Fluid flow-induced deformation dynamics under biaxial loading in anisotropic (layered) analogues of unconventional reservoirs: insights from physical model experiments and numerical simulations Authors: P N R L Sudhishna, Tridib Kumar Mondal, Sourav Mondal Published: 2026 ABSTRACT AbstractUnderstanding fluid-driven fracture propagation in anisotropic, layered geological formations is critical for optimizing resource recovery in unconventional reservoirs. In this study, we present a novel experimental and numerical investigation into fluid flow-induced deformation dynamics under biaxial loading using analogue models of shale and reservoir rock. A custom-designed low-pressure biaxial compression system is developed to simulate realistic subsurface conditions while allowing real-time, high-resolution visualization of fracture evolution in low-modulus elastic and viscoelastic materials. Gelatin-based analogues, representing both isotropic and anisotropic formations, are employed to mimic geological heterogeneity and elasticity. Fracturing experiments are conducted using water, SAE 140 oil, and oil–toluene mixtures injected at a constant rate of 1 ml/min under controlled loading conditions (confining load: 1.23 bar; top load: 0.52 bar). The evolving fracture geometries are analysed using ImageJ, focusing on metrics such as aspect ratio and fracture length. Numerical simulations based on a phase field damage model are performed to predict fracture initiation and propagation, showing strong agreement with experimental results and qualitative consistency with the analytical Khristianovic–Geertsma–de Klerk (KGD) model. Our integrated approach provides critical insights into the role of fluid viscosity, heterogeneity and loading conditions on fracture dynamics. The experimental setup also supports studies relevant to CO₂ sequestration and gas hydrate dissociation. This work advances the mechanistic understanding of fracture processes in layered formations, offering insights for field-scale hydraulic fracturing and sustainable subsurface resource management.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-55767-4.Keywords: Biaxial loading, Layered heterogeneous media, Fracturing, Phase field damage, Unconventional reservoirSubject terms: Energy science and technology, Engineering, Solid Earth sciences Introduction Shale rocks with high porosity, often contain significant hydrocarbons; however, their extraction is challenging due to their low permeability1. Understanding fracture propagation and fluid flow behaviour in such media is therefore critical, as enhanced recovery often relies on induced fractures to improve permeability. Hydraulic fracturing has become the primary technique to enhance and recover oil and gas, particularly in unconventional shale reservoirs2,3. Beyond oil and gas recovery, hydrofracturing is also applied in subsurface applications such as geothermal energy extraction and nuclear waste disposal under confined stress conditions. In geothermal energy systems, rock permeability is increased in hot, dry rocks and low-permeability formations to facilitate heat extraction4,5. Similarly, in the context of CO2 sequestration, the hydraulic fracturing technique studies are used to examine fracture evolution in host rocks, ensuring long-term containment and stability of deep geological repositories6,7. The propagation of fluid induced fractures under various loading conditions such as uniaxial, biaxial, and triaxial stress, is fundamental in understanding the deformation of rock and fracture evolution8,9. Recent studies are mostly based on laboratory investigations, have primarily focused on high-modulus materials such as sandstone, concrete blocks, and resin specimens of varying sizes under a biaxial compression test apparatus10–14. Similarly, several authors have investigated fracture growth in a sandstone and resin samples using a conventional triaxial compression system integrated with Computer Tomography (CT) scanning equipment and biaxial loading frames equipped with X-ray transmission windows to visualize the 3D fracture formation over experimental time15–17. However, a major research gap persists regarding the continuous monitoring of fracture propagation or formation evolution over time, particularly in low modulus or heterogeneous mediums. Different fracture morphologies, such as wing fractures, nonlinear fractures, and disk-shaped fractures, commonly form under biaxial compression across both homogeneous and layered rock samples, regardless of host rock type11,13,14,18. With increasing load, nonlinear fractures typically appear on the surfaces of layered blocks categorized into tensile opening in intact rock, shear and tensile fractures along pre-existing joints, and mixed-mode fractures resulting from combined tensile and shear stresses19. Extensive experimental and numerical research has addressed three-dimensional fracture growth under biaxial loading. Validation of observed fracture patterns in lab and field studies has often relied on numerical modelling methods such as finite element and boundary element methods20,21, extended finite element method (XFEM), finite element (FE) techniques13,22,23, and the Cohesive Zone Method (CZM)24,25. Each of these approaches has specific advantages and limitations in capturing 3D fracture propagation. The Finite Element Method (FEM) and the Extended Finite Element Method (XFEM) provide robust frameworks for simulating complex 3D fracture geometries, but their application to 3D domains requires significant computational resources. However, Boundary element methods are well known for their ability to simulate large-scale models and computational efficiency, but are less suitable for solving heterogeneous materials and non-linear constitutive behaviour. The Cohesive Zone Method (CZM) provides detailed insights into fracture zones and initiation mechanisms but requires predefined crack paths and may become computationally demanding in 3D applications. Recent phase field model approaches have significantly improved the ability to simulate hydraulic fracture propagation in complex geological formations. From previous studies, it is evident that the authors have investigated the interactions of fluid fracture interface interactions in the poroelastic homogeneous rock using a variational phase field approach54,56; meanwhile, others, have analysed fracture initiation, rock failure behaviour, and multi-layer propagation under anisotropic stress fields55. These numerical simulations will provide the energy-based fracture criteria and preferentially propagated fracture pathways; however, they rely on assumptions about material interface properties and the evolution of fracture toughness58–60. By comparing numerical simulations with the analogue experiments in the present framework, the experiments provide an opportunity to directly visualize the evolution and growth of fluid fracture in a heterogeneous medium under controlled biaxial stress conditions. Thus, these observations will provide a physical validation and complementary insights for future phase field simulations. The layered gelatin analogue medium is structurally heterogeneous, consisting of isotropic layers (each homogeneous) with contrasting Young’s moduli, rather than being intrinsically anisotropic like shale. The current study offers significant advantages for modelling hydraulic fracture propagation in the matrix, particularly its ability to capture fracture initiation and complex propagation behaviour in phase field modelling. In this framework, fracture evolution is implicitly represented by a scalar damage field, which is governed by a characteristic length scale parameter (lint) and a fracture aperture. The direct measurement of fracture aperture, mesh resolution, and a parameter that is particularly, sensitive to permeability can be complicated by diffuse representation, leading to uncertainty in aperture-dependent flow calculations. These challenges have been addressed in this study through the adopted modelling strategy, which balances numerical robustness and physical realism in simulating coupled solid-fluid interactions. However, these models failed to predict the complex fracture behaviour or pattern observed in experiments. The phase field model has recently been adopted to simulate fracture initiation and propagation under applied stress conditions. This approach effectively captures fracture evolution even in complex media such as laminated layered formations or sandstone samples from unconventional reservoirs, showing improved alignment with experimental observations18,26–28. Despite advances, a significant research gap remains in the real-time monitoring of fluid-driven fractures under biaxial compression at laboratory scale. Most prior work has focused on analysing the final fracture pattern rather than the temporal evolution of fractures. To address these limitations, the present study introduces a novel, indigenous low-pressure biaxial loading system specifically designed to investigate the dynamic evolution and capable of real-time visualization of three-dimensional fluid-driven fractures in analogue materials mimicking geological media. While previous experiments or relevant studies predominantly relied on homogeneous rock samples under uniaxial or triaxial confinement, but the developed biaxial system can operate under both uniaxial and biaxial loading, enabling us to systematically isolate the coupled effects of stress non-uniformity and material heterogeneity on fracture initiation and penetration arrest, deviation of the fractures across the layers. Thus, the system accommodates low Young’s modulus samples (< 1 MPa), including plastics, polymers, resin, and viscoelastic materials. Gelatin is chosen as the analogue medium in the present work, due to its excellent transparency and mechanical properties, closely replicating geological formations, enabling clear visualization of internal layering and facilitates the observation of fracture initiation, shape, and propagation direction during deformation. Different gelatin composition (contrasting Young’s modulus) is used for different layers (creating a gradient or inducing heterogeneity in the material properties). Furthermore, by fluid injection with varied viscosities (water, oil, and oil-toluene mixtures for creating the hydro fracture), further help in resolving fluid-solid coupling effects from interface-controlled mechanics. This approach reveals the previously, unresolved mechanisms such as material