METHOD FOR TESTING A DAMAGE TOLERANCE PROPERTY OF A PART MADE OF AN ALUMINIUM ALLOY

Abstract

The invention involves a method for testing a damage tolerance property in an aluminum alloy part with the following steps: measure at least one property representative of a the part's tensile strength; use the property measured in step a) as input datum (x.sub.i) of a neural network estimator; estimate, using the estimator, the representative property of the part's tensile strength; the method being characterized in that it includes: consideration of an acceptance threshold and comparison of the property estimated at step c) to the acceptance threshold, taking into account a confidence interval; based on the comparison: consider that the part passes the test; or consider that the part does not pass the test.

Claims

1. A method for testing a damage-tolerant property in an aluminum alloy part, the part being in the shape of a metal sheet or extruded profile, said method comprising a) Measuring at least two properties resulting from tensile testing of the part, in the L (longitudinal) and/or ST (short transverse) and/or LT (long transverse) directions, the properties being chosen from: yield strength; and/or tensile strength; and/or elongation at rupture; b) take into account the thickness of the part; c) using properties measured during a) along with a thickness taken into account during b) as input parameters (x.sub.i) for a neural network estimator; d) Estimating, using an estimator, the property representative of a part's damage tolerance ({circumflex over (z)},) e) Taking into account an acceptance threshold ({tilde over (z)}, K.sub.IC and a confidence interval (□) and comparing the property estimated ({circumflex over (z)}, K.sub.IC during d) to the acceptance threshold, taking into account the confidence interval; f) based on the comparison: consider whether said part passes the test; or consider whether said part does not pass the test.

2. The method according to claim 1, in which when, during f), a part fails the test, the method comprises g) measuring damage tolerance property of the part from a test specimen taken from said part.

3. The method according to claim 1 in which c) also comprises taking into account a concentration of at least one alloy element in the aluminum alloy.

4. The method according to claim 1, in which the a) comprises measuring elongation at rupture and in which c) involves taking into account elongation at rupture thus measured.

5. The method according to claim 4, in which a) comprises measuring yield strength and in which c) comprises taking into account yield strength thus measured.

6. The method according to claim 1, in which c) also comprises consideration of a hardness property.

7. The method according to claim 1, in which a) consists of only measuring elongation at rupture and yield strength, and in which c) comprises consideration of elongation at rupture and yield strength thus measured.

8. The method according to claim 1 in which: when d) comprises an estimation of the property representative of a damage tolerance of a part according to the L-T (Longitudinal—Long Transverse) directions, c) comprises, at least, consideration of one or more properties measured during a), in direction L; when d) comprises an estimation of a property representative of a damage tolerance of the part according to the T-L (Long Transverse—Longitudinal) directions, c) comprises, at least, consideration of one or more properties measured during a), in direction LT; when d) comprises an estimation of the property representative of a damage tolerance of the part according to the S-L (Short Transverse—Longitudinal) directions, step c) comprises, at the least, consideration of the properties measured during step a), in direction ST.

9. The method according to claim 1 in which c) involves consideration of properties measured during a), in several directions.

10. The method according to claim 9, in which c) involves taking into account three properties measured in a), in three different directions.

11. The method according claim 1, in which the damage tolerance property is an apparent intensity factor or a critical value of an intensity factor also called fracture toughness, or an effective stress intensity factor for a predetermined effective crack extension, using optionally 60 mm (K.sub.R60).

12. The method according to claim 1, in which the damage tolerance property is a ballistic limit velocity (V50).

13. The method according to claim 1, in which the aluminum alloy is an alloy of the 2XXX series, or the 7XXX series, or the 6XXX series, or the 5XXX series.

14. The method according to claim 1, in which c) is implemented by a processing unit, optionally a microprocessor.

Description

FIGURES

[0049] FIG. 1 shows, for different directions, results of fracture toughness measurements according to yield strength, for different test samples. Each point corresponds to a test sample. These results relate to AA2050 type aluminum alloys.

