Method and device for controlling a soldering process and method for automatically optimizing a circuit board

Abstract

The invention provides a method for controlling a soldering process, the method comprising the following steps: f) reading in CAD data regarding a circuit board on which electronic components are to be soldered to through-holes and / or pad fields, and using said CAD data as a circuit board model, the through-holes and / or pad fields each defining a soldering point, g) dividing the circuit board model into a plurality of circuit board segments, each circuit board segment having only a few soldering points, h) determining, for each individual circuit board segment, the distribution of an electrically conductive material contained in the circuit board segment, once in the plan view and once in a side view, so that a horizontal distribution and a vertical distribution are obtained, i) determining a minimum value and / or maximum value of at least one control parameter for a soldering process for soldering the soldering points, such that at least one predefined quality parameter is satisfied for all soldering points, the horizontal distribution and the vertical distribution of each segment being taken into account for this purpose, j) carrying out a soldering process, the at least one control parameter satisfying the determined minimum value and / or maximum value.

Description

[0001] Method and device for controlling a soldering process and

[0002] Method for automatically optimizing a printed circuit board

[0003] The present invention relates to a method and a device for controlling a soldering process and a method for automatically optimizing a printed circuit board.

[0004] When developing printed circuit boards, they are initially developed in several iterations between a digital design model in the form of CAD data and a prototype production of a physical circuit board. During development, predetermined design guidelines and a pre-defined standard set of manufacturing process parameters, which are usually determined empirically, are often taken into account. The production of physical prototypes usually leads to considerable waste, incurs costs and considerably delays the entire development process. In “Deep Learning Modelling for Composite Properties of PCB Conductive Layers” Stoyan Stoyanov et.al., International Conference on Thermal, Mechanical and Multi-Physics Simulation and Experiments in Microelectronics and Microsystems (EuroSimE), 2022, the simulation of printed circuit boards is presented so that a model of the conductive tracks orthe circuit board with its thermo-mechanical properties.

[0005] The invention is based on the object of creating a method and a device for controlling a soldering process as well as a method for automatically optimizing a printed circuit board to achieve defect-free solder joints, wherein a soldering process is efficiently controlled in such a way that no defective solder joints occur or a printed circuit board is optimized in such a way that no defective solder joints occur during a predetermined soldering process.

[0006] According to a first aspect, a method for controlling a soldering process is provided, comprising the following steps: a) reading in CAD data of a printed circuit board to which electronic components are to be soldered to through-holes and / or PAD fields, and using this CAD data as a printed circuit board model, wherein the through-holes and / or PAD fields each define a soldering point; b) dividing the printed circuit board model into several printed circuit board segments, wherein each printed circuit board segment has only a few soldering points; c) determining the distribution of an electrically conductive material contained in the respective printed circuit board segments for the individual printed circuit board segments in a distribution grid; d) determining a minimum value and / or maximum value of at least one control parameter for a soldering process for soldering the soldering points, such that at least one predetermined quality parameter is met for all soldering points;wherein the horizontal distribution and the vertical distribution of the respective segments are taken into account, e) carrying out a soldering process, wherein the at least one control parameter satisfies the determined minimum value and / or maximum value.

[0007] The distribution grid specifies the quantity and / or distribution of the electrically conductive material in the segment. In the simplest case, this is a value. It can also be a mathematical distribution function. A three-dimensional subdivision into a 3D matrix or voxels is also conceivable.

[0008] It is also possible to include a top view and a side view, so that a horizontal distribution and a vertical distribution are obtained.

[0009] For the purposes of the present invention, a “soldering point” is the location of a soldering point on a printed circuit board, whereas the soldering point is a soldering point provided with cured solder, which is used to electrically and mechanically connect a contact of an electronic component to a contact point of a printed circuit board, for example in the form of a via or a PAD field.

[0010] A soldering process at a specific solder joint is primarily determined by the flow behavior of the liquid solder and the heat flow. The invention is based on the finding that the heat flow and thus the temperature distribution in the area of ​​a solder joint are primarily influenced by the electrically conductive material of the circuit board, since electrically conductive material is inherently a good heat conductor. Furthermore, the inventor of the present invention has recognized that the heat flow during a soldering process, which is generally very dynamic when the liquid solder comes into contact with the circuit board, occurs only locally.Therefore, by segmenting a circuit board model into predetermined circuit board segments and determining the distribution of the electrically conductive material contained in the respective circuit board segments for each circuit board segment, both in a top view and a side view, it is possible to create a sufficiently precise description of the circuit board structure to determine or calculate specific parameters of a soldering process at a specific conductor location. Firstly, such a horizontal and vertical distribution of the electrically conductive material in the respective circuit board segment can be generated very quickly and easily. This allows the structure of the electrically conductive material in the circuit board to be described with sufficient precision using a small amount of data.By combining the local limitation due to the segmentation and the specific representation of the circuit board structure, a soldering process can be numerically described very precisely at the individual soldering points of a circuit board segment, and specific parameters can be derived. In the method according to the first aspect of the present invention, a minimum value and / or a maximum value is determined for at least one control parameter of the soldering process based on predetermined quality parameters for the soldering points or for the soldering points. The soldering process is then carried out such that the at least one control parameter fulfills the determined minimum value and / or maximum value, thereby ensuring that the predetermined quality parameter for the soldering points of the circuit board is fulfilled.

[0011] According to a second aspect of the present invention, a method for automatically optimizing a printed circuit board to achieve defect-free solder joints is provided. This method comprises the following steps:

[0012] A) Reading in CAD data of a printed circuit board to which electronic components are to be soldered to through-holes and / or PAD fields, and using this CAD data as a printed circuit board model, where through-holes and / or PAD fields each define a soldering point,

[0013] B) Dividing the circuit board model into several circuit board segments, each circuit board segment having only a few soldering points,

[0014] C) Determination of the distribution of an electrically conductive material contained in the respective circuit board segments for the individual circuit board segments in a distribution grid,

[0015] D) determining at least one quality parameter for each solder joint formed at each soldering point by a soldering process with at least one predetermined control parameter, taking into account the horizontal distribution and the vertical distribution of the respective segments,

[0016] E) Checking whether the quality parameters for each solder joint meet a predetermined quality threshold, and if this is not the case, then either the circuit board is redesigned at the corresponding solder joint according to predetermined rules so that the quality parameter for the respective solder joint meets the quality threshold, or at least one control parameter of the soldering process is modified so that the quality parameters for the respective solder joint meet the quality threshold.

[0017] In this method according to the second aspect of the present invention, as in the first aspect of the present invention, the insight that the fluid-mechanical and thermal effects during a soldering process essentially only act locally is utilized. Furthermore, by describing such local circuit board segments using the horizontal and vertical distribution of the electrically conductive material, a sufficiently precise description is obtained to accurately represent the highly dynamic soldering process numerically. This allows for a very simple and rapid optimization of a circuit board or the parameters of the soldering process used to solder this circuit board.

[0018] Preferably, such an optimized circuit board is manufactured according to the optimized CAD data. CAD data generally defines the structure of a circuit board, including all lines and insulating intermediate layers. Within the scope of the present disclosure, the CAD data may also be a reduced data set compared to conventional CAD data, which, for example, only defines the shape of the conductor tracks, as is known from Gerber data. The Gerber data constitutes an industry standard used for exchange between individual manufacturing companies for the manufacture and testing of circuit boards. The CAD data or Gerber data may be used indirectly as a circuit board model; however, it is expedient to process the CAD data or Gerber data specifically so that they are particularly suitable for a circuit board model.

[0019] Gerber data is a data format for exchanging descriptions of the conductive paths in the individual layers of a printed circuit board, which are described using basic geometric shapes. Compared to CAD data, Gerber data is a reduced data set and does not contain information about the thickness of the individual layers of the circuit board. There are also other exchange formats, such as ODB++, which contains more information than Gerber data but still less than CAD data.

[0020] The CAD data or the data according to the exchange formats are further processed, for example, to generate images in a specific format for the individual layers of the conductor tracks, which images can be used to determine a minimum and / or maximum value of at least one control parameter for the soldering process according to the first aspect or to determine at least one quality parameter for solder joints formed at specific soldering points by a soldering process with at least one predetermined control parameter according to the second aspect. In addition to the images of the conductor tracks, information on the thicknesses of the individual layers of the circuit board and / or the materials used for the layers of the circuit board and for the conductor tracks can also be taken into account.

