Portland cement CEM I 52.5 R complying with European standard EN 197-1 was used in this study. Quartz sand with a particle size of 0–2 mm was used as a fine aggregate. To optimize the particle size distribution and increase the packing density, silica fume and quartz powder were also introduced. The fly ash (Class F) used in this study was rich in silicon and aluminium. The particle size distributions of these materials measured with a Mastersizer 2000 (Malvern Panalytical Ltd.) are given in Figure 1, and the chemical compositions determined by X-ray fluorescence are shown in Table 1. The high performance superplasticizer of MasterGlenium ACE 430 based on a polycarboxylic ether was used to disperse the UHPC mixtures. The steel fiber used in this study had a length of 12.5 mm and a diameter of 0.2 mm.

The mixture recipes are shown in Table 2 and were optimized by the modified Andreasen & Andersen packing model (Funk and Dinger 1993; Chen et al., 2019; Li et al., 2022). The water/binder (w/b) ratio was fixed at 0.21, and steel fibre content is set as 2 vol%. C1 is a typical UHPC mixture, and C2 is the blended mixture with 20 wt% fly ash replacing cement. The starting materials were first dry-mixed by a high-performance mixer (Type EL5 Eco, Maschinenfabrik Gustav Eirich GmbH and Co. KG) in the counter-current mode for 2 min. The water previously mixed with half the amount of the precalculated superplasticizer was then added, and the mixture was homogenized for 3 min in counter-current mode at a speed of 300 rpm. After a pause of 1 min, the remaining superplasticizer was added, and the mixing was restarted for another 3 min. Afterwards, the steel fiber was gradually introduced in 2 min. The mixture temperature was monitored throughout the mixing process, avoiding premature solidification caused by excessive temperature. Prismatic specimens 160·40.40 mm³ were cast for compressive and flexural strength tests without any mechanical consolidation. For the samples prepared for microstructure characterization, quartz sand and steel fibre were eliminated. The freshly cast samples were sealed at 20 ± 2°C for 24 h, and then the hardened samples were demoulded followed by 24 h of pre-curing in water at 20 ± 2°C. One batch of samples were placed in water at 20 ± 2°C for 28 days as the reference samples. Another batch of samples were loaded by the temperature-pressure load at 200°C with saturated steam pressure 1.55 MPa for 10 days, simulating the working environment of hot water pressure accumulators. For the samples subjected to the long-term temperature-pressure load, the temperature was heated from room temperature to 200°C in ∼1 h and kept for 10 days. For the microstructure analyses, after cooling to room temperature, the inner parts of samples were extracted and immersed in isopropanol for 1 day, followed by freeze-drying.

A three-point bending test (centre-point loading) was performed on prismatic samples 40·40 160 mm³ at a span length of 100 mm to evaluate the flexural properties. An MTS load frame was used to apply a load at a controlled displacement rate of 0.2 mm/min. The mid-span deflection of the specimen was recorded by linear variable differential transformers (LVDTs), and the average values from three samples were reported. The flexural strength can be calculated by the peak load in the load-deflection curve. The flexural toughness was evaluated by energy absorption capacity, which was calculated as the area of load versus the deflection curve up to the deflection point of 3 mm, according to ASTM C348. Compressive strength was loaded on a surface of 40·40 mm² of the broken samples after a flexural test at a loading rate of 2.4 kN/s, according to ASTM C349. The average value of six samples for each batch was reported as compressive strength. A statistical analysis by t-test was used to analyse the mechanical results, and a significance level of p ≤ 0.05 was chosen.

The mineralogical composition analysis was performed by X-ray diffraction (XRD) using an Empyrean PANalytical diffractometer with CuKα radiation (k = 1.540,598 Å). The measurements were performed at 40 kV and 40 mA in continuous mode with a resolution of 0.0131° and speed of 0.0172°/s for a 2θ range from 3° to 65°. HighScore Plus software equipped with an ICSD database was used to identify the mineralogical phases.

