Computer-implemented method for training an AI module to detect defects in an adhesive bond of an ultrasonic sensor

Abstract

The present invention relates to a computer-implemented method (100) for training an AI module (202) for detecting defects in an adhesive bond of an ultrasonic sensor, wherein the method (100) comprises the steps of: providing (S1) on a data carrier a data set which has at least one data entry of a manufacturing parameter at the adhesive bond of the ultrasonic sensor and a quality parameter of the adhesive bond, generating (S2) a modified training data set based on the data set, wherein the generation of the modified training data set comprises the steps of: forming (S3) an input data set based on at least one data entry of the manufacturing parameter, forming (S4) an output data set based on the quality parameter, wherein the method (100) further comprises the step of: training (S5) the AI ​​module based on the modified training data set.

Description

State of the art

[0001] The present invention relates to a computer-implemented method for training an AI module for detecting defects in an adhesive bond of an ultrasonic sensor and to a vehicle.

[0002] Currently, there are a multitude of different solutions for detecting defects in adhesive bonds in the automotive sector. Due to the increasing number of components to be bonded, especially sensors on the vehicle, and the increased quality requirements, the need for innovative and robust methods for detecting faulty contacts is continuously growing.

[0003] The constant reduction in vehicle weight to reduce fuel consumption as well as increasing competition are creating cost pressure, so that cheaper and more efficient components for vehicles are in greater demand. Disclosure of the invention

[0004] The inventive computer-implemented method for training an AI module for detecting defects in an adhesive bond of an ultrasonic sensor, having the features of claim 1, has the advantage over the known method that the detection rate of faulty adhesive bonds can be significantly increased without increasing the error rate, in particular the rate of false positives. Furthermore, the detection accuracy can be refined so that even slight defects in the adhesive bond can be reliably detected.

[0005] Further preferably, the method can also be used to detect a delimitation between the ultrasonic sensor and the adhesive point.

[0006] This is achieved according to the invention in that the computer-implemented method for training an AI module for detecting defects in an adhesive bond of an ultrasonic sensor comprises the steps: - Providing, on a data carrier, a measured value data set which has at least one data entry of a manufacturing parameter of the adhesive bond of the ultrasonic sensor and a quality parameter of the adhesive bond, - Generating a modified training dataset based on the measured value dataset, wherein the generation of the modified training dataset comprises the steps: ◯ Forming an input data set based on the at least one data entry of the manufacturing parameter, ◯ Forming an output data set based on the quality parameter, the method further comprising the step of: - Training the AI ​​module based on the modified training dataset.

[0007] In other words, with the help of AI-supported evaluation of production parameters, the evaluation of the data already recorded can be improved. For example, the production parameter 16 can include electrical and physical and / or more measured values ​​for the respective adhesive bond or the respective ultrasonic sensor. More preferably, the temperature can be added to these parameters. Based on this data set, an evaluation can be carried out in order to be able to determine the quality parameters of the adhesive bond of the ultrasonic sensor. The quality parameter can in particular be a successful or unsuccessful bonding of the ultrasonic sensor element in a predetermined bonding position. More preferably, the quality parameter can also be a delimitation or similar. By creating a corresponding input data set based on the production parameters already recorded orFrom the 16 physical and electrical parameters and the temperature, an output data set can be generated that describes the quality parameter. Using this modified training data set, which is based on the measured data set, an AI module can be trained. This module is designed to determine defects in the adhesive bonds or a delimitation based on the output data set, in particular the quality parameter.

[0008] The subclaims show preferred developments of the invention.

[0009] Preferably, the method further comprises the step: - Preparing the data entry of the manufacturing parameter by removing incomplete data entries and / or by scaling data entries to increase comparability.

[0010] An advantage of this embodiment is that potentially incomplete data entries in the measured value data set can be excluded from further considerations. Further preferably, these can be scaled accordingly based on the respective existing manufacturing parameters in order to increase comparability between a large number of measured value data sets and manufacturing parameters.