[0021] In step d) and / or step D) of the procedures described above, all soldering points can also be taken into account.

[0022] The determination of a minimum value and / or maximum value of at least one control parameter for a soldering process for soldering the soldering points, so that at least one predetermined quality parameter is met for certain soldering points, and the determination of at least one quality parameter for soldering points formed at certain pixels by a soldering process with at least one predetermined control parameter, wherein the horizontal distribution and the vertical distribution of the respective segments are taken into account, is also referred to below as soldering point determination.

[0023] Such solder point determination is preferably carried out with a trained ML system (machine learning system).

[0024] By determining the respective soldering process at the soldering points segment by segment and by the very simple numerical description of the structure of the electrically conductive material, it is possible to determine the desired values ​​very quickly in such an ML system. Conventional numerical methods for describing a soldering process without using an ML system and without the above-mentioned segmentation and structuring of the distribution of the electrically conductive material in the circuit board require approximately 1 to 2 hours, up to several days, to calculate the soldering process at a single soldering point. Using the method according to the invention, the corresponding parameters at the corresponding soldering points can be determined in fractions of a second.

[0025] For each segment, a solder pseudo-image can be created, which are image-like representations of non-image data. These solder pseudo-images use color channels to convey information: The red channel shows a top view of the segment, the green channel a side view, and the blue channel encodes various control parameters such as soldering temperature and nozzle size as color values ​​scaled from 0 to 255.

[0026] A solder pseudo-image can then be processed in a processing model to generate one or more control parameters. The processing model is, for example, a neural regression network. The processing model consists of several layers, including input, intermediate, and output layers, and processes tensors, which are essentially data processing structures. The input tensor typically contains the solder pseudo-image and possibly other information, while the result tensor in the output layer represents the value(s) of the control parameter.

[0027] The behavior of the processing model is influenced by model parameters that can be optimized during training. For training, for example, a pre-trained model such as

[0028] VGG 16 is used. By refining pre-trained models, transfer learning, where the weights of the pre-trained model serve as initialization for the new task, the model can be trained efficiently. Bayesian hyperparameter optimization can refine the processing model by determining the best hyperparameters for performance. The processing model is trained using annotated data.

[0029] Training with annotated data is also called supervised learning.

[0030] Several or all solder points of the circuit board can each have a through-hole for accommodating a pin of a component, and the at least one quality parameter is the height of a solder penetration in the respective through-hole for a pin of a component to be attached inserted into the through-hole. It has been shown that the solder penetration can be determined very precisely using the method according to the invention, so that a specific height of the solder penetration can be used as a very reliable parameter.

[0031] Preferably, when determining a minimum value and / or maximum value of at least one control parameter for a soldering process according to claim 1 or when determining at least one quality parameter for each soldering point formed at each soldering point by a soldering process with at least one predetermined control parameter according to claim 2, thermal parameters of the component having the pin at the respective soldering point are taken into account, which are either calculated or estimated based on the geometry of the respective component or have been empirically determined in advance on the respective component. The calculated or estimated thermal parameters of the components can also have been determined in advance and stored, for example, in a database.

[0032] In the methods explained above, several predetermined quality parameters and / or several control parameters can be taken into account simultaneously.

[0033] The soldering process underlying these methods can be carried out using a soldering wave. In particular, this can be a locally limited soldering wave, such as that used in selective wave soldering.

[0034] The processes explained above can also be used in reflow soldering, where a suitable quality parameter is the curvature of the circuit board at the soldering point, which represents the deformation of the circuit board due to thermal influences.

[0035] The predetermined quality parameter(s) can be selected from the following group:

[0036] Height of a solder penetration, no bridge between adjacent solder joints, no open solder joint, no solder balls, no solder plugs, sufficient wetting, i.e. suitable wetting angle <90

[0037] Curvature of the circuit board, thermal solderability.

[0038] The quality parameters can thus be parameters represented by numerical values, such as the height of a solder via, which is represented by a corresponding height specification. The quality parameters can also be digital values, in particular "true" and "false," which describe the presence or absence of a bridge, an open solder joint, or a solder plug.

[0039] Thermal solderability is a measure of the solderability of a through-hole plating. It is a weighted summation of the copper content in the vertical direction. The higher the thermal solderability value, the less suitable the soldering process is at that location.

[0040] Thermal solderability does not describe the metallurgical feasibility of materials, i.e., their physical and chemical properties and their interactions during soldering. Thermal solderability is based on the distribution and proportion of copper in the vertical layer structure.

[0041] The copper content is taken into account across all layers. The copper content can be specified either as a percentage of copper in the respective layer of the respective circuit board segment or a predetermined area around the via, as a volume in the circuit board segment or in a predetermined area around the via, or as an area in the circuit board segment or in a predetermined area around the via, whereby the area assumes a uniform layer thickness, which indirectly also specifies a volume. The predetermined area is preferably a circular area arranged concentrically around the via. For each layer, the copper content is multiplied by a specific weighting factor in order to take into account the influence of the copper on thermal solderability.The result of this summation thus represents a measure of solderability, which directly reflects the thermal efficiency for the soldering process.

[0042] The weighting can be determined through measurements and / or simulations. The weighting increases with the distance from the solder contact point, since a high amount of thermally conductive copper in the distant layers favors heat distribution during soldering and is therefore more relevant than in the layers close to the solder contact point, which are inherently warmer. The weighting can now be such that, for example, it increases linearly with the distance to the solder contact point. In principle, quality parameters can also be formed from other quality parameters, for example by combining the quality parameters. Alternatively, mathematical operations such as addition or filter algorithms can be used to generate new quality parameters from existing ones.

[0043] The control parameters may include one or more of the following control parameters:

[0044] Wetting time

[0045] Solder bath temperature

[0046] - Preheating parameters

[0047] - Temperature of the circuit board

[0048] Nozzle diameter / area of ​​heat input

[0049] - Flow conditions of the solder nozzle (solder current, wave height, wave pump power)

[0050] - Composition of soldering materials (flux, solder alloy).

[0051] When determining a minimum value and / or maximum value of at least one control parameter for a soldering process according to the first aspect, or when determining at least one quality parameter for each solder joint formed at a soldering point by a soldering process with at least one predetermined control parameter according to the second aspect, one or more of the following input parameters can be taken into account:

[0052] Lot type

[0053] - Type of non-conductive material of the circuit board

[0054] - Type of conductive material of the circuit board.

[0055] The plurality of circuit board segments preferably have no more than 500 soldering points and preferably no more than 150 soldering points and in particular no more than 100 soldering points and / or a maximum area of ​​30 cm 2 and preferably a maximum area of ​​15 cm 2 and in particular a maximum area of ​​10 cm 2The criteria of the number of solder points and the area of ​​the circuit board segments can be applied independently or in combination. Circuit boards for computer server motherboards often have several hundred thousand solder points. An area with 500 solder points can constitute a small segment, especially if the solder points are formed in the form of a ball grid array. For circuit boards where the solder points are primarily formed by through-hole plating, it may be expedient to limit the circuit board segments to a small number of solder points, such as 150 or 100 solder points, or possibly even just 50 or 30 solder points.Both aspects have the following method steps in common: a) Reading in CAD data of a printed circuit board to which electronic components are to be soldered to through-holes and / or pad fields, and using this CAD data as a printed circuit board model or printed circuit board twin, wherein the through-holes and / or pad fields each define a soldering point, b) Dividing the printed circuit board model into several printed circuit board segments, c) Determining the distribution of an electrically conductive material contained in the respective printed circuit board segments for the individual printed circuit board segments in a distribution grid, e) Determining one or more parameters that describe the soldering process and / or the printed circuit board and / or the soldering point.

[0056] Using steps a) to e), any parameter can be determined very quickly, which indicates a characteristic value for the soldering process and / or the circuit board and / or the solder joint. This method can also be used to examine a soldering process and / or a circuit board with regard to its behavior during a soldering process and / or one or more solder joints.

[0057] If a parameter is determined in step e) that describes the solder joint, such as the solder penetration, then a specific soldering process is assumed, which is taken into account when determining this parameter.