[0050] FIG. 2A is a schematic representation of a neural network architecture.

[0051] FIG. 2B represents the main steps of a procedure which is the subject of the invention.

[0052] FIG. 3A represents fracture toughness values (critical value of the stress intensity factor) estimated by a neural network type model according to measured values.

[0053] FIG. 3B shows a histogram of the relative fracture toughness estimation errors.

[0054] FIG. 4A represents consideration of a confidence interval, forming a safety margin, in the definition of an acceptance criterion, the latter corresponding to a fracture toughness value.

[0055] FIG. 4B illustrates, for different part thicknesses, the consideration of a confidence interval, forming a safety margin, with respect to an acceptance criterion.

[0056] FIG. 4C represents a definition of a confidence level, defined from the confidence interval, according to a risk level.

[0057] FIGS. 3A to 4C were established from estimates of the fracture toughness of type AA2050 aluminum alloys.

[0058] FIG. 4D illustrates, for different part thicknesses, the consideration of a safety margin with respect to a specification, using estimates of the fracture toughness of type AA7050 aluminum alloys.

[0059] FIG. 5A shows results of damage tolerance measurements according to yield strength for different test samples. In FIG. 5A, the considered damage tolerance property is an effective stress intensity factor for an effective crack extension of 60 mm (K.sub.R60).

[0060] FIG. 5B shows a histogram of damage tolerance property estimation errors (K.sub.app) on different types of aluminum alloy.

[0061] FIG. 5C shows a histogram of estimation errors of another damage tolerance property (K.sub.R60) on different types of aluminum alloy.

DISCLOSURE OF PARTICULAR EMBODIMENTS

[0062] The critical stress intensity factor, denoted Kw, sometimes referred to as fracture toughness, is determined according to a test protocol defined in standard ASTM E399-12, mentioned in the prior art. A pre-cracked specimen is subjected to an increasing load. The crack has an opening, whose progression is measured according to the load applied to the test specimen. A curve, representing the load applied according to opening, is obtained, according to which a stress intensity factor K.sub.Q is determined, this latter corresponding to an intersection of the aforementioned curve and a line of predetermined slope. Under certain conditions, specified in paragraph 9 of the aforementioned standard, the stress intensity factor K.sub.Q corresponds to a valid measurement of the critical stress intensity factor K.sub.IC. When these conditions are met, it is considered that the critical stress intensity factor K.sub.IC characterizes the material, being independent of the geometry of the test specimen considered. This magnitude, which corresponds to the fracture toughness in plane stress, is also referred to here simply by the term “fracture toughness”.

[0063] The apparent stress intensity factor at break K.sub.app, which corresponds to the fracture toughness in plane stress, is obtained by establishing a curve referred to by the term “R curve”, according to a test protocol defined in standard ASTM E561. The R curve represents changes in the critical stress intensity factor K.sub.C for crack growth, according to crack length, under a monotonic and increasing stress. The R curve allows a determination of the critical load for an unstable break. A stress intensity factor K.sub.CO can also be determined by assigning an initial crack length, before the load is applied. The apparent stress intensity factor at break K.sub.app is the K.sub.CO factor corresponding to the test specimen that was used to establish the R curve. The K.sub.R60 coefficient is the effective stress intensity factor for an effective crack extension of 60 mm.

[0064] The ballistic limit velocity is defined, for example, in the NF A 50-800-2 and 3 (2014) or MIL-STD-662 (1997) standards. This is the velocity at which the probability of armor plate penetration is 50%. The ballistic limit velocity is the mean of an even number of impact velocities, at least 4, half of which are protections, and the second half are non-protections. It is determined by calculating the mean velocity reached by the projectiles on impact resulting from taking the same number of results with the highest velocities corresponding to partial penetration and those results with the lowest velocities corresponding to complete penetration. Complete penetration occurs when the impacting projectile or any fragment (of the projectile or test specimen) pierces a thin control slab located behind the test specimen.