heterogeneity, deviation of fracture at the interface, layer interface reactivation under low modulus or toughness conditions for hydro fracture evolution. In contrast to numerical phase field approaches, which requires the calibration of fracture energy and regularization parameters like, length scale parameter (lint), the present framework, provides controlled experimental evidence to evaluate fracture transition mechanism and offers quantitative validation for modelling approaches. So, the novel contributions of this study are: (i)Direct and real-time visualization of the three-dimensional evolution of fluid fractures over time in a layered elastic analogue medium subjected to confined biaxial loading conditions, Direct and real-time visualization of the three-dimensional evolution of fluid fractures over time in a layered elastic analogue medium subjected to confined biaxial loading conditions, (ii)Systematic investigation of the coupled role of elastic heterogeneity, modulus contrast, and fluid viscosity effects influence fracture behaviour, Systematic investigation of the coupled role of elastic heterogeneity, modulus contrast, and fluid viscosity effects influence fracture behaviour, (iii)Physical insight into fracture penetration and deviation transitions across the layered domains. The present study demonstrates that fluid viscosity and elastic heterogeneity jointly control fracture propagation, with intermediate and high-viscosity fluids promoting stable and controlled fracture growth through layered media, under low-pressure conditions. The developed phase field models exhibited strong predictive capability, with deviations of only 7–15%, compared to experimental observations. By integrating experimental observations with numerical modelling, this work provides insights into fluid-driven fractures in layered analogue media. Moreover, the developed system has broad applicability across diverse fields, including carbon dioxide sequestration, oil and gas exploration, geothermal energy, wastewater management, petroleum engineering, and research and development activities. Physical insight into fracture penetration and deviation transitions across the layered domains. The present study demonstrates that fluid viscosity and elastic heterogeneity jointly control fracture propagation, with intermediate and high-viscosity fluids promoting stable and controlled fracture growth through layered media, under low-pressure conditions. The developed phase field models exhibited strong predictive capability, with deviations of only 7–15%, compared to experimental observations. By integrating experimental observations with numerical modelling, this work provides insights into fluid-driven fractures in layered analogue media. Moreover, the developed system has broad applicability across diverse fields, including carbon dioxide sequestration, oil and gas exploration, geothermal energy, wastewater management, petroleum engineering, and research and development activities. Research methods Mathematical background The phase field approach describes the evolution of the fluid fracture and propagation, which is obtained from Griffiths’ fracture theory. The mathematical theory of the deformation and damage phenomena on account of the hydrodynamic stress is inspired and adapted from the work of Chen et al.29. Griffiths’ theory for hydraulic fractures Let Ω ℝd (d {1,2}) denote a bounded domain with an external boundary and an internal boundary Γ representing the fracture interface (Fig. S-1a, supplementary material)). We assume that the intrinsic length scale (lint) parameter of the phase field is significantly larger than the fracture size; and the solid is a linear elastic material (Fig. S-1a, supplementary material)). The displacement of the domain Ω at time t is denoted by u (x, t) ℝd, where x is the position vector. The displacement field satisfies both the Dirichlet boundary conditions, ui (x, t) = gi(x,t), on and the Neumann conditions on . The body force b(x, t) ϵ ℝd acting on body Ω, is contributed by the other mechanical or thermal stress. The boundary is subjected to load f(x, t). In the fracture theory proposed by Griffith’s and later formalized by Francfort and Marigo (1998)30, for fracture initiation and propagation, it is considered that the energy required to create a fracture per unit area is equal to the critical fracture energy density (known also as critical energy release rate). The total potential energy for a deforming porous medium can be defined by Eq. 131, where, p: Ω x [0, T]ℝ is the fluid pressure, α is the Biot coefficient and is the strain tensor. The elastic energy density in a linear elastic solid and the strain tensor is defined by Eqs. 2 and (3), where, are the Lame constants. Phase field theory of fracture energy The phase field scalar variable is defined to approximate the fracture surface Γ, as shown in Fig. 1b, where indicates a fractured region, and no fractured region. The fracture surface density function through the phase field model can be defined as32, (a) Stress-Strain profile of various gelatin concentration samples; (b) rheology profile of the injected oil at different temperatures. where lint is the characteristic length scale, that governs the width of the diffused phase field transition region, and it is a dimensional quantity. In this study, lint is set to 2 mm, which also influences the regularized shape of the fracture. In the present work, the phase field parameters, such as fracture energy and length scale parameter , are considered same across all layers. This approach isolates the effects of elastic modulus contrast on fracture deviation and penetration, and to avoid over parameterization, specifically because experimentally measured fracture toughness values for each gelatin layer are not available. So, the performed sensitivity analysis in this study, confirms that this assumption does not affect the qualitative fracture behaviour reported in supplementary (Fig.S2). Now the fracture energy Eq. 1 is calculated by using Eq. 4 and represented by Eqs. 5, The evolution of the phase field is governed by the variational approach based on the elastic energy associated with the fracture. To ensure that fractures occur only under tensile stress, the elastic energy density has been split into tensile and compressive components. As a result, fracture propagation is predominantly driven by the tensile stress, while compressive stresses have no influence on the fracture growth process. Typically, the elastic energy density is governed by tensile stress and affected by the phase field. Accordingly, the total elastic energy density, as shown in Eq. (2), can be expressed in Eq. 6. where k is a model parameter and prevents the positive part of the elastic energy density from disappearing as well as restricts numerical singularity when the phase field . Furthermore, the model parameter k should satisfy . Incorporating these expressions in the Griffith’s equation, total energy is represented by Eqs. 7, By applying the first variational approach to the Eq. 7, energy function (E) achieves a minimum (~ 0), the following set of governing equations are obtained which are subjected to the initiation and propagation of crack for and time . For the rate dependent fracture study, a viscous regularization damage function is introduced to the phase field and is represented in the energy Eqs. 8 and 9. where and is damage evolution function and is the characteristic time scale. The Cauchy stress tensor can be defined as, where I is the unit tensor, . The variational method emphasises that the irreversibility requirement has to ensure that fracture cannot be restored under both loading and unloading conditions. A new field is introduced known as history-strain field H(x,t), as For compression and unloading conditions, the history field H to satisfy the Kuhn-Tucker, By replacing by in Eq. 9, the strong form is attained, Besides, the zeroth order variation of the functional E fulfils the Neumann boundary conditions, where, is the outward normal vector. The initial and boundary conditions to be satisfied are defined below, For the initial crack or pre-existing fractures, the initial phase field value (ϕ) = 1 is set. Governing equations for the flow field The domain Ω is decomposed into three distinct subregions based on the phase field variable for the fluid flow which is categorized into three parts: undamaged area damaged area (fracture) and transition area . This classification is defined using phase field of two predefined threshold values and , where represents the low value of phase field where the material is considered to be undamaged, meanwhile, stands for high values of phase field, above which the material is considered fully fractured. If (where, , then the subarea is known as the undamaged area and for the damaged area . Also, for the transition area. The fluid flow medium within the deformed solid matrix is governed by the linear momentum conservation. For the undamaged area (reservoir domain) , the law of mass conservation is expressed by, where, is the density of the fracturing fluid, is the storage coefficient, v is the fracturing fluid velocity, and is the source term. The volumetric strain is defined as, . The fluid density and biot coefficient can be expressed as here and represents the fluid densities and biot coefficients in the undamaged and damaged areas. The storage coefficient can be expressed as, where , c, and are referred to as porosity, fluid compressibility, and bulk modulus. The fluid compressibility factor c can be defined as , and denotes reservoir and fractured domain fluid compressibilities. For undamaged area . Similarly, for fracture domain . Now, for the whole domain, Darcy’s law equation is valid and then Darcy’s velocity( v) can be described by Where, is the effective permeability, which can be expressed as, and the is the fluid viscosity. Now, flow field expression can be expressed in terms of fluid pressure, as shown in Eq. 20, The permeability values reported here were chosen from previous studies34 a range of values provide a correct order of magnitude for soft poroelastic materials such as gelatin and hydrogel analogue materials, as specific concentration and layer data for the present analogue samples are not available. The permeability in this framework is iteratively calibrated for the range of values m2 presented in Table 1. As, phase field formulation produces a diffuse fracture representation, the physical aperture cannot be uniquely