Method for ascertaining cracking, delamination and / or degradation of circuit board

Abstract

The invention relates to a method for determining cracking, delamination and / or degradation of a circuit board. Additional additional conductor tracks are applied to the upper side of the circuit board. The geometry of the additional conductor track used as the antenna is designed in such a way that a quarter-wavelength converter representing the HF antenna is formed in conjunction with the carrier substrate of the circuit board. At least one characteristic frequency with a minimum value of the reflection coefficient is determined, and the minimum value has a strong correlation with the geometry of the conductor track. The ascertained reflection coefficients and transmission scattering parameters are summarized in the frequency map by means of frequencies or frequency ratios and the location of the damage or crack. The transmission scattering parameters are evaluated by means of a neural network, the neural network being trained with the transmission scattering parameters and information about the location and size of the damage / damage. The trained neural network determines damage parameters and damage positions of the damage / damage of the circuit board via a vector network analyzer while taking into account transmission scattering parameters.

Description

Technical Field

[0001] The present invention relates to a method for determining cracking, delamination and / or degradation of a circuit board having a plurality of conductor lines, especially in the early stages in which cracking, delamination and / or degradation occur. Furthermore, the present invention relates to the use of this method for on-site testing of a circuit, especially a circuit board, using high-frequency signals. Background Art

[0002] US 8,013,600 B1 relates to a mountable eddy current sensor for on-site detection of fatigue cracks on a surface and below the surface. A battery-operated eddy current sensor for surface inspection and for detecting defects below the surface is proposed, which includes a first eddy current coil arranged flat and a second eddy current coil also arranged flat. A signal amplifier and a digital-to-analog converter are provided, which generate an output signal for the flat first eddy current coil. An analog-to-digital converter is provided, which obtains a signal from the flatly arranged second eddy current coil and digitizes it, thereby generating an input signal, wherein the mentioned analog-to-digital converter, analog-to-digital converter and digital synthesizer are synchronized. Furthermore, a circuit for determining the voltage ratio and phase shift between the input signal and the output signal is provided. The mentioned circuit includes a flatly arranged first eddy current coil, a flatly arranged second eddy current coil and a bridge circuit, which is configured such that the bridge circuit detects a change in differential impedance between the flat first eddy current coil and the second eddy current coil. Furthermore, a battery is provided, which supplies electrical energy to the digital synthesizer, digital-to-analog converter and analog-to-digital converter and to the switching elements for determining the voltage ratio and phase shift.

[0003] EP 3 861 361 B1 relates to a method and device for monitoring the reliability of an electronic system. According to a predetermined transmission quality metric, the transmission quality V of a signal transmitted to the electronic system through a wired transmission route is repeatedly measured at different measurement time points t. At each measurement time point, the associated measured transmission quality V is compared with a respectively assigned reference value of the transmission quality predetermined according to the transmission quality metric. The value of the reliability index assigned to the corresponding measurement time point is determined according to the result of the comparison performed at this measurement time point, wherein the transmission quality metric is defined as a measure of the size of a sub-range of a one-dimensional or multi-dimensional operating parameter range of the electronic system. In which, the electronic system operates reliably according to a predetermined reliability criterion.

[0004] DE 10 2021 005 497 A1 relates to a system and method for ensuring the material fatigue quality of a substance by storing data with the aid of a learning system. A storage and / or logic circuit system is proposed, in particular for ensuring material fatigue quality, which has a system and / or method having at least one neural network, in particular at least one switchable resistive structural element, which has a controlled variable resistance, in particular a resistive structural element. The resistive structural element changes the electrical conductivity characteristics of a component, in particular at least one memristor. In particular, in the storage and / or logic circuit, a plurality of applications are provided, in particular by means of at least one component memristor.