[0011] More preferably, generating the modified training dataset further comprises the step: - Develop five criteria based on the manufacturing parameters using a principal component analysis.

[0012] An advantage of this embodiment is that, with the help of principal component analysis, a variety of manufacturing parameters can be used to detect adhesive bonding defects or similar issues. For example, the 16 electrical and physical properties of the sensor, including adhesive bonding and temperature, can be incorporated into the principal component analysis to generate five criteria or variables based on these, which can significantly simplify further data processing.

[0013] More preferably, generating the modified training dataset further comprises the steps: - Forming a first range of values ​​based on the first criterion and a first limit value, - Forming a second range of values ​​based on the second criterion and a second limit value, - Selecting all manufacturing parameters that fall within the first value range and / or the second value range to form a subset.

[0014] An advantage of this embodiment is that the first criterion and the second criterion are selected using the threshold value such that they reflect a high probability that a defective adhesive bond or a defective adhesive bond or a defective sensor attachment may be present. In particular, the first threshold value and / or the second threshold value can be used to distinguish between a first range that has a particularly high probability of detecting an adhesive defect and a further range that has a particularly low probability of detecting a defective bond.

[0015] More preferably, generating the modified training dataset further comprises the steps: - Forming a third range of values ​​based on the third criterion, a third limit and the subset, - Forming a fourth range of values ​​based on the fourth criterion, a fourth limit and the subset, - Selecting each entry of the subset that falls within the first range, the second range, the third range, and the fourth range to form a subset.

[0016] An advantage of this embodiment is that, using the third value range and the fourth value range, the probability of false positive detection for detecting incorrect bonding can be significantly reduced. In particular, correlations between the first value range and the second value range, and between the third value range and the fourth value range, can be utilized to further reduce the number of false positive detections.

[0017] More preferably, generating the modified training dataset further comprises the steps: - Forming a first group if a value of an entry of the subset is greater than the fifth criterion, - Form a second group if the value of the entry of the subset is less than the fifth criterion.

[0018] An advantage of this embodiment is that pseudo-errors due to outliers in the value range can be excluded for further considerations.

[0019] More preferably, generating the modified training dataset comprises the steps: - forming a region, wherein the region is bounded by a first comparison value and a second comparison value, - Determining a position of the adhesive bond defect if a value of the second group falls within the region.

[0020] One advantage of this design is that processing defects can be detected with high precision using ultrasound. This can be achieved by determining the cause of the defect, allowing for rapid response and the elimination of the defect in the production line in a short time, preventing the production of more defective sensors. Furthermore, by determining the position of the adhesive bond defect and dividing the detection areas into several smaller areas, the error rate can be minimized, as it allows for focus on small areas and fine-tuning the threshold values ​​for the criteria.

[0021] More preferably, generating the modified training dataset further comprises the steps: - Generate a variety of regions, - Identify a second group for each of the regions, - Determining the position of the adhesive bond defect in each of the plurality of regions.

[0022] An advantage of this embodiment is that, particularly in a production line for ultrasonic sensors, it can be divided into a plurality of regions. Thus, every position of every defect in the adhesive bond can be determined for each region.

[0023] More preferably, generating the modified training dataset further comprises the steps: - Providing at least one further production parameter, - Adjust the second group based on the additional manufacturing parameter.

[0024] An advantage of this embodiment is that additional manufacturing parameters and / or sensor data can be used to detect possible misdetections or false positives, so that the overall detection performance can be further improved.

[0025] A further aspect of the invention relates to a vehicle which is configured to carry out steps of the method as described above and below and / or has an AI module which is configured to carry out steps of the method as described above and below. Short description of the drawings

[0026] Embodiments of the invention are described in detail below with reference to the accompanying drawings. In the drawing: Fig. 1 and Fig. 2 a flowchart illustrating steps of the method according to an embodiment, Fig. 3 shows a vehicle according to an embodiment, Fig. 4 shows a block diagram illustrating the operation of the method according to an embodiment, Fig. 5 shows a block diagram illustrating the operation of the method according to one embodiment. Embodiments of the invention

[0027] Preferably, in all figures, all elements, units and / or steps are provided with the same reference numerals.

