Method for detecting production error of components in manufacturing
By dimensionality reduction and reconstruction of sensor data, the reconstruction error is calculated, and the problem of component production error detection is solved, and the detection sensitivity and testing efficiency are improved.
Patent Information
- Application Number
- CN202411379761.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-09-29
- Filing Date
- 2024-09-29
- Publication Date
- 2025-04-01
AI Technical Summary
During mass manufacturing of components, performance deviations caused by production errors can lead to poor quality or ineffective final products, and prior art is difficult to detect and resolve these errors effectively.
By providing sensor data with at least two dimensions, dimensionality reduction is performed to extract features, reconstruct the data after dimensionality reduction, calculate the reconstruction error, and detect the production error of the component based on this error.
This method can effectively detect the production error of components, improve detection sensitivity, simplify end-of-line testing, and reduce development consumption and testing time.
Smart Images
Figure CN119740002A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for detecting production errors of a component in manufacturing. Furthermore, the invention relates to a computer program, a device and a storage medium for this purpose. Background Art
[0002] In the case of mass production of components, thousands to millions of identical products are manufactured on partially or fully automated production lines. Depending on the complexity, the components can consist of more or fewer parts. These parts have in particular manufacturing tolerances, and the interaction of the parts with each other may also be subject to certain tolerances. If the product is, for example, a radar sensor for a motor vehicle, which can include a circuit board, an antenna, a SoC (System on Chip) and a plurality of low-frequency components, then, for example, tolerances on material properties apply to the circuit board, tolerances on main dimensions apply to the antenna, and tolerances on performance apply to the SoC, which tolerances may arise from the semiconductor circuit. Tolerances may also exist in the relative position between the antenna and the circuit board.
[0003] Due to tolerances, deviations from the desired performance may occur, which may lead to a poor quality or even non-functional final product. In particular, the goal of series development is to create a stable manufacturing process in which these tolerances are under control and most components have performance within the specification. However, errors may occur when manufacturing or assembling components, which may cause the components to be non-functional or outside the specification. If the manufacturing process is stable enough, the number of these components with errors is not very high, for example. If the component is, for example, a radar sensor for the automotive field, the antenna may be defective, the etching process on the circuit board may be performed incorrectly, or the SoC may provide less or more performance than described. There are situations in which the company that manufactures the component is legally obliged to guarantee a certain performance, for example in the case of components whose failure may endanger human life. For this reason, in particular, end-of-line (EOL) tests are performed to find components with production errors.
[0004] These EOL tests are performed, for example, in such a way that one or more parameters associated with the performance or requirements of the component are measured. If the component is, for example, a radar sensor for the automotive sector with requirements for angular recognition of targets, the performance can be estimated by measuring the antenna diagram of the radar sensor in a controlled environment such as a measurement room, and different performance indicators such as angular error values or correlation values can be calculated in a post-processing step.
[0005] In view of the fact that the inspection of production errors is usually particularly error-oriented during EOL testing, i.e., since 100% of the requirements for components cannot be tested, knowledge of the production errors is usually required in order to determine the post-processing steps for finding the corresponding defective components. This can lead to the need for high development costs to develop new EOL tests in order to find possible new production errors. The more complex the product, the more possible production errors may occur and the more tests may be required to find defective components. For example, each of the post-processing steps requires a certain amount of time. The time can be reduced with corresponding development costs, but even if all EOL tests only last for a moment, this time may be significant in production lines where the cycle time is only a few seconds.