[0065] Structural element: a structural element of a mechanical construction is a piece for which static and/or dynamic mechanical properties are particularly important for the integrity of the structure. In an aircraft construction, these include, among others, the components of the fuselage, the wings, the tail unit and the vertical stabilizer.

[0066] In relation to the tensile tests, the terms sens travers, sens long (L), sens travers-long (TL), sens travers-court (TC) are defined in the NF EN 485 standard. They correspond respectively to the Anglo-Saxon designations Longitudinal (L), Long Transverse (LT or T) and Short Transverse (ST or S). In the following paragraphs, we shall use the acronyms L, LT and ST.

[0067] For damage resistance tests, the L-T, T-L and S-L directions are defined in standard ASTM E399-12, paragraphs 3.1.3.2 and 3.1.3.4. The first letter corresponds to a direction normal to the crack plane. The second letter corresponds to the crack growth direction. The following nomenclature is used in these designations: L=longitudinal; T=Long Transverse; S=Short Transverse.

[0068] The invention applies to aluminum alloys, and in particular to series 2XXX, 7XXX or 5XXX aluminum alloys. The alloys are named according to the nomenclature defined by The American Aluminum Association. The invention allows the testing of a piece made of aluminum alloy, and more precisely the testing of the part's damage tolerance property. The part can be a sheet, or some other type of part.

[0069] The invention takes advantage of a very large number of aluminum alloy parts having undergone precise mechanical or chemical characterizations, among which: [0070] chemical composition; [0071] thickness; [0072] mechanical tensile properties, in particular the tensile yield strength, the ultimate tensile strength, the relative elongation, or the elongation at rupture; [0073] properties representative of the damage tolerance, for example the fracture toughness (K.sub.IC), the apparent intensity factor at break (K.sub.app) or the intensity factor K.sub.R60 or the ballistic limit velocity (V50) defined beforehand.

[0074] For example, the inventors had access to damage tolerance data relating to 6200 parts made of AA2050 type alloy, and this according to the LT, TL and ST directions. This represents very important characterization data. They also had tensile strength test data in the L, LT and ST directions.

[0075] FIG. 1 shows a scatter plot, each point representing a test sample taken from a part. For each test sample, the fracture toughness (K.sub.IC—ordinate axis) was represented according to the yield strength (abscissa axis). The gray level of each point corresponds to a direction as previously defined, the legend being as follows: [0076] L-T: fracture toughness according to L-T, yield strength according to L; [0077] T-L: fracture toughness according to T-L, yield strength according to LT; [0078] S-T: fracture toughness according to S-T, yield strength according to ST; [0079] L-T 50 mm: fracture toughness according to L-T, yield strength according to L—sheet thickness equal to 50 mm.

[0080] There is a degree of correlation between yield strength and fracture toughness.

[0081] The inventors have developed an algorithm for estimating properties characterizing the damage tolerance, according to input parameters representative of the tensile strength. For this, some of the characterization results available were used to form a learning set used to parameterize the algorithm. Another part of the available characterization results were used to form an algorithm test set after its parameterization. 80% of the available data were used to form the learning set. 20% of the available data were used to form the test set.

[0082] The algorithm used is a neural network consists of an input layer, comprising the input parameters x.sub.i, an intermediate layer, or hidden layer, and an output layer, forming the magnitude to be estimated, in this case a damage tolerance property, e.g. fracture toughness. The intermediate layer forms a hidden layer, comprising y.sub.j nodes, or neurons. For each node y.sub.j, and for each input datum x.sub.i, there is a weighting factor w.sub.i,j determined during the learning phase. The inventors programmed the algorithm in the MATLAB® environment, software supplied by the company The Mathworks, by implementing the “ANN Toolbox” module. Learning serves to determine, among other things, the weighting factors of the hidden layer. In the example considered, the hidden layer has 30 nodes. Each node is linked to an input datum by a weighting factor and a bias.