resolved with the internal length scale parameter; therefore, permeability cannot be applied using as aperture based cubic-law. Similarly, in accordance with the standard practice, hydraulic fracture propagation through phase field modelling, matrix and fracture zone permeability is selected from literature-supported ranges are used and iteratively adjusted to match the experimentally observed fracture evolution, initiation time and propagation behaviour. Thus, the numerical results provide insights into stress redistribution, pressure induced loading conditions, and stresses around the fracture-tip that are not directly observable in the laboratory setup through imaging process alone. Table 1Representative parameter values employed in the numerical simulations.ParametersDimensionsBox size (cm)10 cm x 10 cm x 10 cmNumber of layers1–5Layer height (cm)Box size/no. of layersPermeability (k) m2 34 KIC (fracture toughness)59 Pa.m0.535Young’s Modulus (E)50 kPa (5% gelatin)100 kPa (10% gelatin)Poisson’s ratio0.45 [36]Biot coefficient (α)0.8 Representative parameter values employed in the numerical simulations. m2 34 50 kPa (5% gelatin) 100 kPa (10% gelatin) Computational details In the present study, numerical simulations enhance experimental findings by expanding a wide range of parameters beyond lab-scale limitations and providing insights into the internal parameters, including stress distribution at the fracture tip and damage evolution, which are challenging to measure during laboratory experiments. This integration helps improve the reliability and applicability of findings by bridging the gap between laboratory and field-scale conditions. With this aim, numerical simulations are carried out using the phase field method implemented within the COMSOL Multiphysics v6.3® platform [License #16078497], utilizing the CFD, Structural Mechanics, and Nonlinear Structural Materials modules. A two-dimensional square domain measuring 0.1 m × 0.1 m is constructed (Fig. S-1, supplementary material), with a narrow inlet channel of dimensions 0.001 m × 0.004 m positioned centrally along the top boundary. The domain’s material properties, listed in Table 1, are used as input parameters for the simulation. Boundary conditions include vertical loading on the top surface (to represent lithostatic pressure) and horizontal confinement on the sides (simulating in-situ stress). The computational mesh comprises 30,000 quadrilateral elements, with a maximum element size of = 4 mm (Fig. S-1, supplementary material). No initial fractures are introduced in the domain. A mesh independency check is conducted by varying the maximum element size evaluating its effect on maximum fracture width and length over time as shown in Fig. S-1 (supplementary material). The simulations are particularly sensitive to the intrinsic length scale of the phase field model, denoted as , which governs the transition zone of damage. A sensitivity analysis for the model parameter , is performed and represented in Fig. S-2 (supplementary material). Following previous studies33, this parameter is set as /2. To accurately capture steep gradients—especially near the fluid injection point, where stress concentration is highest-a refined mesh is applied in that region. This mesh refinement is essential for enhancing solution accuracy. Non-uniform meshing is employed throughout the domain, a common strategy in numerical modelling to maintain low Courant numbers, promote solution stability, and ensure convergence of the simulation. Experimental design and procedure Materials In this study, a custom-designed biaxial loading apparatus is utilized to explore the progression of fractures within a soft medium. The experimental material consists of commercially available gelatin derived from porcine skin, possessing a bloom strength in the range of 200–225 g, chosen for its elasticity and high optical clarity. To enhance both the transparency and mechanical strength of the gelatin matrix, sucrose is incorporated into the mixture. For the fluid injection tests, two types of dyed fluids are prepared: SAE 140 oil (procured from Reliance Industries Limited) blended with an oil-soluble dye (LOBAChemie), and distilled water tinted with liquid food colouring (PAPILON brand). These fluids allow for clear visualization of fracture paths. A standard 18-gauge butterfly needle (Top brand-winged infusion set) with a rounded (flat) tip is employed as the injection port (with a depth of 0.4 cm, inserted into sample) to deliver the fluids into the gelatin block, enabling controlled initiation and monitoring of fracture development. Gelatin analogue Gelatin has been widely utilized in earlier studies as an analogue material for investigating fluid-induced fractures, dyke intrusions, and magma-driven crack propagation35,37–43. Scaling these laboratory analogue models to field conditions is achievable through appropriate mathematical modelling techniques. The mechanical behaviour of gelatin is characterized using a Universal Testing Machine (UTM) in accordance with ASTM D638 standards44. Stress–strain data obtained from these tests are employed to determine the Young’s modulus of the gelatin samples (Fig. 1a). The Poisson’s ratio value (reported in Table 1) of the gelatin used in this study is assumed to be 0.45as we are using gelatin powder made from porcine skin from previous literature35,36 and was not measured directly in our experiments. Generally, for gelatin materials, Poisson’s ratio can be taken from 0.45 to 0.5. It should also be noted that other parameters, such as fracture toughness and viscoelastic relaxation characteristics, which are important for capturing the evolution of fracture, were not experimentally determined here but referred from previous literature35,45. Future work will be focused on the measurement of these parameters to further strengthen the accuracy of the fracture propagation studies in gelatin-based analogue models. Sample preparation A solid elastic matrix for homogeneous and heterogeneous media is prepared using various gelatin, water, and sucrose concentrations. Initially, we prepare the gelatin solution by heating water to 60 °C and slowly adding gelatin powder. Once gelatin powder is thoroughly mixed in the water, we add sucrose to enhance its transparency and strength. The solution is poured into a 10 cm × 10 cm × 10 cm mold to form a layer of the desired height and is then stored at 5 °C for approximately 14 h. We continue the process with different concentrations of gelatin in the solution, resulting in varying stiffness for the solidified gelatin layers. Since the mechanical properties of gelatin are sensitive to temperature, experiments are conducted at a controlled laboratory temperature of 22–25 °C. Description of Biaxial Loading Setup In this study, we customized the experimental setup to better suit our research objectives. A schematic of the setup is shown in Fig. 2, illustrates the laboratory-scale biaxial loading system used to conduct hydraulic fracturing experiments aimed at investigating fracture dynamics in both homogeneous and anisotropic samples. The setup comprises three main components: the confinement subsystem, the top loading subsystem, and the fluid injection subsystem. (a) Schematic of biaxial loading system, developed indigenously for low pressure material fracture testing. (b) Photograph of the indigenously developed biaxial loading system in laboratory scale. Step 1: Confinement subsystem. We used a PMMA block of dimensions 24 cm x 24 cm x 20.5 cm to place the sample and observe the visibility of fractures formed inside the sample under loading conditions. Three parts, each with a different dimension, compose this block. We used a PMMA block of dimensions 24 cm x 24 cm x 20.5 cm to place the sample and observe the visibility of fractures formed inside the sample under loading conditions. Three parts, each with a different dimension, compose this block. We make sure the 12 cm x 12 cm block space accommodates the nylon plate, and then we place the hydraulic press with a piston on it and secure it with nuts and bolts, this component will serve as a load to the sample. We make sure the 12 cm x 12 cm block space accommodates the nylon plate, and then we place the hydraulic press with a piston on it and secure it with nuts and bolts, this component will serve as a load to the sample. Particularly, the sample size that can be considered in this setup is 10 cm x 10 cm x 10 cm; height of the sample is variable. This loading system receives air pressure from a compressor, connecting a T-junction to the outlet of the compressor. The other side is connected as an inlet to the block and a pressure gauge consisting of a sample piece. Particularly, the sample size that can be considered in this setup is 10 cm x 10 cm x 10 cm; height of the sample is variable. This loading system receives air pressure from a compressor, connecting a T-junction to the outlet of the compressor. The other side is connected as an inlet to the block and a pressure gauge consisting of a sample piece. A load sensor at the base of the PMMA block continuously monitors the applied load throughout the testing process. A load sensor at the base of the PMMA block continuously monitors the applied load throughout the testing process. Step 2: Fluid injection subsystem. The fluid injection subsystem consists of a syringe pump (Cole Palmer) with a steel syringe of 100 ml hold-up volume. It can be operated from very low flow rates (0.01 µl/min) to high flow rates (up to 50 ml/min). One end of the pressure gauge is connected to the syringe outlet, while the other end functions as the inlet to the piston, enabling the system to pump fluid at variable flow rates. The pressure gauge is used to monitor the pressure within the fluid line. At the desired flow rate, fluid is delivered through a needle attached to a nylon plate, which penetrates the sample under load to induce fractures. The fluid injection subsystem consists of a syringe pump (Cole Palmer) with a steel syringe of 100 ml hold-up volume. It can be operated from very low flow rates (0.01 µl/min) to high flow rates (up to 50 ml/min). One end of the pressure gauge is connected to the syringe outlet, while the other end functions as the inlet to the piston, enabling the system to pump fluid at variable flow rates. The pressure gauge is used to monitor the pressure within the fluid line. At the desired flow rate, fluid is delivered through a needle attached to a nylon plate, which penetrates the sample under