loop trial or a high-fidelity validated simulator would settle whether the predicted improvements appear under real feedback.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that replacing a one-time offline Koopman model with a recursively updated one, and adding a safety corridor based on historical data, lets model predictive control track time-varying industrial dynamics while avoiding unsafe actions. In the advisor-mode evaluation, this combination produces predicted state trajectories that stay closer to setpoint, with Cpk values above historical operations across all tested batches. The quantitative claim is about potential rather than a closed-loop demonstration, but the authors frame it as evidence that the controller outperforms historical manual operation.

Load-bearing premise

The reported improvement rests on trusting the rEDMDc model's one-step predictions of what the plant would have done under the MPC's suggested actions; if those predicted states differ from the real plant responses, the Cpk gains are model artifacts rather than actual performance.

Editorial extensions

If this is right

  • If the central claim is right, data-driven Koopman models need no full retraining when process conditions drift; recursive updates keep them current.
  • Historical operating data becomes a reusable safety resource, not just a static training set for an offline model.
  • The safety-performance trade-off becomes self-tuning: constraints tighten under model uncertainty and relax as confidence grows.
  • The approach extends to any batch process with recorded state-input histories and slowly varying dynamics.
  • Cpk-based scoring can support 'advisor mode' comparisons when closed-loop plant experiments are infeasible.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural testable extension is to replace the actual historical action (used as the reference in this study) with the KNN-derived reference from other batches, to see whether the improvement survives when the corridor is not anchored to the exact historical input.
  • The confidence metric relies on the trace of the RLS covariance matrix; a more informative uncertainty measure based on prediction variance would likely give tighter, better-calibrated corridors.
  • The advisor-mode comparison implicitly assumes the historical operator's actions were themselves reasonable; if historical operations were demonstrably poor, even a modest model would show large Cpk gains.
  • The method is applicable beyond tobacco, for example to food processing, chemical batch reactors, or other rotating-drum and drying processes with similar time variation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes HPC-AK-MPC, an adaptive Koopman Model Predictive Control framework that combines recursive Extended Dynamic Mode Decomposition with control (rEDMDc) for online model updates and a Historical Process Constraint (HPC) mechanism that uses historical data to build an adaptive safety corridor for the control actions. The method is applied to a tobacco loosening and conditioning process and evaluated in an 'advisor mode' simulation on 16 historical batches. The central claim is that, compared with historical operations, the proposed controller significantly improves the Process Capability Index (Cpk) of three key quality variables across all tested batches.

Significance. If the reported Cpk improvements were backed by credible closed-loop or validated-model evidence, the framework would be of practical interest: the combination of online Koopman identification with historical-data-based safety constraints is a plausible approach for time-varying industrial processes, and the paper provides reproducible parameter settings and a coherent QP formulation. However, the central performance claim rests entirely on an unvalidated counterfactual simulation in which the controller's own model generates the predicted trajectories that are then scored. As a result, the significance of the claimed experimental demonstration is not established by the present manuscript.

major comments (3)
  1. [IV-B, step 3 and Eq. (29)] The central performance claim is unsupported because the advisor-mode evaluation is self-referential. In Section IV-B, step 3, the controller uses its internal rEDMDc model to predict x_pred,t+1 from the recommended u_mpc,t, and the Cpk comparison in Eq. (29) is computed from this predicted trajectory. Since the same model is optimized by the MPC, the predicted trajectory will tend to look good even if the model is inaccurate; no held-out one-step prediction validation of the rEDMDc model against actual historical transitions is reported. The 'significant improvement in Cpk' therefore reduces to a self-consistency check and is not evidence of real-plant improvement.
  2. [IV-A and Eq. (10)] The model in Eq. (10) is a first-order Markov model over the instantaneous lifted state and current input, yet Section IV-A states that the process has a 5-8 minute transport delay. With 1-second sampling, the effect of an input at time k on the outlet quality variables appears only 300-480 samples later, so the term B_k[u_k] in Eq. (10) cannot represent the causal effect of a control action on the next measured outlet state. The advisor-mode counterfactual therefore relies on a structurally implausible model for the dominant process dynamics.
  3. [IV-A and III-C] The seven-dimensional input vector u_k includes two measured disturbances, Inlet Moisture and Cumulative Water Added, which are treated as system inputs 'for modeling purposes.' In the MPC formulation of Section III-C and in the advisor-mode recommendation step, these disturbances are optimized as decision variables, allowing the QP to recommend physically unrealizable changes to quantities the plant cannot actuate. This further undermines the credibility of the predicted 'would-be' next states.
minor comments (5)
  1. [Section I] There is a typo in the Introduction: 'Ifinite-horizon' should be 'finite-horizon'.
  2. [Algorithm 1] In Algorithm 1, line 7, u_ref,hist is accumulated but never initialized; please initialize it to zero before the loop.
  3. [III-A] Reference [20] is used to support the statement that polynomial dictionaries approximate the dominant Koopman-invariant subspace, but [20] is a textbook on recursive least squares; a more appropriate citation to the Koopman/EDMD literature would be preferable.
  4. [III-C] The sentence before Eq. (24) says the structures are derived by recursively applying (24), but Eq. (24) is the result of the recursion; the sentence should refer to Eq. (22) or otherwise be reworded.
  5. [IV-C] In Figures 4-6, axis labels, units, and legend details are not fully described in the text; please ensure the figures are self-contained for the reader.

