A deep reinforcement learning method for training a jellyfish-like swimmer to effectively track a moving target in a two-dimensional flow was developed. This swimmer is a flexible object equipped with a muscle model based on torsional springs. We employed a deep Q-network (DQN) that takes the swimmer’s geometry and dynamic parameters as inputs, and outputs actions that are the forces applied to the swimmer. In particular, an action regulation was introduced to mitigate the interference from complex fluid–structure interactions. The goal of these actions is to navigate the swimmer to a target point in the shortest possible time. In the DQN training, the data on the swimmer’s motions were obtained from simulations using the immersed boundary method. During tracking a moving target, there is an inherent delay between the application of forces and the corresponding response of the swimmer’s body due to hydrodynamic interactions between the shedding vortices and the swimmer’s own locomotion. Our tests demonstrate that the swimmer, with the DQN agent and action regulation, is able to dynamically adjust its course based on its instantaneous state. This work extends the application scope of machine learning in controlling flexible objects within fluid environments.

Numerical studies on the statistical properties of irregular waves in finite depth have to date been based on models founded on weak nonlinearity; as a consequence, only lower-order (usually third-order) nonlinear interactions have thus far been investigated. The present study performs numerical simulations with a fully nonlinear, spectrally accurate model to investigate the statistics of irregular, unidirectional wave fields in finite water depth initially given by a Texel, Marsen and Arsloe spectrum. A series of random unidirectional wave fields are considered, covering a wide range of water depth. The wave spectrum and statistical properties, including the probability density function of the surface elevation, exceedance probability of wave crests and occurrence probability of extreme (rogue) waves, are investigated. The importance of full nonlinearity in comparison with third-order results is likewise evaluated. The results show that full nonlinearity increases kurtosis and enhances the occurrence probability of large wave crests and rogue waves substantially, in both deep water and finite water depth. Therefore, we propose that full nonlinearity may contribute significantly to the formation of rogue waves. Furthermore, to account for the effects of higher-order nonlinearity on modulational instability, we analyse the relationship between the Benjamin–Feir index (BFI) and maximal excess kurtosis. Our results show a strong linear relationship i.e. $({\mathcal{K}}_{max}-3)\propto {\textrm{BFI}}$, in contrast to $({\mathcal{K}}_{max}-3)\propto {\textrm{BFI}}^2$ based on the assumptions of weak nonlinearity, a narrow-banded spectrum and deep-water conditions. Above, $\mathcal{K}_{max}$ is the maximal kurtosis.

Waves transport particles in the direction of wave propagation with the Stokes drift. When the Earth’s rotation is accounted for, waves induce an additional (Eulerian-mean) current that reduces drift and is known as the anti-Stokes drift. This effect is often ignored in oceanic particle-tracking simulations, despite being important. Although different theoretical models exist, they have not been validated by experiments. We conduct laboratory experiments studying the surface drift induced by deep-water waves in a purpose-built rotating wave flume. With rotation, the Lagrangian-mean drift deflects to the right (counterclockwise rotation) and reduces in magnitude. Compared with two existing steady theoretical models, measured drift speed follows a similar trend with wave Ekman number but is larger. The difference is largely explained by unsteadiness on inertial time scales. Our results emphasise the importance of considering unsteadiness when predicting and analysing the transport of floating material by waves.

The dynamics of self-excited shock train oscillations in a back pressured axisymmetric duct was investigated to deepen the understanding of the isolator/combustor coupling in high-speed propulsion systems. The test article consisted of an internal compression inlet followed by a constant area isolator, both having a circular cross-section. A systematic back pressure variation was implemented by using a combination of aerodynamic and physical blockages at the isolator exit. High bandwidth two-dimensional pressure field imaging was performed at $8\,{\rm kHz}$ repetition rate within the isolator for different back pressure settings. The acquisition rate was considerably higher than the dominant frequency of the shock train oscillations across the different back pressure settings. The power spectral density of the pressure fluctuations beneath the leading shock foot exhibited broadband low frequency oscillations across all back pressures that resembled the motions of canonical shock–boundary layer interaction units. A node in the vicinity of reattachment location that originated the pressure perturbations within the separation shock was also identified, which further ascertained that the leading shock low frequency motions were driven by the separation bubble pulsations. Above a threshold back pressure, additional peaks appeared at distinct higher frequencies that resembled the acoustic modes within the duct. However, none of the earlier expressions of the resonance acoustic frequency within a straight duct agreed with the experimentally observed value. Cross-spectral analyses suggested that these modes were caused by the shock interactions with upstream propagating acoustic waves that emanate from the reattachment location, originally proposed for transonic diffusers by Robinet & Casalis (2001) Phys. Fluids 13, 1047–1059. Feedback interactions described using one-dimensional stability analysis of the shock perturbations by obliquely travelling acoustic waves (Robinet & Casalis 2001 Phys. Fluids 13, 1047–1059) made favourable comparisons on the back pressure threshold that emanated the acoustic modes as well as the acoustic mode frequencies.

