Due to its high power density and low-speed stability, the dual-variable electrohydraulic direct-drive system (DEHDS) has been widely used in heavy-load industrial scenarios, which mainly consists of a servomotor, a variable-displacement pump and a hydraulic motor. Existing load-matching control methods achieve energy-efficient operation via pump displacement regulation, while the dual-variable coupled disturbance and flow mismatch induced by displacement regulation significantly degrade drive accuracy. For that, this article develops a hierarchical control framework for the coordinated achievement of energy-efficiency improvement and control-precision preservation in DEHDS, which integrates an upper-layer pump displacement planner and a lower-layer motor speed controller (MSC). The PDP performs constrained look-ahead displacement planning to improve energy efficiency while limiting the excitation of the dual-variable coupled disturbance. Based on this, upper-layer-regulation-induced coupling effects are then incorporated into the MSC, which is designed based on neural adaptive compensation control to achieve synchronous learning and rejection of matched and mismatched lumped disturbances, including dual-variable coupled dynamics, varying loads, and model uncertainties. Experiments under dynamic loading verify that the proposed framework provides a favorable efficiency-precision tradeoff. Compared with the fixed-displacement MSC, it reduces power loss and motor heat generation by 17.45% and 28.08%, while the average tracking error increases by only 0.028°. Meanwhile, compared with the conventional load-matching method, it achieves better overall tracking accuracy.
Wind turbine blade inspection is increasingly supported by deep learning and vision-language models (VLMs), which typically produce diagnostic results through end-to-end prediction or visual-language reasoning. However, these methods remain insufficient for establishing an autonomous reasoning–action–observation loop characteristic of human expert diagnosis, in which the workflow is adaptively adjusted as diagnostic information is progressively acquired. To address this gap, DiagAgent is developed as an autonomous diagnostic paradigm with adaptive workflow orchestration, in which a large language model dynamically organizes and revises diagnostic actions according to the evolving diagnostic state. A reflective multitool collaboration mechanism further integrates complementary information from heterogeneous inspection tools, with YOLO and a VLM serving as the primary visual tools, and triggers reflection when their outputs are conflicting or uncertain. In addition, dual-level visual retrieval provides category prototypes and historical defect cases as visual references for the reassessment of uncertain cases. Extensive experiments on two autonomous aerial vehicle-based wind turbine blade datasets demonstrate that DiagAgent consistently outperforms conventional deep learning methods and zero- and few-shot VLMs, particularly in fine-grained defect category diagnosis, validating its effectiveness and potential for autonomous wind turbine blade inspection in industrial environments.
In this article, the problem of online active fault diagnosis is investigated for linear parameter-varying systems, where the input is designed to separate output zonotopes under different system modes so as to accelerate the diagnosis process. A novel index is proposed to quantify the separation tendency between zonotopes, and it is proven that nonoverlapping can be guaranteed once the value of this index exceeds 1. Based on this index, a new observer design method is developed to further accelerate zonotope separation. To address multiple fault scenarios, an online test index is introduced to determine whether an operational mode can be immediately excluded, and a novel weight-allocation strategy is formulated based on this index so that the separation tendency of output estimation sets obtained in previous steps can be precisely quantified and utilized to further enhance the separation tendency in subsequent steps. Building upon these results, the input signal is designed to maximize the overall separation tendency, thereby improving diagnostic performance. A case study demonstrates that the proposed method achieves successful fault diagnosis in fewer steps than existing methods.
The high penetration of converter-interfaced renewable energy sources accentuates the spatial distribution characteristics of frequency response in power systems. Moreover, different subarea systems with varying proportions of nonsynchronous generators exhibit varying tolerances to frequency deviations and rate of change of frequency. Therefore, there are differentiated zonal critical inertia requirements (ZCIRs) within the system under disturbance, which cannot be solved by existing methods. To this end, this article proposes an estimation framework of ZCIRs for power systems based on spatial differences of frequency response. First, to further enhance frequency security in this situation, we introduce a novel concept of zonal critical inertia requirements estimation (ZCIRE). Then, to serve ZCIRE, we define key indicators, i.e., composite frequency dynamic index and ZCIRs. On this basis, combining affinity propagationand hierarchical clustering algorithms, we achieve the target of partitioning of the power system. Furthermore, we establish a preset order simplified power-frequency response model for each subarea system above, and employ output error system identification to determine the model parameters precisely. Subsequently, a ZCIRE model is constructed based on frequency dynamic security indexes constraints and reserve capacity constraints for each subarea system, so as to obtain the critical inertia requirements of the system. Finally, the rationality and superiority of the proposed ZCIRs estimation framework are validated in a modified IEEE-39 and IEEE-118 system.
This article proposes a novel singularity-avoidance prescribed performance control (PPC) framework based on control barrier functions (CBFs) to address the boundary-induced singularity problem of conventional PPC for disturbed nonlinear systems. The key idea is to dynamically regulate the prescribed envelope through a CBF-based quadratic programming safety filter. In this way, the forward invariance of the admissible transformation region is rigorously guaranteed, preventing the transformed system from approaching the singular surface. To enhance robustness against disturbances, a robust CBF condition is constructed by incorporating finite-time disturbance estimates, while a continuous-time model predictive controller is synthesized in the transformed unconstrained coordinates to achieve high-precision tracking. Rigorous stability analysis is established for the resulting closed-loop system. Simulation results and experimental validations on motor drives demonstrate the effectiveness and robustness of the proposed method under severe disturbances.
