Experimental and Numerical Analysis on the Thermo-Mechanical Stresses Triggering the Onset of Fire-Induced Concrete Spalling

Fire represents one of the most severe conditions experienced by any structure and it is different from other catastrophic conditions in a way that it interferes with the load bearing capacity of the structure by reducing the strength of the material. Concrete by virtue of its high specific heat and low thermal conductivity, naturally resists the heat flow in case of fire. However, concrete suffers from spalling at high temperatures, characterized as progressive or violent flaking of concrete surfaces. Spalling can have serious consequences as it reduces the effective cross section area of the concrete member and further exposes the underlying reinforcements directly to high temperatures. Spalling is a complex phenomenon, influenced by several factors such as types of aggregates, cement, member shape, moisture content, reinforcements, rate of heating, concrete strength, applied stresses and loading conditions [1‐18]. However, these factors are often interrelated, making it further difficult to predict the onset of spalling and design against it. The literature suggest that fire-induced spalling is mainly governed by two mechanisms, (i) thermal stress cracks [19‐21] and (ii) pore pressure generation [10, 22, 23]. In the thermal stress theory, stresses are generated due to high thermal gradients in concrete and restrained thermal expansion. In real-life, structural components are always under some loading and experience some degree of restraints. Stresses originating from dead and live loads act together with thermal stresses to initiate spalling. The pore pressure theory is based on the rise in pore vapour pressure due to moisture movement and development of a moisture saturated zone at some depth away from the heated zone. However, it is still debatable which of these two mechanisms dominate spalling response at different mix designs and heating conditions.

While several attempts have been made to measure the pore pressure during heating and the study of influencing factors in the process [8, 11, 12, 24, 25], significant variance are often observed ranging from very low to high values of pressure and from no spalling to high degree of spalling. In some cases, concrete spalled with very low pore pressure while in other cases, even at very high-pressure spalling was not observed. Some even suggest that no direct relationship could be established between the pore pressure measurements and spalling severity. While past experimental studies have provided valuable insight into the spalling response of concrete, significant challenges exist when comparing experimental results with numerical models. The inconsistency observed in spalling responses between heating tests in laboratories without loads/constraints and real concrete components or structures emphasizes the importance of reproducing realistic loading conditions during experimental investigations. Numerical models dealing with thermo-hygro-mechanical problem exists in literature, however, the complex nature of the problem and the modelling parameters involved requires several assumptions, making it difficult to incorporate in the design process. Abubaker and Davie [26] studied the effect of test configurations and restraints using a thermo-mechanical numerical model. Authors observed that restraints significantly affected the stress states inside concrete and inferred that the compression zone near the heated face is key to spalling initiation. Li and Zhang [27] performed experiments on ultra-high-performance concrete (UHPC) and compared the effects of restraints numerically by using a thermo-mechanical model and concluded the thermal induced tensile stress as main contributor to crack initiation.

Among the factors responsible for spalling, the heating rate typically has a proportional relationship with spalling severity. Extensive spalling is often observed in case of hydrocarbon fire compared with the standard fire or ISO-834 [28]. Similar response was observed in a past study [29] where four different heating rates were compared, and consistently higher spalling was observed in case of higher heating rates. Tests performed by Jansson and Boström [30] on self-compacting concrete showed no effect of fire severity on the depth of spalling.

This paper aims to study the behaviour of high-strength concrete exposed to hydrocarbon and standard fire curve experimentally using Heat-Transfer Rate Inducing System (H-TRIS) test method [31] and compare the influence of thermo-mechanical stresses using finite element software ABAQUS at the time of spalling. Hydrocarbon fire has a very high heating rate compared to standard fire exposure, generally used in case of tunnel fires. The two heating conditions (namely, hydrocarbon and standard fire), represent contrasting yet relevant scenarios, allowing for a nuanced exploration of spalling behaviour in different fire environments. Clear differences in terms of spalling initiation time, spalling type and spalling depth were observed in the experiments. Further, numerical analysis is performed using the finite element software ABAQUS to study the thermo-mechanical behaviour and stress-state at the time of spalling. The effect of pore pressure is not considered in this analysis. The temperature profile is compared between the experimental observations and numerical results with very good accuracy. Effects of externally applied loads are further investigated in the two heating cases focusing on the changes in stress-states in uniaxial and biaxial loading conditions.

Six tests were conducted, three each for hydrocarbon and standard fire curve exposure. While all the samples were casted with thermocouples embedded in them, the thermocouple at 5 mm depth of H8 sample was damaged during casting. For H5 sample, no spalling was observed for 30 minutes exposure of standard fire. The details of the tests are mentioned in the Table 3.

All the samples exposed to hydrocarbon fire started spalling between 88–103 seconds and 551–560 seconds for standard fire. The similarity of results for each replicate tests shows the repeatability of the test method and is one of the main advantages of using H-TRIS test method as it accurately and precisely controls the thermal boundary condition at the exposed face. Two samples were tested 3 months after casting while rest four samples were tested after 6 months. No significant impact of the concrete age was observed on spalling characteristics. It is important to note that the initiation of spalling in case of hydrocarbon fire was surface level and not explosive in nature. Since the test was stopped at first instance of spalling, further nature of spalling (explosive and/or progressive) could not be ascertained. However, in case of standard fire, the spalling initiation event was explosive in nature with a loud bang. Figure 4 shows the concrete samples exposed to standard fire at the time of explosive spalling.