Thermogravimetric (TG) analysis was carried out using a TG 209 F3 Tarsus^® (NETZSCH) in a nitrogen atmosphere with a 30 ml/min flow rate. 10 ± 1 mg of powder sample was used for each measurement. The temperature was increased from 20 to 1,000°C at a rate of 10°C/min.

A Zeiss Gemini SEM500 Nano VP equipped with an EDX detector was used to observe hydrates and the interface between the matrix and steel fibres. The small pieces used for SEM analysis were cut from the inner part of the UHPC samples with steel fibres. A low-viscosity epoxy resin was used to saturate the sample pieces and the surface of the hardened samples was then polished down to 1 μm using successive diamond abrasives. The chemical compositions of hydrates in autoclaved samples were measured by point element analysis. In addition, the element mapping was also conducted to study the distributions of the elements.

Due to the chemical reactions activated by the high temperature, the matrix pore structure is inevitably varied. Mercury intrusion porosimetry (MIP) using a Porotec porosimeter (Pascal 240/440) was carried out on small cubic samples. Approx. 2 g of each sample was used in the experiments under 22°C to characterize the porosity and pore size distribution. The Mercury surface tension and contact angle were assigned to 0.48 N/m and 140°, respectively. The intrusion pressure was up to 400 MPa with an increase/decrease speed of 3–7 MPa/min.

The compressive strength of the samples cured under different conditions is shown in Table 3. After normal curing for 28 days, the compressive strengths of C1 and C2 are 178.0 and 169.9 MPa, respectively. The partial replacement of cement by fly ash leads to a slight decrease in compressive strength, which is different from the influence of fly ash in normal concrete. In normal blended concrete, fly ash usually causes a low early age strength due to its low activity, while the late strength may be improved, benefiting from the pozzolanic reaction providing additional gel product (Kovács 1975; Feldman et al., 1990). However, in UHPC with a relatively low w/b ratio, cement hydration and pozzolanic reaction compete with water, and the active cement clinker phases take precedence over fly ash. Furthermore, silica fume, generally used in UHPC to optimize the particle size distribution and densify the microstructure, possesses a higher pozzolanic activity than fly ash. The pH in the liquid environment can be effectively decreased by silica fume, which also restricts the reaction of fly ash. This phenomenon is also reported in other blended UHPC mixtures (Yu et al., 2015; Zhang et al., 2017; Chang and Zheng 2020).

After the 10 days of autoclaving, the compressive strength of autoclaved C1 is 195.5 MPa, which is higher than that of normally cured C1 (t-test, p = 0.0228). Similarly, the autoclaved C2 also presents a higher compressive strength of 218.0 MPa than normal cured C2 (t-test, p = 0.0057). It is well reported that several hours of autoclaving can significantly activate the pozzolanic reaction between silica and portlandite, leading to more C-S-H being formed, which is beneficial to the mechanical properties. However, long-term autoclaving can induce the transformation of amorphous C-S-H into crystalline phases, resulting in the volume reduction of hydrates and pore coarsening (Glasser and Hong 2003; Hong and Glasser 2004; Meller et al., 2007; Kyritsis et al., 2009b). Abd Elrahman et al. reported that the compressive strength of the high performance concrete (HPC) consisting of cement (CEM I 32.5 R), quartz powder, quartz sand, and coarse aggregate is significantly decreased by long-term autoclaving (200°C) (Abd Elrahman and Hillemeier 2015). According to the experience in oil and geothermal well cementing under hydrothermal conditions, α-dicalcium silicate hydrate Ca₂(SiO₃OH)OH (α-C₂SH) is usually encountered, causing microstructure degradation. To avoid this problem, additional silica (approx. 35-40 wt%) is introduced to reduce the Ca/Si ratio to approx. 1.0, leading to other C-S-H phases, such as tobermorite Ca₅Si₆O₁₆(OH)₂·4(H₂O) and xonotlite Ca₆(Si₆O₁₇) (OH)₂, depending on the temperature and effective Ca/Si ratio (Colston et al., 2005; Nelson and Guillot 2006; Kyritsis et al., 2009a). In this way, the UHPC components quartz powder and silica fume make it robust in long-term autoclaving to prevent strength degradation, although the addition of these materials initially aims to enhance the packing density and microstructure. Note that the physicochemical properties (size and crystallinity) of silica-containing materials should be considered, as the silica-containing materials with different activity can change the hydrates assemblage. For example, Krakowiak et al. (Krakowiak et al., 2018) conducted a comparative study on the crystalline α-quartz with different sizes at 200°C. It was found that gyrolite Ca₈(Si₄O₁₀)₃(OH)₄·7H₂O and 11 Å tobermorite dominate in the samples containing coarser quartz, while poorly crystalline C-S-H predominates in the samples with fine quartz. Besides, the morphology of the hydrates can be also affected by the silica accessibility. The sample with active silica fume presents the granular xonotlite (major phase), while the sample with quartz powder generates needle-like xonotlite (major phase) accompanied by higher mechanical strength (Eilers et al., 1983).