Circularity Check

1 steps flagged · score 8.0 of 10

Cpk gain is an in-model self-consistency check: the rEDMDc model both picks the control and produces the 'predicted trajectory' used as evidence.

  1. fitted input called prediction [Section IV-B (Advisor Mode), step 3, and Performance Metric; cf. Eq. (10), (21)-(24)]
    "Predict Potential Outcome: Using its internal, up-to-date rEDMDc model, the controller predicts what the next state, x_pred,t+1, would have been if its recommended action u_mpc,t had been applied instead of the historical action u_actual,t."

    u_mpc,t is selected by minimizing the model's own predicted tracking error (Eq. (21)), and the counterfactual next state is then generated by that same rEDMDc model (Eqs. (10), (22)-(24)). The Cpk comparison uses this model-generated 'predicted trajectory' as the evidence for improvement. Hence the exercise can only show that the optimizer can make its own fitted linear model look closer to setpoint; it is a self-consistency check, not an independent validation. No held-out one-step prediction accuracy of the rEDMDc model is reported, so the predicted improvements are not tied to real process response. The result is forced by construction: the evaluation oracle and the controller share the same fitted parameters.

full rationale

The derivation chain itself (rEDMDc update, QP formulation, HPC constraints) is not circular: the equations are standard and no load-bearing result is imported from the authors' own prior work. The circularity is entirely in the 'validation' step. The advisor-mode procedure uses the internally updated rEDMDc model both to choose u_mpc (via the QP, Eq. (28)) and to predict the 'would-be' next state x_pred,t+1; the Cpk comparison then contrasts the actual historical trajectory with this same model's prediction. Because the optimizer minimizes predicted deviations from setpoint, the predicted trajectory is, by construction, the model's own best-case scenario, and any measured 'improvement' is a statement about the model's self-consistency rather than about the real plant. Additionally, the HPC forces the advice to stay within 10% of the historical input, and the 5-8 min process delay is not represented in the 1-s first-order lifted model; these reinforce that the findings are not independent empirical evidence. We therefore flag one central circular/self-evaluative step and score it 8. If the authors had validated the rEDMDc model against held-out real one-step transitions, the advisor-mode result would be meaningful; no such validation is reported.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard RLS assumptions, on the approximation quality of a fixed polynomial lift, and critically on the unverified assumption that the advisor-mode predicted states equal real plant responses. The HPC also assumes that historical actions are safe and that similar states have similar safe controls. These are domain assumptions, and the advisor-mode one is ad hoc and load-bearing.