Particle suspensions in confined geometries can become clogged, which can have a catastrophic effect on function in biological and industrial systems. Here, we investigate the macroscopic dynamics of dense suspensions in constricted geometries. We develop a minimal continuum two-phase model that allows for variation in particle volume fraction. The model comprises a ‘wet solid’ phase with material properties dependent on the particle volume fraction, and a seepage Darcy flow of fluid through the particles. We find that spatially varying geometry (or material properties) can induce emergent heterogeneity in the particle fraction and trigger the abrupt transition to a high-particle-fraction ‘clogged’ state.

Drops in a shear flow experience shear-induced diffusion due to drop–drop interactions. Here, the effects of medium viscoelasticity on shear-induced collective diffusivity are numerically investigated. A layer of viscous drops suspended in a viscoelastic fluid was simulated, fully resolving each deforming drop using a front-tracking method. The collective diffusivity is computed from the spreading of the drop layer with time, specifically a one-third scaling, as well as using an exponentially decaying dynamic structure factor of the system of drops. Both methods led to matching results. The surrounding viscoelasticity was shown to linearly reduce the diffusion-led spreading of the drop layer, the effect being stronger for less deformable drops (low capillary number). Because of the competition between the increasing effect with capillary number (Ca) and the decreasing effect with Weissenberg number (Wi), collective diffusivity vanishes at very low Ca and high enough Wi. The physics behind the hindering effects of viscoelasticity on shear-induced diffusion is explained with the help of drop–drop interactions in a viscoelastic fluid, where shear-induced interaction leads to trapping of drops into tumbling trajectories at lower Ca and higher Wi due to viscoelastic stresses. Using the simulated values, phenomenological correlations relating the shear-induced gradient diffusivity with Wi and Ca were found.

In recent years, integrating physical constraints within deep neural networks has emerged as an effective approach for expediting direct numerical simulations in two-phase flow. This paper introduces physics-informed neural networks (PINNs) that utilise the phase-field method to model three-dimensional two-phase flows. We present a fully connected neural network architecture with residual blocks and spatial parallel training using the overlapping domain decomposition method across multiple graphics processing units to enhance the accuracy and computational efficiency of PINNs for the phase-field method (PF-PINNs). The proposed PINNs framework is applied to a bubble rising scenario in a three-dimensional infinite water tank to quantitatively assess the performance of PF-PINNs. Furthermore, the computational cost and parallel efficiency of the proposed method was evaluated, demonstrating its potential for widespread application in complex training environments.

In this study, the statistical properties and formation mechanisms of particle clusters that consider the influence of particle–wall interactions in particle-laden wall turbulence are systematically investigated through wind tunnel experiments. In the experiments, two particle release modes, including particle top-releasing mode (Case 1) and particle locally laying mode (Case 2), were adopted to establish varying conditions with different particle–wall interaction strengths. The Voronoï diagram method was employed to identify the particle clusters, and the impact of particle–wall interactions on the characteristics of the clusters was analysed. The results indicate that particle–wall interaction is the predominant factor in the formation of particle clusters in the near-wall region. Under Case 1 and Case 2, the maximum concentration of particles in the clusters could reach nearly five times the average particle concentration; however, the clusters with large particle numbers ($N_C\gt 5$) in Case 1 tended to form near the wall and the vertical velocities of these clusters were greater than the average velocities of all particles. In contrast, under Case 2, clusters with large particle numbers exhibited a higher probability of occurrence further from the wall and the vertical velocities of these clusters were lower than the average velocity of all particles. Furthermore, this study found that the presence of particle clusters in these flows significantly alters the flow field properties surrounding them, implying that a region of high strain and low vorticity constitutes an essential but non-sufficient condition for the generation of particle clusters in wall turbulence.