The influence of autoclaving on sample behaviour under flexural testing is shown in Figure 2. After normal curing, the flexural strengths of C1 and C2 are 26.4 and 24.0 MPa, respectively, indicating that the addition of fly ash also decreases the flexural strength. However, once subjected to autoclaving, the flexural strength of autoclaved C1 significantly decreases to 18.8 MPa (t-test, p = 0.0022, compared with normally cured C1), although its compressive strength is increased by autoclaving. A similar phenomenon was also reported in (Yazıcı et al., 2009; Chen et al., 2018; Yazıcı et al., 2013), probably because the bond strength of hydrates is decreased by long-term autoclaving due to the transformation of amorphous C-S-H into more ordered phases. The energy absorption capacity was calculated to evaluate the flexural toughness as shown in Table 3. It can be seen that normally cured C1 and C2 present comparable energy absorption capabilities (15.4 kN⋅mm and 15.7 kN⋅mm for normally cured C1 and C2, respectively). However, the energy absorption capacity of autoclaved C1 is significantly decreased to 13.2 kN⋅mm, consistent with the decrease of flexural strength after autoclaving. It seems that the mixture of C1 can ensure the robust compressive strength, but cannot prevent the detrimental effect on flexural properties caused by long-term autoclaving. Krakowiak et al. reported that the strength retrogression is associated with a change of the proportions of C-S-H gel and crystalline C-S-H in the matrix (Krakowiak et al., 2015).

Interestingly, the addition of fly ash in autoclaved C2 increases the flexural strength to 24.8 MPa (t-test, p = 0.0279, compared with autoclaved C1; t-test, p = 0.3243, compared with normally cured C2), while the energy absorption capacity is also improved to 16.3 kN⋅mm. The results imply that the influence of long-term autoclaving on UHPC is highly related to the mixture composition. The partial replacement of cement by fly ash can mitigate the degradation of mechanical properties caused by long-term autoclaving. The similar effect was also reported in the literature (Abd Elrahman and Hillemeier 2015) that the addition of fly ash mitigates the strength degradation caused by long-term autoclaving. As the fly ash used in this study is rich in Si and Al, the partial replacement of cement by fly ash decreases the effective Ca/Si ratio in the matrix, leading to the C-S-H phases with a low Ca/Si ratio, associated with high mechanical strength. In addition to fly ash, metakaolin as a material rich in Si and Al is also studied in combination with cement and/or silica powder under hydrothermal conditions (Kuzielová 2018; Kuzielová et al., 2019). The increase of Al in the matrix can accelerate the formation of hydrogarnet. Such phase may degrade the microstructure and mechanical strength due to its high density. However, in the matrix with a low Ca/Si ratio, the addition of Al-rich materials can stabilize tobermorite at temperature above 150°C, which benefits mechanical strength (Eℓ-Hemaly et al., 1977; Biagioni et al., 2016; Chen et al., 2017). For example, the compressive strength of the Portland cement pastes is increased by the appropriate addition of metakaolin under hydrothermal conditions (Bu et al., 2016). However, the content of Al should be controlled because excessive Al, which exceeds the upper limit of the incorporation of Al into tobermorite, leads to the formation of hydrogarnet (Kyritsis et al., 2009a; Kyritsis et al., 2009b; Meller et al., 2009b). Therefore, the phases assemblage and microstructure are further discussed in the following sections.

