Method and system for training a machine learning model to determine component lifetime-critical characteristics

Abstract

The invention relates to a method for training a machine learning model to determine component lifetime-critical features of a printed circuit board (PCB) populated with electronic components and installed in the housing of an electronic product, the method comprising the steps of: providing (S1) first component lifetime data of the PCB populated with electronic components, which are acquired in an uninstalled state of the PCB by a thermomechanical stress test; providing (S2) second clustered component lifetime data of a plurality of electronic products, in each of whose housings a PCB populated with electronic components is installed; providing (S3) a respective housing factor for at least each product group, in particular by comparing the first and second component lifetime data;Providing (S4) a multitude of features of the respective electronic components and / or the respective printed circuit board and / or the respective electronic product; training (S5) the machine learning model based on the first and second component lifetime data and / or the respective package factor and based on the multitude of features to determine component lifetime-critical features for each product group; and providing (S6) the trained machine learning model.

Description

[0001] The invention relates to a method and a system for training a machine learning model to determine component lifetime-critical characteristics of a printed circuit board (PCB) populated with electronic components and installed in the housing of an electronic product. The invention further relates to a method for determining the housing factor of a PCB populated with electronic components, to be installed in the housing of an electronic product, using a trained machine learning model. State of the art

[0002] Many populated printed circuit boards (PCBs) experience thermo-mechanical stress when exposed to field conditions. The thermo-mechanical stress occurring at the solder joints stems, firstly, from the relative expansion of the various components and / or the PCB, due to differences in their coefficients of thermal expansion. In mounted PCBs, the solder joints can also be subjected to further stress, for example, from the PCB bending within its housing, again resulting from the difference in the coefficients of thermal expansion between the PCB and the housing. This can lead to multiaxial stress on the solder joints.

[0003] To ensure the reliability of the solder joints under such stress, experiments are conducted, including endurance tests with temperature cycling under specific conditions. To minimize the experimental effort, these tests are often performed on an unpopulated, i.e., free, but populated, circuit board.

[0004] To account for the influence of PCB deflection during the design process, so-called "enclosure factors" are also used. These factors attempt to estimate the increase in stress on the solder joint due to the multiaxial stress state. Furthermore, these enclosure factors can be used to compare the tolerable cycles of the component's solder joint on the PCB with tests performed on a bare PCB. This is done by comparing the tolerable cycles on the bare PCB under defined test conditions, such as a cyclic temperature change between a minimum temperature T and a maximum temperature T. min and a maximum temperature T max , as well as the required cycles to meet the demands of the field of application under the same test conditions. This can be calculated, for example, using a Coffin-Manson approach. Here, the tolerable cycles are reduced by the housing factors.

[0005] From TW 1 625 634 B, a method for estimating the stress of an electronic component is known. First, the electronic component is provided. The electronic component comprises a first component, a second component, and a plurality of solder joints. Each of the conductive solder joints has opposing surfaces. The two surfaces are connected to each other and contact the first and second components. Using the method, a stress value of the first solder joint is calculated under a test condition parameter. Furthermore, a stress value of each second solder joint is calculated under the test condition parameter according to a first calculation formula.

[0006] Although methods for estimating stresses on components and / or solder joints of populated printed circuit boards, especially those caused by thermo-mechanical bending, are known, further development is still needed.

[0007] The invention is therefore based on the objective of providing an improved method.

[0008] The problem is solved by a method for training a machine learning model to determine component lifetime-critical features of a printed circuit board (PCB) assembled with electronic components and installed in a housing of an electronic product, according to the features of claim 1. The problem is solved by a method for determining a housing factor of a PCB assembled with electronic components, to be installed in a housing of an electronic product, using a trained machine learning model according to the features of claim 9. Disclosure of the invention

[0009] According to a first aspect, a method for training a machine learning model to determine component lifetime-critical features of a printed circuit board populated with electronic components and installed in a housing of an electronic product is proposed, the method comprising the steps: - Providing, in particular standardized, initial component lifetime data for the printed circuit board populated with electronic components, which are recorded in an unassembled state of the printed circuit board by a thermomechanical stress test; - Providing, in particular standardized, second clustered component lifetime data for a large number of electronic products, in each of whose housings a printed circuit board populated with the electronic components is installed; - Providing a respective housing factor for at least each product group, in particular by comparing the first and second component lifetime data, especially for each of the clustered product groups; - Providing a wide range of features of the respective electronic components and / or the respective printed circuit board and / or the respective electronic product; - Training the machine learning model based on the first and second component lifetime data and / or the respective housing factor and based on the multitude of features to determine component lifetime-critical features for each product group; and - Deploying the training machine learning model.