[0083] FIG. 2A is a schematic representation of the architecture of a neural network such as that implemented by the inventors. The network has 3 layers: [0084] the input layer IN, comprising the input variables x.sub.i. the index i is an integer between 1 and N.sub.i, N.sub.i, being an integer greater than or equal to 2. Each input variable is: [0085] either a mechanical tensile property, in particular the relative elongation and/or the yield strength and/or the tensile strength; [0086] or the thickness of the part; [0087] or a mass fraction of an alloying element, for example a mass fraction of Cu or a mass fraction of Li. [0088] the hidden layer HI, comprising nodes (or neurons) y.sub.j. The index j is an integer between 1 and N.sub.j, N.sub.j being an integer greater than or equal to 5 or 10. In this example N.sub.j=30. [0089] the output layer OUT, comprising an output variable {circumflex over (z)}. In our example, the output layer has only one output variable {circumflex over (z)}, which corresponds to the damage tolerance property estimated by the algorithm. This may be an estimate of the fracture toughness in plane stress (K.sub.IC), but also of other parameters, for example the fracture toughness in plane stress K.sub.app or K.sub.R60.

[0090] Each node of the intermediate layer is linked to each input datum. In FIG. 2A, not all the relationships have been shown, for the sake of clarity.

[0091] The algorithm is implemented by a data processing unit, for example a microprocessor, connected to a memory comprising the algorithm and its parameterization. The algorithm uses measured physical data, corresponding to the input parameters x.sub.i mentioned above.

[0092] Each node y.sub.j is assigned a weighting factor associated with an input variable x.sub.i. Thus, each weighting factor is associated with an input datum x.sub.i and with a node y.sub.j. Each node also has a bias value w.sub.0,j. The weighting factors along with the bias w.sub.0,j, of each node are determined during the learning phase. Each node y.sub.j implements an activation function f.sub.j, such that:

[00001]$\begin{array}{cc}\left[\mathrm{Math}\right]& \\ y_j=f_j\left(w_{0,j}+\underset{i}{.Math.}{w}_{i,j}{x}_i\right)& \left(1\right)\end{array}$

[0093] In the architecture implemented by the inventors, each activation function f.sub.j is a hyperbolic tangent function. The values of each node y.sub.j are combined to form the value of the output variable {circumflex over (z)}.

[0094] The algorithm having been parameterized by the training set, tests aimed at evaluating the precision of the algorithm, that is to say the difference between the measured fracture toughness and the estimated fracture toughness, were carried out. FIG. 3A shows a curve, each point of which corresponds to a test sample. The abscissa and ordinate axes correspond respectively to fracture toughness (K.sub.IC) values measured and estimated by the algorithm. We observe that the scatter around the line of equation y=x is weak. The curve in FIG. 3A was created considering test samples in all directions and of different thicknesses, ranging from 30 to 200 mm.

[0095] FIG. 3B corresponds to a histogram of relative errors (in %) between the fracture toughness values respectively measured and estimated by the algorithm. This histogram serves to evaluate an estimation uncertainty of the algorithm. The standard deviation σ.sub.KIC of this histogram is estimated at 1.3 MPa√m, making it possible to quantify the estimation uncertainty.

[0096] FIG. 4A is a graph representing on the abscissa a required fracture toughness value, this value being for example derived from a specification by a constructor. On the ordinate, fracture toughness values estimated by the algorithm are represented. The solid line corresponds to the line of equation y=x. A dotted line has been drawn whose deviation from the line y=x corresponds to consideration of a margin of error, forming a confidence interval. According to a usual definition, the confidence interval corresponds to n times the standard deviation σ.sub.KIC, where n is a positive real number. n can be for example equal to 2.