The mineralogical compositions of the mixtures after different treatments are characterized by XRD, as shown in Figure 3. 10 wt% rutile was added as an internal standard to quantify the mineral content. However, due to the severe peak overlaps, the quantification result is unreliable and not presented here, but the rutile peaks can also work as an indicator to roughly estimate the phase evolution. For normal cured samples, the typical hydration products ettringite, portlandite, and amorphous C-S-H can be clearly identified. However, the hydrates in autoclaved samples are substantially changed. Both portlandite and ettringite disappear in autoclaved C1 and C2. Ettringite is known to be unstable at high temperatures and decompose with formation of phases like sulphate-AFm Ca₄Al₂O₆(SO₄)·14H₂O and potentially bassanite CaSO₄.0.5H₂O, depending on the chemical composition of the matrix and the temperature-pressure conditions. At even higher temperatures, bassanite decomposes to anhydrite CaSO₄, and sulphate-AFm disappears, forming katoite (Ca₃Al₂(SiO₄)_x (OH)_4(3-x); 0 ≤ x ≤ 3) and sulphate-containing phases (Vladimír and Vepek 1975; Meller et al., 2009a). In this study, the peaks of katoite are identified in both autoclaved C1 and C2, while its intensity in autoclaved C1 is stronger than that in autoclaved C2. Besides, hydroxylellestadite Ca₁₀(SiO₄)₃(SO₄)₃(OH)₂ as a sulphate-containing phase forms in autoclaved samples, which the sulphate originates from the decomposition of ettringite (Sauman and Vavrin 1980; Vladimír 1992).

According to the C-S-H diagrams by Taylor (Taylor 1964), Hong and Glasser (Hong and Glasser 2004), and Meller et al. (Meller et al., 2005; Meller et al. 2009b; Meller et al. 2007), the initially formed amorphous C-S-H gel converts to crystalline phases at hydrothermal conditions. The type and formation rate of newly formed crystalline phases vary with the temperature and the composition of raw materials. Scawtite Ca₇(Si₆O₁₈) (CO₃)·2H₂O, which is stable at 140–300 °C (Harker 1965), is observed in autoclaved C1. According to the SEM point element analysis (Section 3.4), the Ca/Si ratio of the hydrates in autoclaved C1 ranges between around 1.0–2.0, consistent with the formation of scawtite. In such chemical environment, crystalline C-S-H xonotlite is usually encountered at 200°C (Hong and Glasser 2004). However, there is no apparent signal of xonotlite identified in autoclaved C1, probably because the carbonation transforms xonotlite to scawtite (Nelson 1979; Eilers et al., 1983; Grabowski and Gillott 1989) and the densified UHPC matrix restricts the crystallization of xonotlite which is discussed later. It is reported that scawtite can strongly affect the matrix performance and cause brittleness (Berger 1979; Eilers et al., 1983). This may contribute to the low flexural strength of autoclaved C1.