free parameters (8)
  • Forgetting factor lambda_f = 0.995
    Controls the trade-off between tracking speed and noise sensitivity in rEDMDc; hand-chosen, no sensitivity analysis. Section IV.B.
  • Ridge regularization coefficient lambda_reg = 1e-3
    Regularization in EDMDc least-squares; hand-chosen. Section IV.B.
  • Lifting dictionary order = 2 (10-dimensional lifted space)
    Second-order polynomial lift chosen to approximate Koopman subspace; no comparison with other dictionaries. Section IV.B.
  • Prediction and control horizons = Hp=15, Hc=7
    MPC horizons hand-tuned; no sensitivity study. Section IV.B.
  • MPC weights Q, R, S = Q=diag([10,100,1]), R=1e-2*I7, S=0
    Emphasize outlet moisture tracking; hand-chosen. Section IV.B.
  • HPC corridor parameters in advisor mode = 10% relative deviation, 1% absolute floor
    Simplified HPC implementation uses alpha_base=0.1 and delta_abs=0.01*range; beta_adapt and confidence not exercised. Section IV.B.
  • Initial parameter matrix Theta_0 = Offline batch fit on 32 batches
    Initialization for RLS; affects early behavior of the adaptive model. Section IV.B.
  • Covariance normalization P_max = trace(P0)
    Normalization for the confidence metric; hand-set in Section III.B.
assumptions (4)
  • domain assumption Regressor sequence is persistently exciting, true parameters vary slowly, and process noise is bounded.
    Used to claim UUB of rEDMDc parameter estimates in Section III.A. Not verified with data.
  • domain assumption A finite dictionary of lifting functions approximately spans a Koopman-invariant subspace.
    Basis of the linear lifted model in Eq. (10); Section III.A.
  • ad hoc to paper The rEDMDc model's one-step prediction equals the real plant response to the proposed control action.
    Core of the advisor-mode evaluation in Section IV.B, step 3. This makes the validation circular and is unverified.
  • domain assumption Historical state-action pairs are successful and similar states have similar safe controls.
    Underpins the HPC safety corridor in Section III.B.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Online Learning Control Strategies for Industrial Processes with Application for Loosening and Conditioning." pith.science (2026). https://pith.science/paper/HNR2HXE4

@misc{pith2026250608983,
  author       = {Pith},
  title        = {Pith review of: Online Learning Control Strategies for Industrial Processes with Application for Loosening and Conditioning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HNR2HXE4}},
  note         = {Machine review of arXiv:2506.08983}
}
read the original abstract

This paper proposes a novel adaptive Koopman Model Predictive Control (MPC) framework, termed HPC-AK-MPC, designed to address the dual challenges of time-varying dynamics and safe operation in complex industrial processes. The framework integrates two core strategies: online learning and historically-informed safety constraints. To contend with process time-variance, a Recursive Extended Dynamic Mode Decomposition (rEDMDc) technique is employed to construct an adaptive Koopman model capable of updating its parameters from real-time data, endowing the controller with the ability to continuously learn and track dynamic changes. To tackle the critical issue of safe operation under model uncertainty, we introduce a novel Historical Process Constraint (HPC) mechanism. This mechanism mines successful operational experiences from a historical database and, by coupling them with the confidence level of the online model, generates a dynamic "safety corridor" for the MPC optimization problem. This approach transforms implicit expert knowledge into explicit, adaptive constraints, establishing a dynamic balance between pursuing optimal performance and ensuring robust safety. The proposed HPC-AK-MPC method is applied to a real-world tobacco loosening and conditioning process and systematically validated using an "advisor mode" simulation framework with industrial data. Experimental results demonstrate that, compared to historical operations, the proposed method significantly improves the Process Capability Index (Cpk) for key quality variables across all tested batches, proving its substantial potential in enhancing control performance while guaranteeing operational safety.

Figures

Figures reproduced from arXiv: 2506.08983 by the authors.

Figure 2
Figure 2. The proposed closed-loop control architecture of the HPC-AK-MPC. IV. Experimental Study A. Industrial Process Description and Data Preprocessing The proposed control strategy is applied to a HAUNI loosening and conditioning cylinder, a key piece of equipment in the primary processing line of a cigarette factory. The physical system consists of a large, inclined, rotating drum fitted with internal guide plates. Dried… view at source ↗
Figure 3
Figure 3. Flowchart of the "advisor mode" simulation framework. [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Advisor mode results for Furnace Temperature (top) and Water [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Advisor mode results for Outlet Moisture (top) and Water Flow control [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Comparison of Process Capability Index (Cpk) for historical vs. MPC [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

22 extracted references · 16 canonical work pages

  1. [1]

    Data-Driven Inverse Optimal Control for Continuous-Time Nonlinear Systems

    H. J. Asl and E. Uchibe, “Data-Driven Inverse Optimal Control for Continuous- Time Nonlinear Systems,” Mar. 19, 2025, arXiv: arXiv:2503.09090. doi: 10.48550/arXiv.2503.09090

    work page Pith review arXiv doi:10.48550/arxiv.2503.09090 2025

  2. [2]