load to induce fractures. Step 3: Loading the subsystem. The loading system consists of a piston height with a dimension of 34.5 cm and a diameter of 3.5 cm. The piston has a liquid inlet, and the piston is placed in a cylinder, which is attached to a square plate of dimensions of 24 cm x 24 cm. The hydraulic press seals the cylinder and piston with oil. A nylon plate of size 12 cm x 12 cm and a thickness of 2 cm is attached to the square container, which holds the piston and is connected through screws and bolts and a needle of 0.05 cm diameter and height of 1 cm. This needle facilitates the injection of fluid into the sample. The loading system consists of a piston height with a dimension of 34.5 cm and a diameter of 3.5 cm. The piston has a liquid inlet, and the piston is placed in a cylinder, which is attached to a square plate of dimensions of 24 cm x 24 cm. The hydraulic press seals the cylinder and piston with oil. A nylon plate of size 12 cm x 12 cm and a thickness of 2 cm is attached to the square container, which holds the piston and is connected through screws and bolts and a needle of 0.05 cm diameter and height of 1 cm. This needle facilitates the injection of fluid into the sample. A nozzle links the fluid line to the piston while a plate above the piston, connected to a compressor and pressure gauge, applies confining loads, enabling both biaxial and uniaxial loading configurations. A nozzle links the fluid line to the piston while a plate above the piston, connected to a compressor and pressure gauge, applies confining loads, enabling both biaxial and uniaxial loading configurations. Experimental test procedure and rheology analysis of the injected fluid All components are mounted on an aluminium guide plate, featuring a hydraulic jack connected to the piston and capable of withstanding pressures up to 10 bar. Now, we attach the Perspex block to the piston and place it into the confinement plate, along with the weighing balance. The plate is movable, and the hydraulic jack attaches the hydraulic piston to the bottom. The hydraulic jack, filled with hydraulic oil, has a capacity of 0.5 tons. The setup mechanism involves first inserting the sample into the Perspex block and then sealing it with a piston. Place this total compartment onto the weighing balance and then onto the confinement plate. The weighing balance (up to 150 kg) gives a reading of the total weight of that component. Now, connect the compressor line, as well as the fluid injection line, to the hydraulic press plate and piston. Adjust the air pressure and fluid inlet flow rate accordingly. Using the handle of the hydraulic jack, pump the oil until the hydraulic piston begins to move the plate, ensuring that the piston connected to the fluid line comes into contact with the top confinement plate. Thus, the top load is applied to the sample through the piston movement, and the nylon plate with a needle is automatically placed into the sample to a depth of 0.4 cm, selected to ensure consistent fracture initiation across all experimental runs. Simultaneously, the weighing balance measures the top load while the hydraulic jack facilitates the piston movement. Ultimately, the fluid injection and loading conditions will fracture the sample, allowing two high-resolution cameras positioned at the setup side and front views to record the fracture as shown in Fig. 2. In the present experimental configuration, direct viewing or imaging of fracture evolution or propagation, is not feasible in top or bottom views, due to practical constraints associated with the loading frame, the sealing of the injection system, and the need to maintain stable biaxial boundary stresses during fluid injection. Two cameras were set up on the accessible other views (faces), enabling continuous visualization of fracture evolution and propagation over time throughout the experiment. Meanwhile, this setup cannot directly capture out-of-plane fracture growth in top- and bottom-view images. However, incorporating the utilization observations of other views, the subsequent anticipated three-dimensional fracture morphology, and complementary phase‑field simulations provides a robust framework for interpreting the dominant fracture propagation modes. It may appear counter-intuitive here, that since the applied load on the top surface is the smallest; therefore, the normal direction of the horizontal plane corresponds to the minimum principal stress, which can typically would lead fracture propagation in the horizontal direction. However, it may be noted that the vertical loading magnitude is relative to the confined load. So, even if the top loading is apparently low, it is relatively not small compared to the confined loading. Thus, observed propagated fracture in the layered matrix in the present study is analysed using ImageJ software (https://downloads.imagej.net/fiji/latest/fiji-win64.zip), at different time instants46. Since the fluid injection into the gelatin sample occurs at 15 °C, so the physical properties of the fracturing fluids including water, SAE140 oil, and a mixture of oil (70%) and toluene (30%), were measured at the same temperature, as shown in Table 2. The rheology of the fracturing fluids (particularly the SAE140 oil) is measured by Anton Paar MCR 402 Rheometer using the conical plate CP 100 arrangement at different temperatures (Fig. 1b). Table 2Physical properties of fracturing fluids.Sl. no.Injecting fluidsDensity (ρ) (g/ml)Viscosity ()(mPa.s) at 15 °C1SAE 140 oil0.90425042Water1.013Mixture of oil (70%) and toluene (30%)0.891050 Physical properties of fracturing fluids. Results and discussions In this study, we used water, oil, and a mixture of oil and toluene as fracturing fluids at a constant flow rate of 1 ml/min under a continuous load of confining load of 1.23 bar and a top load of 0.52 bar in a biaxial loading system to determine fluid-driven fractures in various solid elastic mediums such as homogeneous, heterogeneous (layered) of dimensions 10 cm x 10 cm x 10 cm. The injection flow rate was fixed to 1 ml/min, as our main objective was to investigate fracture propagation in various heterogeneous media rather than studying the effect of flow rate. It is to be noted that initial trial runs performed with lower and higher flow rates are not reported here, which resulted in excessive fluid line pressures and unstable fracture formation, whereas at 1 ml/min, stable and reproducible conditions were ensured for all experimental conditions. Pressure was not measured during the experiments because, at these very low flow rates and with soft gelatin analogue materials, the recorded pressures were extremely low and negligible. Therefore, the comparison with numerical predictions focuses on fracture length, aperture, and propagation patterns. The experimental results and model predictions are validated against the classical Khristianovic–Geertsma–de Klerk (KGD) model47,48 for all the scenarios. The phase field model predictions agree with the KGD predictions, and details are presented in the Figs. S-3 through S-5 (supplementary material). Fluid-driven fractures with water under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a homogeneous medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) Comparison plot of aspect ratio evolution between the experimental data and numerical model approach illustrating model validation. Fluid-driven fractures with water under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a heterogeneous (1:5 ratio layers) alternate layered medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) Aspect ratio over time between experimental results and numerical model predictions, representing model validation with a deviation of 15%. Evolution of fluid fractures with low viscosity fluid ( = 1 mPa·s) in different mediums under biaxial stress conditions We have investigated fluid-induced fractures with low viscosity of fluid (water) in homogeneous and heterogeneous solid elastic matrices of varied gelatin compositions. Figure 3 shows the propagation of fluid induced fractures with water in a homogeneous matrix of 10% (A) gelatin concentration with a young’s modulus (E) of 100 kPa at a flow rate of 1 ml/min. The propagated fractures in both side and front views at different time instants are analysed. The matrix compresses to a length of 9.6 cm under the applied confining and top load. Figure 3a and b illustrate the evolution of experimental fracture images in front and side views at various time intervals. In the front view, the matrix compresses to a height of 9.6 cm and elongates up to 12.2 cm under loading, while in the side view, it compresses to 7.5 cm and elongates to 9.92 cm. The observed difference in the matrix dimensions across different views is attributed to variations in frame size during image capturing. The fracturing fluid facilitates propagating the fracture to a depth of 1.82 cm to 3.9 cm, with time intervals ranging from 0.067 min to 0.167 min. Owing to the low fluid viscosity, fracture propagation occurs rapidly, causing matrix failure along the zy-plane of observation and resulting in an asymmetrical fracture shape at the final stage. We compared the experimental results with the phase model approach and the von Mises stress profile for evolved fracture, as shown in Fig. 3c and d. The stress is measured in terms of kPa. In this case, we discovered that the model captured the experimental fracture at various time instances. The model considers a uniform applied stress for fracture initiation and propagation. Figure 3e compares the fracture length propagated on an elastic medium over time. Initially, the fracture length appears consistent in both the front and side views. However, as time progresses, it increases significantly in the front view, reaching up to 3.9 cm, compared to a smaller extension observed in the side view. Figure 3f depicts the aspect ratio of the fracture. The aspect ratio is calculated by ratio of vertical length (major axis, a) to horizontal length (minor axis, b) of the fracture. In the front view, the fracture appears to have an elliptic shape, eventually leading to a circular shape. Figure 3g, represents a comparison plot between the experimental and phase field model aspect ratio over time. At the end, the experimental results are in close agreement with the model with a deviation error of approximately 7.5%. Furthermore, fluid-induced fractures are examined in a layered, heterogeneous medium using two different experimental setups. In the first setup, Layer