Tobermorite as the only crystalline C-S-H phase is identified in autoclaved C2. The partial replacement of cement by Si-rich fly ash increases the ratio of silica to cement, leading to a reduced Ca/Si ratio, which favours tobermorite formation in autoclaved C2. This effect is further confirmed by SEM analysis in Section 3.4. Due to the reduced Ca/Si ratio in autoclaved C2, katoite decomposes as an important intermediate of tobermorite, which coincides with the result that the peaks of katoite in autoclaved C2 are weaker than that in autoclaved C1. Another intermediate of tobermorite under hydrothermal conditions is hydroxylellestadite (Matsui et al., 2011). Accordingly, anhydrite as a decomposition product of hydroxylellestadite is identified in autoclaved C2, indicating the transformation of hydroxylellestadite to tobermorite. Furthermore, the addition of fly ash increases the Al content in the matrix (see Section 3.4), and the incorporation of Al into tobermorite can improve its thermal stability, preventing the conversion of tobermorite to xonotlite and scawtite under long-term autoclaving (Shaw et al., 2000a; Hong and Glasser 2004). Compared with scawtite, katoite, and hydroxylellestadite, tobermorite has a low density and thus a high solid volume. The transformation of katoite and hydroxylellestadite to tobermorite and the prevention of tobermorite to scawtite densify the UHPC microstructure, benefiting the high mechanical strength of autoclaved C2. The benefiting effect of tobermorite on microstructure and mechanical strength is also confirmed in the fabrication of calcium silicate boards (Chen et al., 2017; Wang et al., 2019), autoclaved aerated concrete (Mitsuda et al., 1992; Shams et al., 2021), and well cementing (Nelson and Guillot 2006).

Except for the crystalline phases in autoclaved C1 and C2, a diffuse peak at ∼29.4° 2θ is observed, indicating that the poorly crystallized C-S-H remains until the end of autoclaving without conversion to crystalline phases. The XRD peak broadening mainly originates from the small crystallite size and a low-ordered crystal structure (Young 2002). This phenomenon of poorly crystallized C-S-H remaining after autoclaving is also observed in the literature (Matsui et al., 2011; Krakowiak et al., 2018), which has two explanations. The high solid volume fraction of the matrix provides limited space for crystallization and restricts the transformation of poorly crystallized C-S-H into crystalline phases. Besides, the C-S-H precursor with Ca/Si < 1 contains long and cross-linked chains of silicate tetrahedra, which is relatively hard to transform into crystalline phases, as a larger number of bonds need to be reorganized compared to the C-S-H precursor with Ca/Si > 1 (Chen et al., 2004; Nonat 2004; Richardson 2004). In this study, the UHPC mixtures have a low w/b ratio and the densified microstructure of matrix provides insufficient space for crystallization, suppressing the transformation of poorly crystallized C-S-H to crystalline phases in autoclaved C1 and C2. This effect is infrequent in other hydrothermal conditions which the mixtures possess relatively higher w/b ratios (Abd Elrahman and Hillemeier 2015; Cheng et al., 2021). Furthermore, in autoclaved C2, the partial replacement of cement by fly ash decreases the Ca/Si ratio of the matrix and thus leads to the formation of C-S-H precursor with a low Ca/Si ratio which has less convertibility to crystalline phases. Therefore, compared with autoclaved C1, more poorly crystallized C-S-H is maintained in autoclaved C2, without conversion to crystalline phases, which strengthens the bond property and provides higher flexural strength. This effect is further discussed in Section 3.3 and Section 3.4 , considering the results of TG and SEM analysis.

TG and DTG diagrams of the samples after different curing treatments are shown in Figure 4. For normally cured samples, the first peak up to ∼230°C of the differential curve corresponds to the water loss in ettringite and C-S-H gel (Myers et al., 2015; Guo et al., 2017). The following small peak from around 260–370°C belongs to the dehydroxylation of aluminium hydroxide from ettringite columns. AFm (Al₂O₃-Fe₂O₃-monophase) also loses the layer water in this temperature range (Scrivener et al., 2016). The peak centred at ∼400°C is assigned to the portlandite dehydroxylation, and the carbonate decomposition occurs at ∼640°C. The last peak at ∼840°C is related to the dehydroxylation and recrystallization of C-S-H to wollastonite (Ca₃Si₃O₉). It can be seen that the partial replacement of cement by fly ash slightly decreases the hydration products of ettringite and amorphous C-S-H in normally cured C2, which leads to the relatively low mechanical strength of normally cured C2, compared with normally cured C1.