[0010] It is understood that the steps according to the invention, as well as further optional steps, do not necessarily have to be carried out in the sequence shown, but can also be carried out in a different sequence. Furthermore, additional intermediate steps may be provided. The individual steps may also comprise one or more sub-steps without thereby departing from the scope of the method according to the invention.

[0011] According to a second aspect, a system for training a machine learning model to determine component lifetime-critical characteristics of a printed circuit board populated with electronic components and installed in a housing of an electronic product is claimed, the system comprising an evaluation and / or computing device which is designed to perform the following steps: - Providing (S1) of, in particular standardized, initial component lifetime data of the printed circuit board populated with the electronic components, which are recorded in an unassembled state of the printed circuit board by a thermomechanical stress test; - Providing (S2) of, in particular standardized, second clustered component lifetime data for a large number of electronic products, in each of whose housings a printed circuit board populated with the electronic components is installed; - Providing (S3) a respective housing factor for at least each product group, in particular by comparing the first and second component lifetime data, especially for each of the clustered product groups; - Providing (S4) a multitude of features of the respective electronic components and / or the respective printed circuit board and / or the respective electronic product; - Training (S5) the machine learning model based on the first and second component lifetime data and / or the respective housing factor and based on the multitude of features to determine component lifetime-critical features for each product group; and - Deploying (S6) the trained machine learning model.

[0012] The explanations given for the procedure apply accordingly to the system. It is understood that linguistic modifications of procedurally formulated features can be reformulated for the system according to common linguistic practice, without such formulations needing to be explicitly listed here.

[0013] In addition, the present method can be used to examine, through feature analysis, the commonalities between different product groups that could potentially influence the enclosure factor. Such clustering makes it possible to identify which product groups place stress on the printed circuit board components. The subsequent feature analysis allows the relevant features to be filtered out. An extension of the method is the use of these features as input during the development and / or design of new product generations. The features can preferably be entered and / or provided to obtain an estimate of the enclosure factor that should be assumed for the product. This allows for a further reduction of the risk associated with product release testing.In other words, the present invention enables the prediction of the failure probability of a populated printed circuit board (PCB) in an installed state where the PCB is, for example, mounted in a housing or at least partially surrounded and / or enclosed by a housing or housing component. This prediction can preferably be made based on stress tests of the PCB in which the PCB is not mounted in the housing. Depending on the product generation of the PCB to be tested, corresponding housing factors can be derived via a clustering method and feature analysis, which enable the forecast and / or prediction.

[0014] In this case, electronic products are preferably clustered based on the second set of component lifetime data, which were determined under predefined, and in particular standardized, test conditions. This clustering can be performed, for example, using a k-means approach. Furthermore, to determine the respective product group-specific housing factor, the load capacities and lifetimes from the product tests are compared with those from product tests based on experiments with free-standing (i.e., not housed) printed circuit boards. These product tests are represented by the first set of component lifetime data. Finally, the machine learning model is trained to predict characteristics that have a particularly high influence on lifetime.For this purpose, the machine learning model preferably receives a database of features for each electronic product that this product possesses and that can indirectly or directly affect, in particular thermomechanical, the stress on the printed circuit board installed in the respective housing. The first and / or second component lifetime data can be taken from a database of historical tests for training purposes. The machine learning model thus learns to analyze features depending on the clustered lifetime in order to identify the features of the electronic product and / or a product group with the highest influence on the housing factor. Furthermore, the feature analysis can identify commonalities between the electronic products. This enables it to be used for, or rather,The inference of the trained machine learning model makes it possible to predict the housing factor for new electronic products based on the product-specific characteristics. This allows the new electronic product to be categorized into a product group dependent on its lifespan. The comparison preferably allows the determination of the increased stress on the component's solder joint caused by the printed circuit board's mounting situation within the housing. The comparison is performed in the form of the housing factor for each electronic product or averaged for a clustered product group. The housing factor can be determined for each individual electronic product or averaged for a product group. Clustering makes it possible to compare different product groups and then derive corresponding housing factors for each product group.This allows similarities and differences between the product groups to be identified and potentially used.

[0015] The printed circuit board can be populated with multiple electrical and / or electronic components, for example, surface-mount devices (SMDs). Clustering is preferably based on the crack length of the respective solder joints of the electronic components. The crack length preferably indicates the remaining service life margin for a given residual solder joint integrity. The crack length is preferably determined during destructive testing, particularly by creating samples of the solder joint, to establish the secondary service life data. This is preferably done according to the specific product group.