    Adaptive Data-Driven Control for Linear Time Varying Systems,

    T. Abdalla, “Adaptive Data-Driven Control for Linear Time Varying Systems,” Machines, vol. 9, no. 8, p. 167, Aug. 2021, doi: 10.3390/machines9080167

    work page doi:10.3390/machines9080167 2021

  3. [3]

    Uncertainty-Aware Data-driven Tobacco Loosening and Conditioning Process Moisture Prediction and Control Optimization,

    Y. He, B. Li, Y. Pu, W. Jin, X. Zhou, and C. Jin, “Uncertainty-Aware Data-driven Tobacco Loosening and Conditioning Process Moisture Prediction and Control Optimization,” in 2021 IEEE 10th Data Driven Control and Learning Systems Conference (DDCLS), Suzhou, China: IEEE, May 2021, pp. 600–607. doi: 10.1109/DDCLS52934.2021.9455522

    work page arXiv 2021

  4. [4]

    The added water control system based on neural network model and double parameter corrected for loosening and conditioning cylinder,

    Z. Li, J. Chen, Y. Gao, X. Jia, L. Zhao, and Z. Feng, “The added water control system based on neural network model and double parameter corrected for loosening and conditioning cylinder,” J. Phys. Conf. Ser., vol. 2083, no. 2, p. 022045, Nov. 2021, doi: 10.1088/1742-6596/2083/2/022045

    work page doi:10.1088/1742-6596/2083/2/022045 2021

  5. [5]

    An Operation Mode Optimization Method for Tobacco Loosening and Conditioning Process Based on Batch Clustering and KNN Algorithm,

    W. Yang, C. Xiaofang, P. Xiangyu, W. Yalin, Y. Weichao, and Z. Wei, “An Operation Mode Optimization Method for Tobacco Loosening and Conditioning Process Based on Batch Clustering and KNN Algorithm,” in 2018 Chinese Automation Congress (CAC), Xi’an, China: IEEE, Nov. 2018, pp. 721–726. doi: 10.1109/CAC.2018.8623718

    work page arXiv 2018

  6. [6]

    Construction and application of the prediction model of outlet moisture in the loosening and conditioning process,

    Z. Li et al., “Construction and application of the prediction model of outlet moisture in the loosening and conditioning process,” in International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2022), X. Ye and G. Zhong, Eds., Sanya, China: SPIE, May 2022, p. 35. doi: 10.1117/12.2635392

    work page doi:10.1117/12.2635392 2022

  7. [7]

    Design of an optimal scheduling control system for smart manufacturing processes in tobacco industry,

    X. Liu, J. Li, H. Wang, W. Jia, J. Yang, and Z. Guo, “Design of an optimal scheduling control system for smart manufacturing processes in tobacco industry,” IEEE Access, vol. 11, pp. 33027–33036, 2023, doi: 10.1109/ACCESS.2023.3261883

    work page arXiv 2023

  8. [8]

    A tobacco moisture prediction approach based on VAE and PSO-BiLSTM,

    M. Li, L. Zhu, X. Kong, M. Hu, and Y. Shao, “A tobacco moisture prediction approach based on VAE and PSO-BiLSTM,” in International Workshop on Automation, Control, and Communication Engineering (IWACCE 2024), W. Mou, Ed., Hohhot, China: SPIE, Dec. 2024, p. 66. doi: 10.1117/12.3052456

    work page doi:10.1117/12.3052456 2024

Show all 22 references
  1. [9]

    A Novel Variable Exponential Discrete Time Sliding Mode Reaching Law,

    X. Chen, Y. Li, H. Ma, H. Tang, and Y. Xie, “A Novel Variable Exponential Discrete Time Sliding Mode Reaching Law,” IEEE Trans Circuits Syst II, vol. 68, no. 7, pp. 2518–2522, doi: 10.1109/tcsii.2021.3051904

    2021

  2. [10]

    Control-Coherent Koopman Modeling: A Physical Modeling Approach,

    H. H. Asada and J. A. Solano-Castellanos, “Control-Coherent Koopman Modeling: A Physical Modeling Approach,” arXiv, 2024. [Online]. Available: http://dx.doi.org/10.48550/arXiv.2403.16306

    2024 doi arXiv

  3. [11]

    Machine learning approach to observability analysis of high-dimensional nonlinear dynamical systems using Koopman operator theory,