A contains 10% gelatin concentration while Layer B contains 5%, with alternating layer thicknesses in a 5:1 ratio and a corresponding Young’s modulus ratio of 2:1 (Fig. 4). In the second setup, the configuration is reversed: Layer A contains 5% gelatin and Layer B contains 10%, with alternating layer thicknesses in a 1:5 ratio and a Young’s modulus ratio of 1:2 (Fig. 5). Fluid-driven fractures with water under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a heterogeneous (5:1 ratio thickness) alternate layered medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von-Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) Comparison of aspect ratio evolution between the experimental data and phase field model predictions, demonstrating model validation. Under constant loading conditions, a comparison of Figs. 4 and 5 reveals that the fluid-induced fracture remains confined within the first three layers and fails to propagate further. This behavior is attributed to the higher modulus of the subsequent layers, which impedes fracture propagation. In Fig. 4a and b, the matrix compresses to 5.6 cm in the front view and elongates to approximately 8.2 cm upon applying a load. The change in the matrix height in the images is due to different frame sizes. Fracture propagation is observed up to the third layer from top to bottom of the matrix, rapidly propagating through the high-low-high modulus zones within 0.15 min (6 s). The low-viscosity fluid induces early-stage matrix deformation, which limits vertical fracture propagation. Figure 4a and b present the temporal evolution of fracture length measured from both the front and side views of the specimen, depicting the length of the fractures. Within a time, interval of 0.05 to 0.15 min, after fluid injection, the fracture length increased from 1.27 cm to 2.55 cm in the front view and from 1.75 cm to 3.3 cm in the side view. The larger variation of length in the side view is attributed to the propagation direction relative to the viewing plane, highlighting that fracture growth is inherently three-dimensional rather than confined to a single plane, consistent with the mechanical response of the layered medium. Figure 4c presents the numerical phase field simulation, which aligns well with the experimental observations and demonstrates how fractures initiate within a laminated matrix featuring alternating layers of high and low modulus in a 5:1 thickness ratio. Figure 4d shows the corresponding stress distribution, indicating that the highest stress concentrations occur near the tip of the injection needle. In Fig. 4e and f, the fracture length along the side view (zy-direction) increases over time relative to the front view, while both the fracture shape and aspect ratio decrease, ultimately resulting in an elliptical geometry at later stages. The model-predicted aspect ratio closely matches the experimental data, with a deviation of approximately 15%, as shown in Fig. 4g. This close correspondence between the model and experimental data highlights the model’s effectiveness in capturing the key features of fracture propagation in varied layered heterogeneity. As previously described, in Fig. 5 we examine a modified alternating layered medium by adjusting the layer thicknesses such that layer A (with 10% gelatin concentration) and layer B (with 5% gelatin concentration) are arranged in a 1:5 thickness ratio. The stiffer layer A has a thickness of 3 cm, while the softer layer B is 0.5 cm thick, resulting in a Young’s modulus ratio of approximately 1:2. A biaxial load is applied, compressing the sample to 7.45 cm and extending it to 12.6 cm in the front view (Fig. 5a). The fracture initiated and propagated through the first three layers (Fig. 5a and b). Notably, delamination occurred along the lower modulus layer B at 0.15 min, restricting vertical propagation. The injection of low viscous fluid facilitated rapid lateral fracture growth and eventually led to matrix failure. Over time, the fracture length increased from 1.94 cm to 4.47 cm in the side view, and from 1.9 cm to 4.8 cm in the front view between 0.05 and 0.15 min. Even with a 1:5 thickness ratio between alternating layers, the numerical phase field simulations (Fig. 5c) successfully replicate the experimental observations. The corresponding stress concentration profile is shown in Fig. 5d. In Fig. 5e, the fracture length appears nearly constant in both views over time. At longer durations, the shape and aspect ratio of the fracture, as illustrated in Fig. 5f, progressively decrease, eventually forming an elliptical shape similar to that observed in Fig. 4. Figure 5g compares the fracture aspect ratio between the experiment and the numerical model, showing good agreement with only a 15% deviation over time. When comparing the homogeneous case (Fig. 3) with the heterogeneous cases (Figs. 4 and 5), the overall fracture length remains comparable, regardless of the medium. Evolution of fluid fractures with high viscosity fluid ( = 2500 mPa·s) in different mediums under biaxial stress conditions We investigate the propagation of fractures induced by a high-viscosity fluid (oil) within both homogeneous and heterogeneous elastic solid matrices of varying compositions. The high viscosity of the injected fluid results in stable, anisotropic fracture propagation. Experimental observations and phase field model predictions are benchmarked against the classical Khristianovic–Geertsma–de Klerk (KGD) model, demonstrating strong agreement across all tested configurations (see Fig. S-4 of the supplementary material). Figure 6a and b illustrate the temporal evolution of oil-induced fractures captured in both front (zx-plane) and side (zy-plane) views using ImageJ software. Under compressive load, the matrix compressed vertically to 9.55 cm and elongated horizontally to 12.8 cm along the zx-plane. In the absence of material heterogeneity, the fracture progressed to depths of 2.1–4.7 cm in the front view and 2.3–3.6 cm in the side view, over a period of 0.5 to 2.5 min. The high-viscosity oil led to slow fracture propagation, with matrix failure occurring on the left side after 2.5 min. Fluid-driven fractures with oil under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a homogeneous medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) Aspect ratio as a function of time plotted between experimental results and phase field model predictions. Figure 6c and d show that the numerical phase field simulations accurately replicate the observed fracture growth and stress distribution patterns. Figure 6e presents a temporal comparison of fracture length evolution in both views. A plateau in fracture length is observed until 1.5 min, after which growth resumes, reaching a height of 4.7 cm in the front view by 2.5 min. This directional propagation disparity is attributed to asymmetric stress distribution. Figure 6f illustrates the fracture shape and aspect ratio from both views. While the side view suggests a near-circular fracture profile, the front view reveals significant asymmetry. Figure 6g compares aspect ratio evolution between experiments and simulations, showing a ~ 15% deviation. This discrepancy arises from the model’s limited ability to capture the effects of asymmetric stress fields and the complex fluid–solid interactions occurring under high-viscosity fluid injection. Figures 7 and 8 illustrate fracture propagation induced by high-viscosity fluid in heterogeneous elastic media composed of alternating layers with differing thicknesses and stiffness. Furthermore, oil-induced fractures are examined in a layered, heterogeneous medium using two different experimental setups. In the first setup, Layer A contains 10% gelatin concentration while Layer B contains 5%, with alternating layer thicknesses in a 5:1 ratio and a corresponding Young’s modulus ratio of 2:1 (Fig. 7). In the second setup, the configuration is reversed: Layer A contains 5% gelatin and Layer B contains 10%, with alternating layer thicknesses in a 1:5 ratio and a Young’s modulus ratio of 1:2 (Fig. 8). Fluid-driven fractures with water under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a heterogeneous (5:1 ratio layers) alternate layered medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) Comparison of aspect ratio as a function of time plotted between experimental results and phase field model predictions. Fluid-driven fractures with water under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a heterogeneous (1:5 ratio layers) alternate layered medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) Aspect ratio evolution between experimental results and phase field model predictions, demonstrating the model’s performance and limitations in heterogeneous medium. A comparative analysis of Figs. 7 and 8 reveals that fracture propagation is confined to the initial two layers and fails to penetrate the stiffer upper layers. This confinement is attributed to the elastic modulus contrast between layers, as well as the high viscosity of the injected fluid, which slows fracture advancement. Under applied top loading, the matrix compressed to a height of 7 cm and extended to 10.4 cm in the zx-plane, and from approximately 8.3 cm to 11 cm in the zy-plane. In the front and side views shown in Fig. 7a and b, the fracture length increased from 1.35 cm to 2.7 cm and from 1.28 cm to 2.75 cm, respectively, over a time interval of 0.23 to 0.7 min. Despite of the variations in layer thickness (as in Fig. 8), the fracture remained constrained to the softer (lower modulus) Layer B, progressing laterally. In Fig. 8a and b, the fracture length evolved from 1.5 cm to 1.72 cm in the front view and from 1.7 cm to 2.6 cm in the side view between 0.28 and 0.85 min. Matrix failure occurred shortly thereafter. Compared to the homogeneous case (Fig. 6), both heterogeneous cases (Figs. 7 and 8) display markedly different fracture propagation dynamics. These variations are driven by contrasts in elastic modulus and the influence of fluid viscosity. Figures 7c and 8c present numerical phase field model results that qualitatively replicate the experimental fracture patterns. The associated von Mises stress fields in Figs. 7d and 8d reveal stress concentrations at fracture tips during propagation. Fracture length as a function of time is plotted in Figs. 7e and 8e, showing broadly similar growth trends in both cases. Fracture shape and aspect ratio evolution, shown in Figs. 7f and 8f, reveal an initially circular morphology that becomes asymmetric