We theoretically investigate the small-amplitude broadside oscillations of an annular disk within an unbounded fluid domain. Specifically, we formulate a semi-analytical framework to examine the effects of the oscillation frequency and pore radius on the disk’s added mass and damping coefficients. By leveraging the superposition principle, we decompose the complex original problem into two simpler ones. The force exerted on the disk by the fluid is linked to the solutions of these sub-problems through the reciprocal theorem; the first solution is readily available, while the second is derived asymptotically, assuming a small inner radius. Both solutions are evaluated by transforming dual integral equations into systems of algebraic equations, which are then solved numerically. Building on these solutions, we extract asymptotic expressions for the variations of the quantities of interest in the limits of low and high oscillatory Reynolds numbers. Notably, at high frequencies, we uncover a previously overlooked logarithmic term in the force coefficient expansions, absent in prior scaling analyses of oscillating solid (impermeable) disks. Our findings indicate that, when viscosity plays a dominant role, an annular (porous) disk behaves similarly to a solid one, with reductions in the force coefficients scaling with the cube of the pore radius. We also discover, perhaps surprisingly, that, as inertial effects intensify, the damping coefficient initially increases with the pore radius, reaches a maximum and subsequently declines as the disk’s inner hole enlarges further; at its peak, it can exceed the value for an equivalent solid disk by up to approximately 62 % in the asymptotic limit of extremely high oscillatory Reynolds number. Conversely, the added mass coefficient decreases monotonically with increasing porosity. The decay rate of the added mass in the inertial regime initially scales with the cube of the pore radius before transitioning to linear scaling when the pore radius is no longer extremely small. Although our analysis assumes a small pore radius, direct numerical simulations confirm that our asymptotic formulation remains accurate for inner-to-outer radius ratios up to at least $1/2$.

The shape of a free-surface slump of viscoplastic material supported by an oblique barrier on an inclined plane is investigated theoretically and experimentally. The barrier is sufficiently tall that it is not surmounted by the viscoplastic fluid, and a focus of this study is the largest volume of rigid viscoplastic fluid that can be supported upstream of it. A lubrication model is integrated numerically to determine the transient flow as the maximal rigid shape is approached. Away from the region supported by the barrier, the viscoplastic layer attains a uniform thickness in which the gravitational stresses are in balance with the yield stress of the material. However, closer to the barrier, the layer thickens and the barrier bears the additional gravitational loading. An exact solution for the rigid shape of the viscoplastic material is constructed from the steady force balance and computed by integrating Charpit’s equations along characteristics that emanate from the barrier wall. The characteristics represent the late-time streamlines of the flow as it approaches the rigid shape. The exact solution depends on a single dimensionless group, which incorporates the slope inclination, the barrier width and the fluid’s yield stress. It is shown that the shape is insensitive to the transient flow from which it originates. The force exerted by the slump is calculated for different barrier shapes. The results of new laboratory experiments are reported; these show that although convergence to the final rigid state is slow, there is good agreement with the experimental measurements at long times.

A block of ice in a box heated from below and cooled from above can (partially) melt. Vice versa, a box of water with less heating from below or more cooling from above can (partially) re-solidify. This study investigates the asymmetric behaviours between such melting and freezing processes in this Rayleigh–Bénard geometry, focusing on differences in equilibrium flow structures, solid–liquid interface morphology, and equilibrium mean interface height. Our findings reveal a robust asymmetry across a range of Rayleigh numbers and top cooling temperature (i.e. hysteretic behaviour), where the evolution of freezing shows a unique ‘splitting event’ of convection cells that leads to a non-monotonic height evolution trend. To characterise the differences between melting and freezing, we introduce an effective Rayleigh number and the aspect ratio for the cellular structures, and apply the heat flux balance and the Grossmann–Lohse theory. Based on this, we develop a unifying model for the melting and freezing behaviour across various conditions, accurately predicting equilibrium states for both phase-change processes. This work provides insights into the role of convective dynamics in phase-change symmetry-breaking, offering a framework applicable to diverse systems involving melting and freezing.