[0028] Fig. Figure 1 shows a flowchart illustrating steps of the method 100 according to one embodiment. The computer-implemented method 100 for training an AI module 202 to detect defects in an adhesive bond of an ultrasonic sensor comprises the steps: - Providing S1 on a data carrier, a measured value data set which has at least one data entry of a manufacturing parameter at the adhesive connection of the ultrasonic sensor and a quality parameter of the adhesive connection, - Generating S2 a modified training dataset based on the measured value dataset, wherein the generation of the modified training dataset comprises the steps: - Forming S3 an input data set based on at least one data entry of the manufacturing parameter, - Forming S4 an output data set based on the quality parameter, wherein the method 100 further comprises the step: - Training S5 of the AI ​​module 202 based on the modified training dataset.

[0029] Fig. Figure 2 shows a flowchart illustrating steps of the method 100 according to one embodiment. The computer-implemented method 100 for training an AI module 202 for detecting defects in an adhesive bond of an ultrasonic sensor preferably comprises the same steps S1 to S5 as already described with regard to the Fig. 1. Further preferably, the method 100 further comprises the step of preparing S6 the data entry of the manufacturing parameter. Further preferably, the method 100 preferably further comprises the steps of forming S7 five criteria based on the manufacturing parameters using a principal component analysis during the generation S2 of the modified training data set.

[0030] Further preferably, the method 100 preferably further comprises the steps of forming S8 a first value range, forming S9 a second value range, and selecting S10 all manufacturing parameters during the generation S2 of the modified training data set. Further preferably, the method 100 preferably further comprises the steps of forming S11 a third value range, forming S12 a fourth value range, and selecting S13 each entry of the subset during the generation S2 of the modified training data set.

[0031] Furthermore, the method 100 preferably comprises the steps of forming S14 a first group during generation S2 of the modified training data set, and forming S15 a second group. Preferably, generation S2 of the modified training data set comprises the steps of forming S16 a region, and determining S17 a position of the defect. Further preferably, generation S2 of the modified training data set comprises the steps of generating S18 a plurality of regions, determining S19 a second group for each of the regions, and determining S20 the position of the defect.

[0032] Further preferably, the generation S2 of the modified training data set comprises providing S21 at least one further manufacturing parameter, and adapting S22 the second group based on the further manufacturing parameter.

[0033] Fig. 3 shows a vehicle 200 according to one embodiment. The vehicle 200 is preferably configured to perform steps of the method 100 as described above and below and / or has an AI module 202 configured to perform steps of the method 100 as described above and below.

[0034] Fig. 4 shows a block diagram 300 for instructing the functioning of method 100 according to one embodiment. Block diagram 300 has a starting point 302. Based on starting point 302, a measured value data set can be imported in step 304. More preferably, the measured value data set can be prepared in step 306. In step 308, in particular, the measured values ​​or the data entries of the manufacturing parameters can be scaled to increase comparability. Preferably, in step 310, a principal component analysis can be applied to the scaled and prepared data entries of the measured value data set in order to obtain five criteria in step 312.

[0035] Fig.5 shows a block diagram 400 to illustrate the functioning of the method 100 according to one embodiment. The block diagram 400 has a starting point 402. Following the starting point 402, a measured value data set can be provided in step 404 and possibly processed. In step 406, in particular, the processed measured value data set can be processed using a principal component analysis such that five criteria can be determined. In step 408, in particular, a first value range can be formed based on the first criterion and the first limit value, and a second value range can be formed based on the second criterion and a second limit value. The formation of the first value range and the second value range can be repeated in steps 410, 412, 414 for different regions.If a data entry falls within the first value range, it can be forwarded to case 416 for comparison of the value entry 430 with the fifth criterion. Further preferably, using the comparison value 432 and the comparison value 434, it can be determined whether the respective measured value falls within a first region 436. Further preferably, in the further steps 432 and 434, it can be checked into which region the respective measured value falls.