as the fracture propagates. Finally, Figs. 7g and 8g compare aspect ratio versus time between experimental data and model predictions. A deviation of approximately 15% is observed, with Fig. 7g demonstrating better agreement at early time points than Fig. 8g. This discrepancy arises from the complex interplay of medium heterogeneity, variable permeability, high fluid viscosity, and the anisotropic mechanical behaviour of the layered matrix factors that challenge the phase field model’s predictive capacity. Evolution of fluid fractures with moderate viscosity fluid ( = 1050 mPa·s) in different mediums under biaxial stress conditions In this set of experiments, fractures are induced using a fluid mixture composed of 70% oil and 30% toluene, yielding a combined viscosity of 1050 mPa·s. Fracture propagation is investigated in both homogeneous media and alternately layered gelatin matrices, where layer compositions are varied between 10% (Layer A) and 5% (Layer B) gelatin concentrations in thickness ratios of 5:1 and 1:5. As with previous configurations, both experimental observations and phase field model predictions are compared against the analytical Khristianovic–Geertsma–de Klerk (KGD) model47,48. The phase field simulations demonstrated strong agreement with the KGD model across all tested scenarios. Detailed visualizations and comparative plots for this set of experiments are provided in the supplementary material (Fig. S-5). Figure 9a and b illustrate the progression of fluid-driven fractures in a homogeneous matrix composed of 10% gelatin, characterized by a Young’s modulus ( E) of 1000 Pa and subjected to a constant injection rate of 1 ml/min. The propagation is analyzed in both side and front views across multiple time intervals. Under confining pressure and top loads, the matrix is compressed to heights of 7.6 cm (front view) and 7.85 cm (side view), while elongating up to 11.67 cm and 11.97 cm, respectively. The resulting fractures penetrated to depths ranging from 1.3 to 2.78 cm in the side view and from 1.2 to 2.4 cm in the front view, over a time span of 0.03 min to 0.166 min. Due to the low viscosity of the injected fluid, fracture propagation occurred rapidly, forming predominantly circular geometries. Figure 9c presents a comparison between the experimental observations and the outcomes from the phase field numerical simulations, which effectively captures the fracture evolution. The corresponding von Mises stress distribution shown in Fig. 9d highlights stress concentrations at the fracture tip, consistent with expected fracture mechanics. Fracture length as a function of time is plotted in Fig. 9e, showing steady propagation in both views with minimal deviation. Figure 9f illustrates the evolution of fracture geometry, where the front view reveals an elliptic aspect ratio. Finally, Fig. 9g compares the aspect ratio between experimental observations and model predictions, revealing a deviation of approximately 15%. This deviation is attributed to minor model limitations in capturing fluid-solid interaction dynamics in rapidly evolving fracture fronts. Fluid-driven fractures with mixture of oil (70%) and toluene (30%) under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a homogeneous medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) Comparison plot of aspect ratio evolution between the experimental data and numerical model approach illustrating model validation. In consequence, with the previous investigations, Figs. 10 and 11 illustrate fracture propagation induced by fluid composed of 70% oil and 30% toluene in heterogeneous elastic media composed of alternating layers with differing thicknesses and stiffness using two different experimental setups. In the first setup, Layer A contains 10% gelatin concentration while Layer B contains 5%, with alternating layer thicknesses in a 5:1 ratio and a corresponding Young’s modulus ratio of 2:1 (Fig. 10). In the second setup, the configuration is reversed: Layer A contains 5% gelatin and Layer B contains 10%, with alternating layer thicknesses in a 1:5 ratio and a Young’s modulus ratio of 1:2 (Fig. 11). Fluid-driven fractures with mixture of oil (70%) and toluene (30%) under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a heterogeneous (5:1 ratio layers) alternate layered medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) Comparative plot of aspect ratio evolution from experiments and phase field simulations, highlighting strong agreement and validating the model’s predictive accuracy. Fluid-driven fractures with a mixture of oil (70%) and toluene (30%) under biaxial stress of confining load and a top load of 1.23 and 0.52 bar at 1 ml/min in a heterogeneous (1:5 ratio layers) alternate layered medium. (a) Front view of the fracture progression. (b) Side view of the fracture. (c) Phase field fracture simulations at different time intervals. (d) von Mises stress distribution. (e) Fracture length as a function of time. (f) Aspect ratio of the fracture over time. (g) aspect ratio with respect to time, comparing the experimental data and numerical model simulation results for model validation. Under biaxial loading conditions, the sample compressed to a height of 6.85 cm and elongated to 11.64 cm, as shown in Fig. 10 and 11. The deformation is consistent in both the front view (zx-plane) and side view. In Fig. 10a and b, fracture propagation is observed across the first three layers. Despite the layered heterogeneity, the fracture evolved rapidly in the lateral direction. Between 0.166 and 0.46 min, the fracture length increased from 1.28 cm to 3.14 cm in the front view and from 1.28 to 3.22 cm in the side view. After 0.46 min, the fracture became confined within the high-modulus layer, preventing vertical migration into the lower-modulus layer (A). Conversely, Fig. 11a and b demonstrate that the fracture in the low-modulus layer (B) propagates laterally but becomes arrested at the boundary with the high-modulus layer (A). At approximately 0.46 min, the layer interface opened laterally, allowing fluid to accumulate at the edge of the joint, without any vertical advancement. Beyond this point, the elastic matrix began to fail structurally. Figures 10c and 11c present phase field simulation results, which closely replicate the experimental fracture initiation and propagation patterns. The von Mises stress plots in Figs. 10d and 11d capture localized stress concentrations at the fracture tips, validating the observed failure zones. The time evolution of fracture length is shown in Figs. 10e and 11e. In the front view, fracture length extended up to 4.2 cm, while in the side view it reached 2.52 cm. The fracture shapes and aspect ratios, depicted in Figs. 10f and 11f, indicate decreasing symmetry over time, transitioning toward asymmetric geometries. Figures 10g and 11g compare experimental and phase field model predictions of time-dependent aspect ratio. In Fig. 10g, the model aligns well with experimental trends, showing only a ~ 10% deviation. However, in Fig. 11g, the model exhibits higher deviation due to premature fracture development in the simulation relative to experimental observations. Observations across all experimental cases indicate that, under high confining and low top loading conditions, fracture propagation is more towards the lateral direction than the vertical growth, and fractures are arrested in the heterogeneous matrix. However, fracture growth is not governed by far‑field stress contrast alone at the laboratory scale. Due to the presence of free surface boundary conditions and the finite dimensions of the sample, leading to local stress redistribution and rotation of the principal stress directions, especially when the fracture dimensions become comparable to the sample thickness. The stresses induced near the boundaries within this framework, may locally dominate over the applied stress contrast and promote lateral fracture growth, even where the classical hydraulic fracturing theory would predict vertical propagation61,62. Accordingly, strong pore‑pressure gradients and stress concentrations near the injection inlet can play a key role in controlling the fracture initiation and propagation, leading to orientations governed by local stresses rather than far‑field stresses63,64. In such low absolute stress conditions, typical of laboratory experiments, fracture propagation may further evolve in a three-dimensional and mixed-mode manner such that the final observed fracture geometry may reflect the integrated fracture evolution rather than the initial propagation direction assumed in two-dimensional models65,66. The selected high value of confining load has been chosen to ensure that when a significant top load is applied, the elastic matrix (gelatin) experiences deformation in the initial stages before fluid injection. This approach allows a better understanding of the material’s behaviour under stress. Earlier studies have reported that variations in elastic modulus within layered shales or laminated rocks tend to arrest fractures and promote lateral rather than vertical propagation49,50. Consistent with these findings, our results demonstrate that such heterogeneities in layered media significantly influence the final fracture length under biaxial loading conditions, particularly when injected with a high-viscosity fluid ( = 1050–2500 mPa·s). However, in homogeneous media, the fracture tends to be elliptical, particularly with the further injection of high- or low-viscosity fluid under biaxial loading. Similarly, previous research has shown similar circular shapes in resin and sandstone samples during biaxial fracture studies11. The phase field model adequately captures the overall behaviour of fracture growth. Meanwhile, a few differences are observed compared to the experimental observations, specifically, cases involving low-viscosity fluids or high modulus layers, for example, in Fig. 10. These variations arise from layer heterogeneity, irregular fracture geometries at layer interfaces, the effect of fluid viscosity, and negligible experimental uncertainties. For example, low-viscosity fluids tend to accelerate fracture growth in low modulus layers or in the vertical confinement of high-viscosity cases, which in turn highlights the sensitivity of fracture propagation to local elastic contrasts and fluid–structure interactions. Generally, fluids with a high-viscosity nature tend to yield more stable fractures, whereas low-viscosity fluids often lead to irregular fractures, branching, or