[0058] The modifications to the first and second aspects of the present invention explained above can equally be used in this general analysis method.

[0059] The invention is explained in more detail below using examples illustrated in the drawings. The drawings schematically show:

[0060] Figure 1 shows a soldering system in a block diagram,

[0061] Figure 2 shows a selective soldering device in a block diagram,

[0062] Figure 3 shows a method for generating a soldering pseudo-image in a block diagram,

[0063] Figure 4 Input layer, intermediate layers and output layer of a processing model In a block diagram,

[0064] Figure 5 is a block diagram of a method for training a processing model,

[0065] Figure 6 is a block diagram of a method for controlling a soldering process, and

[0066] Figure 7 is a block diagram of another method for controlling a soldering process.

[0067] A soldering system 1 comprises a soldering device 2 for soldering solder 3 to a printed circuit board 4 and an optimization device 5 for optimizing control parameters 6 from a CAD file 7 (Figure 1). The soldering device 2 is, in particular, a selective wave soldering device 2, which generates a wave of molten solder 3 over which the printed circuit board 4 moves, so that the solder 3 adheres to the required locations on the printed circuit board 4, thereby creating a secure and reliable connection between components and the printed circuit board 4.

[0068] A selective wave soldering device 2 is designed to selectively apply the solder 3 to specific areas of the printed circuit board 4, rather than soldering the entire printed circuit board 4 in one pass, as is the case with traditional wave soldering. This selective process uses a mini-wave or solder beam that is directed only at the desired areas of the printed circuit board 3. The selective wave soldering device 2 has a system that allows a user to specify exactly where on the printed circuit board 4 the soldering should take place. This precise control ensures accuracy and consistency in the soldering process. This method is advantageous for assemblies containing a mix of surface-mount components and through-hole components, as the surface-mount components can be protected from excessive heat. This is advantageous, for example, in high-voltage and high-current applications, which often use such mixed assemblies.

[0069] Selective wave soldering offers the ability to precisely apply solder to specific components or areas, reducing the risk of unwanted connections or short circuits. It is particularly useful for printed circuit boards containing both through-hole and surface-mount components. With this technique, only the through-hole components are selectively soldered, while the surface-mount components are left unaffected.

[0070] By concentrating on specific areas, the amount of wasted solder 3 and flux is significantly reduced. The precision of selective soldering often results in fewer defects, requiring less rework. As printed circuit boards 4 become increasingly complex and component spacing becomes ever smaller, selective soldering can be used to address areas that can be problematic for conventional wave soldering.

[0071] Because only specific areas of the circuit board 3 are exposed to the molten solder 3, the overall thermal stress on the circuit board 4 and its components is reduced. In the long term, the reduction in scrap, rework, and defects can lead to cost savings. High-voltage and high-current components often have specific thermal requirements. Selective wave soldering allows for better control of heat input to specific components, ensuring that temperature-sensitive components are not exposed to excessive heat that could otherwise compromise their performance or service life. High-voltage and high-current applications often include layouts with components that require different soldering conditions. Selective wave soldering offers the flexibility to meet these requirements without the need for multiple soldering processes or machines.

[0072] The wave soldering device 2 comprises a solder bath 8, which comprises a container containing the solder 3 in its molten state (Figure 2). The solder bath 8 is made of a material that can withstand high temperatures, such as stainless steel or other heat-resistant alloys. The solder bath 8 comprises a heating element 9 that keeps the solder 3 at a constant, predetermined temperature and ensures that it remains in a molten state.

[0073] The soldering device 2 further comprises a waveformer for generating the wave of molten solder 3. The waveformer comprises a pump 10 that pushes the molten solder 3 upwards through a nozzle 11, creating a wave-like effect. The height, width, and speed can be adjusted via control parameters 6.

[0074] Another element of the soldering device 2 is a conveyor system 12, which moves the circuit board 4 over the wave of molten solder 3. The circuit board 4 is held at a specific angle so that the components on the circuit board 4 come into contact with the soldering wave. The speed of the conveyor system 12 can be adjusted via corresponding control parameters 6 to control the time the circuit board 4 is exposed to the soldering wave.

[0075] The control parameters 6 are parameters which can be set on the soldering device 2 in order to control a soldering process.

[0076] According to this embodiment, the soldering device 2 comprises a flux system that ensures that, before the circuit board 4 reaches the soldering wave, a layer of flux is applied to the circuit board 4 to clean the surface of the circuit board 4 and improve the adhesion of the solder 3. The flux system can use various application methods, such as spraying, foaming, or dipping.

[0077] After the flux system, the circuit board 4 passes through a preheating system of the soldering device 2, which heats the temperature of the circuit board 4 and its specific components to prevent thermal shock when the circuit board 4 comes into contact with the molten solder 3. The preheating system operates with an infrared heater or a convection heater.

[0078] The soldering device 2 further comprises a soldering device control unit 13 for controlling the operation of the soldering device 2. This can be used to set various control parameters 6 of the soldering device 2, such as the temperature of the solder bath 8, the speed of the conveyor belt or the flow rate of the soldering wave.

[0079] The soldering device control unit 13 comprises several subcomponents, such as a user interface, for interacting with an operator. According to this embodiment, it comprises a screen, as well as input devices such as a keyboard, switches, and / or a touchscreen. A user can enter commands, make settings, and monitor the status of the soldering device 2 via a user interface. The user interface provides real-time information about the operation of the soldering device 2, including the temperature of the solder bath 8, the speed of the conveyor belt, the flow rate of the solder wave, and other relevant data. The soldering device control unit 13 further comprises a microcontroller, which, according to other embodiments, may also be a computer or other computing unit.The microcontroller receives inputs from users and external inputs from other devices, as well as from various sensors 14 located in the soldering device 2. Based on these inputs, it sends commands to the various components of the soldering device 2 to control their operation. The microcontroller is programmed with specific algorithms that allow it to make decisions and adjust the device's operation as needed.

[0080] The soldering device 2 further includes several sensors 14 that monitor the operation of the soldering device 2, such as temperature sensors in the solder bath 8, in the preheating system, as well as speed sensors on the conveyor belt and flow sensors 14 in the wave-forming mechanism. The sensors 14 provide real-time data to the soldering device control unit 13, allowing it to make adjustments as needed to ensure optimal operation.

[0081] The soldering device 2 includes several actuators 15 that perform actions based on commands from the soldering device control unit 13, such as a motor that drives the conveyor belt, pumps 10 that control the solder flow in the wave-forming mechanism, and heating elements 9 in the solder bath 8 and the preheating system. The actuators 15 respond to command signals from the soldering device control unit 13 and adjust their operation to maintain the desired operation.

[0082] A CAD file (CAD = Computer Aided Design) 7 for a printed circuit board 4 (PCB = Printed Circuit Board) is a digital representation of the printed circuit board layout. The CAD file 7 contains detailed information about the design and configuration of the printed circuit board 4, including the placement of components, the routing of electrical connections, and the dimensions of the printed circuit board 4. The CAD file 7 is created using special software for printed circuit board design. The CAD file 7 includes a circuit board outline that defines the dimensions and shape of the printed circuit board 4. The type and placement of components can also be stored in the CAD file 7, specifying the position of all components to be mounted on the printed circuit board 4. This includes both through-hole and surface-mount components.Each component is represented by a footprint that indicates the size, shape and orientation of the component as well as the position of the pins or pads.

[0083] CAD file 7 also includes information about the wires that electrically connect the individual components. This is represented by the conductive traces 16. CAD file 7 specifies the width, shape, and position of each of the conductive traces 16. Drill holes are also listed in CAD file 7, with CAD file 7 specifying the position, size, and type of all holes. These include holes for mounting through-hole components as well as holes for mounting the circuit board 4 itself.

[0084] Another component of CAD file 7 is the layers. The circuit board 4 can also be multi-layered, and CAD file 7 contains information about each layer. These include copper layers containing the conductor tracks 16 and pads, as well as insulating layers that separate the copper layers.

[0085] The CAD file 7 may also contain information about a solder mask, which is a layer of this insulating material that protects the conductor track 16 and prevents solder breaks, as well as information about silk screen, which is a layer of text and symbols that provides information about the components and connections.

