Attenuation and Frequency Characteristics of Acoustic Waves in Steel and Synthetic Fiber-Reinforced Concrete: 3D-PCT and Unsupervised Pattern Recognition

Abstract

In heterogeneous materials such as concrete, deterioration of the elastic wave—which acoustic emission technique (AET) is based on—is one of the research objects in the field. While many studies reveal that the wave is deteriorated due to the concrete content and deterioration of AE signals causes erroneous data interpretation, a limited number of them have suggested eliminating the effects of this problem. For this reason, contributing to the existing literature, this paper proposes to correct AE signals for fiber-reinforced concrete, which is a highly heterogeneous material, by 3D-PCT (Parameter Correction Technique) developed with new approaches in the authors’ previous study for concrete. First, the attenuation properties of concrete samples, including different types and amounts of fibers, were revealed within this scope. Contour maps showed that the type and amount of fiber are effective on elastic wave attenuation. Then, the samples were tested under flexure, and AE results were compared with mechanical findings after parameter correction. The effectiveness of the proposed correction method was verified by separating fiber activities from concrete cracking activities for the first time in the literature with weighted peak frequency and partial power. In this way, by successfully matching the fiber activities, which were revealed after the correction, with the crack development times obtained from frequency-based unsupervised pattern recognition, it was seen that a more accurate AE interpretation could be made with parameter correction. Moreover, corrected AE parameters also provided to propose a new inference for identifying a relationship between the amplitude and energy loss of the AE signals and the type of damage.

1. Introduction

Acoustic emission technique (AET) is one of the structural health monitoring methods and provides information about the fracture processes within a material. Concrete is the most heterogeneously structured construction material, and thus it has been a matter of curiosity how the elastic wave is affected in this medium until it reaches a sensor considering the practical application in the field. Accordingly, numerous studies reveal that the wave is deteriorated due to the inhomogeneity of concrete content. While Phillippidis and Aggelis (2005) investigated the effects of aggregate size on attenuation [1], Aggelis and Shiotani (2008) revealed that wave velocity is not affected by only damage size but also the amount of the damage [2] and Aggelis and Shiotani (2012) pointed out that the shape of inhomogeneity has a strong influence on attenuation, as well [3]. In addition, Asadollahi and Khazanovich (2019) indicated that wave attenuation in concrete also depends on air voids, and cubic aggregates are more attenuative than round aggregates [4]. Thirumalaiselvi and Sasmal (2019) presented hydratation duration-wave velocity and wave propagation distance-attenuation relationships for cement paste, cement mortar, and concrete specimens [5]. The main reason of these cases, stated by Ramaniraka et al. (2020) is that elastic wave attenuation increases as the impedance differences between two materials increase [6]. On the other hand, these signal distortions alter the wave velocity, which is a significant factor affecting AE evaluation results [7]. For instance, Aggelis et al. (2009) pointed out that wave velocities increased after cement injection on large concrete surfaces [8]. Mpalaskas et al. (2014) revealed that ultrasonic wave velocity is related to damage content in artificial damaged cement-based materials [9], and Nogueria and Rens (2018) evaluated the effects of grains on velocity and attenuation of the elastic wave in light EPS concrete [10]. In other words, in a medium having more discontinuities, the wave decelerates and causes a more distorted signal to be recorded by the sensor. While this phenomenon can be useful for tomographic visualization of concrete [11], it gives erroneous results for AET about the source because of all these effects on AE signal parameters [12]. As the medium heterogeneity or wave propagation path length increases, amplitude, energy, and duration of the AE signal decrease [13,14], frequency changes [13,15], and inaccurate results are obtained as to the type and location of the damage due to changes in rise time and arrival time [3,12]. Li et al. (2018) identified relationships between wave propagation path and concrete properties [16]. Their findings show that an AE signal’s energy, amplitude, duration, and count parameters decrease exponentially, and the attenuation increases linearly with the water/cement ratio. Gollob and Kocur (2021) indicated that propagation path significantly affects the accuracy of estimated source location [17]. Wu et al. (2021) showed that AE ring count and rise time gradually decrease with the propagation distance, and the specimens with a larger aggregate particle size experience a relatively rapid increase [18]. Liu et al. (2022) found that the different propagation modes of acoustic waves in concrete have different attenuation, related to the position relationship between the source and the sensor, the propagation path, and path quality [19].