[0097] This graph illustrates how the uncertainty associated with the fracture toughness estimate can be taken into account to ensure compliance with a requirement resulting from a specification. n×σ.sub.KIC corresponds to the confidence interval applied so as to take into account the uncertainty associated with the estimate. Thus, if {tilde over (K)}.sub.IC corresponds to an acceptance threshold of fracture toughness, defined in a specification, and if corresponds to the estimated fracture toughness resulting from the algorithm, compliance with the specification can be such that: [0098] if

[00002]$\begin{array}{cc}\left[\mathrm{Math}\right]& \\ {\tilde{K}}_{\mathrm{IC}}+n\times {\sigma }_{\mathrm{KIC}}\le {\hat{K}}_{\mathrm{IC}}& \left(2\right)\end{array}$ [0099] then the test sample is considered to conform to the acceptance threshold defined in the specification: the part thus passes the test. [0100] If

[00003]$\begin{array}{cc}\left[\mathrm{Math}\right]& \\ {\tilde{K}}_{\mathrm{IC}}+n\times {\sigma }_{\mathrm{KIC}}>{\hat{K}}_{\mathrm{IC}}& \left(3\right)\end{array}$ [0101] then the test sample is considered not to conform to the acceptance threshold defined in the specification: based on the estimate, the part fails the test.

[0102] Thus, in general, it is possible to evaluate, from the test set, a statistical indicator σ.sub.KIC representative of the scatter of fracture toughness estimates with respect to the exact fracture toughness values. The statistical indicator serves to define a confidence interval ϵ=n×σ.sub.KIC which is: [0103] either added to the acceptance threshold , in which case the value +ϵ is compared to the estimate . [0104] or subtracted from the estimate , in which case the value −ϵ is compared to the acceptance threshold .

[0105] FIG. 3B shows a generalization of this method on test samples of different thicknesses Th, with each thickness corresponding to a specification of fracture toughness .sub.Th. Each threshold value {tilde over (K)}.sub.IC,Th is increased by ϵ. Test samples with an estimated fracture toughness less than +ϵ are considered non-compliant. The confidence interval may depend on the thickness.

[0106] Test samples considered to be non-compliant may be subject to an experimental determination of their fracture toughness, in order to determine their compliance or non-compliance with the specification defining the acceptance threshold. The experimental measurement is carried out by taking a test specimen from the test part. The fracture toughness value resulting from the experimental measurement is then again compared to the acceptance threshold .

[0107] It is understood that the method serves to avoid carrying out an experimental fracture toughness measurement for all the parts such that:

[00004]$\begin{array}{cc}\left[\mathrm{Math}\right]& \\ {\tilde{K}}_{\mathrm{IC}}+\mathrm{.Math.}\le {\hat{K}}_{\mathrm{IC}}& \left(4\right)\end{array}$

[0108] The higher the value of n, the lower the percentage of avoided tests. FIG. 4C represents the percentage of tests avoided according to a confidence level associated with the confidence interval E, this latter corresponding to n×σ.sub.KIC. A 100% confidence level leads to an experimental measurement being carried out on all the test samples.

[0109] FIGS. 3A, 3B and 4A to 4C were created considering a type AA2050 aluminum alloy. FIG. 4D is analogous to FIG. 4C, considering a type AA7050 aluminum alloy.

[0110] FIG. 2B summarizes the main steps of a method for testing a part according to the invention.

[0111] Step 100: determination of the input parameters x.sub.i of the algorithm. All or part of the input parameters are determined experimentally.

[0112] Step 110: implementation of the algorithm, such as to obtain a value of the output variable {circumflex over (z)} corresponding to an estimate of the damage tolerance property considered. In the examples given in connection with FIGS. 3A, 3B, 4A to 4C, {circumflex over (z)}={circumflex over (K)}.sub.IC

[0113] Step 120: consideration of a confidence interval ϵ and of an acceptance threshold {tilde over (z)}. Comparison of the estimated property {circumflex over (z)} with the acceptance threshold {tilde over (z)} taking into account the confidence interval ϵ. This latter is either subtracted from the estimated property {circumflex over (z)} or added to the acceptance threshold {tilde over (z)}.