[0086] The CAD file 7 may also contain special instructions for the soldering process, such as the areas that should not be soldered or instructions for handling delicate components.

[0087] According to this embodiment, the circuit board 4 is a printed circuit board (PCB) and comprises a flat, rigid plate made of non-conductive material, such as fiberglass or plastic. The circuit board 4 may also already include conductive tracks 16, solder joints 17, and vias 18 that electrically connect various components.

[0088] The circuit board 4, which is processed in the soldering device 2, is usually already populated with electrical components. These components can include resistors, capacitors, transistors, diodes, integrated circuits, or connectors. The components are attached through holes in the circuit board 4 or rest on the solder pads. The component leads or their pins extend into individual layers of the circuit board 4 or even to the other side of the circuit board 4.

[0089] In conjunction with a wave soldering device 2, the circuit board 4 is typically constructed with through-hole components. This means that the leads and pins of the components extend through to the other side of the circuit board 4. The components have leads that are inserted into the holes on the circuit board 4 and then soldered on the other side of the circuit board 4. The wave soldering process is particularly well suited for soldering this type of component.

[0090] In this embodiment, the following control parameters 6 are adjusted.

[0091] The temperature of the solder 3 used in the soldering process is called the soldering temperature. It must be high enough for the solder 3 to melt and form a connection without damaging the components or the circuit board 4. The ideal soldering temperature depends on factors such as the type of solder alloy used and the specific components being soldered.

[0092] The preheat temperature refers to the temperature to which both the circuit board 4 and the components are heated before the solder 3 is applied. Preheating prevents damage to the components, improves solder flow, and effectively activates the flux.

[0093] The solder contact time indicates how long the solder (3) remains in contact with the circuit board (4) and the components. It should be sufficient to establish a connection, but not so long that it causes damage. The ideal duration of the soldering process depends on factors such as the preheat temperature of the solder and the components involved.

[0094] The size of the nozzle 11 used in wave or selective wave systems determines either a metal wave or a controlled metal jet for precise application in wave or selective wave soldering. The choice of nozzle size is crucial, as it should be tailored to both the size and complexity of the printed circuit boards 4 and their components. Differently sized nozzles 11 are used for soldering different types of printed circuit boards (PCBs).

[0095] The solder alloy refers to the composition of the solder 3 itself. There are solder alloys such as tin-lead, tin-silver-copper, and others. The choice of solder alloy affects factors such as the melting point, wetting properties, and mechanical strength of the solder joints 17.

[0096] The type of component design refers to the components that are to be connected to the circuit board 4 by soldering. Different components may have requirements regarding the temperature or contact time during the soldering process.

[0097] The annular ring width refers to the width of a ring surrounding a hole on a printed circuit board (PCB) 4. The annular ring width plays a role in determining the strength of each solder joint (17). It is important to control this parameter to achieve robust soldering. Heat transfer involves transferring heat from the solder (3) to the printed circuit board (4) and its components.

[0098] The soldering system 1 also includes an optimization device 5 which optimizes the control parameters 6 for the soldering device 2 based on the CAD file 7.

[0099] The optimization device 5 comprises a segmentation module 19, which generates a segment file 20 from the CAD file 7, a weighting module 21, which generates a soldering pseudo-image 23 from individual segments 22 of the segment file 20, and a parameter module 24, which generates control parameters 6 for the individual segments 22 and forms a parameter map 25 therefrom.

[0100] According to this exemplary embodiment, the optimization device 5 is a device that includes a computing unit on which the optimization can be performed. The computing unit can be identical to the soldering device control unit 13 or can be independent of it, such as a computer or a server. If the computing unit is not identical to the soldering device control unit 13, the computing unit is connected to it at least via data connections in order to send corresponding control parameters 6 to it.

[0101] The optimization software, which is explained in more detail below, is stored and executable on the computing unit of the optimization device 5. The software is comprised of several modules that exchange data via channels. The channels are logical data connections between the individual modules. The modules can be either software or hardware modules.

[0102] For example, there may be several computing units that represent the individual modules, with the corresponding information then being exchanged between the individual modules.

[0103] A first such module of the optimization device 5 is the segmentation module 19, which divides the circuit board layout defined in the CAD file 7 into different segments 22.

[0104] In the simplest case, the segmentation process can be defined by a predetermined grid and divide the circuit board 4 into predetermined segments 22. These segments 22 can, for example, be equally sized square areas with an edge length of 22.4 mm.

[0105] However, it has proven advantageous that the respective intended soldering points 17 are taken into account during segmentation, so that there is one soldering point 17 in the center for each segment 22. Each soldering point 17 is thus assigned to a segment 22. In this case, it must be accepted that certain areas of the printed circuit board 4 are imaged multiple times, particularly when two soldering points 17 are close to one another, for example, with the size selected above of an edge length of 22.4 mm and a distance between the soldering points 17 that is less than 11.2 mm, an overlap area must be accepted.

[0106] Other areas of the circuit board 4, however, are not detected in the individual segments 22 if no soldering point 17 is provided in these areas.

[0107] The individual segments 22 are summarized in a segment file 20 and forwarded to a weighting module 21.

[0108] Each segment 22 is later treated separately during the soldering process, which means that individual control parameters 6 are used for each segment 22. However, depending on the soldering device 2, several segments 22 can be soldered simultaneously.

[0109] The weighting module 21 processes the segment file 20 with the individual segments 22. The weighting module 21 imports the 3D segments 22 of the circuit board 4 from the CAD file 7, from which vertical and horizontal 2D projections of the segments 22 are then created (Figure 3). According to this initial example, the horizontal resolution is 0.1 mm per pixel and the vertical resolution is 0.035 mm per pixel. For each individual segment 22, several solder pseudo-images 23 are generated. A solder pseudo-image 23 is an image-like representation of data that does not necessarily have to be in image form. It is simply a way of visualizing and processing non-image data in the same way as an image. Since machine learning algorithms are highly advanced, particularly in the field of image processing, solder pseudo-images 23 can be easily processed in this way.Each color channel in this soldering pseudo-image 23 (red 23', green 23", blue 23'") represents different information regarding the circuit board 4 and the soldering process.

[0110] According to this embodiment, the red channel 23' represents a horizontal projection of the segment 22. It provides a top view of the segment 22 and shows the arrangement of the components and conductor tracks 16. It thus shows a horizontal distribution of the heat-conducting elements within the segment.

[0111] The green channel 23" represents the vertical projection of segment 22. It provides a side view of segment 22 and shows the vertical distribution of the components and conductor tracks 16. It thus shows a vertical distribution of the heat-conducting elements within the segment.

[0112] The blue channel 17'" comprises different stripes (5' to 5"") in the soldering pseudo-images 23 of the various segments 22. These represent various input parameters 35. In this exemplary embodiment, the input parameters 35 are threshold values ​​of the desired solder penetration and standard values ​​of the various control parameters 6, such as the soldering temperature, the preheating temperature, the soldering contact time, the size of the nozzle 11, the solder alloy, the type of component, the annular width, and the heat transfer.

[0113] Each stripe of an input parameter 35 is scaled as a color value from 0-255.

[0114] If individual input parameters 35 are unknown, this can be represented, for example, by a completely black or white area.

[0115] In other embodiments, it is conceivable to enter the input parameters 35 into the parameter module 24 in a different way.

[0116] Each soldering pseudo-image 23 is stored in a database and sent to the parameter module 24.

[0117] The parameter module 24 generates the control parameters 6 for the soldering of each segment 22 from the soldering pseudo-images 23.

[0118] Depending on the embodiment, all, some specific or only one control parameter 6 can be output.

[0119] Depending on how and where the heat-conducting elements, such as the conductor tracks 16, are distributed horizontally and vertically in the segment 22 and how high the solder penetration is desired, the soldering contact time is adjusted.

[0120] The parameter module 24 comprises a processing model 26 in which the solder pseudo-images 23 are read individually and independently of one another.

[0121] The processing model 26 is a neural regression network, also referred to below as a neural network, which determines the control parameters 6 as output values ​​from the input solder pseudo-images 23. A value of a control parameter 6 is assigned to each type of solder pseudo-image 23. The regression network is constructed with multiple layers. The neural network comprises several intermediate layers 33, an input layer 27, and an output layer 28 (Figure 4).

