Vipulanandan p–q model to predict the effect of pressurised water on the strengths of gypsum rock in Kurdistan Region, Iraq

a, b, c

linear model parameters

b

width of the tested beam

d

depth of the tested beam

E_i

initial modulus of elasticity

k

initial model parameter for gypsum rock

L and D

length and thickness of a sample, respectively

N

number of experimental data points

P

confining water pressure

P_max

load at a given point (maximum)

p and q

material parameters

R²

coefficient of determination

t

soaking time

$x¯$

mean of calculated data

x_i

calculated data from the model

$y¯$

mean of the actual test data

y_i

actual test data

ϵ

axial strain: %

σ

compressive stress

σ_c

compressive strength

σ_f

flexural strength

σ_fo

initial flexural strength

Gypsum (CaSO₄·2H₂O) rocks cover more than 20% of the earth’s surface, with 7 million km² covered by highly soluble gypsum-bearing rocks (Dreybrodt et al., 2002). Construction failures and geological problems have globally resulted from the presented properties of gypsum content (GC), two examples being the surface collapse of roads and bridge in Ripon, UK (Cooper and Saunders, 2002) and the 400 m³ sinkhole and 5·5 m land surface drop in the north-eastern side of Mosul in Iraq (Al-Layla and Thabet, 1990). Gypsum is a soluble mineral deposit resulting from evaporation of natural water. It can transform to anhydrite by losing its hydration water, or anhydrite can transform to gypsum by the addition of water (Ingebritsen and Sanford 1999; Jaworska, 2010). The conversion of anhydrite to gypsum may cause over 60% increase in the volume of the solid phase, which causes lots of engineering problems (Salih et al., 2015). The average uniaxial compressive strength of gypsum rock has been found to be in the range 9–16 MPa (HaiFei et al., 2009; Liang et al., 2012; Shafiei et al., 2008) as shown in Figure 1. The influence of saturation on the compressive properties of gypsum rock has been considered by several research studies; usually with increasing soaking time (i.e. increasing degree of saturation), the strength of gypsum rock decreases (Heidari et al., 2012; Salih et al., 2015; Vásárhelyi and Ván, 2006). Water is suggested to be a significant factor influencing the stability and mechanical properties of gypsum rock (Dusseault, 2011; Mohammed and Mahmood, 2018a, 2018b; Salih et al., 2015). Among the various notable problems investigated by various research studies, gypsum rock can experience formation of karst features (caves and sinkholes), due to continuous dissolution (Johnson, 2005; Shafiei et al., 2008; Yılmaz, 2001). The dissolution of gypsum is dependent on the surface area exposed to water. The solubility of gypsum is greater in distilled water than in water containing more calcium (Ca²⁺) and sulfate (SO₄²⁻) ions (Fengxiang et al., 1983). In water saturated with calcium sulfate (CaSO₄), gypsum solubility was found to be about 1·83–2·5 g/l at 20°C, the average of which (2·2 g/l) is approximately 1/160 of the solubility of ordinary salt (360 g/l) and 1·5 times the solubility of calcium carbonate (CaCO₃) (1·5 mg/l) (Bell, 1994; Gumusoglu and Ulker, 1982; Yılmaz, 2001). Gypsum is susceptible to rapid dissolution whenever there is an active movement of groundwater, which is unsaturated with calcium sulfate (Salih et al., 2015). Gypsum, like calcium carbonate and salt, dissolves reversibly, but anhydrite does not. When anhydrite is dissolved, it forms a solution of calcium sulfate which, at common temperatures and pressures, is in equilibrium with the solid phase of gypsum, but not with anhydrite. If disequilibrium of the solid–solvent system occurs, gypsum precipitates. This is due to the instability of anhydrite under normal surface and shallow subsurface thermobaric conditions (Klimchouk, 1996).

Leaching of gypsum is an essential point for the dissolution process, which has been also studied. Unbonded gypsum leaches out quickly due to the water percolating through the fissures at the boundary between gypsum and the surrounding rock. Rock collapse in the roofs of tunnels and mines often occurs in bending conditions (Fengxiang et al., 1983). Elizzi (1976) and Ali (1979) studied the difference between compression and tension strain in four-point mode bending tests for gypsum rock. The studies showed that the compressive strain from the four-point bending test for a gypsum rock sample was greater than the tension strain. Efimov (2009) highlighted the method of bending testing. The result showed that the three-point bending test for gypsum rock showed higher tension compared with the four-point bending test. Based on the information from various research studies performed on different types of rock, the size (L/D, where L and D are the length and thickness of the sample, respectively) for the compression test varied from 0·5 to 3. The different beam sizes (L × W × D) used for flexural strength tests found in the literature are summarised in Table 1.