[0005] In applications in the automotive field, the electronic circuits usually arranged on a circuit board are exposed to very strong thermomechanical loads. This can lead to premature failures. A large amount of experimental work has been carried out in order to predict such failures, examine the aging and damage processes and derive material and lifetime models. It is known from "Early Detection of Interconnected Degradation by Continuous Monitoring of HF Impedance" by D. Kwon, M. H. Azarian, M. Pecht, IEEE Transactions on Device and Materials Reliability, Vol. 9, No. 2, June 2009 that the high-frequency TDR coefficient can be used to predict degradation phenomena in advance. Here, the so-called "skin" effect is used for advance selection prediction. The phenomenon of the current increasing on the upper side of the conductor is fully utilized. Since cracks almost always occur on the surface of the conductor, by making full use of high frequencies, changes in the conductor characteristics can be more clearly identified within the scope of the above-mentioned "skin" effect. Summary of the Invention

[0006] According to the invention, a method is proposed for determining cracking, delamination and / or degradation of a circuit board having a plurality of conductor tracks, in particular in the early stages in which cracking, delamination and / or degradation occur, wherein the following method steps are carried out:

[0007] a) applying additional conductor tracks to the upper side of the circuit board,

[0008] b) designing the geometry of the additional conductor tracks to be used as an antenna such that, in combination with the carrier substrate of the circuit board, a quarter-wavelength transformer (QWT) is formed, which is used as an HF antenna;

[0009] c) determining at least one characteristic frequency at which the reflection coefficient has a minimum value, which minimum value has a strong correlation with the geometry of the conductor tracks,

[0010] d) Summarize the reflection coefficients and transmission scattering parameters S obtained according to method step c) using the frequency and the location of damage / deterioration in the frequency diagram 11 、S 21 ,

[0011] e) Evaluate the transmission scattering parameters S through a neural network 11 、S 21 wherein the neural network is trained using the transmission scattering parameters S 11 、S 21 and information on the location and size of the damage / deterioration / crack; and

[0012] f) The trained neural network predicts the damage parameters and the damage location taking into account the transmission scattering parameters obtained by a vector network analyzer (VNA).

[0013] The transmission scattering parameters according to method step f) are different from the transmission scattering parameters S 11 、S 21 according to method step e). The training data set and the inference data set should be distinguished. During training, defects such as delamination occurring between circuit board layers or cracks on conductor lines are characterized and registered by their size, e.g., in terms of location and length. This information is assigned to the corresponding VNA measurements and forms the training data set. In the second phase, in the so-called inference phase, the trained neural network is used to monitor the circuit board and the conductor lines. If a defect occurs during actual operation, the neural network can be used to estimate the state and location of the defect. By chance, the length and location of the crack may happen to be the same as the corresponding parameters of the training set, but this is usually not the case.

[0014] In an advantageous expansion of the method according to the invention, simultaneously / parallel to the execution of method step c), damage / deterioration is determined by determining a strong increase in the amplitude of the reflected signal by means of a vector network analyzer (VNA). The occurring defects change the reflection characteristics. This change is particularly evident in the characteristic frequency of the λ / 4 transformer. A small shift in this frequency leads to a significant change in the amplitude of the reflection coefficient.

[0015] In a further advantageous expansion of the method according to the invention, in addition to the presence of damage / deterioration, its location and the relationship to the HF antenna are also determined.

[0016] Advantageously, in accordance with the method according to the invention, the transmission scattering parameters S 11 、S 21 can be selected according to method step d) such that these transmission scattering parameters are complex coefficients that are frequency-dependent and represent the ratio of the reflected signal to the input signal and the ratio of the output signal to the input signal.

[0017] In an advantageous refinement of the method according to the invention, according to method step d), the "time domain response" signal (TDR) is obtained by inverse FFT of the transmission scattering parameter S 11 . The time domain response signal (TDR) can be used to calculate the position of the defect by inverse FFT.

[0018] In an advantageous refinement of the method according to the invention, the amplitudes of the transmission scattering parameters S 11 and S 21 are selected as frequency-dependent complex coefficients for predicting crack growth in advance.

[0019] The method according to the invention is furthermore characterized in that, according to method step d), in the frequency diagram, the transmission scattering parameters S 11 and S 21 that are dependent on the frequency are plotted on the x-axis, and the crack position is plotted on the y-axis. This represents the correlation between the transmission scattering parameters S 11 and S 21 and the damage location, which can thereby be significantly promoted.

[0020] Advantageously, in the method according to the invention, the frequency diagrams generated according to method step d) are approximately opposite to each other. One curve trace represents the illustration of the transmission scattering parameter S 11 , and the other curve trace corresponds to the illustration of the transmission coefficient D 21 .