[0122] The individual layers process tensors. The tensors are data. Each layer receives input tensors and outputs output tensors. The input tensor of input layer 27 is also called the initial tensor. The output tensors of input layer 27 and the intermediate layers 33 form the input tensors of the subsequent layers in the neural network. The output tensor of output layer 28 is also called the result tensor 34 and represents the corresponding value of the control parameter(s) 6.

[0123] The initial tensor includes the previously described solder pseudo-image 23.

[0124] The output layer 28 in the neural network consists of a single neuron.

[0125] However, it is also conceivable that the neural network consists of several neurons if several output values, ie control parameters 6, are to be determined simultaneously.

[0126] The input layer 27 of the regression network accepts the solder pseudo-images 23 generated from the segmented CAD files 7 of the printed circuit board 4.

[0127] The intermediate layers 33, also called convolutional layers, are designed to automatically and adaptively learn spatial hierarchies of features from the solder pseudo-images 23. These layers apply a series of filters to the input image to create feature maps that highlight the areas of the image most relevant to the prediction task. The network further includes several activation functions. The activation functions in the neural network are mathematical functions that determine the output of a neuron, or node, given an input or a series of inputs. They introduce nonlinearity into the output of a neuron, allowing the neural network to learn and represent complex patterns and relationships in the data.

[0128] The activation functions used here are the SIGMOID function, which takes any real number and returns an output value that falls within the range 0 to 1, or a hyperbolic tangent function (TANH). The TANH function is similar to the SIGMOID function but returns an output value that falls within the range -1 to 1, and the Rectified Linear Unit (RELU) function, which returns the input directly if it is positive and 0 otherwise.

[0129] Other activation functions are also conceivable, such as the leaky Relu function, which takes a small slope for negative values ​​instead of a flat slope. The intermediate layers 33 can also include pooling layers, which are used to reduce the spatial dimensions of the corresponding tensors and thus reduce the computational complexity of the network. Common pooling types are max pooling and average pooling.

[0130] Several layers can also be fully connected. Fully connected layers take the high-level features learned by the convolutional layers and combine them to predict the control parameters. These layers are called fully connected because every neuron in this layer is connected to every neuron in the previous layer.

[0131] The behavior of the processing model 26 is determined by model parameters 30. These model parameters 30 are optimized during training.

[0132] When training the processing model 26, a pre-trained model, such as a VGG 16, is used. The VGG 16 is a convolutional neural network introduced by K. Simonyan and Zisserman in their paper "Very Deep Convolutional Networks for Large Scale Image Recognition." It comprises 16 layers. VGG 16 is particularly widely used in image recognition tasks. Compared to other models, it comprises smaller convolutional layers and a less complex structure. It was pre-trained on the ImageNet dataset.

[0133] This pretraining is used to transfer learning the learned weights as initialization for training a new task. Such transfer learning is also applied here to generate the corresponding control parameters 6 from the solder pseudo-images 23.

[0134] In this example, the pre-trained network is first optimized using so-called Bayesian hyperparameter optimization. Hyperparameters are the model parameters that are defined before the learning process begins, such as the learning rate, the number of layers in a neural network, or the number of neurons.

[0135] The goal of hyperparameter optimization is to find the combination of hyperparameters that yields the best model performance on a validation set. Bayesian hyperparameter optimization is distinguished by its use of a probabilistic model to predict which hyperparameters will perform best.

[0136] Once the hyperparameters have been found through Bayesian hyperparameter optimization, the neural network can perform transfer learning. In this case, the training was set to four fully connected layers with up to 1024 neurons. In transfer learning, a pre-tested model, such as this VGG 16, is used, and the last layer(s) of the pre-trained model, which are specific to the original task, are then removed. New layers are then added to the model to meet the requirements of the new task, i.e., generating control parameters.

[0137] According to this embodiment, the weights of the pre-trained layers are frozen, or set so that they cannot be trained. This means that the weights of this layer(s) are not updated during training of the new task. This allows the model to retain the learned feature representations from the pre-trained model.

[0138] After initial training with frozen pre-trained layers, the model is fine-tuned. This involves unfreezing some or all of the pre-trained layers and continuing training. This allows the model to adapt the pre-trained feature presentations to be more suitable for the new task.

[0139] The model is trained on the training set (Figure 5). For a training step, annotated data 30 is input into the processing model 26.

[0140] Annotated data 30 or an annotated data set is a data set that includes both initial tensors and a corresponding target tensor 31 for each initial tensor.

[0141] Thus, a solder pseudo-image 23 of a segment 22 with one or more input parameters 35 is entered as the initial tensor.

[0142] The target tensor 31 is similar in structure to the result tensor 34, but was not generated by this processing model 26.

[0143] For this purpose, printed circuit boards 4 are manufactured for learning purposes with the corresponding segments 22 and with various control parameters 6 before the actual training. After production, the quality of the control parameters 6 is assessed by one or more experts based on the height of the solder penetration.

[0144] If the control parameters 6 have generated an acceptable solder penetration, these control parameters 6 are assigned as a target tensor 31 to the corresponding solder pseudo-image 23. They then form a data point for the annotated data 30. The annotated data 30 can also have the same solder pseudo-image 23 multiple times, provided there are multiple combinations of control parameters 6 that enable an acceptable solder penetration.

[0145] The processing model 26 now generates a result tensor 34.

[0146] The output result tensors 34 are used together with the given target tensors 31 to calculate a target function 32.

[0147] Using the objective function 32, new model parameters 30 for the processing model 26 are calculated.

[0148] During training, the objective function 32 is optimized. The objective function 32 is either a loss function to be minimized or a gain function to be maximized. Since new model parameters 29 are calculated based on the objective function 32, the processing model 26 as a whole is also optimized.

[0149] A stochastic gradient descent method is used here to optimize the objective function 32. However, other methods are also conceivable.

[0150] A large number of steps are required to train the processing model 26. For this purpose, a corresponding amount of annotated data 30, i.e., solder pseudo-images 23 with associated control parameters 6, must be read into the processing model 26.

[0151] In order to determine all control parameters 6, the annotated data 30 must also include solder pseudo-images 23 that generate control parameters 6 in the respective areas.

[0152] It may be advantageous to use provoked soldering pseudo-images 23, i.e. images that intentionally show corresponding areas.

[0153] In principle, the processing model 26 is more reliable the more it has been trained with solder pseudo-images 23 of different control parameters 6 and values ​​of control parameters 6.

[0154] The control parameters 6 determined by the neural network for the individual segments 22 are then used to optimally solder the corresponding segments 22 in the soldering device control unit 13. Since the neural network has been trained to generate a desired solder penetration for the respective segments 22 of the specified control parameters 6, the control parameters 6 are determined in such a way that they deliver a solder penetration with an acceptable height, even for new segments 22 unknown to the neural network.

[0155] The method for controlling a soldering process is explained below. In particular, control parameters 6, which influence the heat flow to the solder joints 17, are optimized.

[0156] The process begins with step S1 (Figure 6).

[0157] In the next step (S2), the CAD file 7 is sent to the soldering device 2 and to the optimization device 5.

[0158] The soldering device 2 receives the information from the CAD file 7 as to where which soldering points 17 are located.

[0159] In the optimization device 5, the CAD file 7 is used in the segmentation module 19 to generate several segments 22. These segments 22 are stored in a segment file 20 and sent to the weighting module 21.

[0160] In alternative embodiments, the individual segments 22 can also be sent individually to the weighting module 21.

[0161] This is followed by step S4, in which solder pseudo-images 23 of the individual segments 22 are created in the weighting module 21. The creation of such solder pseudo-images 23 has already been explained in detail above.

[0162] In the following step (S5), the individual control parameters 6 are determined by the parameter module 24, which comprises a neural network, as explained in detail above.

[0163] Thereafter, in step S6, the determined control parameters 6 are sent to the soldering device 2.

[0164] In the subsequent step S7, the circuit board 4, which is represented by the CAD file 7, is soldered with the control parameter 6 determined in the optimization device 5.

[0165] The process ends with step S8. It may be that some control parameters 6, such as the solder bath temperature, cannot be individually adjusted for each soldering point during the process. In this case, a compromise can be found between the individual optimization steps, for example, by calculating an average or median.