In this study, non-linear models were used to predict the long-term stress–strain behaviour of gypsum rock subjected to different confining water pressures. The effect of the confining water pressure on the dissolution of the GC was also studied. At least three samples were tested for each condition.

The stress–strain behaviours of strain-softening materials such as concrete, glass-fibre-reinforced polymer concrete, fine sand grouted with sodium silicate, sulfate contaminated clay soil and cement mortar were predicted using the Vipulanandan p–q model (Mebarkia and Vipulanandan, 1992; Mohammed and Vipulanandan, 2014; Vipulanandan and Paul, 1990). Vipulanandan and Usluogullari (2011) modelled the stress–strain behaviour of Portland cement-stabilised sand using the p–q model. Vipulanandan and Mohammed (2015a) used the Vipulanandan p–q model for predicting the piezoresistive behaviour of smart cement modified with iron oxide nanoparticles. Also, the Vipulanandan p–q model was used to predict the electrical resistivity against curing time of smart oil well cement (Mohammed and Vipulanandan, 2015; Vipulanandan and Mohammed, 2015b).

The overall objective of this study was to investigate the effect of long-term soaking time and confining water pressure on the compressive and flexural strengths of gypsum rock. The specific objectives are as follows

• investigation of the effect of the soaking time and confined water pressure on the dissolution of GC

• the long-term compressive strength and flexural strength behaviour of normal gypsum rock samples and pre-saturated samples subjected to different confining water pressures

• quantification of the long-term stress–strain relationship of gypsum rock under different confining water pressures

• correlation between the compressive strength and flexural strength of gypsum rock.

X-ray diffraction (XRD) was used to characterise the chemical composition of gypsum rock. Specimens for XRD were prepared from an air-dried gypsum rock sample. The sample (≈2 g) was placed in an acrylic sample holder which was about 3 mm deep. The sample was analysed by using parallel-beam optics with copper (Cu) Kα radiation at 40 kV and 30 mA. The sample was scanned for reflections (2θ) in the range 0–80° at a step size of 0·02° and a 2 s count time per step (Vipulanandan and Mohammed, 2015b).

Thermogravimetric analysis (TGA) curves, % weight loss with temperature (TGA) and its derivative (% weight change/°C – derivative thermogravimetry (DTG)), were recorded using Setaram TGA 500 apparatus at a heating rate of 10°C/min for a sample of about 20 mg (Khalil and Gad, 1972). TGA and DTG curves for the field gypsum rock were obtained. The test sample was loaded in a platinum pan (three-quarters full). This was followed by introduction of nitrogen (N₂) gas into the TGA compartment for 5 min to purge the likely oxygen in the environment of the system. After purging, the sample was heated in the nitrogen atmosphere from room temperature to the maximum of 800°C (Vipulanandan and Mohammed, 2015b).

In this study, an area in northern Iraq (Kurdistan Region) in Sulaymaniyah City (elevation = 966 m, 0495932 38S and 3945243 UTM) that is rich with massive layers of gypsum rock was selected. Sufficient block samples for all the laboratory tests were collected. The site features a massive gypsum rock layer. The gypsum rock locally contains marl or clay impurities within cracks. Gypsum rock samples were collected from the mined layer of gypsum. Based on the information from the literature (Table 1), the collected samples were cored in NX size (54 mm dia.) and L/D was selected to be 2·5 (ASTM D 7012-10 (ASTM, 2010a)). The field rock samples were washed to remove the fine materials and were kept in an oven at 25°C for 48 h in order to obtain constant weight. Oven-dried (25°C) compression and flexural samples were kept inside a vacuum desiccator to be evacuated for 3 h. The sample preparation and testing procedures are explained in Figure 2.

Steel pressure vessels were used to allow the simulation of water pressures. The pressure vessels are cylindrical containers; the base and side-walls were made of stainless steel, while the top was made from a specific plastic material suitable for vessel conditions as shown in Figure 3. One inlet valve located on the top surface was used to introduce air to pressurise the vessel. An outlet valve was used to extract the water from the vessel. Samples could be placed into and removed from the vessel through the removable gate located at the top centre of the vessel. Plastic pipes, special connections and a pressure meter were prepared to be used with the vessels as shown in Figure 3. All the tests were performed at the room temperature of 24 ± 1°C.