[0021] In the method according to the invention, in the training phase, the transmission scattering parameters S 11 and S 21 obtained by means of a vector network analyzer and the description of the crack state and its position on the circuit board are fed to a neural network formed as a 1D-CNN (one-dimensional convolutional neural network).

[0022] In an advantageous refinement, in the method according to the invention, a large number of measurements of the circuit board with damage / damage are fed to the neural network, so that the neural network has an effective database for performing comparison operations.

[0023] It is provided in the method according to the invention that the trained neural network predicts the crack state and the crack position by means of the transmission scattering parameters S 11 and S 21 obtained in the vector network analyzer.

[0024] The method according to the invention is further characterized in that by means of the transmission scattering parameters S 11 and S 21Data arrangement before being transmitted to the neural network, numerically identifying the length of the conductor line under discussion in the identifiable section and the conductor line itself.

[0025] The invention also relates to the use of this method for on-site testing of electronic circuits, especially printed circuit boards, using high-frequency signals with a trained neural network. Although high-frequency signals are used to determine the transmission scattering parameters S 11 、S 21 , printed circuit boards designed for low-frequency signals can also be fully monitored.

[0026] Advantages of the invention

[0027] Through the solution proposed according to the invention, by using a trained neural network, the experimental workload for generating models of the material and service life of printed circuit boards subjected to strong loads can be significantly minimized. Depending on the degree to which the used neural network is trained, multiple damage samples, crack samples, and degradation samples of the printed circuit board or the conductor lines on the printed circuit board can be collected as comparison materials, so that the prediction probability that can be obtained as a result from the neural network within the range of the output signal is very accurate and within the range of 90% and above. The obtained results can be used again as the input for data-driven reliability methods, and in addition as the initial basis for virtual design and used within the scope of the finite element method.

[0028] By combining artificial intelligence methods, the fatigue probability of materials and components can be predicted, which has already identified the occurrence of damage in advance, so that faults in a complete printed circuit board, such as in power electronics devices, can be identified in advance, and before the damage causes cracking, delamination, or similar severe effects, timely replacement of relevant components can be initiated, for example, within the scope of a workshop visit or by bringing a vehicle, for example, specifically into the workshop.

[0029] Self-diagnosis can be performed when the electronic device is turned on. When an error message appears, it is prevented that the connection of high-voltage or high-power signals causes greater damage. For electronic engineers or electricians, information about the location of the defect point can be completely valuable in order to visually confirm the formation of the defect as well. When a complete system failure is expected, it may be possible to notify the user based on the status of the defect. Based on this information, the user can already make decisions. These decisions can be made, for example, in such a way that it is determined whether preparations for replacement or repair are necessary, whether it makes sense to wait for a discount event or a new version, or whether replacement must be carried out earlier than the set possibility. In larger investments, timely financial preventive measures can also be prepared in order to provide sufficient liquidity at the time of replacement. Brief description of the drawings

[0030] The embodiments of the invention are explained in detail with the aid of the drawings and the following description. Among them:

[0031] Figure 1 shows a perspective top view of a circuit board with conductor tracks and additional conductor tracks serving as HF antennas,

[0032] Figure 2 shows a schematic diagram of a field test assembly for, for example, a circuit board to be tested,

[0033] Figure 3 shows a perspective view of the circuit board being contacted,

[0034] Figure 4 shows the course of the magnitude of the reflection coefficient plotted via the frequency ratio,

[0035] Figure 5 shows an enlarged view of a damage (shown here as a crack) in the conductor track,

[0036] Figure 6.1 、 6.2 、6.3 shows a frequency diagram for transmitting the scattering parameters S 11 、S 21 of,

[0037] Figure 7 shows the training phase of a neural network,

[0038] Figure 8 shows, according to Figure 7 the prediction phase of the trained neural network,

[0039] Figure 9 shows the segmentation of the conductor tracks on the upper side of the circuit board,

[0040] Figure 10 shows a comparison of the original data, the training augmented data, and the validation augmented data. DETAILED DESCRIPTION

[0041] In the following description of the embodiments of the present invention, the same or similar elements are denoted by the same reference numerals, where, in individual cases, the repeated description of these elements is dispensed with. The drawings only schematically show the subject matter of the present invention.