[0006] The inputs for the EOL test, for example a large number of angle points based on antenna diagrams and various measurement parameters of the system on chip (SoC) as well as the adaptation for each transmitter and the noise and cross-coupling parameters for each receiver, can be correlated. For each individual measurement, there are in particular a plurality of post-processing steps or error detection steps, each of which can be designed for other error modes. However, for example in the prior art, the correlation between different measurement data is not used. If a new error type is found, a new post-processing step or error detection step is therefore in particular required. If there is a defective component, multiple limit values may be exceeded, but the sensitivity of the individual limit values may be low. That is, if there is a defective component with a small error, it is disadvantageous that none of the limit values may be exceeded. Summary of the invention
[0007] The subject matter of the invention is a method having the features of claim 1, a computer program having the features of claim 8, a device having the features of claim 9 and a computer-readable storage medium having the features of claim 10. Further features and details of the invention are derived from the respective dependent claims, the description and the drawings. Features and details described in conjunction with the method according to the invention naturally also apply in conjunction with the computer program according to the invention, the device according to the invention and the computer-readable storage medium according to the invention, and vice versa, respectively, so that with respect to the disclosure for the individual inventive aspects, reference is always made to one another or can be made to one another.
[0008] The subject matter of the invention is in particular a method for detecting production errors of a component during production, comprising the following steps:
[0009] - providing sensor data having at least two dimensions, wherein respective dimensions of the sensor data each comprise measurement data about the component,
[0010] - performing a dimensionality reduction of the sensor data, wherein at least one feature is extracted based on at least two dimensions of the sensor data,
[0011] - reconstructing the reduced-dimensionality sensor data based on the at least one extracted feature so as to provide reconstructed sensor data,
[0012] - determining a reconstruction error based on a comparison of the sensor data with the reconstructed sensor data,
[0013] - Detecting production errors of the component based on the determined reconstruction errors.
[0014] Production errors may be different types of defects of components, which are reflected in the sensor data of the corresponding components. In the case where the component is a radar sensor, for example, the antenna may be defective, the etching process on the circuit board may be performed incorrectly, or the SoC (system on chip) may provide less or more performance than described. The measurement data can be obtained from various types of measurements of the component or when the component is used. In the case of a radar sensor, the measurement data can be obtained, for example, from measurements of different angle points of the antenna diagram or using different measurement parameters. Dimensionality reduction can, for example, reduce the number of dimensions present in the provided sensor data to the number of extracted features. Therefore, the extracted features can represent the dimension of the sensor data after dimension reduction. For components, based on reconstruction errors, in particular, it is determined to what extent the sensor data deviates from a normal value or a reference value, which is given in the case of a component without production errors. Through this method, production errors can be advantageously detected based on sensor data with at least two dimensions. In other words, multiple input parameters, i.e., at least two measurement data, are considered in combination through this method, thereby advantageously improving the detection sensitivity.
[0015] Optionally, it is also conceivable that determining the reconstruction error comprises the following steps:
[0016] - calculating a Euclidean distance between the sensor data and the reconstructed sensor data in order to determine the reconstruction error based on the calculated Euclidean distance.
[0017] The Euclidean distance may also be referred to as the Euclidean distance and preferably corresponds to the length or value of the connecting vector of two points or vectors. Therefore, the distance between points in the n-dimensional space can be calculated using the Euclidean distance, whereby the distance may currently correspond to the reconstruction error.
[0018] Furthermore, it is optionally provided that detecting a production error comprises the following steps:
[0019] - defining a threshold for said reconstruction error,
[0020] - comparing the reconstruction error with a defined threshold value in order to detect production errors of the component.
[0021] Thus, the threshold value can advantageously be defined individually according to a given application. For example, in the case of a certain type of component, a sufficiently good quality can still be present even with higher deviations. The threshold value can also be determined and defined by trial and error. Furthermore, the threshold value can also be defined automatically based on past detection of production errors of the component.
[0022] Furthermore, within the scope of the invention it can be provided that the method is performed using a machine learning model, preferably a neural network, more preferably an autoencoder, such as a variational autoencoder. By using a machine learning model, in particular a neural network such as an autoencoder, a continuous improvement can advantageously be provided, since the machine learning model can optimize its weights based on new data to which the machine learning model is applied, in order to enable a more precise detection of production errors of components.