The signals of the autoclaved samples are quite different. The first water loss peak up to around 250°C significantly decreases. This is because ettringite decomposes under autoclaving condition and a part of the C-S-H gel transforms into more ordered phases, losing the excessive gel water after autoclaving. Furthermore, the water loss of autoclaved C2 in this temperature range is more significant than that of autoclaved C1. This is consistent with the discussion in Section 3.2 that due to the partial replacement of cement by fly ash, more poorly crystallized C-S-H with a low Ca/Si ratio remains in autoclaved C2, providing more loosely bound water. It is reported that the fracture toughness is positively related to the specific surface area of the solid building blocks of the matrix (Krakowiak et al., 2015). The maintenance of a higher amount of poorly crystallized C-S-H in autoclaved C2 provides a higher specific surface area, leading to the enhanced flexural properties. Besides, tobermorite formed in autoclaved C2 loses its four interlayer water molecules per formula unit between 50 and 250°C (Shaw et al., 2000b; Biagioni et al., 2016). The addition of fly ash increases the Al content in matrix, and the uptake of Al in C-S-H is thus accelerated, which may also increase the water content in C-(A-)S-H (Gartner et al., 2017). The sharp peak of portlandite dehydroxylation is not observed in autoclaved samples. Instead, a broad and flat peak from 250 to 600°C appears in both autoclaved C1 and C2. It is reported that the remaining molecular water of tobermorite (except the preceding four loosely bound water molecules) is lost in this temperature range (dehydroxylation process) (Shaw et al., 2000b; Biagioni et al., 2016). However, katoite decomposes at above 260°C (Rivas-Mercury et al., 2008), and scawtite decomposes at around 400°C (Marincea et al., 2001). Therefore, for autoclaved C1, the water loss from 250 to 600°C is mainly contributed by the decomposition of katoite and scawtite, while for autoclaved C2, the water loss in this temperature range mainly originates from the dehydroxylation of tobermorite. Interestingly, compared with autoclaved C2, autoclaved C1 presents a higher mass loss at ∼640°C, which is probably due to the more calcite being decomposed from scawtite. The last broad peak from ∼800°C to the end of heating is assigned to the C-S-H dehydroxylation in autoclaved C1 and C2, for which the temperature is higher than that in normally cured C1 and C2. Due to the pozzolanic reaction, the C-S-H formed in autoclaved samples possesses a lower Ca/Si ratio characterized by higher polymerization degree and longer silicate chains, requiring more energy to dehydroxylate and reorganize into wollastonite (Chen et al., 2004; Nonat 2004; Richardson 2004).

To further clarify the influence of the long-term autoclaving on UHPC, SEM element mapping is carried out. For autoclaved C1 (Figure 5A), a quartz particle with a blurred surface is observed, which is also reported in oil well cement systems (Palou et al., 2013; Krakowiak et al., 2015) and other kinds of cementitious products treated by hydrothermal curing (Peters et al., 1978; Rashad et al., 2012). Silica reactivity is proportional to its available surface area and crystallinity (Eilers and Nelson 1979; Matsui et al., 2011; Krakowiak et al., 2018). The presence of silica with high reactivity, e.g., silica fume, restricts the reaction of silica with low reactivity, e.g., quartz powder. In this study, the highly reacted surface of quartz particles indicates that the silica fume in UHPC probably has been exhausted by the active pozzolanic reaction under the long-term autoclaving.