[0042] Figure 1 shows a perspective view of a circuit board 10 to be tested having a plurality of conductor tracks 18 and additional conductor tracks 20 representing HF antennas 22. Figure 1 The shown circuit board 10 has a plurality of conductor tracks 18 on its upper side 12, namely a first conductor track 18.1, a second conductor track 18.2, and optionally further non - shown nth conductor tracks 18.n. The circuit board 10 includes a carrier substrate 16 and a lower side 14. From according to Figure 1It is also obtained in the perspective top view that the additional conductor line 20 serving as the HF antenna 22 extends on the upper side 12.

[0043] The conductor line 10 or its carrier substrate 16 has a long side 28 and a wide side 30. With respect to the long side 28, the long side is approximately bisected by the reference line 26. From Figure 1 It is also obtained that the first conductor line 18.1 has a damage in the form of a crack 24 here. With respect to the reference line 26, the damage point or crack 24 is located near the coordinates of the crack position 36 along the Y direction from the reference line 26. The reference numeral 38 refers to the first connection side, and the HF antenna 22, i.e., the additional conductor line 20, is contacted at least on one side through the first connection side. In addition, from according to Figure 1 the perspective top view, it is obtained that the HF antenna 22 detects or absorbs the transmission scattering parameters S 11 、S 21 32.

[0044] Figure 2 A general overview of the measurement and evaluation configuration is shown, where, for example, the circuit board 10 described in Figure 1 is used as the circuit board 50 to be tested. The crack 24 or impending damage in the material of the circuit board 10 or the conductor line 18 is detected by a vector network analyzer (VNA) 42. The vector network analyzer 42 schematically shown in, for example, Figure 2 can operate in the frequency range from 100 MHz to 8.2 GHz. The occurrence of the crack 24 or damage point in the conductor line 18 is located, where these conductor lines 18 extend adjacent to the additional conductor line 20 representing the high-frequency antenna (HF antenna) 22. An input signal 46 is generated by a frequency source 44 provided in the vector network analyzer 42, and the input signal 46 is fed to the circuit board 50 to be tested. The reflected signal 48 arrives at one or more receivers 54. The output signal 52 that also leaves the circuit board 50 to be tested also arrives at the receivers 54 and is output on an output unit 56 in the form of a personal computer with a display.

[0045] In the illustration according to Figure 3 it is basically correspondingly recognizable that the first conductor line 18 extending on the upper side 12 of the carrier substrate 16 of the circuit board 10 has a damage in the form of a crack 24. The damaged conductor line 18.1 extends at a small distance from the additional conductor line 20 serving as the HF antenna 22. It is connected to the connection terminal via a connection line 58 on the first connection side 38. In a similar manner, this also applies to according to Figure 1 the first conductor line 18 has a damage in the form of a crack 24. The damaged conductor line 18.1 extends at a small distance from the additional conductor line 20 serving as the HF antenna 22. It is connected to the connection terminal via a connection line 58 on the first connection side 38. In a similar manner, this also applies to according to Figure 3Opposite short end sides of the circuit board 10 in the perspective view. A plurality of conductor lines 18 extend on the upper side 12 of the circuit board 10, and only some of the conductor lines are shown here. Positions 28, 30 refer to the long side 28 and the wide side 30 of the circuit board 12.

[0046] Figure 4 Different courses of the magnitude 62 of the reflection coefficient are shown, which is shown in a manner plotted via the frequency ratio f / f 0 (see position 60). The magnitude 62 of the reflection coefficient (Γ value) extends to the value 1, where different coefficients Z L / Z 0 .

[0047] Figure 5 An enlarged view of the damage in the form of a crack 24 on the upper side 12 of the circuit board 10 is shown. Here, for example, the first conductor line 18.1 selected is torn on one side, where Figure 5 the crack 24 shown enlarged has a crack length 66 and a width 64 schematically shown perpendicular to the crack. In the illustration according to Figure 5 , the first conductor line 18.1 is not yet completely broken, but at most about half of the conductor line material is torn. Such pre-damage shown, for example, in Figure 5 may lead to a complete tear or complete interruption of the first conductor line 18.1 during further operation, which results in impaired function of the circuit board 10.