[0114] Based on the comparison: [0115] Step 130: acceptance of the test part; or [0116] Step 140: Experimental determination of the damage tolerance property z, based on which the part is either accepted (step 130) or rejected (step 150).

[0117] The algorithm used in step 110 has previously been trained, using learning test samples. Learning is the subject of a step 90, during which the number of hidden layers, the number of nodes per hidden layer and the activation functions associated with the nodes of the hidden layer are defined. Learning comprises an optimization, used to define the weighting factors for each given pair of input x.sub.i-node y.sub.j along with the bias w.sub.0,j associated with each node.

[0118] For AA2050 alloy sheets, the inventors were able to estimate the predictive power of the various input variables x.sub.i considered. Table 1 shows, for each input datum, the predictive power. This is a real number between 0 and 1, quantifying the relative importance of each input datum in the estimation of the result.

TABLE-US-00001 TABLE 1 Input variable x.sub.i Predictive power Elongation at rupture 0.88 Yield strength 0.57 Thickness 0.36 Tensile strength 0.16 Li (%) 0.08 Cu (%) 0.07

[0119] The input variables with the greatest influence on the estimation of fracture toughness are therefore the elongation at rupture, yield strength and thickness. The mass fractions of lithium and copper are of almost negligible significance in determining the mass fraction.

[0120] The inventors have developed an estimator, similar to that described above, capable of estimating intensity factors such as the apparent stress intensity factor K.sub.app or the intensity factor K.sub.R60 defined beforehand. FIG. 5A represents a variation of the yield strength (abscissa axis) according to the intensity factor K.sub.R60. Each cross corresponds to a test sample. As in FIG. 1, there is a certain correlation between the value of the yield strength and the value of the intensity factor K.sub.R60.

[0121] FIGS. 5B and 5C represent a histogram of the relative errors (in %) of relative estimates of the apparent stress intensity factor and of the intensity factor K.sub.R60. These histograms were produced from different types of alloy, which explains a greater scatter of the relative error values represented in FIG. 1B. In each of these figures, the dark gray levels (a) correspond to relative error values obtained using learning test samples, while the lighter gray levels (b) correspond to relative error values obtained using test samples. A straight line has also been shown corresponding to a zero relative error.

[0122] According to one embodiment, the input parameters of the estimation algorithm can include hardness values measured on the test part.

[0123] Further tests were performed to determine the influence of the input parameters on the accuracy of the fracture toughness estimate {circumflex over (K)}.sub.IC. In these tests, the estimates of the fracture toughness were separated in the respective directions L-T, T-L and S-L. The directions considered were also taken into account when carrying out the tensile tests.

[0124] In one series of tests, the products analyzed were AA2050 T851 alloy sheets. The parameters of these tests are shown in Table 2A. The results of these tests are listed in Table 2B.

[0125] In Table 2A, the columns correspond respectively to the following data: [0126] first column “Ref”: reference of the test; [0127] second column “Nb data”: number of test samples considered. 80% of the test samples form the model learning set. 20% of the test samples form the test set; [0128] third column “Thickness”: consideration (X) or non-consideration of the thickness in the model; [0129] fourth column “Widening”: consideration (X) or non-consideration of a widening factor when widening is carried out during rolling; [0130] fifth column “Orient”: consideration of the orientation, as detailed in connection with tests 7 and 8 [0131] sixth, seventh and eighth columns: consideration (X) or non-consideration of properties resulting from tensile tests (UTS, TYS and A% respectively) in the L direction. [0132] ninth, tenth and eleventh columns: consideration (X) or non-consideration of properties resulting from tensile tests (respectively UTS, TYS and A%) in the LT direction. [0133] twelfth, thirteenth and fourteenth columns: consideration (X) or non-consideration of properties resulting from tensile tests (respectively UTS, TYS and A%) in the TS direction.