[0166] Individual control parameters for each soldering point in selective wave soldering are the solder contact time and the flux application quantity.

[0167] A second embodiment is explained below, wherein identical elements as in the first embodiment are provided with the same reference numerals. The above explanations apply to identical elements unless otherwise stated below.

[0168] The second embodiment involves a reflow soldering process instead of a selective wave soldering process. In particular, the bending or warping of a circuit board during reflow soldering is monitored.

[0169] Reflow soldering is a technique used to connect electronic components to printed circuit boards (PCBs). The process involves using solder to bond the component pads to flat contact areas on the board, called pads. The process begins with the application of solder paste—a combination of solder particles and flux—to pads on the board where the components are to be electrically connected and mechanically secured. To ensure accuracy, a stencil, for example, may be used for application. Once the solder paste is applied, the electronic components are positioned on the board so that their leads align with the areas containing the solder paste.

[0170] After the components are placed, a controlled heating process takes place in an oven. This heating process follows a defined profile consisting of several stages: preheating, insertion, melting, and cooling. During the preheating and soaking phases, the entire circuit board is heated as evenly as possible to activate the flux contained in the solder paste. It can also be useful to apply different levels of heat to individual sections of the circuit board, for example, by individually adjusting the intensity of a hot air stream.

[0171] During the reflow phase, the solder paste reaches its melting point, allowing it to melt and create connections by wetting both the component leads and the pad areas. Finally, during the cooling phase, the temperature is lowered, solidifying the solder and creating robust and reliable solder joints.

[0172] Reflow soldering offers the advantage of allowing the simultaneous soldering of multiple SMD components, which significantly improves efficiency in high-volume production. The process guarantees good solder joints thanks to the controlled conditions provided by the reflow oven. However, it is important to control the thermal profile and apply the solder paste precisely to avoid problems such as solder bridges, voids, cold solder joints, or poorly connected solder points.

[0173] During the reflow soldering process, the bending of printed circuit boards (4), often referred to as "warping," is a widespread problem. This occurs when the printed circuit board (4) is deformed due to uneven thermal expansion and contraction. This deformation can result from a combination of factors, such as the material properties of the printed circuit board (4), its design features, the uneven distribution of electrically conductive material, such as copper, within the electrically insulating material, and the thermal profile that develops in the printed circuit board during soldering.

[0174] One of the reasons for warpage is the temperature behavior during the soldering process of the PCB. 4 When materials used in printed circuit boards are heated, they expand. When they cool, they contract. However, the different materials within a PCB can expand to different degrees when heated. Therefore, during the heating phase of the soldering process, different layers or sections of the PCB can expand to different degrees. Similarly, during the cooling phase, these layers will contract to different degrees. This uneven expansion and contraction can lead to stresses within the PCB, causing deformation—warpage.

[0175] The speed at which temperature changes occur during soldering affects warpage. When a printed circuit board (PCB) is heated or cooled rapidly, certain sections may not have enough time to expand or contract evenly. The different distribution of highly thermally conductive material, which is typically the electrically conductive material of the PCB, and the electrically and thermally insulating material can play a role here. This can lead to stresses in the material. Likewise, rapid cooling can trap these stresses. The result is a warped PCB.

[0176] The design and layout of a printed circuit board, combined with the reflow oven settings, can result in uneven temperature distribution across the surface of the board. 4 Components and copper traces can influence how heat is absorbed and dissipated on the board. Areas with higher component density or thicker copper layers can retain heat longer than thinner areas. The uneven temperature distribution can cause areas of the board to temporarily expand and contract, which can lead to warping. If significant flexion occurs, it can cause a variety of problems, including component misalignment, solder joint defects, and even electrical failures, affecting the reliability and functionality of the finished product.

[0177] These factors can be optimized by the method described here in the optimization device 5 in order to at least minimize warping.

[0178] Other factors that can be taken into account during optimization are: no open solder joints to ensure an electrical connection, uniform heating of the circuit board to ensure low thermal loads are also applied to components.

[0179] In this embodiment, the parameter module 24 in the optimization device 5 uses two machining models 26, a macro model and a micro model, each having different resolutions, to evaluate the global and local effect of the heat distribution on the deformation.

[0180] In the first embodiment, the influence of a substantially central heat source within a segment 22 was analyzed. During reflow soldering, heat is introduced globally across the entire circuit track, and warping of a circuit board is an effect based on the temperature distribution over a larger area of ​​the circuit board. It is therefore expedient to analyze a larger area of ​​the circuit board and consider the heat dissipation between adjacent segments. Therefore, it is also expedient to consider the heat distribution of the entire circuit board 4.

[0181] The micromodel corresponds to, or is at least similar to, the processing model of the first embodiment. The output of the micromodel and the macromodel are fed to a higher-level global module.

[0182] The micromodel analyses the heat distribution or heat flow in the circuit board within the individual segments, and the macromodel analyses the heat distribution or heat flow with less precision across the entire circuit board. The two analyses are combined in the global model. As explained in more detail below, the macromodel takes into account the influence of the global heat flow on the individual segments of the micromodel. In contrast to the first exemplary embodiment, the analysis does not primarily evaluate a point-like area of ​​the circuit board; rather, taking into account the curvature of a circuit board requires the observation of a larger area in the plan view of the circuit board. The macromodel operates at a coarse resolution and is used to evaluate the global heat distribution and its effect on warping under thermal influences.The model uses a solder pseudo-image 23 of the entire circuit board 4 as input. The segmentation module 19 described above can also be used here, but the entire circuit board 4 is selected as the segment 22. The weighting module 22 then generates a solder pseudo-image 23 of the entire circuit board 4 as described above.

[0183] The resolution of the solder pseudo-image 23 can be fixed or variable. In the case of a variable resolution, the model can process circuit boards 4 of any size by adjusting the resolution so that the image is represented in a fixed number of pixels. This allows the model to process circuit boards 4 of different sizes using the same pre-trained neural network. The selected resolution is communicated to the model as a feature, allowing the model to assess the size of the circuit board 4.

[0184] As described above, the macromodel uses a pre-trained neural network, such as the VGG 16, and is fine-tuned using Bayesian hyperparameter optimization to optimize the soldering parameters at the global level. The result of the macromodel is a set of global control parameters 6 that minimize deformation, i.e., warping, across the entire printed circuit board 4.

[0185] For the annotated data 30, printed circuit boards 4 are manufactured for learning with various control parameters 6. During and / or after production, the curvature of the printed circuit boards 4 is determined as a quality characteristic of the control parameters 5. The curvature, which represents a three-dimensional deformation, can be detected, for example, using a 3D scanner, such as a stereo camera, or a laser scanner, i.e., a light source for emitting a light line and a camera that operates according to the light-section method.

[0186] The 3D scanner is placed in the soldering device 2, specifically at the position where the greatest curvature is expected. In the reflow process described here, this is the location where the reflow phase, referred to as the reflow oven, is performed. Multiple 3D scanners can also be distributed throughout.

[0187] The measurements of the three-dimensional surface can be output as a height map. Experts then use this height map to assess the quality of the control parameters.

[0188] If the control parameters 6 have generated no or an acceptable curvature, these control parameters 6 are assigned as target tensor 31 to the corresponding solder pseudo-image 23. The macromodel outputs optimized control parameters 6.

[0189] It is fundamentally conceivable that a calculated curvature is also output, provided that the model has been trained to do so.

[0190] The micromodel operates at a finer resolution and focuses on specific components, such as ball grid arrays, whose soldering success is susceptible to thermal distortion. The micromodel takes as input a solder pseudoimage 23 of a local segment 22 of the printed circuit board 4 around a specific component, as described above in the first embodiment.

[0191] The position of segment 22 is also entered as input parameter 35.

[0192] The micromodel also uses a pre-trained neural network and is fine-tuned using Bayesian hyperparameter optimization. However, the micromodel is trained to optimize the control parameters at a local level, focusing on the specific requirements of the component.

[0193] For the annotated data 30, control parameters 6 found to be optimal are again selected for the corresponding pseudo-images 23.

[0194] The result of the micromodel is a set of local control parameters 6 that minimize the deformation around the specific component.

[0195] This approach with two processing models 26 enables sophisticated control of the soldering process to minimize thermal distortion. Both processing models 26 can be trained in such a way that the curvature is minimized for each input.