Considering all these effects of source position and propagation path, distorted AE signals should be corrected before analysis or automatically if the structure is monitored real-time in the field. To eliminate the effects of these problems on AE signals, Al-Jumaili et al. (2015, 2018) suggested parameter correction technique (PCT) for a two-dimensional material, carbon fiber-reinforced polymer [20,21]. Their findings showed the ability to use all sensors’ data to improve accuracy and avoid losing data. With the motivation of this issue, the authors Tayfur and Alver (2018, 2019, 2020) developed 3D-PCT for damage detection in concrete and reinforced concrete [22,23,24]. Differently from the previous studies, we used a new approach-based source localization algorithm that considered the medium’s heterogeneity, corrected whole AE signals, recalculated signal parameters, and conducted different parameter analyses [25]. According to the results, it was seen that deterioration of AE signals causes to presume the AE activity having more shear effectiveness and the energy is a more deteriorated parameter than the amplitude. To make this correction, the attenuation properties of the material must first be known, and relationships must be established using these properties. In addition to these innovations, this paper proposes to correct AE signals for fiber-reinforced concrete, a highly heterogeneous material. To characterize signal features of such different types of fiber activities and to verify the effectiveness of the proposed correction method, unsupervised pattern recognition was used, which is also used in the sub-branches of structural engineering such as earthquake engineering, building safety, and performance evaluation, especially for structural health monitoring and damage assessment. Clustering has been chosen as the validation method in this study, as it gives accurate results about crack modes and AE activity types such as matrix cracking or fiber breakage [26,27,28,29,30,31,32,33]. Accordingly, Das et al. (2019) used pattern recognition to cluster cracking modes in steel fiber concrete and cement-based composites [34]. There are also many studies on which parameters to use for clustering AE signals. For example, in carbon fiber-reinforced polymer, any relationship was not observed between damage mechanisms and AE energy [35]. The literature also mentioned that the amplitude of the signal (related to the sensor location) is related to the magnitude of the activity rather than its type [30]. However, the frequency characteristics of the signal show a significant relationship with the damage type [36]. Studies show that matrix cracking can be differentiated at low frequencies and fiber breakage at high frequencies. Again, in these studies, instead of clustering based on the single frequency value of the signal such as center, average, or peak, it was observed that using weighted peak frequency (WPF), which is calculated as the geometric mean of the center and peak frequencies, is healthier and the frequency characteristics of the signal are better revealed. The parameter in which WPF is evaluated together in clustering is also suggested as partial power (PP). PP is defined as the percentage contribution of the signal to a specific frequency range in the frequency spectrum, and multiple PP can be calculated for different frequency ranges [37]. Sause et al. (2012), Sause (2016), Bohmann et al. (2018), Kelkel et al. (2018), Tabrizi et al. (2019), and Qasim Ali et al. (2019) used WPF-PP relations for clustering AE activities [26,37,38,39,40,41,42].

For this reason, contributing to the existing literature, this paper proposes to identify how the elastic wave is affected by the type and amount of fiber in concrete and to correct AE signals for fiber-reinforced concrete. For this, first, attenuation properties of concrete samples, including different types and amounts of fibers were revealed by 3D-PCT using the change in the main AE parameter “amplitude”. Here, acoustic pulses having different amplitudes with the same frequency and wave shape generated by a signal generator were sent to the points in the grids created on all samples. Colors were assigned to amplitude values recorded from each point, and 3D attenuation maps were created. Then, the samples were tested under flexure, and AE results were corrected using information from 3D contour maps and compared with mechanical findings. Results were verified by separating fiber activities from concrete cracking activities in the literature with weighted peak frequency and partial power for the first time. In this way, by matching the fiber activities, which were revealed after correction, with the origination times obtained from frequency-based unsupervised pattern recognition, the effectiveness of the proposed correction method on AE interpretation for fiber-reinforced concrete was investigated.