Recent studies reveal the central role of chaotic advection in controlling pore-scale processes including solute mixing and dispersion, chemical reactions, and biological activity. These dynamics have been observed in porous media (PM) with a continuous solid phase (such as porous networks) and PM comprising discrete elements (such as granular matter). However, a unified theory of chaotic advection across these continuous and discrete classes of PM is lacking. Key outstanding questions include: (i) topological unification of discrete and continuous PM; (ii) the impact of the non-smooth geometry of discrete PM; (iii) how exponential stretching arises at contact points in discrete PM; (iv) how fluid folding arises in continuous PM; (v) the impact of discontinuous mixing in continuous PM; and (vi) generalised models for the Lyapunov exponent in both PM classes. We address these questions via a unified theory of pore-scale chaotic advection. We show that fluid stretching and folding (SF) in discrete and continuous PM arise via the topological complexity of the medium. Mixing in continuous PM manifests as discontinuous mixing through a combination of SF and cutting and shuffling (CS) actions, but the rate of mixing is governed by SF only. Conversely, discrete PM involves SF motions only. These mechanisms are unified by showing that continuous PM is analogous to discrete PM with smooth, finite contacts. This unified theory provides insights into the pore-scale chaotic advection across a broad class of porous materials and points to design of novel porous architectures with tuneable mixing and transport properties.

Axisymmetric turbulent boundary layers are of great significance in industry and the fluid dynamics community. In this paper, direct numerical simulations of an axially developing axisymmetric turbulent boundary layer along a slender cylinder are performed. Periodical suction and blowing perturbation are used to trigger the transition from laminar inflow to turbulent flow downstream, resulting in the boundary layer thickness varying from 7 to 13 times the cylinder radius, and the friction Reynolds number varying from 300 to 510. Turbulence statistics including wall friction coefficient, mean velocity profile and Reynolds stresses are obtained. The turbulence intensities are weakened compared with the planar turbulent layer, and the inter-component energy transfer is also inhibited. A curvature-weighted transformation is proposed, and the transformed Reynolds stresses and mean velocity deficit collapse well with the planar case in the near-wall region. The velocity streaks and vortical structures are explored. The wall-normal variation of the mean spanwise spacing of low-speed streaks is greatly influenced by the cylindrical geometry. Quasi-streamwise vortices dominate the near-wall region, and the arch vortices are prevalent in the outer region. The prograde hairpin vortices can be commonly observed.

The reactivity of transverse waves in detonations of methane, oxygen and nitrogen are experimentally assessed using MHz rate schlieren and chemiluminescence imaging. In these highly unstable mixtures, the mode of wave propagation is more complex than what is described by the cellular instability model that is conventionally used for weakly unstable mixtures. Behind the low-speed leading shock in unstable waves, the processed gas remains essentially unreacted until transverse waves reach this region. In highly unstable waves, the transverse waves have a range of reactivity, that is rates of reaction in the flow immediately behind the wave. In this study, we present examples of transverse waves for near-limit detonations and analyse four cases in detail. In some cases, these waves appear to be essentially non-reactive or cause very slow reaction. In other cases, the transverse waves can be highly reactive. In the most extreme example, the transverse wave is propagating at the Chapman–Jouguet speed with a small reaction zone, i.e. a transverse detonation. A reactive oblique shock model is used to approximate the triple-point configuration of this case as a double-Mach reflection, which shows good agreement with the images. The reaction evolution along path lines is analysed using detailed reaction mechanisms and considerations about flow-field unsteadiness. Length scales of the energy release and expansion processes within the reaction zone region are used to explain the observed modes of wave propagation and interaction.

This paper presents an experimental application of reactive control to jet installation noise based on destructive interference. The work is motivated by the success of previous studies in applying this control approach to mixing layers (Sasaki et al. Theor. 2018b Comput. Fluid Dyn. 32, 765–788), boundary layers (Brito et al. 2021 Exp. Fluids 62, 1–13; Audiffred et al. 2023 Phys. Rev. Fluids 8, 073902), flow over a backward-facing step (Martini et al. 2022 J. Fluid Mech. 937, A19) and, more recently, to turbulent jets (Maia et al. 2021 Phys. Rev. Fluids 6, 123901; Maia et al. 2022 Phys. Rev. Fluids 7, 033903; Audiffred et al. 2024b J. Fluid Mech. 994, A15). We exploit the fact that jet–surface interaction noise is underpinned by wavepackets that can be modelled in a linear framework and develop a linear control strategy where piezoelectric actuators situated at the edge of a scattering surface are driven in real time by sensor measurements in the near field of the jet, the objective being to reduce noise radiated in the acoustic field. The control mechanism involves imposition of an anti-dipole at the trailing edge to cancel the scattering dipole that arises due to an incident wavepacket perturbation. We explore two different control strategies: (i) the inverse feed-forward approach, where causality is imposed by truncating the control kernel, and (ii) the Wiener–Hopf approach, where causality is optimally enforced in building the control kernel. We show that the Wiener–Hopf approach has better performance than that obtained using the truncated inverse feed-forward kernel. We also explore different positions of the near-field sensors and show that control performance is better for sensors installed for streamwise positions downstream in the jet plume, where the signature of hydrodynamic wavepacket is better captured by the sensors. Broadband noise reductions of up to 50 % are achieved.