[0023] It is also possible that the machine learning model is trained based on the following steps:
[0024] - providing reference data having at least two dimensions, wherein the respective dimensions of the reference data each comprise measurement data for a component free of production errors,
[0025] - performing a dimensionality reduction of the reference data, wherein at least one reference feature is extracted from the reference data based on at least two dimensions of the reference data,
[0026] - reconstructing the reduced-dimensionality reference data based on at least one extracted reference feature,
[0027] - determining a reconstruction error based on a comparison of said reference data with the reconstructed reference data.
[0028] Here, the steps of dimensionality reduction, reconstruction of the reduced-dimensional reference data and determination of the reconstruction error can be performed until the reconstruction error does not exceed a defined limit value. Expressed in simplified terms, the machine learning model should learn by means of training: the number of dimensions of the reference data is reduced by extracting at least one reference feature, so that the reference data can subsequently be reconstructed as accurately as possible based on the extracted reference feature. In particular, it is provided that the measurement data of the reference data and the measurement data of the sensor data are respectively determined in the same manner. For example, in the case where the component is constructed as a radar sensor, the measurement data of the corresponding dimensions in the reference data and in the sensor data can be obtained from measurements of the same angular points of the antenna diagram of the component.
[0029] Preferably, it can be provided that the component is a radar sensor and that the measurement data of the corresponding dimension of the sensor data are obtained from measurements performed by and / or at the radar sensor. The measurement data can be obtained, for example, from measurements of input signal strength, reflected signal, resolution, effective range, frequency stability and / or noise level.
[0030] Furthermore, it is conceivable that the method is carried out within the scope of a quality test of production, in particular an end-of-line test. The method can be advantageous in particular in the case of an end-of-line quality test, since in this case the focus is on being able to determine whether the corresponding component has a production error, i.e. is defective. The method can advantageously allow a basic decision to be made about the presence of a production error based on a large number of input variables, wherein the detection can advantageously be very sensitive by combining a large number of input variables. Since in the case of an end-of-line test it may not be important to differentiate more finely how exactly the errors of the components are, the method can be particularly advantageous in this case.
[0031] The subject of the present invention is also a computer program, in particular a computer program product, comprising instructions which, when executed by a computer, cause the computer to perform the method according to the invention. The computer program according to the invention thus brings the same advantages as described in detail with respect to the method according to the invention.
[0032] The subject matter of the present invention is also a device for data processing, which is set up to carry out the method according to the invention. For example, a computer can be provided as the device, which executes the computer program according to the invention. The computer can have at least one processor for executing the computer program. A non-volatile data memory can also be provided, in which the computer program is stored and from which the computer program can be read out by the processor for execution.
[0033] The subject of the present invention may also be a computer-readable storage medium having a computer program according to the present invention and / or comprising instructions which, when executed by a computer, cause the computer to carry out the method according to the present invention. The storage medium is designed, for example, as a data memory, such as a hard disk and / or a non-volatile memory and / or a memory card. The storage medium may, for example, be integrated into a computer.