Bild vergrößern

Figure 5 shows images of the samples post-spalling and the spalled debris were collected after spalling as shown in Figure 6. Temperature distribution was measured using the in-depth thermocouples at different depth shown in the following section. Consistent results were obtained between the replicate tests confirming the validity of the testing procedure. One out of three samples exposed to the standard fire curve did not spall while all the testing conditions were the same which indicates the heterogenous behaviour of concrete in fire. During testing of standard fire exposure, water was observed to be leaking from the sides and the top of the sample through cracks presumably thermally induced as can be seen from Figure 4. It is important here to note that water leaking was observed in all the samples exposed to standard fire exposure.

Bild vergrößern

Bild vergrößern

The general-purpose finite element software, ABAQUS [36] is employed to model the thermo-mechanical problem. Concrete is modelled using a continuum-based damage plasticity material model known as Concrete Damaged Plasticity (CDP). The model is modified by incorporating the temperature-dependent properties of concrete enabling it to capture the coupled thermo-mechanical behaviour of concrete due to applied load or restraints. The yield function of CDP is proposed by Lubiner et al. [37] and further modified by Lee and Fenves [38].

The CDP model has been used successfully in literature to simulate the thermo-mechanical behaviour of heated concrete. The model uses isotropic damaged elasticity along with isotropic tensile and compressive plasticity to simulate the inelastic behaviour of concrete [36]. The failure is governed by compressive crushing and tensile cracking of concrete. For the compressive regime, stress–strain data in the plastic range is provided in the form of inelastic strain at different temperature ranges. Corresponding damage data is also provided in terms of temperature-dependent damage parameter and inelastic strain in the post yield region. The inelastic strain can be computed by subtracting the elastic strain from the total strain, \({\widetilde{\epsilon }}_{c}^{in}={\epsilon }_{c}-{\epsilon }_{0c}^{el}\), where\({\epsilon }_{0c}^{el}={\sigma }_{c}/{E}_{0}\), as shown in Fig. 7. ABAQUS computes the plastic strain from inelastic strain input using the Eq. (2) [36],

Bild vergrößern

Similarly, tensile cracking is modelled using tension stiffening by defining strain-softening behaviour in the post-peak regime. Temperature-dependent inelastic strain and post-failure stress values are required along with tensile damage parameters. The cracking strain is computed by subtracting the elastic strain from the total strain, \({\widetilde{\epsilon }}_{t}^{ck}={\epsilon }_{t}-{\epsilon }_{0t}^{el}\), where \({\epsilon }_{0t}^{el}={\sigma }_{t}/{E}_{0}\) as shown in Fig. 7; and ABAQUS internally computes plastic strain from cracking strain using the Eq. (3) [36],where E₀, d_c, d_t are the elastic stiffness of the undamaged material, compression damage parameter and tension damage parameter respectively. The calculated plastic strain must be positive and/or increasing with increasing cracking strain or inelastic strain.

The constitutive model proposed by Kodur et al. [39] is implemented to model high-strength concrete at elevated temperatures described in equations (4) – (7). The reduction factors for elastic modulus at high temperatures are taken from [40, 41] and the tensile strength reduction factors from [42] shown in Figure 8.

Bild vergrößern

Transient thermal strain approximation is considered in cases where loading is applied using the method given in [26]. Temperature dependent thermal properties such as conductivity, density, specific heat and thermal expansion are also incorporated into the model [43, 44]. The net heat flux was calculated by accounting for the thermal losses in a 1-D heat flow case and was applied in the model as surface heat flux. Coupled thermo-mechanical analysis was performed in ABAQUS with temperature dependent brick elements (C3D8 T) with trilinear displacement and temperature. A fine mesh of 2.5 mm was used near the heated face up to 50 mm to accurately model the high thermal gradients and then 5 mm mesh was employed for the remaining 150 mm. A Steel plate was used to support the specimen in vertical direction with general contact interaction (penalty friction as tangential behaviour and hard contact as normal behaviour) and gravity load was used to make sure of contact stability during the initial steps as shown in Figure 9. It was not possible to measure strains experimentally due to the extreme heat near the exposed face, therefore the model was validated successfully with experimental temperature results showing a decent match with all the experimental replicates and cracking behaviour during testing of standard fire exposure.

Bild vergrößern

This section is broadly categorized into three parts, in the first part the numerical model is validated using the experimental in-depth temperature distribution results, in the second part, the result of the thermo-mechanical analysis is discussed at the time of spalling and in the third part, the effect of externally applied load is investigated.