[0196] While the macro model focuses on global heat distributions spanning circuit boards and thus minimizes warping, the micro model concentrates on individual segments 22.

[0197] The vertical projection of the solder pseudo-image 23 plays a rather minor role in the macro model, since the projection depicts almost all layers as thermally conductive. This is due to the fact that the projection is projected onto an area relatively small compared to the surface of the circuit board 4. The influence is correspondingly small in the macro model and can also be omitted in other embodiments.

[0198] In the micromodel, however, the vertical projections can certainly have a greater influence. Using only the macromodel would compute too little information about the local solder joints. Conversely, the micromodel alone is not capable of determining control parameters 5 that prevent warping.

[0199] Depending on which control parameters 6 are output in the macro and micro models, the respective results may partially contradict each other. To do this, both sets of control parameters 6 are loaded into the global module, which then combines the two sets of control parameters 6 based on predefined criteria and outputs a single set.

[0200] According to this embodiment, this global module is also a processing module with a neural network, which has been trained to output a single set of optimized control parameters 6 based on the input control parameters 6 of the macro and micro models.

[0201] Alternatively, different control parameters 6 can be output for the macro and micro models, so that they do not contradict each other and only need to be combined.

[0202] Alternatively, a processing model may be provided that includes the functionality of both the macromodel and the micromodel.

[0203] The input data of this processing model always includes at least a solder pseudo-image of the entire circuit board, a solder pseudo-image of the segment to be examined and the location of this segment on the circuit board.

[0204] Depending on the application, the output data includes a) control parameters to control the soldering process, b) optimization parameters to optimize the circuit board, c) analysis parameters that describe the curvature and / or temperature distribution.

[0205] To a)

[0206] When determining optimal control parameters for controlling the soldering process, parameters can also be entered as input data that define the soldering machine for which the control parameters are to be determined and, if necessary, individual control parameters that are defined in advance.

[0207] Regarding b) When determining optimization parameters to optimize the printed circuit board so that it can then be manufactured according to the optimization parameters, parameters can be entered as input data that define the soldering machine to be used and / or the control parameters to be used.

[0208] To c)

[0209] When determining analysis parameters that describe the curvature and / or heat distribution, parameters that define the soldering machine to be used and / or the control parameters to be used can be entered as input data.

[0210] The annotated dataset for training the processing model comprises a set of input data and a set of output data. In cases a) and b), such a set of input data and output data qualifies as annotated data if it has been previously determined that the output data is obtained when the input data is applied and, at the same time, the required quality criteria for curvature and / or temperature distribution are met. This can be verified in advance with real experiments in which the curvature and / or temperature distribution are measured and compared with the corresponding quality criteria.

[0211] The annotated dataset for training the processing model comprises a set of input data and a set of output data, whereby in c) the input data additionally includes the parameters that define the soldering machine used and / or the control parameters used therein. The output data are the measured values ​​for the curvature (e.g., in the form of a linear or planar height profile, spline functions, or similar) and / or temperature distribution in the circuit board, which is measured directly on the circuit board to create the annotated data.

[0212] The processing model is trained using the annotated data sets so that after training, the processing model can be used to determine the parameters according to a) to c) in order to output the corresponding target parameters (= output data) for a specific segment of the circuit board for any circuit board.

[0213] By taking the entire circuit board into account, heat flows between the individual segments are also included in the result, and by taking each individual segment into account, the heat flows can be influenced precisely at a specific location, whereby the different layers of the circuit board are also reliably recorded and evaluated.

[0214] It may also be useful to further develop this processing model so that the solder pseudo-images of all segments and their position data are input simultaneously. In this case, it is not necessary, but possible, to omit the solder pseudo-image of the entire circuit board.

[0215] In an alternative embodiment, a quality criterion of good control parameters 6 of the macro and micro models is a temperature development over time at certain points of a test circuit board 4.

[0216] For the annotated data 30, circuit boards 4 are manufactured for learning with different control parameters 6.

[0217] During production, the temperature development of the printed circuit boards 4 is determined as a quality characteristic of the control parameters 5. For this purpose, the temperature development over time is measured at specific points on a test printed circuit board 4. This can be done, for example, using thermocouples or thermal imaging cameras.

[0218] If the control parameters 6 have not produced any or an acceptable temperature development, these control parameters 6 are assigned as target tensor 31 to the corresponding soldering pseudo-image 23.

[0219] The macro model outputs optimized control parameters 6.

[0220] While the first embodiment focuses in particular on control parameters 6 that regulate the heat supply to the solder joint 17, the second embodiment concerns control parameters 6 that regulate the heat distribution of the entire circuit board 4 in the macro model and the entire segments 22.

[0221] Thus, further embodiments are also possible that include features of both described embodiments. In such embodiments, the respective parameter modules 22 can include corresponding processing models 26 as described above, or new processing models 26 can each include combined features as previously listed. In the embodiments described above, optimized control parameters 6 are output. However, it is also conceivable in principle that other parameters are output instead or in addition.

[0222] These can, for example, be design parameters that describe the design of the printed circuit boards 4.

[0223] Under certain circumstances, it may be useful to adapt the design of the printed circuit boards 4 to achieve the quality characteristics described above (height of the soldering access, curvature and / or heat distribution).

[0224] This can be done, for example, by making the conductor tracks 10 thinner or wider.

[0225] It is also conceivable that the size or position of the solder joints 17 or the position and width of the conductor tracks 16 are adjusted.

[0226] Once the parameter module 24 has processed and determined the optimal design parameters for a specific circuit board design, these control parameters 6 can be reintegrated into the original CAD file 7 (dashed line in Fig. 1). They can then be easily retrieved during the manufacturing process.

[0227] By saving the design parameters in the CAD file 7, a feedback loop can be established to optimize the electronic manufacturing process. Further input here can be the evaluation of a PCB designer, who checks the control parameters 6 for plausibility.

[0228] The optimization process can also take into account constraints or specifications that limit the control parameters 6 for other reasons. For example, certain conductor tracks 16 may be responsible for grounding. The position and width can be designed more freely here than for conductor tracks 16 that carry signals and are easily susceptible to external influences. The position of the solder joints 17 can also depend on the geometries of the components.

[0229] Further additional or alternative outputs can be calculated values ​​of the quality parameters. The solder pseudo-images 23 are entered into the respective processing model, with the control parameters 6 being included as input parameters 35. Then, for example, the expected height of the solder penetration, the curvature, and / or temperature distribution are output.

[0230] In the following, a further method for controlling a soldering process is explained as a further exemplary embodiment, wherein identical elements as in the first exemplary embodiment are provided with the same reference numerals. The above explanations apply to identical elements unless otherwise stated below.

[0231] The method begins with step S9 (Figure 7). Steps S10 and S11 correspond to steps S2 and S3 of the method already described above.

[0232] In step S12, control parameters 6 are defined. These defined control parameters 6 are generic in nature and are based, for example, on an estimate by qualified personnel. Pseudo-images of the individual segments are created in step S13 as described above in step S4.

[0233] In the following step S14, a quality parameter is determined. For this purpose, one of the processing models 26 described above is used, which, however, has been programmed and trained to accept the soldering pseudo-images 23 and control parameters 6 as input and output the quality parameter as output.

[0234] In principle, however, it is also conceivable that instead of using a processing model, 26 a classic algorithm, also called hard-coded, is used to determine the quality parameter. This hard-coded programming is rule-based. The algorithm comprises certain rules and algorithms that deterministically always produce the same result for the same input.

[0235] The quality parameter is the one previously used above to determine the good control parameters 6 for the annotated data. This can be, for example, the height of the solder penetration or a curvature of the circuit board. It is also possible for the temperature distribution of the circuit board 4 to be output. In other words, the respective processing model 26 calculates the height of the solder penetration, the curvature of the circuit board, and / or the heat distribution for a segment and / or for the entire circuit board 4 from the solder pseudo-image 23 and the control parameters 6.

[0236] The quality parameter can also be thermal solderability, as explained above. The weighting can be such that, for example, it increases linearly with the distance to the solder contact point. For example, if there are four layers, the weightings of the layers can be assigned 0.1, 0.2, 0.3, and 0.4, starting with the layer closest to the solder via.