[0134] In Table 2B, the columns correspond respectively to the following data: [0135] first column “Ref”: reference of the test; [0136] second, third and fourth columns: estimation (X) or non-estimation of fracture toughness respectively in the L-T, T-L and S-L directions; [0137] fifth column: standard deviation of the estimation error; [0138] sixth column: standard deviation of the relative estimation error; [0139] seventh column: mean relative estimation error.

[0140] Data for estimation errors, shown in the fifth, sixth and seventh columns, are obtained as a result of the fracture toughness estimate with each test set.

[0141] In tests 1, 2 and 3, three estimators were developed and used, each estimator being respectively dedicated to the estimation of fracture toughness according to the directions L-T, T-L and S-L, from tensile strength data respectively measured according to the L, LT and ST directions. Each estimator took into account the thickness of the part. During test 1, the estimator was configured using tensile strength data measured along the L direction. The estimator thus configured was used to estimate the fracture toughness along the L-T directions. In test 2, the estimator was configured using tensile strength data measured in the LT direction. The estimator thus configured was used to estimate the fracture toughness according to the T-L directions. In test 3, the estimator was configured using tensile strength data measured in the ST direction. The estimator thus configured was used to estimate the fracture toughness according to the S-L directions. The estimates are correct, with an mean error of less than 4%.

[0142] In tests 4, 5 and 6, three estimators were developed and used, respectively serving to estimate the fracture toughness according to the three directions L-T, T-L and S-L, from tensile strength data measured in the three directions L, LT and ST. The estimators also took into account the thickness of the part.

[0143] It is observed that tests 4, 5 and 6 lead to reduced estimation errors compared to tests 1, 2 and 3. This shows that, in order to estimate a fracture toughness value in a given direction (for example the L-T direction), it is preferable to have input parameters, resulting from tensile tests, not in a single direction (in this case L for direction L-T), but in three directions (L, LT and ST). Thus, taking into account tensile strength data measured in multiple directions, such as two or three different directions, reduces the error in estimating fracture toughness, regardless of the directions considered for estimating fracture toughness.

[0144] In tests 7 and 8, the same estimator was used, developed with tensile test data from three directions (L, LT and ST). In test 8, the thickness of the part and the post-rolling widening factor were taken into account. During test 7, part thickness was not taken into account. When implementing the estimator, the operator selected the directions (L-T, T-L or S-L), corresponding to column 6 of the table: [0145] when the selected direction is L-T, the input parameters are the tensile strength properties measured in the L direction; [0146] when the selected direction is T-L, the input parameters are the tensile strength properties measured in the LT direction; [0147] when the selected direction is S-L, the input parameters are the tensile strength properties measured in the ST direction.

[0148] We see that taking part thickness into account reduces the estimation error.

[0149] It follows from the above that it is optimal to take into account the thickness of the test part, along with the tensile strength properties in various directions. This reduces the estimation error and increases the repeatability of the measurements.

[0150] In one series of tests, the products analyzed were AA7050 T7451 alloy sheets. The parameters of these tests are shown in Table 3A. The results of these tests are listed in Table 3B. The columns of tables 3A and 3B contain the same data as the columns of tables 2A and 2B respectively.

[0151] Tests 9, 10 and 11 are similar to tests 1, 2 and 3 previously described in relation to Tables 2A and 2B.

[0152] Test 12 is similar to test 8, as the widening factor was not taken into account. Test 12 serves as a benchmark for comparison to tests 13 to 17. In tests 13 to 17, the estimator was configured and used without taking into account at least one of the parameters considered in test 12: [0153] in test 13, part thickness was not taken into account; [0154] in test 14, elongation at rupture was not taken into account; [0155] in test 15, tensile strength was not taken into account; [0156] in test 16, yield strength was not taken into account; [0157] in test 17, tensile strength and yield strength were not taken into account.