[0016] The training of the machine learning model is essentially based on two experimentally determined or already determined datasets. The first component lifetime data is obtained during release tests for a standardized printed circuit board (PCB) or for different PCBs. Preferably, the test involves determining the number of cycles that can be withstood on the respective PCB at defined temperatures. For example, the PCB in question is cyclically subjected to temperature fluctuations in a climate chamber between a minimum and a maximum temperature until a complete failure of an electronic component occurs due to a crack in at least one solder joint. The respective cycles thus describe temperature changes during the tests. Product-specific tests are also performed to determine the second component lifetime data.The tests preferably focus on the tolerable cycles at the product level, for electronic products with a printed circuit board (PCB) mounted in a housing. For each product, the tolerable cycles until electrical failure (i.e., 100% cracking of the solder joint in an electronic component) can be determined for a combination of test conditions, such as minimum and maximum temperatures. This component lifetime data is then clustered, meaning that similar lifetime results from electronic products are grouped together. By comparing these results with tests on a bare PCB, the housing factor for these clustered groups can then be estimated.

[0017] According to one embodiment, the method further comprises the following steps: - Determining the component lifetime-critical characteristics for each product group using a Pareto analysis, and - Output of the component lifetime-critical characteristics that, based on Pareto analysis, have the greatest influence on the respective, especially product group-specific, housing factor.

[0018] In other words, the process includes a step in which specific characteristics of product components that are critical for the component lifetime are identified. This identification is preferably performed for each product group. The method used to determine these critical characteristics is Pareto analysis. Alternatively, a principal component analysis (PCA) method, for example, can be used.

[0019] After the determination, the procedure outputs those critical characteristics that have the greatest influence on a specific factor. This factor is referred to as the "housing factor." It is particularly emphasized that this housing factor can be "product group-specific," indicating that it can vary depending on the product group. The combination of these steps makes it possible to identify critical characteristics of components in different product groups that could affect the component lifespan, and then to highlight or output these characteristics, especially with regard to their influence on the housing factor of the respective product group.

[0020] According to one embodiment, providing the second clustered component lifetime data involves clustering the second component lifetime data of a multitude of electronic products using a clustering method to identify lifetime-dependent product groups. In principle, several clustering methods are conceivable, which need not necessarily be mentioned here. In principle, several different clustering methods can also be applied, for example, to enable quality control.

[0021] According to one embodiment, the clustering method comprises a k-means clustering method and / or a Gaussian process clustering method and / or a random forest clustering method. Other clustering methods are also conceivable, without needing to be mentioned here.

[0022] According to one embodiment, the second clustered component lifetime data are clustered based on a crack length resulting at a respective solder joint of the electronic components with the circuit board due to thermomechanical stresses.

[0023] A crack length in a solder joint of a populated printed circuit board (PCB) preferably refers to the circumference or extent of a break or separation in the solder or the connection that attaches a component to the PCB. Cracks in solder joints can be caused by various factors. Firstly, thermal stress in the form of temperature fluctuations can cause the PCB and its components to expand and contract. These repeated thermal cycles can cause cracks in the solder joints over time. Furthermore, mechanical stress in the form of physical loads, such as bending or vibration, can cause cracks in solder joints. The crack length is a critical measurement because it can affect the integrity and reliability of the solder joint and thus the entire PCB. A crack can cause the electrical connection to break down or become unreliable.In many applications, especially in safety-critical systems, it is therefore important to detect and prevent or repair cracks before they lead to a failure.

[0024] According to one embodiment, the respective housing factor is influenced by the multitude of features of the respective electronic components and / or the respective printed circuit board and / or the respective electronic product, wherein the features include a type of housing and / or a housing material and / or a connection of the printed circuit board to the housing, in particular screwed or plugged, a connection of the housing and / or the printed circuit board to a heat conduction component and / or a connection of the housing to a motor or other heat-generating component and / or thermal information of an electronics of the printed circuit board and / or thermal conditions and / or housing-external influencing factors, in particular stresses applied to the housing from the outside.In general, the enclosure factor can be influenced by all characteristics of the electronic product and / or the circuit board and / or the electronic components and / or other external influences that have an indirect or direct effect on the load on the circuit board.