2. Acoustic Emission Technique (AET)

Acoustic emission is the release and propagation of energy due to fracture in a stressed material. Based on this natural phenomenon, AET, which can be described as detecting, and analyzing elastic waves by appropriate sensors, has been used to examine different facilities such as structures, nuclear plants, high-pressure pipes or tanks and airplanes. Since attenuation and damage type classification analyses are made using information about an AE signal, some characteristics of an AE signal should be mentioned (Figure 1a). Among these parameters, maximum amplitude (or amplitude)—peak voltage reached in the signal—was used in this study to evaluate the attenuation properties of the specimens. This parameter is related to the magnitude of the source. On the other hand, clustering was conducted using some seconder parameters associated with peak and centroid frequencies (Figure 1b) obtained after fast Fourier transformation (FFT) [43] to classify damage types.

As it was mentioned in Section 1, studies in the literature show that using weighted peak frequency (WPF) (Equation (1)) [42] for clustering AE activities is a good approach. Here, the center frequency (f_center) is the mass center of the frequency spectrum, and the peak frequency (f_peak) is the highest frequency value achieved. The parameter that WPF is evaluated together in clustering is partial power (PP). PP is the percentage contribution of specific frequency ranges in the frequency spectrum. Multiple PP can be calculated for different frequency ranges (Equation (2)) [39].

$$Weighted peak frequency \left(WPF\right)=\sqrt{f_{center}·f_{peak}}$$

(1)

$$Partial power \left(PP, \%\right)=\frac{{{\displaystyle \int }}_{f_i}^{f_j}{\tilde{U}}^2\left(f\right)df}{{{\displaystyle \int }}_0^{f_{max}}{\tilde{U}}^2\left(f\right)df}$$

(2)

where $\tilde{U}\left(f\right)$ is the FFT of the signal $U\left(t\right)$.

2.1. K-Means Unsupervised Pattern Recognition for AE Clustering

K-means is a well-known unsupervised clustering algorithm for clustering AE data. It classifies a set of data into number of clusters by defining their centroids. As the primary positions of the clusters’ centroids move away from each other, clustering precision increases. Afterwards, each point is associated with the nearest centroid and relevant cluster. This procedure is repeated until the objective function “J” is the minimum using Equation (3) by Kaufmann and Rousseeuw (2008) [44]:

$$J={\displaystyle \sum }_{m=1}^k{\displaystyle \sum }_{n=1}^l\Vert x_n^{\left(m\right)}-c_m\Vert {}^2$$

(3)

where k and m are number of clusters, n and l are number of cases, and $\Vert x_n^{\left(m\right)}-c_m\Vert {}^2$ is the distance between the point and the center of the cluster.

2.2. 3D Parameter Correction Technique (3D-PCT)

The method is based on the construction of relationships between transmitted and received signal parameters. First, a region is created and is divided into meshes on the tested material. Then, artificial AE sources with different input amplitudes are generated at each mesh node, and AE parameters are recorded for each sensor. Relationships between the transmitted and received signal parameters are constructed. Any needed amplitude value can be interpolated by using transmitted and received values of artificial AE sources. These distributions can be scattered by using contour maps and the attenuation factor of any position is determined by interpolating the node parameters. Then, AE data obtained from the tests are localized and their relevant input vs. output relations are determined by contour maps created. Finally, the parameters of AE signals are corrected by using appropriate relations of their locations and relevant sensors recording.