A specific procedure for saturating the normal gypsum rock samples was considered (Hawkes and Mellor, 1970). Distilled water was added into the desiccator, and the evacuation process was continued for 24 h (Ali, 1979; Rauh et al., 2006). The weight of the samples was measured in real time every 6 h using a 0·001 g electronic balance to check if a constant weight for the samples was obtained. After 12 h, the samples’ weight was constant. The first step involved saturating the gypsum rock samples under atmospheric pressure in containers with a known volume of water. The samples were soaked inside the containers for durations of 35, 70 and 105 d; the change in the samples’ weight and the electrical conductivity (EC) of the water-dissolved gypsum were recorded at 7 d intervals. Three samples with the same size, colour and origin were considered for each time interval. Then, the second step was conducted by using pressure vessels to saturate the samples and applying three different levels of confining water pressures of 0·2, 0·35 and 0·5 MPa to the pre-saturated samples.

A confining water pressure of 0·5 MPa is equivalent to a 50 m high water column above the sample, simulating the conditions in the reservoir of hydraulic structure projects, which directly affects the gypsum-rich substrates near the surface under the foundation. The flow diagram of the laboratory programme is shown in Figure 2.

Cylindrical specimens 54 mm in diameter and 135 mm high (2·13 inches dia. × 5·30 inches high) were tested at a predetermined controlled loading rate. Compression tests were performed on the normal gypsum rock and pre-saturated samples at 35, 70 and 105 d of soaking using a hydraulic compression testing machine (Instron, UK, 5584 universal testing machine) with a displacement rate of 1 mm/min and a loading rate of 0·5 mm/min. A circumferential extensometer was used to measure the radial strain of the specimens as shown in Figure 4.

Based on the various research studies performed on the flexural beam rock samples, various dimensions (L × W × D) of the flexural rock samples were used (Table 1). In this study, the four-point bending test (ASTM, 2010b) with dimensions (L × W × D) of 140 mm × 40 mm × 20 mm was performed as shown in Figure 4(b). The loading rate was selected to be 0·075 MPa/s. The flexural strength (σ _f) of the samples was calculated using the equation

where σ _f is the flexural strength (MPa); P _max is the load at a given point (maximum) on the load–deflection curve (N); L is the support span (mm); b is the width of the tested beam (mm); and d is the depth of the tested beam (mm) as shown in Figure 4(b).

A conductivity probe was used to measure the conductivity of the contaminated distilled water with dissoluble gypsum. The conductivity measuring range was from 0·1 to 1000 μS/cm. The device was calibrated using different standard solutions.

In order to determine the accuracy of the model predictions, both the coefficients of determination (R ²) and the root mean square error (RMSE) for the model predictions as defined in Equations 2 and 3 were quantified.

where y _i are the actual test data; x _i are the calculated data from the model; $y¯$ is the mean of the actual test data; $x¯$ is the mean of calculated data; and N is the number of experimental data points.

The field rock sample contained gypsum (2θ peaks at 14·96, 21·00, 26·80, 36·68, 39·64, 42·52, 50·28 and 55·48°) and quartz (SiO₂) (2θ peaks at 60·02 and 76·21°) as shown in Figure 5. Similar results were obtained by Koukouzas and Vasilatos (2008). Based on the XRD test results, the GC in the field sample was 96%.

Using TGA, the weight loss at the rate of weight change with temperature was obtained for field gypsum rock. Weight loss in gypsum rock was analysed in four temperature ranges as shown in Figure 6. The hemihydrate was formed between 100 and 200°C when the heating rate was 10°C/min, but complete conversion below 100°C occurred when heating was very slow. A few minutes of heating above 250°C or longer between 100 and 125°C resulted in the formation of soluble anhydrite (γ-CaSO₄); further heating around 360°C gave anhydrite (Khalil and Gad, 1972). Weight loss for the field gypsum rock between 25 and 120°C was 5·70%, and it was 0·75% when the temperature changed from 120 to 400°C. The weight loss for gypsum rock was 0·39% when the temperature changed from 400 to 600°C as shown in Figure 7. The total weight loss for the gypsum rock at 800°C was 7·53%, as shown in Figure 6.