[0025] According to one embodiment, the thermomechanical stress test includes a temperature cycle test. A temperature cycle test is a method in which an object or component (in this case, possibly an electronic component or a printed circuit board) is repeatedly subjected to temperature changes between at least two specified temperatures. The goal is often to evaluate the reliability and robustness of the test object under thermomechanical stress. During this test, rapid temperature changes are repeatedly applied to simulate the thermal expansion and contraction of the materials from which the test object is made. This can induce cracks, solder joint failures, or other mechanical and thermal defects, especially in areas with dissimilar materials that expand and contract differently.

[0026] According to one embodiment, the first and / or second component lifetime data are preferably updated continuously or iteratively with further component lifetime data resulting from stress tests, wherein preferably a cluster adjustment of the product groups is performed based on the updated first and / or second component lifetime data, wherein preferably a housing factor adjustment is performed based on the cluster adjustment, and wherein the machine learning model is retrained, in particular continuously or iteratively, based on the updated first and / or second component lifetime data and / or the cluster adjustment and / or the housing factor adjustment. In this way, a continuous improvement in the prediction accuracy of the machine learning model can be achieved.

[0027] Furthermore, a method for determining a housing factor of a printed circuit board equipped with electronic components to be installed in a housing of an electronic product is claimed by means of a machine learning model trained according to the method according to any embodiment, wherein the method comprises: - Providing a wide range of features of the respective electronic components and / or the respective printed circuit board and / or the respective electronic product; and - Determining a package factor of the printed circuit board using the trained machine learning model based on the multitude of features.

[0028] This eliminates the need for costly and time-consuming destructive testing of new electronic products, as the enclosure factor can preferably be estimated based on tests performed on a bare circuit board. This has a cost-effective impact on product development.

[0029] According to the invention, a computer program with program code is also claimed to execute at least parts of the method according to the invention in one of its embodiments when the computer program is executed on a computer. In other words, according to the invention, a computer program (product) is claimed to comprise instructions that, when the program is executed by a computer, cause it to execute the method / steps of the method according to the invention in one of its embodiments.

[0030] According to the invention, a computer-readable data carrier containing the program code of a computer program is also proposed in order to execute at least parts of the method according to the invention in one of its embodiments when the computer program is executed on a computer. In other words, the invention relates to a computer-readable (storage) medium comprising instructions which, when executed by a computer, cause the computer to execute the method / steps of the method according to the invention in one of its embodiments.

[0031] The described designs and further training courses can be combined as desired.

[0032] Further possible embodiments, developments and implementations of the invention also include combinations of features of the invention described previously or subsequently with regard to the exemplary embodiments that are not explicitly mentioned. Short description of the drawings

[0033] The accompanying drawings are intended to provide a further understanding of the embodiments of the invention. They illustrate embodiments and, in conjunction with the description, serve to explain the principles and concepts of the invention.

[0034] Other embodiments and many of the aforementioned advantages become apparent with reference to the drawings. The elements depicted in the drawings are not necessarily shown to scale.

[0035] It shows: Fig. 1 a schematic flowchart of an embodiment of the method according to the invention.

[0036] In the figures of the drawings, the same reference symbols designate the same or functionally equivalent elements, parts or components, unless otherwise stated.

[0037] Fig.Figure 1 shows a schematic flowchart of a procedure for training a machine learning model to determine component lifetime-critical features of a printed circuit board populated with electronic components and installed in a housing of an electronic product.

[0038] The method can be carried out in any embodiment, at least partially, by a system 100, which may comprise several components not shown in detail, for example, one or more provisioning units and / or at least one evaluation and computing unit. It is understood that the provisioning unit may be configured together with the evaluation and computing unit, or it may be different from it. Furthermore, the system may comprise a storage unit and / or an output unit and / or a display unit and / or an input unit.

[0039] According to the invention, the computer-implemented method comprises at least the following steps: In step S1, initial component lifetime data, particularly standardized data, are provided for the printed circuit board populated with electronic components. This data is obtained in an unassembled state of the printed circuit board through a thermomechanical stress test. In step S2, second clustered component lifetime data, in particular standardized data, is provided for a large number of electronic products, in whose respective housing a circuit board populated with the electronic components is installed; In step S3, a respective housing factor is provided for at least each product group, in particular by comparing the first and second component lifetime data, especially for each of the clustered product groups. In step S4, a large number of features of the respective electronic components and / or the respective circuit board and / or the respective electronic product are provided. In step S5, the machine learning model is trained based on the first and second component lifetime data and / or the respective housing factor and on the basis of the multitude of features to determine component lifetime-critical features for each product group. In step S6, the training machine learning model is provided. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] TW 1 625 634 B

[0005]

