Proton exchange membrane fuel cells (PEMFCs) undergo continuous performance degradation during long-term operation, and accurate lifetime prediction is essential for improving their reliability and health management. However, most existing prognostic methods mainly rely on stack-level health indicators and rarely account for the heterogeneous aging behavior among individual cells inside the stack. To address this issue, this paper investigates cell aging heterogeneity and its application to stack lifetime prediction from the perspective of polarization resistance evolution. First, electrochemical impedance spectroscopy data are fitted using an equivalent circuit model, and the total polarization resistance of each cell is extracted as the key variable for characterizing cell-level degradation. Statistical features are then constructed from the cell-level resistance distribution to describe both the average aging level and the aging heterogeneity inside the stack. Analyses of two stack datasets show that significant aging heterogeneity exists in both stacks, and that the increase in degradation dispersion in the later stage is mainly caused by the accelerated deterioration of a few abnormal cells rather than by the uniform degradation of all cells. Correlation analysis further indicates that cell-level aging features exhibit weak synchronous correlation with the stack-level health indicator but show clearer association with the degradation increment. On this basis, a 50 h-ahead degradation prediction framework is developed using a gradient boosting decision tree model, in which stack-level health information is progressively fused with average aging and heterogeneity features. Comparative experiments on two different dynamic conditions show that incorporating cell-level aging information significantly improves prediction accuracy. In particular, μRp provides the primary performance gain, while CVRp offers additional refinement and, in some cases, achieves performance comparable to the best model. These results demonstrate that the explicit incorporation of cell aging heterogeneity is effective for enhancing fuel cell stack lifetime prediction.

Accurate lifetime prediction has become a critical research focus in the field of hydrogen fuel cell technology advancement towards commercialization. Data-driven methods represent a mainstream approach for lifetime forecasting, their performance generally relies on extensive historical data. To reduce dependence on large datasets and enhance model generalization, this study proposes a hybrid prediction method that integrates physical mechanisms with data-driven methods. The proposed approach incorporates a fuel cell voltage degradation mechanism model as a physical constraint within the data-driven framework. The hybrid model consists of three core components. Firstly, a gradient boosting decision tree (GBDT)-based data-driven module that uses operating time and voltage as inputs to generate initial voltage predictions. Secondly, a voltage degradation mechanism model has been established based on the kinetics equations of the membrane and catalytic materials, providing theoretical voltage estimates as physical constraints. Finally, a bidirectional long short-term memory (Bi-LSTM) network for error correction, which uses prediction deviations from the first two stages as inputs to effectively minimize systematic errors through compensatory adjustments. The hybrid method has been validated under operating conditions that were static, quasi-dynamic, and dynamic with varying training durations. The results indicate that its accuracy is consistently superior to predictions generated directly by the GBDT algorithm. In data-scarce scenarios characterized by steady-state operations with only 300 h of training, the proposed method achieved reductions of 56.2 % in the root mean square error (RMSE) and 64.4 % in mean absolute percentage error (MAPE), thereby validating its superior generalization capability and data efficiency. Among all the evaluated operating conditions, the optimal RMSE and MAPE values were achieved under the dynamic working condition with a 200-h training duration, reaching 0.0047 and 0.09 %, respectively.

Retractable worm robots excel in navigating confined spaces that are inaccessible or harmful to humans, making them an innovative solution for inspection tasks in aerospace craft inspections, planetary exploration, and industrial maintenance. This article introduces RW-Robot, a novel retractable robot featuring a codesigned mechanical structure and control strategy to improve locomotion speed without sacrificing terrain adaptability. The robot utilizes a cascaded M-Canfield parallel mechanism, combining rigid stability with significant deformability for directional pointing and segment retraction. To mitigate rigidity limitations, a spatial locomotion strategy is proposed, synthesizing rectilinear gait and configuration adaptivity. First, a dual-mode bioinspired rectilinear gait integrates caterpillar-inspired cautious movement and inchworm-inspired rapid progression, enabling dynamic mode switching based on task urgency. Second, a foothold model predictive contouring control (MPCC-F) algorithm optimizes the robot's configuration and joint solutions, effectively balancing locomotion speed and terrain compliance, particularly during sharp turns. Simulation and experimental validation demonstrate high-speed locomotion (0.34 body length per second), agile turning capabilities (radius of 0.32 body length), and slope climbing up to 75(degrees), confirming its effectiveness for adaptive inspection tasks. Comprehensive comparisons with related worm robots highlight RW-Robot's superior directional flexibility and rapid inspection capabilities, indicating its suitability for urgent or time-sensitive inspections in confined environments.