[0036] Further preferably, if a measured value or an entry according to which the data record falls into a different region because it was not successfully checked in the first block 408, this can be forwarded in the second block 410, case 418, to the third and fourth value range 424. If the respective value falls into the third and fourth range 424, this can become part of a subset. Further preferably, the subset can be compared against the fifth criterion 430 in order to be able to fall into the second group, for example. Using the comparison values ​​432 and 434, it can be determined into which region of the respective range the measured value falls.

[0037] For example, a measured value may fall within the third range, which in case 420 is forwarded to the third and fourth value range 426. Based on the comparison with criterion 430, the second group can be formed again. A further comparison with the comparison values ​​432 and 434 can be used to determine a relevant region 436. More preferably, the measured value may fall within a fourth range 422. In the fourth range, in particular, the third value range and the fourth value range can be compared in step 428. More preferably, the measured value can be compared with the fifth criterion 430 in order to be able to determine the relevant region again based on the comparison values ​​432 and 434.

Claims

[1] Computer-implemented method (100) for training an AI module (202) for detecting defects in an adhesive bond of an ultrasonic sensor, the method (100) comprising the steps of: - Providing (S1) on a data carrier a data set which has at least one data entry of a manufacturing parameter at the adhesive connection of the ultrasonic sensor and a quality parameter of the adhesive connection, - generating (S2) a modified training dataset based on the dataset, wherein the generation of the modified training dataset comprises the steps: - forming (S3) an input data set based on at least one data entry of the manufacturing parameter, - Creating (S4) an output data set based on the quality parameter, - wherein the method (100) further comprises the step: - Training (S5) the AI ​​module based on the modified training dataset. [2] The method (100) of claim 1, wherein the method (100) further comprises the step of: - Preparation (S6) of the data entry of the manufacturing parameter by removing incomplete data entries and / or by scaling data entries to increase comparability. [3] Method (100) according to one of the preceding claims, wherein the generation (S2) of the modified training data set further comprises the step: - Develop (S7) five criteria based on the manufacturing parameters using a principal component analysis. [4] The method (100) of claim 3, wherein generating (S2) the modified training data set further comprises the steps of: - forming (S8) a first value range based on the first criterion and a first limit value, - forming (S9) a second range of values ​​based on the second criterion and a second limit value, - Selecting (S10) all manufacturing parameters which fall within the first value range and / or the second value range to form a subset. [5] The method (100) of claim 4, wherein generating (S2) the modified training data set further comprises the steps of: - Forming (S11) a third range of values ​​based on the third criterion, a third limit and the subset, - Forming (S12) a fourth range of values ​​based on the fourth criterion, a fourth limit value and the subset, - Selecting (S13) each entry of the subset which falls within the first value range, the second value range, the third value range and the fourth value range to form a subset. [6] The method (100) of claim 5, wherein generating (S2) the modified training data set further comprises the steps of: - Forming (S14) a first group if a value of an entry of the subset is greater than the fifth criterion, - Forming (S15) a second group if the value of the entry of the subset is less than the fifth criterion. [7] The method (100) of claim 6, wherein generating (S2) the modified training data set further comprises the steps of: - forming (S16) a region, wherein the region is limited by a first comparison value and a second comparison value, - determining (S17) a position of the defect of the adhesive bond if a value of the second groups falls within the region. [8] Method (100) according to one of claims 6 to 7, wherein generating (S2) the modified training data set further comprises the steps of: - Generating (S18) a multitude of regions, - Determine (S19) a second group for each of the regions, - Determining (S20) the position of the adhesive bond defect in each of the plurality of regions. [9] Method (100) according to one of claims 6 to 8, wherein generating (S2) the modified training data set further comprises the steps of: - Providing (S21) at least one further production parameter, - Adjusting (S22) the second group based on the additional manufacturing parameter. [10] Vehicle (200) which is configured to carry out steps of the method (100) according to one of the preceding claims and / or has an AI module (202) which is configured to carry out steps of the method (100) according to one of the preceding claims.

Images