Claims

[1] Method for training a machine learning model for determining component life-critical characteristics of a printed circuit board mounted with electronic components in a housing of an electronic product, the method comprising the steps: - Providing (S1) of, in particular standardized, first component lifetime data of the circuit board populated with the electronic components, which are recorded in an unassembled state of the circuit board by means of a thermomechanical stress test; - Providing (S2) second clustered component lifetime data, in particular standardized, of a plurality of electronic products, in the respective housing of which a printed circuit board populated with the electronic components is installed; - Providing (S3) a respective housing factor for at least each product group, in particular by comparing the first and second component lifetime data, in particular for each of the clustered product groups; - Providing (S4) a plurality of features of the respective electronic components and / or the respective circuit board and / or the respective electronic product; - training (S5) the machine learning model on the basis of the first and second component lifetime data and / or the respective housing factor and on the basis of the plurality of features to determine component lifetime-critical features for each product group; and - Providing (S6) the trained machine learning model. [2] The method of claim 1, further comprising the steps of: - Determine the component lifetime-critical characteristics for each product group using a Pareto analysis, and - Output of the component lifetime-critical characteristics which, based on the Pareto analysis, have the greatest influence on the respective, particularly product group-specific, housing factor. [3] The method of any preceding claim, wherein providing the second clustered component lifetime data comprises clustering the second component lifetime data of the plurality of electronic products by a clustering method for determining lifetime-dependent product groups. [4] The method according to claim 3, wherein the clustering method comprises a k-means clustering method and / or a Gaussian process clustering method and / or a random forest clustering method. [5] Method according to one of the preceding claims, wherein the second clustered component lifetime data are clustered on the basis of a crack length resulting at a respective solder joint of the electronic components with the circuit board due to thermomechanical stresses. [6] Method according to one of the preceding claims, wherein the respective housing factor is influenced by the plurality of features of the respective electronic components and / or the respective printed circuit board and / or the respective electronic product, wherein the features comprise a type of housing and / or a housing material and / or a connection of the printed circuit board to the housing, in particular screwed or plugged, a connection of the housing and / or the printed circuit board to a heat conduction component and / or a connection of the housing to a motor or another heat-generating component and / or heat conduction information of an electronics of the printed circuit board and / or thermal conditions and / or influencing factors external to the housing, in particular stresses applied to the housing from the outside. [7] A method according to any one of the preceding claims, wherein the thermomechanical stress test comprises a temperature cycling test. [8] Method according to one of the preceding claims, wherein the first and / or second component lifespan data are preferably updated continuously or iteratively by further component lifespan data resulting from load tests, wherein preferably a cluster adjustment of the product groups is carried out on the basis of the updated first and / or second component lifespan data, wherein preferably a housing factor adjustment is carried out on the basis of the cluster adjustment, and wherein the machine learning model is retrained, in particular continuously or iteratively, on the basis of the updated first and / or second component lifespan data and / or the cluster adjustment and / or the housing factor adjustment. [9] Method for determining a housing factor of a printed circuit board to be installed in a housing of an electronic product and populated with electronic components by means of a machine learning model trained according to the method according to one of the preceding claims, the method comprising the steps: - Providing a variety of features of the respective electronic components and / or the respective circuit board and / or the respective electronic product; and - Determining a package factor of the printed circuit board by the trained machine learning model based on the multitude of features. [10] System for training a machine learning model for determining component life-critical features of a printed circuit board mounted with electronic components in a housing of an electronic product, the system comprising an evaluation and / or computing device designed to carry out the following steps: - Providing (S1) of, in particular standardized, first component lifetime data of the circuit board populated with the electronic components, which are recorded in an unassembled state of the circuit board by means of a thermomechanical stress test; - Providing (S2) second clustered component lifetime data, in particular standardized, of a plurality of electronic products, in the respective housing of which a printed circuit board populated with the electronic components is installed; - Providing (S3) a respective housing factor for at least each product group, in particular by comparing the first and second component lifetime data, in particular for each of the clustered product groups; - Providing (S4) a plurality of features of the respective electronic components and / or the respective circuit board and / or the respective electronic product; - training (S5) the machine learning model on the basis of the first and second component lifetime data and / or the respective housing factor and on the basis of the plurality of features to determine component lifetime-critical features for each product group; and - Providing (S6) the trained machine learning model. [11] Computer program with program code to carry out at least parts of a method according to one of claims 1 to 9 when the computer program is executed on a computer. [12] Computer-readable data carrier with program code of a computer program for carrying out at least parts of a method according to one of claims 1 to 9 when the computer program is executed on a computer.

Images