To address the problems of insufficient transient disturbance rejection capability and degraded steady-state current performance in surface permanent magnet synchronous motor (SPMSM) speed control, this paper proposes a finite-time switching active disturbance rejection control (FSADRC) method. The proposed method consists of a finite-time cascaded extended state observer (FCESO) and a switching control law. The FCESO is composed of a first-layer finite-time ESO and a second-layer linear ESO. The first-layer finite-time ESO and the second-layer linear ESO are designed to provide speed and disturbance estimations for the transient and steady-state stages, respectively, achieving both high disturbance rejection capability, rapid response, and high steady-state current performance. To suppress oscillations during the transition between transient and steady-state modes, a linear smooth switching mechanism is developed to ensure the continuity of the control signal. Experimental results demonstrate that the proposed FSADRC significantly improves transient disturbance rejection capability, convergence speed, and steady-state speed and current performance, thereby validating the effectiveness and robustness of the proposed FSADRC.

Proton exchange membrane fuel cells (PEMFCs) are increasingly being deployed in high-altitude applications such as unmanned aerial vehicles and balloons. However, altitude-induced reductions in atmospheric pressure and oxygen partial pressure significantly alter electrochemical kinetics and mass transfer processes, leading to accelerated degradation and pronounced shifts in aging behavior. As a result, conventional lifetime prediction models calibrated under a single operating condition often exhibit poor generalization across different altitudes. To address this challenge, a physics-guided transferable lifetime prediction framework is proposed for PEMFCs operating under varying altitude conditions. An altitude-corrected degradation indicator (ACDI) is first constructed from electrochemical impedance spectroscopy data and an equivalent circuit model, enabling unified characterization of degradation behavior across different altitudes. Subsequently, an empirical degradation model is employed to capture the long-term aging trend under ground-level conditions, while a diffusion Transformer (DiT) is used to learn the stochastic residual dynamics. By combining trend mapping and residual alignment, the model trained at sea level is transferred to higher altitudes (1000, 2000, and 3000 m) using only partial early-life data. Experimental results show that increasing altitude accelerates degradation and shortens the remaining useful life. The proposed method accurately reconstructs the evolution trajectories of ACDI and provides reliable RUL estimates with narrow confidence intervals across all altitude conditions, thereby validating both the effectiveness of ACDI and the robustness of the proposed transferable prediction framework.

To address the inherent trade-off between disturbance rejection capability and noise suppression performance in the speed-loop control of surface permanent magnet synchronous motors (SPMSMs), an enhanced nonlinear active disturbance rejection control (ENADRC) method is proposed. Conventional high-bandwidth observers improve disturbance rejection capability at the expense of amplified measurement noise and degraded current performance. To alleviate this conflict, an enhanced nonlinear extended state observer (ENESO) is developed to establish coordinated fast and slow disturbance estimation channels. The fast estimation channel is utilized during transient-state to improve disturbance rejection capability and dynamic response performance, while the slow estimation channel is adopted during steady-state operation to suppress the influence of measurement noise on speed and current. In addition, a linear switching mechanism is designed to ensure smooth transition between the two estimation channels and prevent oscillatory phenomena caused by frequent switching. Experimental results demonstrate the effectiveness of the proposed method.

Under high-altitude and long-endurance operating conditions, fuel cell-powered autonomous aerial vehicles (AAVs) suffer significant performance degradation due to insufficient air supply. To address this challenge, this article proposes a novel air management strategy. This strategy integrates a turbocharger for exhaust gas recovery while dynamically regulating the boost air supply strategy, thereby creating synergistic effects to enhance net power performance. First, an air supply system model for variable altitude conditions is established, with subsequent analysis of air compressor and turbocharger operating characteristics. Second, a system net power optimization scheme is proposed to achieve precise search for the optimal net power trajectory. Then, an adaptive fuzzy proportional-integral-derivative (AFPID) control strategy was designed to enable accurate tracking of the optimal trajectory. Experimental results demonstrated that the proposed strategy achieves a significant enhancement in net power output compared to systems without turbocharging.