On the other hand, some calcium aluminoferrite particles (marked as C₄AF) remain unhydrated after the long-term autoclaving, while no silicate clinker is observed, as confirmed by the XRD results. Due to the low w/b ratio in UHPC, cement hardly hydrates entirely at room temperature. However, during the long-term autoclaving, excessive gel water is released through the transformation of amorphous C-S-H into more ordered phases, and the clinker hydration can thus achieve a higher degree, compared with the normally cured samples. Aluminoferrite phase generally has fast kinetics under early hydration once in contact with water. Calcium sulphates, e.g., anhydrite and gypsum, can suppress its hydration, which are used to prevent premature stiffening. Possible suppression mechanisms are the adsorption of sulphate ions at active dissolution sites and the formation of sulphate-AFm around anhydrous particles (Ectors et al., 2013; Morin et al., 2017). In this study, sulphate-AFm is unstable under the long-term autoclaving condition, while a large amount of S still concentrates around the ferrite particles, as shown in S mapping (Figure 5A), which indicates that rather than sulphate-AFm, the newly formed sulphate-containing phase under the long-term autoclaving hinders the hydration of aluminoferrite phase. Furthermore, the retardation of aluminoferrite phase after heat curing is also reported in the literature (Yang and Sharp 2001), and a higher temperature exerts stronger retardation effect (20–100°C). Except for the aforementioned S concentrated around aluminoferrite particles, some S also accumulates in the areas that are initially occupied by other clinker phases and/or hydrates and then filled by the sulphate-containing phase, as shown in Figure 5A. According to the XRD results, the crystalline sulphate-containing phase is identified as hydroxylellestadite. However, it is reported that sulphate can be co-sorbed on the surface and interlayer of poorly crystallized C-S-H together with Ca²⁺ (Barbarulo et al., 2007) and inserted into the structure of crystalline C-S-H under hydrothermal conditions (Mostafa et al., 2009). Therefore, the sulphate-containing phase observed in S mapping could be a mixture of hydroxylellestadite and C-S-H incorporated with sulphate. Note that the areas filled by the sulphate-containing phase seem porous and leaky, which is probably due to the high density (low solid volume) of crystalline hydroxylellestadite. Thus, the persistence of hydroxylellestadite in autoclaved C1 may contribute to the low mechanical strength, compared with autoclaved C2. As for Al mapping (Figure 5A), Al widely distributes in the matrix in addition to the part concentrated around aluminoferrite particles. This is because Al can also incorporate into C-S-H, and hydrothermal curing accelerates the incorporation (Andersen et al., 2006; Jupe et al., 2008; Matsui et al., 2011). Therefore, the widely distributed Al in the matrix mainly originates from the C-(A-)S-H and katoite (decomposition product of ettringite).

Unreacted aluminoferrite particles are also observed in autoclaved C2 (Figure 5B). According to S mapping, the sulphate-containing phase is again observed in autoclaved C2. Similarly, due to the high density (low solid volume) of hydroxylellestadite, the areas occupied by the sulphate-containing phase present porous and leaky. However, the partial replacement of cement by fly ash decreases the content of sulphate, suppressing the formation of hydroxylellestadite. This effect benefits the microstructure densification and thus the high mechanical strength. Besides, the transformation of hydroxylellestadite together with silica to tobermorite increases the solid volume as tobermorite has a relatively low density, which also contributes to the high mechanical strength of autoclaved C2.

Element point analysis on the autoclaved samples is further conducted, as shown in Figure 6. The Si/Ca ratio of hydrates in autoclaved C1 ranges between around 0.5–1.0, consistent with the formation of scawtite as discussed in Section 3.2. The partial replacement of cement by fly ash increases the Si/Ca ratio in autoclaved C2, originating from the increased ratio of silica to cement and the introduced Si by fly ash. Compared with autoclaved C1, autoclaved C2 can thus generate the C-S-H precursor with a higher Si/Ca ratio, which has lower convertibility to crystalline phases. Therefore, more poorly crystallized C-S-H with a high Si/Ca ratio can persist in autoclaved C2 after the long-term autoclaving, improving the microstructure and mechanical strength. Note that the Si/Ca ratio of the tobermorite solid solution is between 1.2 and 1.5 (Biagioni et al., 2015), consistent with the increased Si/Ca ratio of autoclaved C2 as shown in Figure 6. Besides, compared with autoclaved C1, the Al/Ca ratio of autoclaved C2 is also increased, indicating that the Al-rich fly ash can actively participate in the hydrates formation.

