An inductively coupled system composed of two identical linear resistive-capacitive shunted Josephson junction (LRCSJJ) circuits driven by external direct current (DC) sources was modeled and analyzed in this study. The coupling between the junctions was realized through a shared inductor. The existence and nature of equilibrium states were shown to depend critically on the normalized DC bias currents applied to the junctions. It was demonstrated that the coupled LRCSJJ system admits either an unique equilibrium point or no equilibrium points at all, depending on the biasing conditions. A comprehensive linear stability analysis of the equilibrium point was carried out, revealing that its stability is jointly governed by the inductive coupling strength and the magnitude of the normalized DC currents. When a single stable equilibrium point exists, the system operates in an excitable regime. Conversely, when equilibrium points are absent, sustained oscillatory dynamics emerge. The analysis highlights the role of inductive coupling in regulating the balance between dissipation, energy storage, and nonlinear Josephson dynamics, thereby shaping the global behavior of the coupled system. These results provide fundamental insight into the controllable dynamical regimes of inductively coupled Josephson junction (JJ) circuits and may be of relevance for the design of superconducting electronic devices, including neuromorphic circuits and high-frequency oscillators, where excitability and oscillation play a functional role.
Nonlinear dynamical systems operating under explicit constraints may exhibit qualitatively different closed-loop behaviours depending on the interaction between system coupling, feasibility boundaries, and feedback decision mechanisms. In constrained optimization-based control frameworks, such behaviour often manifests as distinct operating regimes and regime transitions that cannot be captured through local linear analysis. This study investigates constraint-induced nonlinear operating regimes arising in nonlinear model predictive control by considering quadrotor trajectory tracking as a representative constrained intelligent dynamical system. A physics-based rigid-body Newton–Euler model is embedded within a receding-horizon optimization framework with explicit actuator saturation and attitude safety constraints. Beyond conventional tracking objectives, the analysis adopts a system-level perspective to examine how nonlinear translational–rotational coupling and constraint activation jointly shape the qualitative structure of the closed-loop response. Comparative numerical simulations are conducted for both mild and aggressive reference maneuvers under varying constraint boundaries. The resulting responses reveal two dominant classes of nonlinear behaviour: constraint-inactive regimes, in which coupling-driven dynamics govern convergence characteristics, and constraint-active regimes, in which feasibility limits reallocate control authority and dominate transient response. Increased maneuver aggressiveness amplifies coupling-dominated effects, whereas tightened constraints induce regime transitions characterised by feasibility-driven dynamics. The results demonstrate that nonlinear model predictive control functions not only as an effective control strategy for constrained trajectory tracking, but also as a structured analytical tool for characterising regime-dependent behaviour in nonlinear intelligent systems. The findings provide insight into performance limitations, stability-relevant behaviour, and design trade-offs arising from the interplay between nonlinear dynamics and constraint geometry.
Physiological acoustic signals generated during respiration exhibit complex and often nonlinear characteristics that pose challenges for reliable analysis, particularly under real-world recording conditions. This study explores an intelligent signal analysis framework for early respiratory disease screening based on the processing of nonlinear respiratory sound patterns acquired via mobile devices. A hybrid deep learning architecture combining convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM) units is employed to capture both spectral structures and temporal dependencies inherent in respiratory acoustics. The proposed framework integrates feature extraction, nonlinear temporal–spectral representation, and server-side inference within a unified system architecture. Mel-frequency cepstral coefficients (MFCC) are used to characterize the acoustic signals, while the hybrid CNN–BiLSTM model learns discriminative patterns associated with different respiratory conditions. Experimental evaluation using a publicly available dataset demonstrates stable classification performance under moderately noisy conditions and varying recording scenarios, indicating robustness to practical acquisition constraints. The results suggest that intelligent learning-based representations can provide effective tools for analyzing nonlinear biomedical signals and supporting early-stage disease screening. Although focused on respiratory sound analysis, the proposed approach illustrates a general framework for intelligent interpretation of complex acoustic signals, which may be extended to other nonlinear signal analysis tasks in health-related and engineering applications.
The rapid digitalization of manufacturing has resulted in the accumulation of extensive archives of industrial technical drawings, particularly in mold manufacturing and aluminum profile design, where the efficient retrieval of geometrically and functionally similar designs has become essential for reducing engineering lead time, minimizing design redundancy, and improving product development efficiency. Conventional document management systems rely primarily on metadata-based indexing and keyword matching, making the geometric content and engineering attributes embedded in technical drawings largely inaccessible during retrieval, leading to unnecessary redesign efforts and increased development costs. To address these limitations, a hybrid similarity detection framework was proposed to integrate deep-learning-based visual feature extraction with parametric engineering information. Visual characteristics were extracted using a pre-trained Residual Network (ResNet) model, specifically ResNet-18, a convolutional neural network (CNN), from which 512-dimensional feature vectors were generated to represent the geometric characteristics of technical drawings. Simultaneously, optical character recognition (OCR) combined with rule-based information extraction was employed to identify critical engineering parameters, including profile number, cross-sectional area, and unit weight. The complementary visual and parametric representations were subsequently fused through a hybrid scoring mechanism to improve retrieval robustness. The proposed framework extends conventional content-based image retrieval by integrating domain-specific engineering knowledge. Experimental evaluation conducted on a real industrial dataset containing 2,390 aluminum profile technical drawings demonstrated high retrieval accuracy while maintaining computational efficiency, achieving a vector database search time of approximately 0.1 s and an end-to-end retrieval time (including feature extraction, parametric analysis, and hybrid ranking) of approximately 0.9 s per query. The results indicate that the proposed framework provides an effective solution for large-scale technical drawing management by facilitating design reuse, reducing repetitive engineering work, accelerating product development cycles, and improving operational efficiency in industrial design repositories.