3. Experimental Study

3.1. Specimens and Materials

Seven concrete and a cement mortar beam specimens (100 × 100 × 600 mm) were designed within the experimental procedure. While one of the concrete specimens was plain, the others contained different types and fractions of fibers, as given in

Table 1. Descriptions of the test specimens.

Specimen	Statement
REF	Conventional plain concrete
S05	Steel fiber-reinforced concrete (0.5% of the volumetric fraction)
S10	Steel fiber-reinforced concrete (1.0% of the volumetric fraction)
Mi05	Micro polyamide fiber-reinforced concrete (0.5% of the volumetric fraction)
Mi10	Micro polyamide fiber-reinforced concrete (1.0% of the volumetric fraction)
Ma05	Macro polypropylene fiber-reinforced concrete (0.5% of the volumetric fraction)
Ma10	Macro polypropylene fiber-reinforced concrete (1.0% of the volumetric fraction)
CM	Cement mortar

. Concrete and fibers were used for the production of the specimens. The conventional concrete mix design is presented in

Table 2. Mix design of conventional concrete (kg/m³).

Amount	Material
	Cement	Water	Aggregate (mm)			Plasticizer
			0–3	5–15	15–25
	295	177	960	394	591	3

. W/C ratio of the mix was 0.6 and the cement was Cimentas CEM I 42.5R type. In addition, crushed limestone aggregates were three classes of 0–3 mm, 5–15 mm, and 15–25 mm. Materials used as different inclusions in concrete (Figure 2) were BetonFiber HE 0735 type hooked steel fibers (S05 and S10), KraTos Micro Synthetic Fiber (Mi05, Mi10), and KraTos Macro Synthetic Fiber (Ma05 and Ma10). Their properties are presented in

Table 3. Properties of fibers.

Fiber	Type	Length	Tensile Strength
BetonFiber HE 0735 Hooked Steel Fiber	Cold-drawn	35 mm	1400 MPa
KraTos Micro Synthetic Fiber	Polyamide 2.6	12 mm	900 MPa
KraTos Macro Synthetic Fiber	Polypropylene	54 mm	550 MPa

.

At the beginning of the specimen production, a dry mixture containing aggregates, cement, and fibers was mixed approximately for 2 min in the mixer. Afterwards, mixing was continued by adding water. Then, a plasticizer was added to the mixture and the concrete was ready. The concrete was poured into pre-lubricated steel prismatic molds and was compressed by rodding technique. After, letting polished specimens for 24 h in the laboratory conditions, they were unmolded and were cured in the standard conditions for 28 days (20 ± 2 °C and 95% relative humidity).

3.2. Test Set-Up

Experimental program consisted of two main stages: (i) Conducting 3D-PCT to identify attenuation properties and (ii) conducting flexure tests to correct AE signals and to cluster different damage mechanisms. As seen in Figure 3, 150 × 100 × 100 mm meshes were created at flexure spans of the specimens and tests were conducted on these volumes which were divided into 50 × 100 × 100 mm grids. 1 V, 5 V, 10 V, and 24 V artificial sine wave sources (frequency of 50 kHz) were generated on each test node by PCI-5412 Signal Generator, which can produce different signal forms such as sine and triangle with 100 MS data per second between −6 V and +6 V interval. A piezoelectric AE sensor connected to the signal generator was used as a transmitter, and it was moved on each point to allow the elastic waves to propagate within the material. All artificial sources were recorded by an 8-channel AE system with six AE sensors by Mistras Group. All sensors were amplified with preamplifiers having 40 dB gain and the threshold was set as 40 dB to eliminate ambient noise. In addition, Hsu-Nielsen Calibration Tests were conducted at the close points of the sensors to check the quality of the detected signal. This procedure was carried out to create 3D attenuation contour maps of the specimens.

In the second stage of the study, flexure tests were conducted on each specimen in accordance with TS EN 12390-5 standard [45] with a net span of 500 mm. The load was applied with a speed of 0.2 mm/min at the center of the mid-span by Shimadzu AG-IS 100 kN Universal Testing Machine (Figure 4). Simultaneously, AE data were recorded during the tests. Afterwards, by using correction relations for the relevant sensor from 3D contour maps (see Section 4.1 for details), localized AE signals were corrected.