Proton exchange membrane fuel cell (PEMFC) is used in several fields due to its high efficiency and environmental friendliness. Durability issues remain one of the main barriers limiting its large-scale commercialization. Although prognostic techniques can optimize operation and extend fuel cell lifetime, their accuracy is often compromised by recoverable fault behaviors. Therefore, addressing this interference has become a crucial challenge in achieving reliable PEMFC prognostics. To tackle these issues, a diagnostic-prognostic hybrid framework integrating data-driven recoverable fault recognition with model-based degradation prediction is proposed. Specifically, a recoverable fault recognition model is developed to determine the operational state of the fuel cell, where data augmentation is employed to mitigate the adverse effects of data imbalance on classification performance. Using a prognostic strategy-guided particle filter (PSG-PF) model for degradation trend prediction and remaining useful life (RUL) estimation. The proposed method is evaluated by the dynamic load aging experimental data of PEMFC. The results show that the proposed method achieves over 98% accuracy in recoverable fault recognition. Meanwhile, by effectively circumventing the impact of recoverable faults, the average relative accuracy of predicted RUL reached 86%. The proposed framework can achieve reliable recognition of recoverable faults and credible RUL prediction under complex operating conditions. It enhances the robustness of PEMFC prognostics, supporting longer lifetime and more stable operation of fuel cells.

Model-free predictive current control (MFPCC) with generalized proportional-integral observer (GPIO) eliminates the impact of flux and resistance, while maintaining excellent steady-state and dynamic performance. However, the control performance of MFPCC is impacted by the observer and inductance parameters. It is difficult for the conventional GPIO with a fixed bandwidth gain to balance the transient performance, disturbance rejection, and noise sensitivity. Therefore, an adaptive GPIO (AGPIO) is proposed to solve this conflict, which utilizes a sigmoid-based adaptive law to determine the bandwidth gains adaptively. In addition, the control gain beta(c) impacts the performance of MFPCC, which is set as the inverse of inductance. To reduce the impacts of inductance, an improved finite-time gradient method (IFGM) is designed, which incorporates the finite-time gradient method (FGM) with adaptive optimization gain. It maintains lower steady-state fluctuations while ensuring rapid identification of the control gains. Comparative experiments on an SPMSM test bench verify the excellent control and harmonic rejection performance under parameter mismatch of the proposed method.

Switched reluctance machines (SRMs) are widely recognized for their low cost and robust structure. However, their practical application is severely limited by excessive vibration and noise, with large radial force ripple being a primary contributor to this issue. To address this issue, this article poses a novel radial force ripple suppression method that integrates model predictive control (MPC) with a newly designed force sharing function (FSF). The FSF is constructed based on a dynamically optimized flux-linkage curve, and enhanced with online compensation to enable real-time optimization. The MPC minimizes the error between the predicted total radial force and the reference force from the FSF, ensuring precise tracking control. To validate the effectiveness of the proposed method, experiments were conducted on a three-phase 12/8-pole SRM, with comparisons to the traditional current chopper control (CCC) method and torque sharing function based direct instantaneous torque control (TSF-DITC) method. Experimental results show that under the speed of 1200 rpm and the load of 4 Nm, the radial force ripple of the proposed method is only 8.13%, which is 23.12% lower than that of CCC and 8.26% lower than TSF-DITC.

Under varying operational conditions, prognostics of degradation for the proton exchange membrane fuel cell (PEMFC) is a vital, intricate endeavor, pivotal for predictive servicing and condition monitoring. Nevertheless, the unpredictability of time-varying operating regimes indices, as well as the limitation of transient lifespan prognosis mechanisms, pose significant challenges in the practical prediction process. To boost the forecasting precision of degradation methods, this study introduces the symphysis dynamic factor (SDF), a single health indicator (HI) created by blending two pieces of information. First, it needs to compute the factor-I deriving from feature parameters by establishing an equivalent circuit model (ECM). Second, a correlation model of aging PEMFC is derived from the semiempirical equation and ECM parameters, enabling the extraction of the correlation coefficient to compute factor-II. Then, the SDF is obtained by combining factor-I with factor-II. Employing the preconfigured current load and thermal conditions as variables, the particle filter (PF) estimates the operating voltage and the SDF, while quantifying the uncertainty of aging factor estimation. Following this, the decoupled echo state network is realized to employ unlimited period prediction, allowing for the estimation of the leftover useful duration. The efficacy and precision regarding the novel aging metric and the combined forecasting technique introduced have been validated under time-varying operating states.