Out of the six samples tested, five experienced spalling during heating. In hydrocarbon fire tests, all three samples spalled, staring very early of the heating regime between 88 s to 103 s. The spalling initiation type in hydrocarbon fire tests were all surface level flaking or chipping, while it is observed in other tests performed in literature that with time it progressively spalls and can be catastrophic for long fire durations. In standard fire exposure, two out of three samples spalled during heating. Both of the samples spalled in an explosive manner with a loud bang and with significant material ejection. The spalling time recorded was very close, between 551 s and 560 s. However, one sample exposed to standard fire, with all other conditions being same, did not spall for a 30 min fire duration. The test was stopped at 30 mins since it is observed that spalling tends to occur in first 10–20 minutes. During tests of all three samples exposed to standard fire, water was seen to escape from the heated sample from the sides and the top through cracks. This observation can be seen through the tensile damage calculated in the numerical models. For hydrocarbon fire however, this water leaking was not observed since the spalling initiation was very fast, with no time for water movement. Using the temperature distribution results from experiments, the numerical model was validated, in absence of strain or displacement data at high temperatures. It can be assumed that the model results are reliable as the thermal field matches very well, the stress distribution follows the general consensus from literature and cracks match accurately with experimental observations. In unloaded samples analysis, progression of stress distribution along the thickness is shown for both hydrocarbon and standard fire exposure. It is important to note the differences in stress-states, since the spalling initiation is different for both heating cases. In hydrocarbon fire, no significant thermal damage is observed at the time of spalling in experiments and numerical models as well. While visible damage was observed in ISO heating through cracks and water leaking, same was seen for numerical model in terms of tensile damage. It can be then interpreted that when spalling happens in moderate or slower heating scenarios, the initiation of spalling is through explosive type of spalling. Whereas, in higher rate of heating conditions, spalling initiates first as surface spalling nature and expected to be followed by potentially further spalling events.

In hydrocarbon fire exposure, the stress-distribution follows the general trend of the unloaded response with a slightly higher compressive peak. In the case of standard fire, in the uniaxial case, the thermal damage occurred around 400 s, while the damage level is reduced for biaxial loading cases. In biaxial loading, damage effects can be seen in the S22 stress contour starting at 250 s in all ratios however in S11 contour the damage is observed in only K=0.75 and K=1 load ratio. Previous literature suggests that under loaded or restrained conditions, spalling severity increases. The compression zone near the heated face has been identified as an important factor which contributes to the increase in spalling severity [26]. In hydrocarbon fire exposure, the peak of compressive stresses near the heated face, in case of K=1 equibiaxial loading is 36.81 MPa, while gradually increasing till uniaxial loading case up to 42.57 MPa. Similarly, for standard fire exposure, the peak in K=1 case is 36.29 MPa and increases to 42.32 MPa in case of uniaxial loading.

This paper presents experimental spalling behaviour of high strength concrete and numerically analyse the thermo-mechanical stresses generated along the thickness when exposed to standard and hydrocarbon fire scenarios. Six samples were tested in total with three each exposed to each fire curve exposure using the H-TRIS test setup. The samples were embedded with thermocouples to study the heat flow for the two heating rates and the numerical models were validated with the experimental temperature distribution. At the time of spalling, the temperature at 5 mm depth was reported between 155–158 °C for hydrocarbon and 276–285 °C for standard fire exposure. The temperature at the exposed surface at the time of spalling calculated numerically is 413 °C and 444 °C for hydrocarbon and standard fire respectively. The thermal gradient is high very close to the heated face for hydrocarbon fire, while in standard fire, the thermal gradient is lower with more volume of heated concrete. Spalling initiates between 88 and 103 seconds for hydrocarbon fire, and between 551 and 560 seconds for standard fire. It is therefore observed that, spalling initiates early for higher thermal gradients. The resulting spalling initiation is surface spalling for hydrocarbon and explosive spalling for standard fire. Since the spalling initiation is very fast for the hydrocarbon curve, the energy absorbed is lesser than in standard fire, resulting in surface spalling.

The general trend of stress distribution (S11, S22 and S33) at the time of spalling for both fire exposure is very similar in nature with compression near the heated face (S11 and S22) and tension at the middle part. The transition zone (S11 and S22) from compression to tension is closely related with the maximum spalling depth. For hydrocarbon fire, the transition occurs between 15 and 17.5 mm and the reported maximum spalling depth is between 5 and 12 mm. Similarly for standard fire, the transition from compression to tension zone occurs between 17.5 and 37.5 mm and the reported maximum spalling depth is between 20 and 28 mm. This overlap clearly indicates there is a correlation between the maximum spalling depth during first spalling event and the thermo-mechanical compression to tension transition zone at the time of spalling. Thermal damage effects are observed in the stress contours of standard fire exposure while it is absent in hydrocarbon fire. Similar effects are observed in the analysis with externally applied load with earlier onset of thermal damage (in S11 and S22 both) in standard fire exposure in uniaxial, K=0.75 and K=1. While the damage effects in S11 are lower in K=0.25 and K=0.5 case. The peak of compressive stresses near the heated face report similar values in both heating cases at time of spalling, confirming the correlation between the spalling and compression zone. Similar peaks of compression zone in cases of different biaxial loading ratios suggest similar spalling response for different biaxial loading ratios. Further experiments incorporating different biaxial loading ratios needs to be performed to investigate the influence of the compression zone near the heated face.