Understanding how bubbles on a substrate respond to ultrasound is crucial for applications from industrial cleaning to biomedical treatments. Under ultrasonic excitation, bubbles can undergo shape deformations due to Faraday instability, periodically producing high-speed jets that may cause damage. While recent studies have begun to elucidate this behaviour for free bubbles, the dynamics of wall-attached bubbles is still largely unexplored. In particular, the selection and evolution of non-spherical modes in these bounded systems have not previously been resolved in three dimensions, and the resulting jetting dynamics has yet to be compared with that observed in free bubbles. In this study, we investigate individual micrometric air bubbles in contact with a rigid substrate and subjected to ultrasound. We introduce a novel dual-view imaging technique that combines top-view bright-field microscopy with side-view phase-contrast X-ray imaging, enabling visualisation of bubble shape evolution from two orthogonal perspectives. This set-up reveals the progression of bubble shape through four distinct dynamic regimes: purely spherical oscillations, onset of harmonic axisymmetric meniscus waves, emergence of half-harmonic axisymmetric Faraday waves and the superposition of half-harmonic sectoral Faraday waves. This stepwise evolution contrasts with the behaviour of free bubbles, which exhibit their ultimate Faraday wave pattern immediately upon instability onset. For the substrate chosen, the resulting shape-mode spectrum appears to be degenerate and exhibits a continuous range of shape mode degrees, in line with our theoretical predictions derived from kinematic arguments. While free bubbles also display a degenerate spectrum, their shape mode degrees remain discrete, constrained by the bubble spherical periodicity. Experimentally measured ultrasound pressure thresholds for the onset of Faraday instability agree well with classical interface stability theory, modified to incorporate the effects of a rigid boundary. Complementary three-dimensional boundary element simulations of bubble shape evolution align closely with experimental observations, validating this method’s predictive capability. Finally, we determine the acceleration threshold at which shape mode lobes initiate cyclic jetting. Unlike free bubbles, jetting in wall-attached bubbles consistently emerges from the side not restricted by the substrate.

Accurately predicting the mean flow properties of wall-bounded turbulence is essential for both fundamental research and engineering applications. In atmospheric boundary layers, the mean flow within the surface layer is typically described by Monin–Obukhov similarity theory (MOST). However, beyond the surface layer, MOST no longer applies as the Coriolis effect becomes significant. To address this issue, this study introduces a novel analytical model for the mean turbulent momentum fluxes and geostrophic wind deficits in nocturnal stable atmospheric boundary layers (NSBLs), which are stably stratified near the surface and transition to neutrally stratified flow above. The model solutions are derived from the Ekman equations using the eddy viscosity approach and a new parametrisation of the flux Richardson number. The solutions show that the geostrophic wind deficits scale with $u_*^2/(hf)$, where $u_*$ is the friction velocity, $h$ is the boundary layer height, and $f$ is the Coriolis parameter. The model’s predictions align closely with recent large-eddy simulation studies, confirming the model’s accuracy. Combined with the geostrophic drag law, the model can reliably predict the wind speed profile above the surface layer of NSBLs. This work marks a significant step in modelling atmospheric turbulence and its fundamental dynamics.