[0048] Figure 6.1 and 6.2 Exemplarily show the selected frequency diagrams 72, 74 for transmitting the scattering parameters S 11 , S 21 in the case of a width 64 of the crack 24 of, for example, 1 mm. The shown magnitudes of the transmission scattering parameters S 11 , S 21 are plotted as a function of frequency. The Y value 36 of the crack position is introduced as an additional parameter, where, for example, in Figure 6.1 and 6.2 the frequency diagrams 72, 74 shown have the frequency plotted on the X-axis and the crack position Y 36 plotted on the Y-axis. The transmission scattering parameters S 11 , S 21 are generated for the corresponding crack lengths, where a crack length of 0.1 mm is shown here. Each of the two corresponding frequency diagrams 72, 74, i.e., the frequency diagram 72 for S 11 and the frequency diagram 74 for S 21 can be understood as a cross-section at a specific value of the crack position Y 36 plotted via the distance 68. Since the energy absorption is relatively low, the transmission scattering parameters S 11 and S21 The sum is substantially constant in the preset frequency: S 11 |f x +S 21 |f x ≈ constant.

[0049] For transmitting the flow parameter S 11 、S 21 The frequency diagrams are substantially opposite to each other. According to the illustration in FIG. 6, from the transmission scattering parameters S 11 、S 21 The variation process (see solid line and dotted line) results in, for example, a first local minimum 84 and an adjacent second local minimum 86 for the transmission flow parameter S 11 There is a first local minimum 80 and a second local minimum 82 arranged adjacent thereto for the transmission scattering parameter S 21 .

[0050] Figure 7 Shows a schematic diagram of the training phase 90 of the neural network 88. The neural network can be formed, for example, as a 1D-CNN (one-dimensional convolutional neural network). In Figure 7 The training phase 90 schematically shown in, the neural network 88 is prepared as follows, that is, a large number of identifiable samples of the training phase 90 are extracted from the circuit board 10 and its conductor lines 18.1, 18.2, 18.n. Information about damage, such as crack 24 and its location, is transmitted to the neural network 88 via transmission 100. The neural network is basically constructed by a plurality of planes 94, and each node 96 is located in these planes. In addition, during the training phase 90 of the neural network 88, starting from the vector network analyzer 42 (VNA), the transmission scattering parameters S 11 、S 21 Are transmitted 98 to the data collation 92. The data collation in turn conveys the collated transmission scattering parameters S 11 、S 21 To the input side of the neural network 88 during the training phase 90 of the neural network. During the training phase 90, the database inventory within the neural network 88 is increased so that a sufficient number of comparison samples are available, with the aid of which an evaluation can be performed. If the neural network 88, for example, in Figure 7 The training phase 90 schematically shown in ends, then it is possible to transfer to the prediction phase 104 schematically shown in Figure 8 .

[0051] In accordance with Figure 7During the training phase 90, a large number of measurements of damaged conductor lines 18.1, 18.2, and 18.n are transmitted to the neural network 88 with the help of samples. Damage, delamination, or cracks 24 in different propagation phases, crack states, and different positions are used to train the yet untrained neural network 88. If the database stock is processed by the neural network 88, then the neural network becomes the trained neural network 102, which can predict the localization of damage or injury in the form of cracks 24 on different conductor lines 18.1, 18.2, 18.n. Since in the trained neural network 102, information about the damage state, injury, or crack 24 and its position is already known in a large number of comparison examples, as Figure 8 shown, only the transmission scattering parameters S 11 、S 21 are fed to the trained neural network 102. With the transmission scattering parameters S 11 、S 21 32, the reflection coefficient, and their further processing and collation in the data collation 92, including multiple planes 94 and multiple nodes 96, the trained neural network 102 can predict the prediction 106 about the crack state or damage state and its position in the conductor lines 18.1, 18.2, 18.n on the upper side 12 of the circuit board 10. In Figure 7 and 8 shown in the data collation 92, the transmitted data, especially the transmission scattering parameters S 11 、S 21 32 transmitted through the transmission 98 are collated so that these transmission scattering parameters are better applicable to machine learning. For this purpose, classification is carried out. For example, the lengths of the conductor lines 18.1, 18.2, 18.n on the upper side 12 of the circuit board 10 are divided into individual segments 108. This is obtained in more detail from the illustration according to Figure 9 . Each segment 108 is designed such that it is numerically identifiable, so that the segment 108 is divided into the first segment 108.1, the second segment 108.2, or the nth segment 108.n. The corresponding conductor lines 18 on the upper side 12 of the circuit board 10 are also identified and can be identified by the index n. Therefore, the damage occurring in the conductor line 18 can be located, whether delamination or crack 24 occurs.

