3.1. Database for Determining ξ(δ) Coefficients
To begin with, three prismatic beams, B
1, B
2, and B
3, which differ in their geometric properties, were calculated in detail for different boundary conditions. For each type of beam, analyzes were performed with three different boundary conditions (cantilever clamped at the left end (C), simply supported beam (SS) and propped cantilever clamped at the left end (PC)). Geometric data of these beams (length
L, width
b and height
h), together with the analyzed boundary conditions, are given in
Figure 2,
Figure 3 and
Figure 4, representing nine basic (without detailed information on crack locations) combinations. The Young’s modulus
E and the axial load
Nx were 30 GPa and 1 MN, respectively. Furthermore, in each beam, transverse cracks were introduced individually at a distance of 0.5 m along each type of beam. At each location, six cracks of different depths were modeled. Relative depths between 0.1 and 0.6 were applied in steps of 0.1. Each crack was modeled on the upper surface of the beams, and as a result, its impact on the beam was reflected by a negative internal moment
MN.
Each 3D FE model consisted of 2000 elements in the axial direction, 30 in the vertical direction and 2 in the transverse horizontal direction. A total of 744 such models were analyzed (where almost 600,000 equations were solved in each analysis). The calculated vertical displacement at the crack point of each analyzed structure allowed the computation of the ratio
ξ(
δ) of the governing parameters of the model. The main results of each analysis were the vertical displacement at the crack location and the reactions in the supports (where they occurred). Ratio values for the same crack depths obtained for different locations were then compared to examine the effect of crack location on the identified ratio
ξ(
δ). The average results from all analyzed locations for each considered relative crack depth for beam type B
1 are given in
Table 1 for cantilever clamped at the left end (B
1C), simply supported beam (B
1SS) and propped cantilever clamped at the left end (B
1PC). In the table, the results for the two additionally analyzed beams with modified lengths, a cantilever of 7 m length, B
1C7, and a simply supported beam of length 6 m, B
1SS6, are also included (they were analyzed only for initial testing of the hypothesis that beam length does not affect the identified values.).
The identified values of
ξ(
δ) were certainly not ideally the same for each relative depth
δ, but their differences were clearly small enough to be easily attributed to the numerical modeling of the structure. Furthermore, since these values were identified at different crack locations
L1, their good agreement indicates that they depend solely on the relative depth of the crack
δ and not on the location of the crack itself. Therefore, the identification process was repeated for beam type B
2, and the obtained average results for all three considered boundary conditions are presented in
Table 2.
Although these values were not directly comparable to those from beam B1 due to the different cross-sectional dimensions, they also showed good agreement for each relative crack depth δ analyzed.
Consequently, the analyses were also performed for beam type B
3, and the results for all three analyzed boundary conditions are presented in
Table 3.
As with the previous beams, the average values of the three structures with different boundary conditions for each relative crack depth were in good agreement.
However, these values were again incomparable with the previously obtained ones due to the different cross-sectional dimensions of the analyzed beams.
Nevertheless, it can be reliably concluded from
Table 1,
Table 2 and
Table 3 that the ratio
ξ(
δ) clearly depends on the relative depth of the crack
δ, but at the same time it is independent of the location of the crack. Yet, in order to enable a direct comparison of the results for different cross-sections, it was necessary to convert the obtained
ξ(
δ) values to a common basis.
3.2. Conversion of Identified ξ(δ) Values into a Common Comparable Platform
In order to further compare the results for different cross-sections, Equation (4) was transformed into the following form in which the internal moment
MN in the numerator was defined by an upgraded definition compared to the simple definition given in Equation (1):
For a rectangular cross-section, the expression takes form:
where
EA represents the axial stiffness of the beam.
By introducing a new dimensionless variable representing the normalized local slope change:
the following dimensionless relation can be obtained:
which now allows a direct comparison of the obtained values for different cross-sections. Therefore, the values from
Table 1,
Table 2 and
Table 3 were transformed accordingly to the common platform and are presented in
Table 4.
Table 4a–c show results for beam types B
1, B
2 and B
3, respectively.
It can be seen from all the tables that the calculated values for each relative depth δ now show a rational agreement.
Nevertheless, for better insight into the dispersion of the results from all obtained values for each relative crack depth
δ, two extreme values (minimum value,
, and maximum value,
) were identified from the 127 values calculated for each relative depth. Further, the mean
and median
values as well as their ratios were also calculated for all relative crack depths,
Table 5.
It can be seen from
Table 5 that the mean and median values (columns 4 and 5) match almost perfectly, as their ratio (column 6) is almost equal to 1 for all relative crack depths. From the last two columns of
Table 5, it is further evident that also the deviations of the extreme values from the mean values (Δ
n and Δ
p) are rather minor, as they do not exceed ±1.7%. Therefore, it is obvious that from an engineering point of view, the same
function can be applied to different cross-sections. Its polynomial approximation is thus given as:
Figure 5 shows the piecewise linear approximation of the
data (black dashed line) and the polynomial approximation derived by Equation (16) (red line).