    S. R. W. Balakrishnan, “Machine learning approach to observability analysis of high-dimensional nonlinear dynamical systems using Koopman operator theory,” 2023

    2023

  4. [12]

    Extended dynamic mode decomposition with learned Koopman eigenfunctions for prediction and control,

    C. Folkestad, D. Pastor, I. Mezic, R. Mohr, M. Fonoberova, and J. Burdick, “Extended dynamic mode decomposition with learned Koopman eigenfunctions for prediction and control,” in 2020 american control conference (acc), IEEE, 2020, pp. 3906–3913. doi: 10.23919/ACC45564.2020.9147729

    2020

  5. [13]

    Learning model predictive control of nonlinear systems with time-varying parameters using Koopman operator,

    Z. Chen, “Learning model predictive control of nonlinear systems with time-varying parameters using Koopman operator,” Appl. Math. Comput., 2024

    2024

  6. [14]

    Model Predictive Traction Control System Based on the Koopman Operator,

    J. K. Hromatko and andor Ile, “Model Predictive Traction Control System Based on the Koopman Operator,” in 2024 28th International Conference on System Theory, Control and Computing, ICSTCC 2024 - Proceedings, Sinaia, Romania, 2024, pp. 26–31. [Online]. Available: http://dx.do...

    2024

  7. [15]

    Robust Koopman-MPC Approach with High-Order Disturbance Observer for Control of Pneumatic Soft Bending Actuators under External Loads,

    J. Wang, B. Xu, J. Liu, Z. Zhao, W. Peng, and A. Song, “Robust Koopman-MPC Approach with High-Order Disturbance Observer for Control of Pneumatic Soft Bending Actuators under External Loads,” IEEEASME Trans. Mechatron., 2025, [Online]. Available: http://dx.doi.org/10.1109/TMEC...

    2025

  8. [16]

    A MPC Performance Degradation Diagnosis Method Based on Receding Feature Horizon,

    W. Jiao, T. Zou, Z. Xie, S. Yang, Y. Dong, and L. Li, “A MPC Performance Degradation Diagnosis Method Based on Receding Feature Horizon,” in 2024 IEEE 20th International Conference on Automation Science and Engineering (CASE), IEEE, 2024, pp. 2314–2319

    2024

  9. [17]

    A Deep Reinforcement Learning Approach to Improve the Learning Performance in Process Control,

    Y. Bao, Y. Zhu, and F. Qian, “A Deep Reinforcement Learning Approach to Improve the Learning Performance in Process Control,” Ind. Eng. Chem. Res., vol. 60, no. 15, pp. 5504–5515, Apr. 2021, doi: 10.1021/acs.iecr.0c05678

    2021 doi

  10. [18]

    Safe Reinforcement Learning With Dual Robustness,

    Z. Li, C. Hu, Y. Wang, Y. Yang, and S. E. Li, “Safe Reinforcement Learning With Dual Robustness,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 46, no. 12, pp. 10876–10890, Dec. 2024, doi: 10.1109/TPAMI.2024.3443916

    2024

  11. [19]

    A Review of Safe Reinforcement Learning: Methods, Theory and Applications,

    S. Gu et al., “A Review of Safe Reinforcement Learning: Methods, Theory and Applications,” May 24, 2024, arXiv: arXiv:2205.10330. doi: 10.48550/arXiv.2205.10330

    2024 doi arXiv

  12. [20]

    P. J. G. Teunissen, Dynamic data processing: recursive least-squares, 2nd ed. Delft [Netherlands]: TU Delft OPEN publishing, 2024

    2024

  13. [21]

    On the design of persistently exciting inputs for data-driven control of linear and nonlinear systems,

    M. Alsalti, V. G. Lopez, and M. A. Müller, “On the design of persistently exciting inputs for data-driven control of linear and nonlinear systems,” IEEE Control Syst. Lett., vol. 7, pp. 2629–2634, 2023, doi: 10.1109/LCSYS.2023.3287133

    2023

  14. [22]

    Continuous state feedback guaranteeing uniform ultimate boundedness for uncertain dynamic systems,

    M. Corless and G. Leitmann, “Continuous state feedback guaranteeing uniform ultimate boundedness for uncertain dynamic systems,” IEEE Trans. Autom. Control, vol. 26, no. 5, pp. 1139–1144, Oct. 1981, doi: 10.1109/TAC.1981.1102785

    1981

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.