[0052] If the number of training examples for training the neural network 88 is too small, then the available data set can be expanded using the following examples, which are obtained from the original database stock and in which, for example, sound or noise is added.

[0053] In an architecture of a neural network 88, for example, formed as a 1D-CNN, the neural network has 32 channels and has filters of size 3. The second and third one-dimensional layers, for example, have 64 channels and filters of size 3. All pooling layers have size 2.

[0054] Figure 9 Segments of respective conductor lines 18 on the upper side 12 of the circuit board 10 are shown. The conductor lines 18.1, 18.2, and 18.n extend between a first connection side 38 and a second connection side 40; in addition, an additional conductor line 20 serving as an HF antenna 22 is shown on the upper side 12, which is connected to the connection sides 38, 40 via connection lines. Segmenting the respective conductor lines 18 into a plurality of segments 108 or segmenting the first conductor line 18.1 into segments 108.1, 108.2 allows for accurate identification and precise localization of the location of damage or breakage in the form of cracks 24 that occur.

[0055] Figure 10 The accuracy of the prediction 106 by the trained neural network 102 is shown in a manner drawn via multiple iterations 110. The reference numeral 112 refers to the raw data obtained from a series of measurements. The reference numeral 114 refers to "enhanced verification", and in addition, the reference numeral 116 refers to "enhanced training". On the specimen, there are antennas connected to a vector network analyzer (VNA) 42 and two routes (n = 2, n = 3) where cracks 24 may occur. These two routes are each divided into 10 uniform segments 108, i.e., a total of 20 segments 108, and are consecutively numbered. The task of the neural networks 88, 102 is to indicate in which of these 20 segments 108 the crack 24 is located. During training and inference, instead of the actual crack location, the segment numbers are processed. Thus, classification is performed instead of interpolation. Figure 10 The time flow of the training phase 90 and the verification process are shown. The number of iterations is shown on the X-axis, and the accuracy is shown on the Y-axis. The training of the untrained neural network 88 is an iterative process. After each training step, a test is performed, in which the transmission scattering parameters 32 of the training set are fed into the neural network to be trained. The response (prediction) of the neural network 88 is compared with the crack location 36 (the segment number of the segment with a crack). What is stored is whether the prediction is correct or incorrect. Here, it is important that the training points and test points come from the training data set, and the untrained neural network 88 is trained using this training data set during the training phase 90.

[0056] Finally, the hits (Treffer) are added up and compared with the total number of test points. This ratio determines the accuracy and can potentially be expressed as a dimensionless number (0 - 1) or as a percentage (0% - 100%). Since the test is performed after each iteration step, its accuracy value can be stored as a function of the number of iterations. Figure 10 The diagram is shown. The same applies to the validation dataset. Importantly here, the validation points are not obtained from the training dataset, but from the validation dataset. The validation results are also obtained from Figure 10 obtained.

[0057] Reference numeral 114 refers to "enhanced validation", and in addition, reference numeral 116 refers to "enhanced training". "Enhanced validation" 114 is a method used during the training of the untrained neural network 88. It is developed to avoid overtraining. Overtraining occurs when too few data points are available for training the untrained neural network 88. Subsequently, it results in the fact that the untrained neural network 88 reaches sufficient accuracy during the test phase of the training phase 90, but produces poor values during validation.

[0058] The invention also relates to the use of the proposed method for on-site testing of a circuit, in particular a circuit board 10, with the aid of high-frequency signals using an evaluation implemented by a trained neural network 102, said circuit board having a plurality of conductor tracks 18.1, 18.2 and 18.n.

[0059] The invention is not limited to the embodiments described herein and the aspects emphasized therein. On the contrary, various modifications can be made within the scope defined by the claims, which are within the scope of the person skilled in the art.