To classify different damage mechanisms of the specimens, AE activities were clustered based on unsupervised pattern recognition using their frequency characteristics (WPF vs. PP). PP is divided into equal ranges considering the higher frequency of the fiber activities [37,38,39,40,41,42]. Accordingly, PP was divided into four ranges as 0–75 kHz (PP1), 75–150 kHz (PP2), 150–275 kHz (PP3), and 275–300 kHz (PP4) considering peak frequency values of the spectrums. K-means clustering was conducted on these distributions and almost the same results were obtained for each PP value.

4. Results and Discussion

4.1. Attenuation Results

First, recorded amplitude values of each sensor at all 16 nodes from four different artificial sources were obtained. Then, by stacking amplitude values of all nodes and interpolating them; 3D volumetric contour maps were created for all sensors and all artificial sources. Concrete heterogeneity was chosen as the experimental variable and the presence of fibers increased the concrete heterogeneity. To examine the effect of this situation on attenuation, 3D-contour maps showing the distribution of elastic waves of 24 V recorded by the sensor S6 are presented in Figure 5. Each contour map has a color scale from 40 dB to 80 dB. This artificial source, corresponding to 108 dB, was recorded at a maximum of 80 dB in REF, while it decreased to 60 dB (S05) and 40 dB (S10) in steel fiber-added specimens. In polyamide and polypropylene fiber-reinforced specimens, it increased up to 90 dB in some places. This situation may be due to the higher density due to over-close aggregates in that region. In this way, waves passing through that path can have higher velocity, as if they were passing through a much more solid area, and therefore attenuation may remain at low levels. All these show that the fiber added to the mixture increases the heterogeneity of the concrete and causes more attenuation of the signals. The type and amount of fiber are also effective in attenuation. With the amount of fiber increases, heterogeneity increases and thus, attenuation increase.

4.2. AE Activities

Before clustering, AE activities were evaluated according to the toughness and fiber types of the specimens (Figure 6). As seen, the lowest amount of AE activity was recorded in REF. The presence of a higher amount of steel fiber increased the toughness. In this case, since the development of cracks is prevented with the bridging mechanisms of fibers, the amount of AE was reduced (S10). However, this behavior changed in synthetic fiber-reinforced specimens. Toughness increased with increasing fiber content in synthetic microfiber-reinforced specimens. However, since these fibers are less rigid and easier to break, the amount of AE increased as more fibers were exposed to damage (Mi05 and Mi10). As for synthetic macrofibers, because they are more rigid compared to microfibers, less AE activity was recorded in Ma05 and Ma10. As expected, more AE activities were obtained as CM had more damage than concrete REF.

4.3. Unsupervised Pattern Recognition Results

WPF vs. PP relationships were examined to classify different fracture mechanisms (related AE sources) in fiber-reinforced concrete (Figure 7). It is seen that the AE activities of the specimens are concentrated at different frequency values, but the common frequency range in all specimens is 50–200 kHz. While only this cluster (Cluster-1) appeared in REF and CM, a new cluster (Cluster-2) at a higher frequency range was observed in the fiber-reinforced specimens. In other words, Cluster-1 activities were recorded from damage mechanisms in cement paste. On the other hand, Cluster-2 attributes to fiber activities [35,46,47,48,49]. Particularly in samples containing 1.0% fiber by volume, much more fiber activities concentrated at 250 kHz were observed due to the higher amount of fiber.