Predicting the degradation of proton exchange membrane fuel cell (PEMFC) is a critical yet complex task under dynamic operating conditions, essential for their prognostics and health management. However, the uncertainty of the dynamic condition index and the localization of the short-term lifetime forecasting mechanism leads to many limitations in the actual predicting process. To improve the actual degradation prediction ability of prognostic methods, the comprehensive aging indicator (CAI) and the hybrid forecasting method which combines real-time estimation and long-term prediction under dynamic working conditions are proposed in this study. To be specific, firstly, the equivalent circuit model (ECM) is constructed to extract the feature parameters. Afterward, the correlation model can be obtained from the semi-empirical equation of fuel cell aging and ECM parameters, and then the initial parameters of the model are extracted. Finally, the working voltage and CAI of PEMFC are estimated by the extended Kalman filter using the pre-planned working current and temperature as input. Then, the long-term prediction is realized by the cascaded echo state network, and the remaining useful life is estimated. The effectiveness and accuracy of the proposed aging indicator and hybrid long-term lifetime prediction method are verified under dynamic working conditions.

The deadbeat predictive current control (DPCC) method for surface mounted magnet synchronous motor predicts the current by an accurate model, which suffers from parameter mismatch and harmonic disturbance. To enhance the dynamic tracking and steady-state predictive precision of DPCC under disturbance, a robust incremental model-based DPCC with a hybrid compensation method (HC-IDPCC) is proposed. First, the IDPCC dynamic model considering disturbances is derived, and the parameter sensitivity is analyzed. Then, a harmonic suppression extended state observer (HSESO) is designed to enhance the harmonic suppression ability and parameter robustness. In addition, an improved prediction error correction method (IPEC) is designed to enhance the dynamic performance, which utilizes the current and cumulative predictive errors to compensate for the disturbances. An improved switching method with a sliding window is designed to identify the dynamic and steady-state of the $q$-axis current and maintain a smooth switch between IPEC and HSESO compensation methods. HC-IDPCC integrates the advantages of IPEC and HSESO, which improves the parameter robustness and harmonic suppression ability while maintaining superior dynamic and steady-state performance. Finally, the effectiveness of the HC-IDPCC is verified by steady-state, dynamic, and switching method experiments.

The control effectiveness of model predictive current control (MPCC) for surface mounted permanent magnet synchronous motor (SPMSM) usually deteriorates with parameter mismatch and model uncertainties under different operation conditions, which lead to large current ripples, current tracking errors, and dynamic response degradation. To obtain excellent steady-state and dynamic performance, a robust MPCC based on a newly designed adaptive switching hybrid cost function is proposed. The adaptive switching hybrid cost function combines the merits of linear extended state observer (LESO) and improved prediction error correction (IPEC) strategies. A LESO with lower bandwidth is designed to be implemented in the steady-state to reduce the tracking errors and ripples of the current. To obtain excellent dynamic response and reduce the current tracking errors, an IPEC with a proportional-integral structure is proposed, integrating the accumulated error into the prediction error correction (PEC) strategy. In addition, an adaptive switching function (ASF) with a sliding window is designed to identify the current state and maintain a smooth switch between the LESO and IPEC. Hence, the proposed method inherits the excellent dynamic response of IPEC and the superior steady-state performance of LESO through the ASF. Finally, the control performance of the proposed method is evaluated by the experiment of parameters mismatch and switching methods.

The proton exchange membrane fuel cells (PEMFC) system is an environment-friendly power conversion equipment which can be employed in different applications. Prediction of the lifetime can help the users take some rewarding actions to extend the service life of the PEMFC system. Health indicators (HIs) serve to signify the extent of degradation under varying operational conditions. Given the constrained nature of dynamic HIs, a novel robust dynamic HI termed the relative power-loss rate (RPLR) is proposed in the paper. Besides the HIs, the output temperature of reactants (i.e., hydrogen and oxygen) are two important degradation-related operating parameters because they could reflect the electrochemical reaction process to some extent. Then the data-driven prediction method of hierarchical echo state network (HESN) is premiere proposed to predict the output temperatures of reactants and the lifespan at the same time. The first echo state network (ESN) is utilized for shortterm temperature prediction, whereas a subsequent cascaded ESN is employed for long-term lifetime forecasting. Incorporating the predicted output temperature as an additional input into the second ESN within the Hierarchical ESN (HESN) structure enhances the accuracy of lifetime prediction. The experimental results reveal that hydrogen or air output temperature significantly impacts PEMFC degradation rates, increasing internal resistance and polarization loss, and causing intermittent temperature fluctuations within the stack. The coupling between hydrogen output temperature and PEMFC aging is particularly notable. Compared to methods without temperature input, the HESN model with multi-step prediction of hydrogen or air output temperature improves convergence speed and accuracy of RUL estimation for commercial development of PEMFC.