The interface between steel fibres and matrix was further observed, as shown in Figure 7. A clear interstice appears between the steel fibres and matrix in autoclaved C1 (Figure 7A). XRD and TG results confirmed that the C-S-H gel converts to more ordered phases, losing the excessive gel water and resulting in solid volume shrinkage. Furthermore, more katoite and scawtite formation in autoclaved C1 also contributes to the decrease of solid volume, which leads to the undensified interface between steel fibres and matrix. However, autoclaved C2 presents a denser interface between steel fibres and matrix (Figure 7B). This is because more poorly crystallized C-S-H with a low Ca/Si ratio remains in autoclaved C2, mitigating the degradation of matrix bonding and the solid volume shrinkage. Besides, the formation of Al-stabilized tobermorite through the decomposition of hydroxylellestadite and katoite in autoclaved C2 can increase the solid volume, which benefits the microstructure densification.

The cumulative porosity curves and pore size distributions of the samples after different treatments are shown in Figure 8. Compared with normally cured samples, the porosities of autoclaved C1 and C2 are significantly decreased by the long-term autoclaving (Figure 8A). According to the literature (Matsui et al., 2011; Krakowiak et al., 2018), large crystals with a high density and crystallinity are usually observed in the mixtures with a high w/b ratio and low solid volume fraction after hydrothermal treatment, accompanied by the reported pore coarsening phenomenon. In this study, the transformation of amorphous C-S-H to crystalline phases is also confirmed. However, in the UHPC with a quite low w/b ratio and dense microstructure, the limited space suppresses the crystallization, leading to poorly crystallized C-S-H without conversion to crystalline phases. Furthermore, the typical UHPC components silica fume and quartz powder are further activated by the long-term autoclaving, generating additional C-S-H, filling the porosity, and densifying the microstructure. Therefore, compared with the normally cured samples, the critical pore entry radii of autoclaved C1 and C2 are also shifted to lower values, as shown in the curves of pore size distribution (Figure 8B). However, this phenomenon is not observed in other hydrothermally cured mixtures with relatively higher w/b ratios, such as that the porosities and gel pores of the blended cement pastes with metakaolin, fly ash, or silica fume increase after hydrothermal curing (150/165/200°C) (Kuzielová et al., 2017; Kuzielová 2018). The quite low w/b ratio and the high amount of silica component favors the microstructure densification of UHPC after long-term autoclaving.

Interestingly, the porosity of autoclaved C2 is lower than that of autoclaved C1. According to the pore size distribution curve, the pore volume of autoclaved C2 is dominated by gel porosity, assuming 10 nm as the dividing value between gel and capillary pores (Berodier and Scrivener 2015; Durdziński et al., 2017). As confirmed by SEM, the partial replacement of cement by fly ash provides additional Si and increases the ratio of silica to cement, resulting in more poorly crystallized C-S-H with Ca/Si < 1. This kind of C-S-H is hard to transform into crystalline phases under autoclaving, which provides more solid volume and prevents the increase of porosity. Therefore, gel pores dominate in the microstructure of autoclaved C2, which mainly originates from the fine internal porosity of hydrates. However, compared with autoclaved C2, a larger fraction of the pores from 3 to 40 nm is shown in the differential curve of autoclaved C1. This part of pores contains gel and capillary porosity, probably because more crystalline hydrates of scawtite, hydroxylellestadite, and katoite form in autoclaved C1, leading to the shrinkage of solid volume.

It is interesting to point out that there is no strict correlation between porosity and mechanical strength for the UHPC under the long-term autoclaving. Compared with normally cured C1, the porosity of autoclaved C1 is significantly decreased, but the flexural strength of autoclaved C1 is far lower than that of normal cured C1. It seems that the type of hydrates can strongly affect the flexural strength. Under the long-term autoclaving, amorphous C-S-H with high bond strength converts to more ordered phases accompanied by bond degradation. This is further confirmed by the result that more poorly crystallized C-S-H is maintained in autoclaved C2 and autoclaved C2 thus presents high flexural strength. However, the compressive strength of autoclaved C1 and C2 can stay robust under the long-term autoclaving. Compared with normally cured C1 and C2, more hydrate formation in autoclaved C1 and C2 builds the solid skeleton, benefiting the compressive strength. Compared with autoclaved C1, the persistence of more poorly crystallized C-S-H with low Ca/Si and tobermorite in autoclaved C2 ensures the higher compressive strength.