[0034] Furthermore, the method according to the present invention can also be implemented as a computer-implemented method. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Further advantages, features and details of the invention are apparent from the following description, in which exemplary embodiments of the invention are described in detail with reference to the accompanying drawings. The features mentioned in the claims and in the description may be essential to the invention individually or in any combination. Among them:
[0036] Figure 1 Schematic visualization showing a method, a device, a storage medium and a computer program according to embodiments of the present invention,
[0037] Figure 2 shows a schematic visual representation of a portion of a method according to an embodiment of the invention,
[0038] Figure 3 A schematic diagram showing sensor data of a large number of components according to an embodiment of the present invention,
[0039] Figure 4 A schematic diagram showing a histogram of measurement data according to an embodiment of the present invention,
[0040] Figure 5 A schematic diagram showing a histogram of other measurement data according to an embodiment of the present invention,
[0041] Figure 6 A schematic diagram showing sensor data of a large number of components based on two extracted features according to an embodiment of the present invention,
[0042] Figure 7 A schematic diagram showing a histogram based on extracted features according to an embodiment of the present invention is shown,
[0043] Figure 8 A schematic diagram showing a histogram based on other extracted features according to an embodiment of the present invention is shown,
[0044] Fig. 9 A schematic diagram showing sensor data of a large number of components after dimensionality reduction according to an embodiment of the present invention,
[0045] Fig.10 A schematic diagram showing sensor data of a large number of components based on two extracted features after dimensionality reduction according to an embodiment of the present invention,
[0046] Fig.11 A schematic visual diagram showing a histogram of reconstruction errors according to an embodiment of the present invention. DETAILED DESCRIPTION
[0047] Figure 1 A method 100 , an apparatus 10 , a storage medium 15 and a computer program 20 according to an embodiment of the invention are schematically shown.
[0048] Figure 1 In particular, a method 100 for detecting production errors of a component 1 in manufacturing according to an embodiment of the present invention is shown. In particular, in a first step 101, sensor data 2 having at least two dimensions 3 are provided, wherein the respective dimensions 3 of the sensor data 2 each comprise measurement data about the component 1. In a second step 102, a dimensionality reduction of the sensor data 2 is preferably performed, wherein at least one feature 4 is extracted based on the at least two dimensions 3 of the sensor data 2. In a third step 103, the reduced-dimensional sensor data 5 are reconstructed, preferably based on the at least one extracted feature 4, in order to provide reconstructed sensor data 6. In a fourth step 104, a reconstruction error is determined based on a comparison of the sensor data 2 with the reconstructed sensor data 2. In a fifth step 105, production errors of the component 1 are detected based on the determined reconstruction error.
[0049] A method for detecting production errors of a component 1 is described below, which method can, for example, provide the following advantages. In particular, the inspection is no longer based on performance indicators or requirements for the component 1, but on the statistical behavior of the component 1. In addition, production errors can be detected regardless of the source of the error. The method according to an exemplary embodiment leads in particular to a single value, in particular a reconstruction error, regardless of how many different measurement data are provided for the sensor data 2. The output, in particular the reconstruction error, can be compared with at least one threshold value that can be specified.
[0050] Often, for calibration or inspection purposes, in particular, a plurality of measurements must be performed at a component 1. These measurements can be represented by points (or vectors) of dimension N, where N is the number of measurement data based on the measurements performed. Starting from the statistics of the plurality of measured components 1, a statistical representation of the component 1 can be obtained in such a way that all points are included in this N-dimensional space. For example, the points for a particular component 1 are not randomly distributed over the N-dimensional space, but are clustered in clusters or follow specific trajectories. This can be attributed to the fact that the various measurements constituting the N-dimensional space for a particular component are correlated with one another.
[0051] If the component 1 is, for example, a radar sensor for the automotive sector, in which the antenna diagrams are measured for different angle points, the amplitudes of two adjacent angle points have a high correlation in particular. If this is the case in the case of the product to be inspected, this fact can be used advantageously to detect production errors of the component 1, i.e., defective components. For example, a defective component 1 has different correlations between the N dimensions 3, which can mean that the defective component 1 is not scattered in the same trajectory / cluster as components without production errors 1′. The method according to an embodiment of the invention preferably uses the correlation between the various measurements of the component 1 in order to separate the components 1 that are outside the statistics (defective components 1) by applying a technique for dimensionality reduction.