3.3. Decoupling the Governing Parameters of the Model
As it became apparent that the same functions
ρ(
δ) and
fN(
δ) could be used for different cross-sections, their separate evaluation was carried out. The research focus was on data on propped cantilevers, as it was clear from the displacement function of cantilever and simply supported beam (Equations (5)–(8)) that only the ratios
ξ(
δ) (but not separate individual functions) could be obtained. At first glance, the four parameters (displacement at the crack location, two vertical reactions, and a bending moment reaction) should allow direct calculation of both model parameters independently for each crack location
L1. However, basic equilibrium in the vertical direction requires that both vertical reactions be of equal magnitude (but in opposite directions), reducing the number of available parameters by one. Furthermore, the equilibrium of moments further implies that the reaction bending moment is equal to the vertical reaction multiplied by the length
L of the beam, which further reduces the number of parameters by one. Therefore, the number of equations available for each crack location was reduced to two, which was still sufficient to determine the two discrete unknowns. The solution for the rotational spring stiffness
KrN from which the function
fN(δ) could be further found, was thus derived as:
and the corresponding value
ρ(
δ) could be further evaluated from Equation (4). Therefore, in the first attempt, the analysis for each crack location provided both discrete values using Equations (4) and (17) sequentially. The results for beam type B
1PC and relative crack depth of
δ = 0.6 are given in the second and third columns of
Table 6 (where the value of
ξ = 8.650680 × 10
−3 was used). The fourth column of the table presents the discrete values of the lever arm function normalized to the value at the first crack.
The last column of the table shows that the mutual matching of the results is quite good (the differences between the results are less than 1%) for the first third of the locations (i.e., up to the first third of the beam length from the clamped end). However, the increase in the discrepancies of the results becomes quite apparent as the crack location L1 approaches the simply supported end. An even greater dispersion of results was evident for shallow cracks, where the zone with noticeable deviations extended further across the complete beam field.
Therefore, in the second attempt, only vertical reactions were used as input data for parameters identification. However, since only one vertical reaction cannot allow identification, vertical reactions from two arbitrary crack locations were used to obtain the two required values for each relative crack depth δ. Thus, each individual reaction was used multiple times, as it could be used in combination with all the remaining reactions for other crack locations of the same relative depth δ. In this way, combining the Nc crack data made it possible to obtain significantly more pairs of solutions than Nc.
Thus, the three considered propped cantilevers of beam types B1, B2 and B3 and their crack locations (35 in overall) gave a total of 188 solution pairs (MN, KrN) for each crack depth δ considered.
The results exclusively for beam type B
1PC and a relative crack depth of 0.6 are given in
Table 7, where all 55 identified
KrN values are presented, ranked in ascending order.
The differences between the results are still noticeable. Nevertheless, a comparison of
Table 7 and
Table 8 shows that the dispersion of the results in
Table 7 is evidently smaller as the deviations of the results from the mean value ranged from −0.902% to 0.778%. Similar, but even slightly smaller discrepancies of range −0.660% and 0.749% were detected for the values of the internal moment
MN. Thus, this approach to determining the two governing parameters not only provided a greater variety of results for
KrN and
MN, but also offered a more stable precision with a mean value that was very close to the median value.
Then, for each type of beam as well as for each relative crack depth
δ, the discrete values of the normalized lever arms
ρ(
δ) were evaluated and further averaged. These values are presented in
Table 8.
The weighted mean value for each relative crack depth
δ was further used in an interpolation procedure to obtain a polynomial approximation function for
ρ(
δ).
Figure 6, which proves the usefulness of the derived lever arm function, shows the piecewise linear interpolation of the original identified discrete values in the black dashed line, while the polynomial approximation function for
ρ(
δ) is plotted in the red line.
With the help of the known functions
and
ρ(
δ), it was further possible to calculate the missing function
fN(
δ) from the definition of the stiffness
KrN of the rotational spring. The required function follows from Equation (14) in the form:
The original function
fN(
δ) from Equation (19), initially given in the fractional form, was further replaced by a simple uniform polynomial function
fNp(
δ) for practical reasons only.
Figure 7 compares the graph of the original function
fN(
δ) (black line) with its simple polynomial replacement
fNp(
δ) (red line). From a comparison of the two graphs, it is evident that the match is generally more than decent, except in the area of very shallow cracks (i.e., δ < 0.1).
For the case without cracks, however, the rotational spring stiffness
KrN(
δ) must be infinite, which consequently requires the function in the definition denominator to be zero. Therefore, in the derivation of the final function, a zero initial condition was additionally included, which resulted in a slightly modified function
fNo(
δ):