Evaluations made above for the mechanical findings, and the frequency distributions and damage type can also verify AE activities: Fiber activities preventing crack formation in S10, which has higher toughness due to the amount of increased steel fiber, increased the number of high-frequency activities (Cluster-2). With the increase of fiber amount in micro synthetic fiber-reinforced specimens (Mi10), toughness of the concrete increased and since these fibers were less rigid than the steel fibers, they were more damaged and the number of AE activities in Cluster-2 increased. Since macrofibers are more rigid compared to microfibers, increasing the amount of fiber (Ma10) did not change the amount of activity in Cluster-2. In general, much more AE activity describing fiber activities due to the higher fiber content and concentrated around 250 kHz were observed in specimens containing 1.0% fiber by volume.

In addition, Figure 8 indicates the origination time of these clustered activities. To evaluate the originations of these activities regarding the mechanical behavior of fiber-reinforced concrete, graphs were divided into three phases as P1 (where the load reaches to ultimate capacity and decreases in a plain concrete), P2 (hardening phase where the load values continue increasing), and P3 (descending slope drawn after second maximum load up to the end of the test) [50]. As can be seen, all the high-frequency activities defined as fiber activity Cluster-2) occurred in phases P2 and P3 and only matrix cracking was encountered up to ultimate load capacity (P1) such as the plain concrete behavior. This proves that AE activities were clustered in accordance with the mechanical behavior of fiber-reinforced concrete.

4.4. Parameter Correction Results

AE activities of the specimens were localized using a new approach-based source localization algorithm that considers the medium’s heterogeneity [25]. Then, localized activities were corrected using relevant input vs. output amplitude relations for their locations. Relations of recorded vs. corrected amplitudes of all specimens were examined using Figure 9. Second-order polynomial curves and equations calculated by regression analysis indicating the most appropriate behavior for these distributions are also presented on the graphs. It is seen that, the corrected amplitudes of the hits around 40 dB (threshold) vary over a wide range. However, this interval narrows as the recorded amplitude increases. This indicates that hits having different amplitudes recorded by the same sensor can be perceived as having equal amplitude due to different attenuation factors. Moreover, as the amplitude of the activity increases, the need for correction decreases. As the recorded amplitude value increased, the proximity to the corrected values increased and the closeness to the actual amplitude value reached up to 90% for the highest amplitudes. In addition, when the closeness of the recorded amplitude values to the corrected values is evaluated, it is observed that the presence of the steel fiber decreases the level of closeness, while the presence of polypropylene and polyamide fibers increases. Actual amplitude values recorded in cement mortar are also closer to recorded ones compared to the reference specimen, which is more attenuative due to the presence of coarse aggregates.

To evaluate the signal amplitude variation according to the concrete content (fiber types and amounts), closeness values between recorded and corrected amplitude were plotted in Figure 10. It is seen that the closeness orders of the specimens are different at each amplitude value. For example, while the highest corrected amplitude value of a signal with a recorded amplitude of 40 dB is obtained in Mi10, the highest corrected amplitude value of a signal with a recorded amplitude of 60 dB is obtained in Ma10. Because although the amount of fiber and material heterogeneity increase, this heterogeneity is not regular among the medium and the attenuation at source location may be lower. For this reason, although the ordering of the signal amplitude according to the contents of the specimens depends on attenuation properties at a particular location, the specific curves of the specimen can be used to determine relations between recorded and corrected amplitudes. In the right graph, the equation y = x^1.03 was proposed for the closeness of the recorded amplitude to the corrected amplitude.