[0052] In Figure 2 it is shown in particular how, according to an embodiment, sensor data 2 with N dimensions 3 is transformed into reduced-dimensional sensor data 5 in a latent space with M extracted features 4 (M < N), where the extracted features 4 can be combinations of the original dimensions 3 and do not necessarily represent physical features of the component 1. Subsequently, a reconstruction back into the original N-dimensional space is preferably performed in order to obtain the reconstructed sensor data 6. Since M < N, some information may be lost during this process and reconstruction errors may occur. The reconstruction error can be determined, for example, by calculating the Euclidean distance between the sensor data 2 and the reconstructed sensor data 6 (reconstruction error = |sensor data 2 - reconstructed sensor data 6|). The challenge may lie in selecting as few dimensions 3 as possible, i.e., the reduced-dimensional sensor data 5, from this latent space and keeping the reconstruction error small for components without production errors 1'. For this purpose, unsupervised learning can be performed based on training using reference data. After training, new components 1 can be tested, for example. The smaller the reconstruction error, the more similar the new component is to the components without production errors 1' used in particular for training the reference data. If a defective component 1, i.e., a component with a production defect 1, is tested, the error is preferably many orders of magnitude higher than in the case of components without production defects 1' and the defective component 1 can be very easily filtered out.
[0053] The above method can furthermore be used for identifying deviations in the manufacturing process. If, for example, the reconstruction error of a component without production errors 1' increases slowly (or rapidly) over time, this may be an indication of a deviation in the manufacturing process. For example, it may then be necessary to recalibrate or maintain the measuring device.
[0054] In the prior art, various methods for dimensionality reduction are known, among which linear methods such as PCA (principal component analysis), FA (factor analysis), LDA (linear discriminant analysis) or truncated SVD (truncated singular value decomposition). Furthermore, nonlinear methods are known, such as kernel PCA, t-distributed stochastic neighbor embedding (t-SNE), multidimensional scaling (MDS), isometric mapping (Isomap), autoencoder neural networks, autoencoders or variational autoencoders.
[0055] Here, according to an embodiment, PCA and autoencoders can be preferred due to the following advantages.
[0056] PCA preferably attempts to project the data into a new linear dimension that maximizes the variance of the data. If the original dimension N has only linear dependencies, the new dimension M, i.e. the extracted features 4, are especially linearly independent. The first dimension calculated may be more important because it contains the most information.
[0057] Advantageously, only a small amount of computational effort is required to perform PCA. Furthermore, PCA can be implemented simply. Projecting the new points into the new dimension M, ie, in particular extracting the features 4 and reconstructing the data in the original N-dimensional space, can require only a small amount of computational effort.
[0058] Furthermore, PCA can be understood deterministically and simply. The new dimension 3 preferably depends only on the input data, i.e. the sensor data 2. There are no random factors, such as parameter initialization, in the calculation. If the calculation is repeated with the same data, in particular the same results are obtained. This is particularly advantageous in the case of components whose randomness is difficult to explain or model.
[0059] According to the following Figures 3 to 10 Explain how a method according to an embodiment of the invention can be used for detecting production errors in a component 1 when the sensor data 2 has N==2 dimensions and M=1 features 4 are extracted for the reduced sensor data 5. Depending on the type of component 1, more dimensions 3 may be required.
[0060] For example, 10,003 components 1 are given. For each component 1, for example, two measurements are performed to provide two measurement data. 10,000 of the components 1' have no production errors and are therefore functioning normally, while three of the components 1 have production errors and are outside the statistics. Figure 3 It can be seen in FIG. 1 that components without production error 1 ′ are distributed along the line x==y of the diagram based on the measurement data, while components with production error 1 are outside this area.
[0061] By means of a method according to an embodiment of the present invention, for example using PCA, features 4 may be extracted. For example, the extracted features 4 may not have physical meaning, but may include information about the component. The extracted features 4 may be evenly ordered so that higher dimensions 3 contain less information.
[0062] exist Figure 3 A two-dimensional example is shown in Figure 4 and 5 A histogram for each of the two dimensions 3 of the corresponding measurement data is shown in . It is noticeable here, for example, that defective components 1 are also included in the statistics, so that in particular it is not possible to distinguish them from components without production errors 1 ′.