The EC is a good tool to measure the capacity of water to conduct electrical current; it is related to the concentration of gypsum dissolved in 1 litre of distilled water as shown in Figure 8. The EC of the water was used to determine the amount of the gypsum (g/l) dissolved in the distilled water. Distilled water does not contain dissolved gypsum so it does not conduct electricity and has an EC of zero, as shown in Figure 8. The EC of the distilled water contaminated with dissolved gypsum was recorded every 7 d. The process of dissolution started from 0 up to 3 g of gypsum powder mixed with 1 litre of distilled water using an electrical blender. Based on the experimental data shown in Figure 8, the relationship in Equation 4 was used to predict the GC in grams per litre from the EC (μS/cm).

According to the conducted relationship in Figure 8, the measured conductivity amount will be deducted from the total GC of soaked samples. After that, the remaining GC is achieved due to the measured dissolved gypsum per cent.

Using the 0·001 g electrical balance, the gypsum loss at the rate of weight loss with soaking time for pre-saturated samples at various confined water pressures was measured. GC (%) decreased with increasing soaking time and confined water pressure as shown in Figure 9. GC decreased from 96 to 89·5%, a 6·8% reduction, when the sample was soaked for 105 d under 0 MPa of confined water pressure as shown in Figure 9. GC decreased from 96 to 85%, an 11·5% reduction, when the rock sample was soaked for 105 d in distilled water under 0·5 MPa of confined water pressure as shown in Figure 9. The weight loss represented by the amount of gypsum lost increased by 37, 43 and 70% when the confining water pressure increased to 0·2, 0·35 and 0·5 MPa at 105 d of soaking, respectively, as shown in Figure 9. This also indicates that the rate of the gypsum dissolved increased with increasing confining water pressure.

The compressive strength and the density of the normal gypsum rock samples were compared with 102 data for the different types of rock collected from literature as shown in Figure 2. With an increase in the soaking time, the compressive strength decreased. For example, it decreased from 19·6 MPa (2843 pounds per square inch (psi)) to 12·1 MPa (1755 psi), which is a 38% reduction, at 35 d of soaking at 0 MPa of confined water pressure as summarised in Table 2. The compressive strength decreased from 19·6 MPa (2843 psi) to 6·3 MPa (914 psi), which is a 68% reduction, at 105 d of soaking at 0 MPa of confined water pressure as summarised in Table 2. With the increase in the water pressure, the compressive strength decreased. The compressive strength decreased from 6·3 MPa (914 psi) to 2 MPa (290 psi), which is a 68% reduction, at 105 d of soaking when the confined water pressure increased from 0 to 0·5 MPa (73 psi), as shown in Figure 10.

With the increase in soaking time, the initial modulus of elasticity decreased. It decreased from 5297 to 3335 and 1600 MPa, 37 and 70% reductions, at 35 and 105 d of soaking, respectively, at 0 MPa of confined water pressure, as shown in Figure 11. Increasing the confined water pressure from 0 to 0·5 MPa decreased the initial modulus of elasticity of the samples by 43 and 50% at 35 and 105 d of soaking, respectively, as shown in Figure 11. The initial modulus of elasticity of the gypsum rocks decreased with increasing soaking time and confined water pressure, as shown in Figure 11.

In this study, the pre-saturated gypsum rock with four different confining water pressures after various soaking times exhibited a strain-softening behaviour, as shown in Figures 12 and 13.

Based on experimental results, the model proposed by Vipulanandan and Paul (1990) was used to predict the stress–strain behaviour of gypsum rock with four different confining water pressures at different soaking times (Equation 5). The model is defined as follows

where σ is the compressive stress, ϵ is the axial strain (%), $σc$ and $ϵc$ are the compressive strength and the corresponding strain, respectively, and p and q are material parameters.

The parameter q was defined as the ratio of the secant modulus at peak stress to the initial tangent modulus. The parameter p was obtained by minimising the error in the predicted stress–strain relationship. Hence, the parameters p and q in Equation 5 were determined based on the stress–strain behaviour of gypsum rock with different soaking times, confined water pressures, RMSEs and coefficients of determination R² and are summarised in Table 2. The parameters p and q are influenced by GC (%), as summarised in Table 2. The shape of the stress–strain curve can be changed based on the p and q values as shown in Figure 14(a).