Figure 11 and Figure 12 show distributions of raw and corrected RA values of AE activities, respectively. RA value is a secondary AE parameter identifying the activity type which can be calculated by dividing rise time over amplitude. AE activities having higher RA values are attributed to shear type. To investigate the effect of parameter correction on damage assessment, corrected RA values and their moving averages (blue dashed lines) were evaluated together with the average frequencies of the signals (Figure 13). As can be seen from Figure 11 and Figure 12, the corrected RA values are lower than the raw RA values. Because RA value is inversely proportional to the amplitude parameter, it decreases as the amplitude value increases after parameter correction. In this case, it can be said that higher RA values are calculated for the signals whose amplitude decreased due to attenuation, thus these activities are perceived as more shear-type. Since fewer AE activities were corrected in REF and CM, their moving average could not be calculated. On the other hand, following changes were observed for the fiber-reinforced specimens: As seen, sudden decreases in the moving average of the average frequency for S05 and S10 were observed after the damage development accelerates. The decrease in average frequency indicates that fiber activities present a shear-type damage signal. This situation can also be determined by the recalculated RA values after the correction. In the corrected RA value distributions, higher RA values are observed which cause jumps in the moving average. Since synthetic fibers are damaged much more than steel fibers, these jumps were seen more in micro and macro synthetic fiber-reinforced specimens. At these moments, the corrected RA values of AE activities indicating fiber activities are more predictive of damage type than the raw values. The activities occurring at these moments are not in Cluster-2. It can be said that these activities are defined as fiber rupturing, different from slipping activities. In this way, slipping activities of the fibers could be distinguished from the cracking activities of the concrete and it was determined that it was characterized by a shear-type AE signal.

Figure 14 shows energy loss versus amplitude loss of the AE activities. As can be seen, AE signals lost their amplitudes at a maximum of 70–80% and at least 60% of their energy. This shows that the energy of an AE signal is distorted more than the amplitude by attenuation. The reason for this situation is that as the wave attenuates, counts exceeding the threshold and the signal’s duration decrease, and the energy decreases accordingly. Considering the amplitude loss of the activities that lost all their energy and were listed at nearly 100% energy loss, the following evaluation can be made: among the activities that lost energy at the same rate, the activities with lower amplitude loss lost more AE duration. Activities with higher amplitude loss, on the other hand, and the same amount of energy loss, loss in AE duration was less. Accordingly, it can be concluded that under the same energy loss, activities with less amplitude loss are characterized by more tensile type, and activities with more amplitude loss are characterized by more shear type. Considering all of these evaluations, a new inference for a relationship between the amplitude and energy losses and the type of damage was proposed as in Figure 15.

5. Conclusions

In AET, an elastic wave-based method, deterioration of the elastic wave until it reaches a sensor, particularly in highly heterogeneous materials such as concrete, is the subject of many studies. This study investigated the correction of AE signals recorded in fiber-reinforced concrete by 3D-PCT. First, attenuation properties of concrete samples including different types and amounts of fibers were revealed. Effects and differences of the concrete content on the damage types and attenuation properties showed that the presence of the fiber increases the concrete heterogeneity and causes more attenuation of the AE signals. The type and amount of fiber is effective in attenuation: steel fibers cause more elastic wave attenuation than synthetic fibers. Then, the samples were tested under flexure, and AE results were compared with mechanical findings after parameter correction. AE hits having different amplitudes recorded by the same sensor can be perceived as having equal amplitude due to different attenuation factors. The presence of the steel fiber decreases the level of closeness between recorded and corrected amplitudes, while synthetic fibers increase it. In addition, as concrete is more attenuative than cement mortar, amplitude values recorded in cement mortar are also closer to the recorded ones. The closeness of recorded amplitude to the corrected amplitude of AE signals in fiber-reinforced concrete was defined with the equation of y = x^1.03.

The effectiveness of the proposed correction method was verified by clustering fiber activities from concrete cracking activities for the first time in the literature with WPF and PP. Cluster-1 (50–200 kHz) was associated with cement paste damage for both concrete, fiber-reinforced concrete, and cement mortar specimens. Cluster-2 (200–250 kHz) was only seen in fiber-inclusive specimens. Fiber activities are distinguished more clearly after parameter correction by recalculating amplitude and RA values. Successfully matched fiber activities were revealed after correction and their origination times obtained from frequency-based unsupervised pattern recognition show that more accurate AE interpretation could be made with parameter correction technique for excessive heterogeneous concrete structures. In addition, an inference was proposed for a relationship between the amplitude and energy losses of the AE signals and the type of damage after parameter correction.