[0063] Figure 6 The space newly created by the extracted features 4 is shown. Based on this, a new histogram of the extracted features 4 can be created (see Figure 7 and Figure 8 ). Here, in Figure 8 In the case of the histogram of one of the extracted features 4 shown in , deviations can be seen, for example, at defective components 1. Defective components 1 therefore have high values for one of the extracted features 4. For this two-dimensional example, production errors of defective components 1 can therefore be detected with the aid of this second histogram. For cases with higher dimensions, additional steps can be performed.
[0064] In the case of higher dimensionality, multiple of the dimensions 3 may be ignored, i.e., dimensionality reduction may be performed on more than one dimension 3. The number of dimensions 3 to be ignored may depend on how strongly the original dimensions 3 are correlated with each other. In the case of very high correlation, ignoring many dimensions 3 may lead to much better results.
[0065] After ignoring the second dimension 3, the components can be represented again in the original space and in the new space, as in Fig. 9 (original space) and Fig.10 This can be seen in (the new space with extracted feature 4).
[0066] In the case of the original space, the reconstructed sensor data 6 of the component can now be compared with the original sensor data 2 in order to determine the reconstruction error. For this purpose, the Euclidean distance can be calculated. The histogram of the reconstruction error is Fig.11 It can be seen that a component with a production error 1 has a significantly higher error than a component without a production error 1 ′. A threshold value can be defined for the maximum permissible reconstruction error in order to detect production errors in the component 1 .
[0067] The higher the original dimension 3 and the higher the linear correlation between the original features 4, the higher the resulting reconstruction error.
[0068] Autoencoders, in particular variational autoencoders, can also be used for dimensionality reduction. The autoencoder preferably projects the raw data, i.e., in particular the sensor data 2, into a new latent space with the extracted features 4 and fewer dimensions. In addition, nonlinear dependencies can be taken into account, so that it may be particularly advantageous to compress components 1 that have nonlinear dependencies between their dimensions.
[0069] The new points can be projected into the new dimension 3, ie the features 4 can be extracted, and the data in the original N-dimensional space can be reconstructed with little computational effort by applying a non-linear function, such as a rectifier function.
[0070] Unlike PCA methods, the latent space of the autoencoder can be differentiated between different training runs due to the possible random initialization of the initial neural network. The latent space for simple autoencoders is notably unordered (e.g. in contrast to variational autoencoders).
[0071] The autoencoder machine learning model can first be trained using a set of typical measurement data of components without production errors 1′. During this training process, in particular the encoder part (the encoder part projects the reference data onto the latent space, i.e. the latent space reduces the reference data, wherein features 4 are extracted) and the decoder part (the decoder part reconstructs the reduced reference data from the latent space, i.e. the reduced space, back to the original dimension 3) are trained. The trained machine learning model can then be applied to the sensor data 2. If the new sensor data 2 follows the same distribution as the reference data, the corresponding reconstructed sensor data 6 are preferably very close to the reference data. In the case of outliers (sensor data 2 with correlations between their dimensions 3 that are different from those in the case of the observed reference data), the autoencoder machine learning model preferably has difficulty in reconstructing the data and therefore results in high reconstruction errors, from which production errors of the component 1 can be detected. For the reconstruction error, a domain-specific metric for the difference (e.g. MSE) can be used.
[0072] According to an embodiment, this is an end of line (EoL) test of a manufacturing line. In this case, in particular, there are a plurality of different measurements for corresponding measurement data, which can be fed to the method according to an embodiment as sensor data 2. In the case of conventional tests in the prior art, for example, each measurement is tested individually without combining the measurements, so that usually at least as many independent calculations as measurements are required, sometimes even more. In the method according to an embodiment of the invention, a plurality of measurement data can advantageously be combined into sensor data 2 and fed into an algorithm for detecting production errors, which can advantageously simplify the implementation, reduce the complexity and increase the sensitivity.