tip instabilities. Experimentally propagated fracture widths are different from numerical results due to complex, asymmetric fracture geometries and limitations in measuring width along the entire fracture length. Because of these variations in the fluid and elastic properties of the layered media, they influence fracture length, aperture, and propagation speed, and partially explaining the observed deviations between numerical simulations and experiments. These findings highlight that the developed fracture plays a dominant role in fluid viscosity and the applied stress conditions over material heterogeneity in governing fracture evolution. These findings enhance our understanding of fracture mechanics in layered media, enabling the design of effective hydraulic fracturing strategies and the development of predictive models suitable for geologically complex formations. Impact of viscosity of the fluid in different elastic mediums The produced fracture patterns explicitly demonstrate the mechanical interpretation of the differences between water, oil, and oil - toluene mixtures. The underlying mechanics of observed fracture patterns arise from the fluid’s viscosity, which helps shift the equilibrium between the fluid pressure distribution within the fracture, the resistance to fluid leak-off in the gelatin matrix under biaxial loading, and stress intensity around the fracture tip. Low-viscosity fluids, such as water, generate a high-pressure front that rapidly transfers pressure to the fracture and leads to a sharp increase in the Mode I stress intensity factor. These result in fast, unstable propagation, tip instabilities, and irregular or asymmetric geometries, as observed in the Figs. 3, 4 and 5. While high viscous fluids such as SAE 140 oil, on the other hand dissipates pressure gradients along the fracture length, reducing local pressure at the crack tip and thereby slowing the fracture front’s acceleration. Thus, it produces slower, more stable propagation, and more symmetric fracture shapes are observed in Figs. 6, 7 and 8. The intermediate viscosity of oil - toluene mixture exhibits transitional behaviour, the pressure front progress steadily, by avoiding the instabilities typical of low viscous fluids, resulting in smoother growth while remaining sensitive to modulus contrasts in layer stiffness (Figs. 9, 10 and 11). Under biaxial loading, as lateral confinement is high, viscous resistance delays the process of fluid injection into high modulus layers and promotes lateral fracture growth. Thus, viscosity plays a key role in whether fractures arrest at modulus interfaces or propagate through the layers. Overall, viscosity does not affect the rate of propagation, but it also actively changes the stress field near the tip, controlling the stability of fracture growth, and governing the balance between the vertical penetration and lateral deviation is more effective in heterogeneous media. These mechanical characteristics governs the redistribution of pressure and modify tip stresses, as reflected quantitatively in the temporal evolution of fracture length across all tested fluids and elastic media due to fluid viscosity. To quantify these viscosity effects across the elastic media, Fig. 12 represents the temporal evolution and growth of fracture length in homogeneous and heterogeneous cases (alternate layered). Temporal evolution of fracture length for different elastic media under injection of fluids with varying viscosities; (a) Homogeneous matrix, (b) Heterogeneous matrix with alternating layer thickness ratio of 5:1, and (c) Heterogeneous matrix with alternating layer thickness ratio of 1:5. The influence of fluid viscosity and elastic heterogeneity on fracture propagation dynamics is illustrated for water ( = 0.001 Pa·s), oil–toluene mixture ( = 1.05 Pa·s), and oil ( = 2.5 Pa·s). Figure 12 illustrates the temporal evolution of fracture length in a solid elastic matrix subjected to varied viscous fluid injection, comparing homogeneous, heterogeneous (5:1), heterogeneous (1:5) altered layered thickness. The fracturing fluid viscosities are water ( 15 °C = 0.001 Pa·s), oil ( 15 °C = 2.5 Pa·s), oil-toluene mixture ( 15 °C = 1.05 Pa·s). Figure 12a illustrates the fracture length over time in a homogeneous elastic medium for fluids of varying viscosities. Low-viscosity fluids (e.g., water) resulted in rapid fracture propagation, forming nearly circular geometries at early time steps. High-viscosity fluids showed similar patterns but progressed more slowly and steadily, ultimately achieving longer fracture lengths at later stages. Fluids with intermediate viscosity, such as the oil–toluene mixture, mirrored the low-viscosity behaviour in both propagation rate and fracture morphology. This demonstrates the critical role of viscosity in early-stage fracture dynamics. Figure 12b presents fracture propagation in a heterogeneous layered medium with alternating stiff (100 kPa) and soft (50 kPa) layers in a 5:1 thickness ratio. In this anisotropic setting, fracture directionality becomes more sensitive to the elastic contrast. Water and the oil–toluene mixture led to moderate fracture propagation, while pure oil (high viscosity) produced more confined fractures, advancing preferentially in the vertical direction due to the resistance of stiffer layers. Notably, oil and the oil–toluene mixture propagated over similar time intervals, suggesting that once fluid viscosity surpasses a certain threshold, mechanical heterogeneity (stiffness contrast and layering) becomes the dominant factor influencing fracture behaviour. Figure 12c explores a second layered system with an inverted 1:5 thickness ratio of stiff (100 kPa) to soft (50 kPa) layers. Fracture propagation trends for the oil and oil–toluene fluids remained consistent with Fig. 12b. However, the low-viscosity fluid displayed accelerated growth at early time steps, likely due to the increased volume of softer material, which facilitates faster fracture expansion. Overall, Figs. 12a to c collectively highlight the interplay between fluid viscosity and layered elastic heterogeneity in controlling fracture length, shape, and growth direction. These findings underscore that intermediate- to high-viscosity fluids, such as oil–toluene mixtures and oil, are promising candidates for controlled and stable fracture propagation, even in mechanically complex environments. Morphology of final fracture patterns in 3D To complement the temporal monitoring of experimental fracture propagation in Sect. 5.1–5.3. Figure 13 represents the final evolved fractures captured in different directions of the gelatin analogue, at the end of the experiment as a representative case. In the top view (xy-plane) as shown in Fig. 13a, branching and intersection of fracture planes were observed, while in other views (Fig. 13b & c), it is clearly visible that the fracture is propagating into other layers. These final images demonstrate that fracture growth in three-dimensional view, evolving beyond a single plane, with visible variations depending on the propagation direction relative to the viewing plane. This consistency between the final fracture patterns and the temporal evolution, demonstrates the reliability of the experimental observations. Thus, this work addresses the key research gap by capturing or visualizing fluid fracture evolution/propagation in anisotropic media under low-pressure biaxial stress conditions. Evolved fracture in a heterogeneous medium captured from three different viewing planes: (a) top view (xy-plane), (b) front view (zx-plane), and (c) side view (zy-plane) representing fracture propagation under biaxial stress. Comparison of the phase field modelling results with (modified) KGD model The primary aim of this study is not to create a new numerical formulation, but rather to showcase the effectiveness of the phase field approach by calibrating and validating it against controlled hydrofracture experiments in layered gelatin models. Unlike earlier studies that mainly concentrated on uniform domains, our focus is specifically on utilizing experimental data from elastic anisotropic or layered heterogeneous analogue materials. To evaluate the predictive capability of the model, the simulated fracture propagation is compared with experimentally measured fracture length, aperture, and propagation time, and further across various elastic media, by providing a quantitative validation of the experimental results against the outcomes of phase field modelling. Additionally, these results are evaluated in conjunction with the well-established Khristianovich-Geertsma-de Klerk (KGD) model47,48, as detailed in the supplementary material. The KGD formulation used here is only as a baseline reference for the fracture behaviour within each layer. Specifically, the Eq. (21) is not a classical KGD solution, it is a KGD inspired, layer-wise plain-strain scaling applied locally within each layer in the layered matrix, assuming homogeneous to each layer, viscosity- dominated analytical benchmark. The constraints of the experimental setup are broadly consistent with the assumptions underlying plane -strain conditions. KGD‑type model and its variants assume plane‑strain conditions, which are applicable for fractures propagating in laterally extensive and vertically confined media, rather than as a model intended to predict the fully evolved fractures in this study. Thus, the fractures develop with a large aspect ratio, so that opening deformation dominates over variations along the propagation direction, within each layer. However, the stiffness contrast between adjacent layers, together with lateral confinement, limits deformation normal to the layering. Consequently, the fracture system as a whole evolves in three- dimensional form; in-plane stresses govern the mechanical response within the interior of each layer. In this context, the KGD‑based scaling is mainly used as a reference for comparing phase field simulations, rather than as an attempt to reproduce the full three-dimensional complexity that develops near layer boundaries. After computing this equation to each layer individually, plotted these layer-specific values together. Consequently, minor non-monotonic features or trends (e.g. Fig. S-5a) can appear due to the irregularity in 3D experimental fracture geometry observed might me due to the arrest or plane deviation and should not be interpreted as analytical KGD behaviour. Although the classical KGD model is based on the assumptions of a homogeneous, isotropic, and