For gypsum rock with different percentages of GC, the parameter p was in the range 0·80–1·81, as summarised in Table 2. Hence, the descending part of the strain-softening stress–strain relationship for the saturated samples was steeper compared to that for the normal gypsum rock samples. For the normal gypsum rock samples with 96% GC, p was 1·81 and was reduced to 0·8 when the GC was 85%, as summarised in Table 2. The parameter p and GC were correlated using Equation 8, and model parameters, RMSE and coefficients of determination (R ²) are summarised in Table 2.

This parameter represents the non-linear behaviour of the gypsum rock up to the peak stress. For gypsum rock with different percentages of GC, the parameter q was in the range 0·47–1·30, as summarised in Table 2, and indicates that the material behaviour is more linear with soaking time and confining water pressure. For the normal gypsum rock sample with 96% GC, q was 0·84 and was reduced to 0·56 when the GC was 85%, as summarised in Table 2. The parameter q and GC were correlated using Equation 8, and model parameters, RMSE and coefficient of determination (R ²) are summarised in Table 2.

Ezeldin and Balaguru (1992) proposed an analytical equation (Equation 7) to generate the stress–strain curve for the normal strength of steel-fibre-reinforced concrete based on the equation proposed by Carreira and Chu (1985) for uniaxial compression of plain concrete. This equation involves the material parameter β, which is the slope of the inflection point at the descending branch of the shear stress relationship.

where σ is the compressive stress, $σc$ and $ϵc$ are the compressive strength and the corresponding strain, respectively, and β is a material parameter.

The stress–strain relationship for gypsum rock in different soaking times and four different confined water pressures and the model predictions are shown in Figures 11 and 12.

For the normal gypsum rock sample, parameter β was 3·38. It decreased to 2·55 when the sample was kept in distilled water for 105 d, as summarised in Table 2. The parameter β varied from 2 to 3 when the soaking time varied between 0 and 105 d under confined water pressure varying from 0 to 0·5 MPa, as summarised in Table 2. The shape of the stress–strain curve can be changed based on the β value, as shown in Figure 14(b).

The compressive strength (σ_c) of the normal gypsum rock sample with a density of 2·3 g/cm³ and GC of 96% was 19·6 MPa, as shown in Figure 11. The compressive stress–strain behaviour of the normal gypsum rock sample was modelled using the Vipulanandan p–q model (Equation 5). The coefficient of determination (R²) was 0·99, as summarised in Table 1. The RMSE was 0·204 MPa, as summarised in Table 2. The model parameters p and q for normal gypsum rock were 1·81 and 0·84, respectively, as summarised in Table 1. The Vipulanandan p–q model (Equation 5) predicted the stress–strain behaviour of the normal gypsum rock sample very well, as shown in Figure 12.

The compressive stress–strain behaviour of the pre-saturated gypsum rock with three different soaking times of 35, 70 and 105 d was modelled using the Vipulanandan p–q model (Equation 5). The coefficients of determination (R ²) were 0·99, as summarised in Table 2. The RMSE varied between 0·051 and 0·134 MPa, as summarised in Table 2. The model parameter p for pre-saturated gypsum rock with three different soaking times varying between 1·1 and 1·5 is summarised in Table 2. The model parameter q for pre-saturated gypsum rock with three different soaking times varying between 1·13 and 1·30 is summarised in Table 2.

The compressive strength (σ _c) of the pre-saturated gypsum rock sample decreased by 50% when the sample was soaked in distilled water for 35 d under 0·5 MPa of confined water pressure, as summarised in Table 2. The compressive stress–strain behaviour of the pre-saturated gypsum rock with three different soaking times of 35, 70 and 105 d was modelled using the Vipulanandan p–q model (Equation 5). The coefficients of determination (R ²) for all samples were 0·99, as summarised in Table 2. The RMSE varied between 0·025 and 0·068 MPa, as summarised in Table 2. The model parameter p for gypsum rock with three different soaking times varied between 0·8 and 1·7, as summarised in Table 2. The model parameter q for the pre-saturated gypsum rock with three different soaking times varied between 0·58 and 0·97, as summarised in Table 2. The Vipulanandan p–q model (Equation 5) predicted the stress–strain behaviour of the pre-saturated gypsum rock very well (Figure 12(d)).

The model parameters σ _c, ϵ _c, p, q and β and initial modulus of elasticity (E _i) were influenced by the GC (%). It is being proposed to relate the model parameters to the independent variables (GC (%)), soaking time (t) (d) and confining water pressure (P) (MPa)) using a linear relationship as proposed.