[0073] The result of the method according to an embodiment of the invention is preferably a single error value based on the reconstruction error. A threshold value may be defined to detect production errors of the component 1. Thereby, the EoL test may advantageously be greatly simplified.
[0074] Advantageously, all measurement data can be taken into account in a combined manner in the sensor data 2 by the method according to the exemplary embodiment, and the reconstruction errors can be used to detect production errors of the component 1 .
[0075] The use of reconstruction errors to detect production errors of the component 1 takes into account, in particular, the correlation between different parameters, i.e., the measurement data, and can advantageously increase the detection sensitivity by a value of at least 100 compared to the prior art performance-based algorithms. This has the advantage, for example, that the same or higher sensitivity can be achieved with fewer measurements than in the case of performance-based algorithms. The effort for calculating the threshold value can also be advantageously reduced, since only one threshold value may be required for the reconstruction error.
[0076] The above explanations of the embodiments describe the invention only within the scope of examples. Of course, the individual features of the embodiments can be freely combined with one another as far as technically reasonable, without departing from the scope of the invention.
Claims
1. A method (100) for detecting production errors of a component (1) during manufacturing, comprising the following steps: - providing (101) sensor data (2) having at least two dimensions (3), wherein a respective dimension (3) of the sensor data (2) respectively comprises measurement data about the component (1), - performing (102) dimensionality reduction of the sensor data (2), wherein at least one feature (4) is extracted based on at least two dimensions (3) of the sensor data (2), - reconstructing (103) the reduced-dimensionality sensor data (5) based on at least one extracted feature (4) in order to provide reconstructed sensor data (6), - determining (104) a reconstruction error based on a comparison of the sensor data (2) with the reconstructed sensor data (2), - detecting (105) production errors of the component (1) based on the determined reconstruction errors.
2. The method (100) according to claim 1, characterized in that: Determining (104) the reconstruction error comprises the following steps: - calculating a Euclidean distance between the sensor data (2) and the reconstructed sensor data (6) in order to determine the reconstruction error based on the calculated Euclidean distance.
3. The method (100) according to any one of the preceding claims, characterized in that Detecting (105) production errors comprises the following steps: - defining a threshold for said reconstruction error, - comparing the reconstruction error with a defined threshold value in order to detect production errors of the component (1).
4. The method (100) according to any one of the preceding claims, characterized in that The method (100) is performed using a machine learning model, preferably a neural network, more preferably an autoencoder.
5. The method (100) according to claim 4, characterized in that: The machine learning model is trained based on the following steps: - providing reference data having at least two dimensions (3), wherein the respective dimensions (3) of the reference data each comprise measurement data for a component free of production errors (1′), - performing a dimensionality reduction of the reference data, wherein at least one reference feature is extracted from the reference data based on at least two dimensions (3) of the reference data, - reconstructing the reduced-dimensionality reference data based on at least one extracted reference feature, - determining a reconstruction error based on a comparison of the reference data with the reconstructed reference data, The steps of reducing the dimension of the reference data, reconstructing the reduced-dimensional reference data and determining a reconstruction error are performed until the reconstruction error does not exceed a defined limit value.
6. The method (100) according to any one of the preceding claims, characterized in that The component (1) is a radar sensor, and the measurement data of the corresponding dimension (3) of the sensor data (2) are obtained from measurements performed by and / or at the radar sensor.
7. The method (100) according to any one of the preceding claims, characterized in that The method (100) is carried out within the scope of a quality test of production, in particular an end-of-line test.
8. A computer program (20) comprising instructions which, when the computer program (20) is executed by a computer (10), cause the computer (10) to perform the method (100) according to any one of the preceding claims.
9. A device (10) for data processing, which is configured to carry out the method (100) according to any one of claims 1 to 7.
10. A computer-readable storage medium (15) comprising instructions which, when executed by a computer (10), cause the computer to perform the steps of the method (100) according to any one of claims 1 to 7.