unconfined linear elastic medium, our approach extends its applicability by treating the stratified medium as a series of individual homogeneous layers, and the results of the KGD model in each layer is stitched to produce the final profile spanning across the multiple layers. Because the KGD model is derived under plane strain conditions, the plane-strain modulus is computed for each layer values , where i=1, 2, 3…etc., layers, and determined the effective modulus for the layered system using the thickness-weighted averaging consistent with fracture orientation to facilitate a transparent comparison while noting that this does not capture interface compliance or toughness contrast. From previous studies, a similar approach has been reported for generalized fracturing scenario in laminated strata with stress contrast56,57. Therefore, KGD is not used as a predictive model for the entire configuration even for the biaxial stress conditions, but rather to benchmark early-stage fracture evolution at the layer scale. The fracture length is computed separately within each layer and then integrated to obtain the final propagation profile. The analytical formulation of the KGD model follows, where h is the fracture length, Q is the injection flow rate, is the Poisson ratio and is the fracture toughness factor. All the experimental cases in this work are validated. We tested scenarios involving water (see Fig. S-3), oil (Fig. S-4), and an oil-toluene mixture (Fig. S-5), as described in the supplementary material. Comparing the KGD model to the phase field simulation, we observed that the model is unable to capture layer heterogeneity after fracture propagation has occurred. At present, it is not feasible to further validate the viscosity-dominated or toughness-dominated regimes. The irregular fracture geometry and anisotropic layered structure of the gelatin medium violate the assumptions of these analytical models. Still, the KGD-based benchmarking provides a reliable basis for assessing the accuracy of the phase field model for fracture evolution in our analogue system. Integrating layered material heterogeneity, fluid–structure interaction, and direct experimental validation offers new insight into hydrofracture growth in soft geological analogues. This approach establishes a modelling strategy that is both quantitatively benchmarked and experimentally grounded. Relevance of the findings to field-scale and engineering applications The outcomes of the current study demonstrate the role of fluid viscosity and heterogeneity of the gelatin analogue mediums for fracture propagation under biaxial stress in controlled laboratory conditions, relevant to real-world applications. For instance, in the CO₂ sequestration scenario, the phase change of CO₂ in the reservoir (from gaseous to supercritical state) results in a substantial reduction in fluid viscosity (from ~ 0.02–0.1 mPa·s)51,52, which is considerably lower than the SAE 140 oil used in our experiments. Despite this difference, the observed fracture behaviour with high viscous oil, such as fracture growth, direction, and impact of elastic heterogeneity, can still provide qualitative insights into how lower-viscosity fluids behave at the field scale. In shale gas reservoirs, natural fractures play an important role in the propagation of fluid-driven fractures through layered section53 with pre-existing fractures. However, our current experiments on intact gelatin layers do not capture these interactions. This discussion frames our findings relevant to CO₂ sequestration and shale gas development scenarios by highlighting their applicability and limitations. Future work will be focused on pre-fractured anisotropic analogue media to mimic actual field-scale conditions and further bridge the gap between laboratory-scale observations and engineering applications. Conclusions This study successfully demonstrates the capabilities of a customized low-pressure biaxial loading system for observing the dynamic evolution of fluid-driven fractures in analogue materials, providing valuable insights into fracture mechanics under controlled stress conditions. With real-time monitoring and 3D visualization, the biaxial loading system offers a powerful tool for capturing fracture propagation, enhancing our understanding of fracture behaviour under biaxial stress regimes. Following are the key outcomes of the study. The experiments highlight the contrasting effects of fluid viscosity on fracture dynamics: water, with its low viscosity, promotes rapid and linear fracture propagation, while oil, a high-viscosity fluid, induces slower, more intricate fracture patterns. We observe the same behaviour when we use a mixture of oil and toluene. The experiments highlight the contrasting effects of fluid viscosity on fracture dynamics: water, with its low viscosity, promotes rapid and linear fracture propagation, while oil, a high-viscosity fluid, induces slower, more intricate fracture patterns. We observe the same behaviour when we use a mixture of oil and toluene. Gelatin is employed as an analogue medium to replicate geological formations. This, combined with quantitative analysis of fracture dynamics and phase field simulations, validates the experimental setup and establishes a foundation for future investigations into fracture behaviour in more complex rock systems. Gelatin is employed as an analogue medium to replicate geological formations. This, combined with quantitative analysis of fracture dynamics and phase field simulations, validates the experimental setup and establishes a foundation for future investigations into fracture behaviour in more complex rock systems. Phase field model predictions for fracture aspect ratio evolution demonstrate a deviation of 7.5–15% across all tested scenarios, due to the minor heterogeneities in the gelatin, approximations in fracture parameters such as viscoelastic effects, fracture toughness, and the assumption of isotropic linear elasticity Despite these factors, this level of agreement confirms the model’s robustness in capturing the complex spatio-temporal evolution of fluid-driven fractures in both homogeneous and layered heterogeneous elastic media, under varying fluid viscosities, layer thicknesses, and mechanical properties. Phase field model predictions for fracture aspect ratio evolution demonstrate a deviation of 7.5–15% across all tested scenarios, due to the minor heterogeneities in the gelatin, approximations in fracture parameters such as viscoelastic effects, fracture toughness, and the assumption of isotropic linear elasticity Despite these factors, this level of agreement confirms the model’s robustness in capturing the complex spatio-temporal evolution of fluid-driven fractures in both homogeneous and layered heterogeneous elastic media, under varying fluid viscosities, layer thicknesses, and mechanical properties. The phase field predictions are further compared against the analytical Khristianovic-Geertsma-de Klerk (KGD) model, showing deviations in the range of 5–7.5%. This comparison supports the model’s ability to capture the fundamental mechanics of fluid-driven fracture propagation under biaxial stress conditions and varying fluid viscosities in elastic media. The phase field predictions are further compared against the analytical Khristianovic-Geertsma-de Klerk (KGD) model, showing deviations in the range of 5–7.5%. This comparison supports the model’s ability to capture the fundamental mechanics of fluid-driven fracture propagation under biaxial stress conditions and varying fluid viscosities in elastic media. Fracture propagation behaviour is strongly influenced by the combined effects of fluid viscosity and medium heterogeneity. Fluids with intermediate to high viscosities, such as oil–toluene mixtures, promote stable fracture growth in layered configurations. These findings indicate that using higher-viscosity fluids can enhance fracture propagation in mechanically complex media under biaxial loading. Fracture propagation behaviour is strongly influenced by the combined effects of fluid viscosity and medium heterogeneity. Fluids with intermediate to high viscosities, such as oil–toluene mixtures, promote stable fracture growth in layered configurations. These findings indicate that using higher-viscosity fluids can enhance fracture propagation in mechanically complex media under biaxial loading. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 List of symbols Body force Young’s modulus, Pa Friction force Critical fracture energy release rate, N/m Fracture length, cm History-strain field Effective permeability, m2 Fracture toughness factor, m0.5 Internal length scale parameter, m Directional vector Pressure, Pa Applied pressure, Pa Injection flow rate, Source term, Kg/m3s Radius of the crack, m Surface Storage coefficient, Pa-1 Time, s Displacement field, m Flow velocity, m/s Position vector Greek symbols Total potential energy Elastic strain energy Compressive elastic strain Tensile elastic strain Negative strain tensor Positive strain tensor Principal strain Maximum size of the mesh element, m Biot coefficient Fracture boundary Fracture surface density Strain tensor Volumetric strain of the reservoir domain Lame parameter Fluid viscosity, Pa·s Effective fluid viscosity, Pa.s Poisson ratio Density, kg/m3 Scalar phase field variable External boundary (domain) Abbreviations Three dimensional American Society for Testing and Materials Computational Fluid Dynamics Carbon Dioxide Cohesive Zone Method Finite Element Techniques’ Finite Element Method Khristianovic–Geertsma–De Klerk PolyMethyl Methacrylate Universal Testing Machine Extended Finite Element Method Author contributions P N R L Sudhishna: Formal analysis, methodology, data curation, Software, Investigation, Validation, Writing - original draft, Writing – review & editing. Sourav Mondal: Conceptualisation, supervision, Resources, Writing - original draft, Writing – review & editing, funding acquisition, Project administration. Tridib Kumar Mondal: Conceptualization, supervision, Writing – review & editing, Project administration. Funding Open access funding provided by Indian Institute of Technology Kharagpur. declaration. No funding is received for the support of this work. Data availability The data underlying this article will be shared on reasonable request to the corresponding author. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data underlying this article will be shared on reasonable request to the corresponding author.