The effects of GC, soaking time and confining water pressure were separated as follows

where k is the initial model parameter for gypsum rock (normal gypsum rock) and a, b and c are the linear model (LM) parameters. The LM parameters were obtained from multiple regression analyses using the least-squares method. The LM parameters and coefficient of determination (R ²) and RMSE are summarised in Table 3.

Stress–strain data from all the finished tests in Table 2 are utilised to produce the corresponding p and q values by regression analysis following Equation 5. Then, these p, q, σ _c and ϵ _c values are used to train Equation 7 to get the value of the parameters k, a, b and c. Secondly, for any other untested gypsum sample provided with GC, soaking time and confining water pressure, the p, q, σ _c and ϵ _c values of that untested gypsum sample will be derived by applying those values of k, a, b and c obtained in the first step into Equation 7. Finally, these new p, q, σ _c and ϵ _c values can be used to predict the stress–strain curve of that untested gypsum sample by following Equation 5.

A 0·01 mm digital vernier calliper was used to measure the volume of the flexural gypsum rock beams with dimensions (L × W × D) of 140 mm × 40 mm × 20 mm. The rate of volume change with soaking time for pre-saturated samples at various confined water pressures was measured. Volume loss in saturated gypsum rock at different confined water pressures was evaluated in three time interval ranges, as shown in Figure 15. The volume reductions (%) of the gypsum rock at 0 MPa confined water pressure were 2·2, 4·7 and 7·3 after 35, 70 and 105 d of soaking, respectively, as shown in Figure 13. The volume loss of the samples increased with increasing confining water pressure at the same soaking time, as shown in Figure 16. Based on the test results, the flexural strength of the samples decreased by 33% for a soaking time of 105 d at 0·5 MPa of confined water pressure, as shown in Figure 16.

Two- and three-parameter hyperbolic relationships (Equation 8) have been used for decades to correlate the changes in material properties with and without additives (Vipulanandan and Krishan, 1993; Vipulanandan and Mohammed, 2015a, 2015b).

The variation in flexural strength with soaking time was represented using the proposed hyperbolic model (Equation 8) and the parameters; the coefficient of determination (R²) and RMSE are summarised in Table 4.

where $σf$ is the flexural strength (MPa), $σfo$ is the initial flexural strength (normal gypsum rock sample) (MPa) and A and B are the flexural strength hyperbolic model parameters.

Based on the test results of the variation in flexural strength (σ _f) with the compressive strength (σ _c), Figure 17 is represented by the following non-linear hyperbolic relationship between flexural strength and compressive strength for the gypsum rocks.

In this study, the effect of soaking time and confined water pressure of pre-saturated gypsum rock was characterised and quantified. Based on the experimental and analytical data and the compressive and flexural strength behaviour of the gypsum rocks, the following conclusions are advanced.

• Based on the TGA test results, the highest weight loss of the sample was observed between 25 and 120°C, which was 5·7%. The total weight loss of the gypsum rock was 7·53% at 800°C.

• GC decreased with increasing soaking time and confined water pressure. The GC decreased by 11% when the sample was subjected to 0·5 MPa confined water pressure for 105 d of soaking.

• The compressive strength of gypsum rock decreased by 50 and 68% when the samples were subjected to 0·5 MPa of confined water pressure for 35 and 105 d, respectively.

• The initial modulus of elasticity of gypsum rock decreased by 70% when the samples were subjected to 0 MPa of confined water pressure for 105 d of soaking. The initial modulus of elasticity of gypsum rock decreased by 43 and 50% when the samples were subjected to 0·5 MPa of confined water pressure for 35 and 105 d, respectively.

• The flexural strength of gypsum rock decreased by 61 and 91% when the samples were subjected to 0·5 MPa of confined water pressure for 35 and 105 d, respectively.

• GC decreased with increasing soaking time and confined water pressure. The GC decreased by 11% when the samples were subjected to 0·5 MPa confined water pressure for 105 d of soaking.

• Based on the RMSE and R², the non-linear Vipulanandan model (p–q model) predicted the stress–strain behaviour of gypsum rock at different soaking times and confined water pressures very well. The model parameters were sensitive to the GC.

• The compressive strength and flexural strength of the normal gypsum rock decreased with increasing soaking time and confined water pressure.