Claims

1. A method for determining cracking, delamination and / or degradation of a circuit board (10) having a plurality of conductor tracks (18.1, 18.2, 18.n), in particular in an early stage of the occurrence of cracking, delamination and / or degradation, the method comprising the following method steps: a) applying a further additional conductor track (20) to the upper side (12) of the printed circuit board (10), b) designing the geometry of the additional conductor path (20) used as an antenna so that, in combination with the carrier substrate (16) of the circuit board (10), a λ / 4 wavelength converter is formed, which serves as an HF antenna (22), c) determining at least one characteristic frequency at which the reflection coefficient has a minimum value (80, 82, 84, 86), said minimum value having a strong dependence on the geometry of the conductor track (18.1, 18.2, 18.n), d) summarizing the reflection coefficients and transmission scattering parameters (S) determined according to method step c) in a frequency diagram (72, 74) using the frequencies and the positions of the damage / impairment (24) 11 , S 21 ), e) evaluating the transmission scattering parameter (S) by means of a neural network (88, 102) 11 , S 21 ),in, The neural network uses the transmission scattering parameter (S 11 , S 21 ) and information about the location and size of the damage / crack (24) are trained, and a neural network (88) is trained, wherein the training data set contains the transmission scattering parameters (S 11 , S 21 ) and information about the location and size of the defect (24), f) Considering the transmission scattering parameter (S) obtained by vector network analyzer (VNA) (42) 11 , S 21 ), the trained neural network (102) describes the damage parameters and the damage location (106).

2. The method according to claim 1, characterized in that Simultaneously / parallel to method step c), an impairment / destruction (24) is ascertained by ascertaining a strong increase in the reflected signal amplitude using the vector network analyzer (VNA) (42).

3. The method according to claim 1 or 2, characterized in that: In addition to the presence of a damage / destruction (24), the position of the damage / destruction and its relationship to the HF antenna (22) are determined.

4. The method according to any one of claims 1 to 3, characterized in that According to method step d) the transmission scattering parameter (S 11 , S 21 ), so that the transmission scattering parameter is a frequency-dependent complex coefficient, and the complex coefficient represents the ratio of the reflected signal (48) to the input signal (46) and the ratio of the output signal (52) to the input signal (46).

5. The method according to any one of claims 1 to 4, characterized in that According to method step d), the transmission scattering parameter (S 11 ) to obtain the "time domain response" signal (TDR signal).

6. The method according to any one of claims 1 to 5, characterized in that The transmission scattering parameter (S 11 , S 21 ) is chosen as a frequency-dependent complex coefficient for early prediction of crack growth.

7. The method according to any one of claims 1 to 6, characterized in that According to method step d), in the frequency diagram (72, 74), the frequency-dependent transmission scattering parameter (S 11 , S 21 ) and plot the crack position on the Y-axis (36).

8. The method according to any one of claims 1 to 7, characterized in that The frequency diagrams (72, 74) generated according to method step d) are approximately inverse to one another.

9. The method according to any one of claims 1 to 8, characterized in that According to method step e), in a training phase (90), a neural network (88, 102) in the form of a 1D-CNN (one-dimensional convolutional neural network) is fed with the transmission scattering parameter S determined by means of a vector network analyzer (42). 11 , S 21 (32) and a description of the crack state and its location on the circuit board (10, 50).

10. The method according to claim 9, characterized in that A plurality of measurements of conductor paths (18.1, 18.2, 18.n) of a printed circuit board (10) having damage / impairment (24) are fed to the neural network (88, 102).

11. The method according to any one of claims 1 to 10, characterized in that The trained neural network (102) is obtained by using the transmission scattering parameter S obtained in the vector network analyzer (42). 11 , S 21 (32) to obtain the crack state and crack location (36).

12. The method according to any one of claims 1 to 11, characterized in that By converting the transmission scattering parameter S 11 , S 21 (32) Data collation (92) prior to transmission (98) to the neural network (88, 102) to numerically identify the position of the conductor lines (18.1, 18.2, 18.n) in the sections (108, 108.1, 108.2, 108.n) and the conductor lines (18.1, 18.2, 18.n).

13. Use of the method according to any one of claims 1 to 12 for on-site testing of an electric circuit, in particular a circuit board (10, 50), using a high-frequency signal or a low-frequency signal.

Images