In sensorless control systems of permanent magnet synchronous motors (PMSMs), the traditional linear extended state observer (LESO) is preferred due to its simplicity and ease of implementation. With the development of PMSM sensorless control systems, the requirements for position estimation performance have increased, and thus, traditional LESOs can no longer meet those needs. To address this issue, this article proposes an estimation method based on an integrally compensated-enhanced linear extended state observer (IC-ELESO) and an improved quadrature phase locked loop (IQPLL) with a third-order LESO. In the back electromotive force estimation scheme, by introducing a compensation loop, the proposed IC-ELESO suppresses DC bias and improves position estimation accuracy compared to traditional LESOs. In the position estimation scheme, the IQPLL combines the third-order LESO with a quadrature phase locked loop (QPLL) to eliminate errors introduced by ramp signals. Finally, a PMSM experimental platform is built to conduct a comparative experiment between the method proposed and the traditional LESO, which verifies the feasibility and superiority of the method proposed in this article.

Deadbeat predictive current control (DPCC) has received widespread attention due to its fast dynamic response and good current tracking accuracy. However, DPCC suffers from internal disturbance, such as parameter mismatch, and external disturbance, such as external load change. Firstly, an anti-disturbance terminal sliding mode controller (ATSMC), consisting of a novel reaching law (NRL) based terminal sliding mode controller (TSMC) and an integral-type terminal sliding-mode observer (ITSMO) is developed to improve the performance of speed loop. ATSMC can accelerate the convergence velocity and suppress the chattering simultaneously. Furthermore, to promote the current tracking performance and anti-disturbance property of DPCC against various unknown disturbances, a super-twisting observer (STO) in discrete time is designed to eliminate the disturbance caused by parameter mismatch of conventional DPCC, and the estimated values are compensated for the voltage vector to enhance the robustness. Finally, experiments are implemented on a 1.7-kW surface-mounted permanent-magnet synchronous motor (SPMSM) experimental platform, and the results validate the superiority of the proposed method compared with conventional methods.

To promote the drive performance of surface-mounted permanent magnet synchronous motor (SPMSM), such as responsiveness, tracking accuracy, and antidisturbance capability, an enhanced deadbeat predictive current control (DPCC) algorithm combining super-twisting terminal sliding mode is proposed in this article. First, a novel SPMSM model considering parameter perturbation is derived. Subsequently, a super-twisting terminal sliding mode control (STSMC) scheme with a super-twisting observer (STO) is proposed to enhance the performance of the speed loop, with stability validated using the Lyapunov theory. In addition, two extended state observers (ESOs) are developed to estimate the predictive error caused by parameter mismatch in the dq axis, respectively, and the estimated values are compensated with feedback to DPCC. Finally, the proposed control methods are implemented on an SPMSM platform and compared with conventional control methods under various operating conditions, and the comparison results prove the superiority.

A major obstacle to achieving large-scale applications of fuel cells in smart cities is the lack of durability of the stack. Accurate prediction of the Remaining Useful Life (RUL) of a fuel cell becomes a key prerequisite for enhancing its durability. However, reversible degradation behavior in fuel cell operation not only challenges the reliability of degradation trend prediction, but also affects the accurate estimation of RUL. To address this issue, this work proposes a reversible degradation detection-identification-avoidance and deep learning co-driven prognostic strategy. Diagnostic techniques are utilized to detect and isolate reversible degradation. The interference of data imbalance on model training is improved by a data augmentation approach incorporating physical knowledge. The prognostic strategy proposed in this paper is validated on a dynamic operating conditions dataset, and the results show an average improvement of 22 % to 62 % in the prognostic horizon matching rate.