[0237] Once such a quality parameter has been determined by the processing model 26, a check is performed to determine whether this quality parameter lies within a desired range. A threshold can be defined in advance that the value of the quality parameter must reach. For example, a predetermined height of the solder penetration can be achieved, or the curvature of the circuit board 4 must not exceed a predetermined value.

[0238] Step S15 queries whether this threshold for the quality parameter has been reached.

[0239] If it was not reached, new control parameters are set in step S12, and the loop starts again. When setting the new control parameters 6, mathematical methods can be used to find the optimized control parameters. One possible method for this is the gradient method, with which the individual control parameters 6 each span one dimension, and a maximum is found using the gradient method, whereby the quality parameter is to be optimized here.

[0240] Finding the optimal control parameters 6 is an optimization in the sense of a multidimensional problem and is to be distinguished here from the optimization module of the optimization device 5.

[0241] If the threshold for the optimization parameter has been reached, step S16 follows, in which the circuit board is soldered with the determined, optimized control parameters 6.

[0242] The method ends with step S17.

[0243] According to a further development, the importance of the gradient can also be taken into account when optimizing the control parameters 6

[0244] According to an additional development, the quality of the calculation in the processing model 26 can be output by scattering the input parameters in order to determine robustness against the variance of the parameters.

[0245] The invention can be briefly summarized as follows: a) reading in CAD data of a printed circuit board to which electronic components are to be soldered to through-holes and / or pad fields, and using this CAD data as a printed circuit board model or printed circuit board twin, wherein the through-holes and / or pad fields each define a soldering point, b) dividing the printed circuit board model into a plurality of printed circuit board segments, c) determining the distribution of an electrically conductive material contained in the respective printed circuit board segments for the individual printed circuit board segments in a distribution grid, e) determining one or more parameters that describe the soldering process and / or the printed circuit board and / or the soldering point.

[0246] If a parameter describing the solder joint, such as the solder penetration, is determined in step e), then a specific soldering process is assumed, which is taken into account when determining this parameter. Based on one or more parameters determined in this way, the soldering process and / or the circuit board can be optimized, or it can be determined whether the soldering process meets one or more predetermined quality criteria. In this process, all aspects explained above and described with reference to the various exemplary embodiments and embodiments can be considered or integrated.

[0247] Reference symbol

[0248] 1 soldering system 19 segmentation module

[0249] 2 soldering device 25 20 segment file

[0250] 3 Lot 21 weighting module

[0251] 4 circuit boards 22 segments

[0252] 5 Optimization device 23 Soldering pseudo-image

[0253] 6 control parameters 24 parameter module

[0254] 7 CAD file 30 25 Parameter card

[0255] 8 Solder bath 26 Processing model

[0256] 9 Heating element 27 Input layer

[0257] 10 Pump 28 Output layer

[0258] 11 Nozzle 29 Model parameters

[0259] 12 conveyor system 35 30 annotated data

[0260] 13 Soldering device control unit 31 Target tensor

[0261] 14 sensors 32 target functions

[0262] 15 actuators 33 intermediate layers

[0263] 16 tracks 34 result tensor

[0264] 17 solder joints 40 35 input parameters

[0265] 18 vias

Claims

Claims 1. A method for controlling a soldering process, comprising the steps of a) reading in CAD data of a printed circuit board to which electronic components are to be soldered to through-holes and / or pad fields, and using this CAD data as a printed circuit board model, wherein the through-holes and / or pad fields each define a soldering point, b) dividing the printed circuit board model into several printed circuit board segments, c) determining the distribution of an electrically conductive material contained in the respective printed circuit board segments for the individual printed circuit board segments in a distribution grid, d) determining a minimum value and / or maximum value of at least one control parameter for a soldering process for soldering the soldering points, such that at least one predetermined quality parameter is met for certain soldering points, wherein the horizontal distribution and the vertical distribution of the respective segments are taken into account, e) carrying out a soldering process,wherein the at least one control parameter satisfies the determined minimum value and / or maximum value., 2. A method for automatically optimizing a printed circuit board to achieve defect-free solder joints, comprising the steps A) Reading in CAD data of a printed circuit board to which electronic components are to be soldered to through-holes and / or pad fields, and using this CAD data as a printed circuit board model, whereby the through-holes and / or pad fields each define a soldering point, B) Dividing the PCB model into several PCB segments, C) Determination of the distribution of an electrically conductive material contained in the respective circuit board segments for the individual circuit board segments according to a distribution grid, D) Determining at least one quality parameter for solder joints formed at specific soldering points by a soldering process with at least one predetermined control parameter, taking into account the horizontal distribution and the vertical distribution of the respective segments, E) Check whether the quality parameters for each solder joint meet a predetermined quality threshold, and if this is not the case, then either the circuit board is redesigned at the predetermined soldering point according to predetermined rules so that the quality parameter for the respective soldering point meets the quality threshold, or at least one control parameter of the soldering process is modified so that the quality parameter for the respective soldering point meets the quality threshold.

3. Method according to claim 2, characterized in that the thus optimized circuit board is manufactured according to the optimized CAD data.

4. Method according to claim 1 or 2, characterized in that the distribution grid comprises a top view and a side view, so that a horizontal distribution and a vertical distribution are obtained.

5. Method according to one of claims 1 to 4, characterized in that in step d) of claim 1 and / or in step D) of claim 2 all soldering points are taken into account.

6. Method according to one of claims 1 to 5, characterized in that the determination of a minimum value and / or maximum value of at least one control parameter for a soldering process according to claim 1 or the determination of at least one quality parameter for solder joints formed at specific soldering points by a soldering process with at least one predetermined control parameter according to claim 2 is carried out with a trained ML system.

7. Method according to one of claims 1 to 6, characterized in that a plurality of soldering points each have a through-hole for receiving a pin of a component and the at least one quality parameter is the height of a solder penetration in the respective through-hole for a pin of a component to be fastened thereto which is inserted into the through-hole.

8. Method according to one of claims 1 to 7, characterized in that that when determining a minimum value and / or maximum value of at least one control parameter for a soldering process according to claim 1 or when determining at least one quality parameter for each soldering point formed at each soldering point by a soldering process with at least one predetermined control parameter according to claim 2, thermal parameters of the component having the pin at the respective soldering point are taken into account, which parameters are either calculated or estimated based on the geometry of the respective component or have been empirically determined in advance on the respective component.

9. Method according to one of claims 1 to 8, characterized in that several predetermined quality parameters and / or several control parameters are taken into account.

10. Method according to one of claims 1 to 9, characterized in that the soldering process on which the method is based is carried out by means of a soldering wave, in particular with a locally limited soldering wave, and in particular the soldering process is a selective wave soldering process.

11. Method according to one of claims 1 to 10, characterized in that for each soldering point for which a minimum value and / or a maximum value of at least one control parameter for a soldering process according to claim 1 or at least one quality parameter for a soldering point formed at the soldering point by a soldering process with at least one predetermined control parameter is formed, a separate circuit board segment is determined, wherein the respective soldering point is preferably located in the center of the respective circuit board segment.

12. Method according to one of claims 1 to 11, characterized in that the predetermined quality parameter(s) are selected from the following group: - height of a soldering access, - no bridge between adjacent solder joints, - no open solder joint - no solder pin, - sufficient wetting - thermal solderability.

13. Method according to one of claims 1 to 12, characterized in that the control parameters comprise one or more of the following control parameters: - Wetting time - Solder bath temperature - Preheating parameters - Temperature of the circuit board.

14. Method according to one of claims 1 to 13, characterized in that when determining a minimum value and / or maximum value of at least one control parameter for a soldering process according to claim 1 or when determining at least one quality parameter for each soldering point formed at each soldering point by a soldering process with at least one predetermined control parameter according to claim 2, one or more of the following input parameters are taken into account: - Lot type - Type of non-conductive material of the circuit board - Type of conductive material of the circuit board.

15. Method according to one of claims 1 to 14, characterized in that the plurality of circuit board segments have not more than 500 soldering points and preferably not more than 150 soldering points and in particular not more than 100 soldering points and / or a maximum area of ​​30 cm 2 and preferably a maximum area of ​​15 cm 2and in particular a maximum area of ​​10 cm 2 has.

Images