[0158] It follows from tests 12 to 17 that failure to take into account tensile strength properties increases the estimation error. It appears particularly optimal to consider the three mechanical traction properties (tensile strength, yield strength and elongation at rupture), along with the thickness of the part. Moreover, the mechanical traction properties must be determined in a direction conducive to the directions considered to estimate fracture toughness: in this case L, LT and ST to estimate fracture toughness in the L-T, T-L and S-T directions respectively.

[0159] The invention serves to avoid systematic use of destructive tests on test specimens, which are reserved for parts whose estimated damage tolerance value is not sufficiently far from the acceptance threshold, with a predetermined level of confidence. It paves the way for a new approach to part testing for high-demand applications, such as vehicles or aircraft.

[0160] In another test series no. 18 (see Table 4), the product analyzed was AA 5083 H131 alloy plates for which the ballistic limit velocity (V50) was estimated taking into account the mid-thickness traction characteristics in the TL direction, i.e. yield strength, tensile strength and elongation measured at mid-thickness in the TL direction. The results obtained are given in Table 4. In Table 4, the columns correspond respectively to the following data: [0161] first column “Ref”: reference of the test; [0162] second column: number of data taken into account, here 448, [0163] third, fourth, fifth and sixth columns: parameters taken into account (X); [0164] seventh column: type of values estimated, here the ballistic limit velocity (V50); [0165] eighth column: standard deviation of the estimation error; [0166] Ninth column: standard deviation of the relative estimation error; [0167] Tenth column: mean relative estimation error.

[0168] It is thus shown that the previously described fracture toughness determination approach applies to the ballistic limit velocity (V50). It is also remarkable that the standard deviation of the relative error is small, indicating an excellent prediction by the method.

TABLE-US-00002 TABLE 2A L L L LT LT LT TS TS TS Nb UT TY A UT TY A UT TY A Ref. data Thickness Widening Orient. S S % S S % S S % 1 6435 X X X X 2 6434 X X X X 3 5544 X X X X 4 6433 X X X X X X X X X X 5 6432 X X X X X X X X X X 6 5543 X X X X X X X X X X 7 18321 X X X X X X X X X X X X 8 18321 X X X X X X X X X X X X X X

TABLE-US-00003 TABLE 2B L-T T-L S-L σ(ε) Mean(ε) Ref. KIC/Kq KIC/Kq KIC/Kq σ(ε) (%) (%) 1 X 1.64 4.65 3.60 2 X 1.02 3.46 2.69 3 X 1.31 5.17 4.02 4 X 1.42 4.03 3.07 5 X 0.91 3.12 2.40 6 X 1.24 4.91 3.80 7 X 1.65 5.31 4.07 X X 8 X 1.26 4.19 3.17 X X

TABLE-US-00004 TABLE 3A Nb L L L LT LT LT TS TS TS Ref. data Thickness Widening Orient. UTS TYS A % UTS TYS A % UTS TYS A % 9 1480 X X X X 10 1480 X X X X 11 1480 X X X X 12 4440 X X X X X X X X X X X X X 13 4440 X X X X X X X X X X X X 14 4440 X X X X X X X X X X 15 4440 X X X X X X X X X X 16 4440 X X X X X X X X X X 17 4440 X X X X X X X

TABLE-US-00005 TABLE 3B L-T T-L S-L σ(ε) Mean(ε) Ref. KIC/Kq KIC/Kq KIC/Kq σ(ε) (%) (%) 9 X 0.94 2.74 2.04 10 X 0.88 2.91 2.16 11 X 1.17 3.63 2.85 12 X 1.23 3.76 2.96 X X 13 X 1.58 4.74 3.70 X X 14 X 1.29 3.95 3.11 X X 15 X 1.34 4.08 3.20 X X 16 X 1.29 3.93 3.08 X X 17 X 1.36 4.16 3.27 X X

TABLE-US-00006 TABLE 4 Inputs Target (σ) Mean UTS TYS A % V50 Value (ε) REF Nb data Th. (LT) (LT) (LT) (m/s) (m/s) (%) (%) 18 448 X X X X X 0.28 0.039 0.05