We investigate the effects of external harmonic forcing on flow through a duct with square cross-section containing two circular orifice plates – a double-orifice cavity – at an operating condition where self-sustained limit cycle oscillations are observed. When the oscillatory flow is periodically forced at a frequency $f_f$ near its natural frequency $f_n$ ($0.9\leqslant f_f /f_n \leqslant 1.1$), it undergoes lock-in and amplitude suppression through synchronous quenching. We observe phase-drifting (or phase-slipping) prior to lock-in that happens via a saddle-node bifurcation. However, when the flow system is forced far from its natural frequency ($0.8\leqslant f_f /f_n\leqslant 0.9$ and $1.1\leqslant f_f /f_n\leqslant 1.4$) lock-in happens via asynchronous quenching through a Neimark–Sacker bifurcation (torus death). In asynchronous quenching, phase-drifting and phase-trapping are observed before lock-in. An asymmetry is present in the synchronization map on forcing either side of the natural frequency, which becomes more pronounced in the asynchronous quenching regime. There is also an observed saturation of the synchronization map for $f_f/f_n\gt 1$ over the range of frequencies explored. Subharmonic synchronization or $1:2$ lock-in with period-two oscillations is also observed when the system is forced near $f_n/2$ ($ 0.49 \leqslant f_f /f_n \leqslant 0.51$). The route to lock-in consists of a three frequency regime where subharmonics of the forcing frequency ($f_f/2$ and $f_f/3$) play an important role in the dynamics. The transition from $1:1$ to $1:2$ lock-in occurs via a de-lock-in regime ($ 0.55 \leqslant f_f /f_n \leqslant 0.65$), where a lock-in boundary is present; i.e. the system delocks after lock-in if the amplitude is raised beyond a critical value. The de-lock-in regime is also characterized by a nonlinear phase drift after de-lock-in and a significant jump in the forcing amplitude for lock-in for $f_f/f_n=0.6$. Amplification is observed for $f_f/f_n\gt 1$ and also in the $1:2$ lock-in and de-lock-in regimes where the total signal power exceeds the unforced system’s power for small increases in forcing amplitude after lock-in. Based on these results, we identify the asynchronous quenching regime for $f_f/f_n\lt 1$ as the optimal frequency range where active control is most effective. Finally, we introduce a reduced-order phenomenological model based on vortex–acoustic interaction dynamics from first principles. The model correctly identifies the four regimes, their dynamics leading to lock-in, and asymmetry and saturation in the synchronization map.

The pressure effects on the mixing fields of non-reacting and reacting jets in cross-flow are studied using large eddy simulation (LES). A hydrogen jet diluted with 30 % helium is injected perpendicularly into a cross-stream of air at four different pressures: 1, 4, 7 and 15 bar. The resulting interaction and the mixing fields under non-reacting and reacting conditions are simulated using LES. The subgrid scale combustion is modelled using a revised flamelet model for the partially premixed combustion. Good agreement of computed and measured velocity fields for reacting and non-reacting conditions is observed. Under non-reacting conditions, the mixing field shows no sensitivity to the pressure, whereas notable changes are observed for reacting conditions. The lifted flame at 1 bar moves upstream and attaches to the nozzle as the pressure is increased to 4 bar and remains so for the other elevated pressures because of the increasing burning mass flux with pressure. This attached flame suppresses the fuel–air mixing in the near-nozzle region. The premixed and non-premixed contributions to the overall heat release in the partially premixed combustion are analysed. The non-premixed contribution is generally low and occurs in the near-field region of the fuel jet through fuel-rich mixtures in the shear layer regions, and decreases substantially further with the increase in pressure. Hence, the predominant contributions are observed to come from premixed modes and these contributions increase with pressure.

Several million years of natural evolution have endowed marine animals with high flexibility and mobility. A key factor in this achievement is their ability to modulate stiffness during swimming. However, an unresolved puzzle remains regarding how muscles modulate stiffness, and the implications of this capability for achieving high swimming efficiency. Inspired by this, we proposed a self-propulsor model that employs a parabolic stiffness-tuning strategy, emulating the muscle tensioning observed in biological counterparts. Furthermore, efforts have been directed towards developing the nonlinear vortex sheet method, specifically designed to address nonlinear fluid–structure coupling problems. This work aims to analyse how and why nonlinear tunable stiffness influences swimming performance. Numerical results demonstrate that swimmers with nonlinear tunable stiffness can double their speed and efficiency across nearly the entire frequency range. Additionally, our findings reveal that high-efficiency biomimetic propulsion originates from snap-through instability, which facilitates the emergence of quasi-quadrilateral swimming patterns and enhances vortex strength. Moreover, this study examines the influence of nonlinear stiffness on swimming performance, providing valuable insights into the optimisation of next-generation, high-performance, fish-inspired robotic systems.