Circuit board quality analysis method, device, equipment and medium based on artificial intelligence

Abstract

The circuit board quality analysis method and device, equipment and medium based on artificial intelligence provided in the present application relate to the field of artificial intelligence technology. In the present application, first, the circuit board welding image and welding equipment operation data of the target circuit board during the production process are obtained; secondly, semantic mining operations are performed on the circuit board welding image and the welding equipment operation data respectively, and the circuit board welding image features corresponding to the circuit board welding image and the welding equipment operation data features corresponding to the welding equipment operation data are output; then, the circuit board welding image features and the welding equipment operation data features are subjected to multi-level semantic aggregation operations, and multi-level circuit board welding features are output; finally, at least based on the multi-level circuit board welding features, the target circuit board quality analysis result corresponding to the target circuit board is analyzed. Based on the above content, the problem of relatively low reliability of circuit board quality analysis existing in the prior art can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a circuit board quality analysis method and device, equipment and medium based on artificial intelligence. Background Art

[0002] The formation process of a printed circuit board (PCB) is usually divided into several key stages: design, manufacturing and assembly. The following are brief steps of the entire circuit board formation process:

[0003] 1. Design stage;

[0004] 2. Manufacturing stage;

[0005] 3. Post-processing and cutting;

[0006] 4. Testing phase;

[0007] 5. Assembly stage;

[0008] Component placement: placing electronic components (such as resistors, capacitors, ICs, etc.) on the PCB through SMT (surface mount technology) or plug-in technology;

[0009] Reflow soldering: SMT components are soldered through a reflow soldering machine to melt the solder to form solder joints;

[0010] Wave soldering: Plug-in components are soldered using wave soldering technology;

[0011] 6. Packaging and shipment.

[0012] Each of the above-mentioned processing stages affects the quality of the circuit board, so it is necessary to analyze the process of the circuit board at each stage to determine its quality. However, in the prior art, when analyzing the welding quality, generally only the presence of a weld is checked after welding to determine the corresponding welding quality, which leads to the problem of low reliability of quality analysis. Summary of the invention

[0013] In view of this, the purpose of the present application is to provide a circuit board quality analysis method and apparatus, equipment and medium based on artificial intelligence to improve the problem of relatively low reliability of circuit board quality analysis in the prior art.

[0014] To achieve the above objectives, this application adopts the following technical solutions:

[0015] A circuit board quality analysis method based on artificial intelligence, comprising:

[0016] Acquire a circuit board welding image collected during the production process of a target circuit board, and acquire welding equipment operation data of a corresponding target welding equipment during the production process of the target circuit board, wherein the circuit board welding image is at least used to characterize image information of the target circuit board during the welding process of electronic components, and the welding equipment operation data is at least used to characterize operation parameters of the target welding equipment when welding electronic components, including at least welding method, temperature and duration;

[0017] Performing semantic mining operations on the circuit board welding image and the welding equipment operation data respectively, and outputting circuit board welding image features corresponding to the circuit board welding image and welding equipment operation data features corresponding to the welding equipment operation data;

[0018] Performing a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features, and outputting a multi-level circuit board welding feature corresponding to the target circuit board;

[0019] At least based on the multi-layer circuit board welding characteristics, a target circuit board quality analysis result corresponding to the target circuit board is analyzed, wherein the target circuit board quality analysis result is at least used to characterize that the target circuit board has quality in the welding dimension of electronic components.

[0020] In a preferred embodiment of the present application, in the above-mentioned circuit board quality analysis method based on artificial intelligence, the step of performing a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features to output the multi-level circuit board welding features corresponding to the target circuit board includes:

[0021] Splitting the circuit board welding image feature according to the color channel to form a first welding image channel feature, a second welding image channel feature and a third welding image channel feature corresponding to the circuit board welding image feature, wherein the first welding image channel feature corresponds to a red color channel, the second welding image channel feature corresponds to a green color channel, and the third welding image channel feature corresponds to a blue color channel;

[0022] Using the first weight, the second weight and the third weight in the image channel fusion unit included in the circuit board quality analysis network, the first welding image channel feature, the second welding image channel feature and the third welding image channel feature are weighted summed and calculated, and the weighted welding image channel feature corresponding to the circuit board welding image feature is output, wherein the circuit board quality analysis network is a neural network formed by training in advance, and the first weight, the second weight and the third weight are network parameters of the image channel fusion unit, which are formed during training;

[0023] Based on the first welding image channel feature, the second welding image channel feature, the third welding image channel feature and the weighted welding image channel feature, a multi-level semantic aggregation operation is performed on the welding equipment operation data feature to output the multi-level circuit board welding feature corresponding to the target circuit board.

[0024] In a preferred embodiment of the present application, in the above-mentioned circuit board quality analysis method based on artificial intelligence, the step of performing a multi-level semantic aggregation operation on the welding equipment operation data features based on the first welding image channel features, the second welding image channel features, the third welding image channel features and the weighted welding image channel features, and outputting the multi-level circuit board welding features corresponding to the target circuit board includes:

[0025] Using a first semantic aggregation unit included in a circuit board quality analysis network, a multi-level semantic aggregation operation is performed on the first welding image channel feature and the welding equipment operation data feature, and a multi-level first aggregation feature corresponding to the target circuit board is output;

[0026] Using the second semantic aggregation unit included in the circuit board quality analysis network, the second welding image channel feature and the welding equipment operation data feature are subjected to a multi-level semantic aggregation operation, and the multi-level second aggregation feature corresponding to the target circuit board is output;

[0027] Using the third semantic aggregation unit included in the circuit board quality analysis network, the third welding image channel feature and the welding equipment operation data feature are subjected to a multi-level semantic aggregation operation, and the multi-level third aggregation feature corresponding to the target circuit board is output;

[0028] Using the fourth semantic aggregation unit included in the circuit board quality analysis network, the weighted welding image channel features and the welding equipment operation data features are subjected to a multi-level semantic aggregation operation, and the multi-level fourth aggregation features corresponding to the target circuit board are output;

[0029] The multi-level first aggregation feature, the multi-level second aggregation feature, the multi-level third aggregation feature and the multi-level fourth aggregation feature are superimposed or averaged according to corresponding levels, and the multi-level circuit board welding feature corresponding to the target circuit board is output.

[0030] In a preferred choice of the present application, in the above-mentioned circuit board quality analysis method based on artificial intelligence, the first semantic aggregation unit included in the circuit board quality analysis network is used to perform a multi-level semantic aggregation operation on the first welding image channel feature and the welding equipment operation data feature, and the step of outputting the multi-level first aggregation feature corresponding to the target circuit board includes:

[0031] In the first level, using the first semantic aggregation unit included in the circuit board quality analysis network, based on the welding equipment operation data features, the semantic aggregation operation is performed on the first welding image channel features to output the first aggregation features of the first level;

[0032] In each other level except the first level, using the first semantic aggregation unit, based on the second aggregation feature of the previous level and / or the third aggregation feature of the previous level, a semantic aggregation operation is performed on the first aggregation feature of the previous level, and the first associated feature corresponding to the other level is output; and based on the first associated feature corresponding to the other level, a semantic aggregation operation is performed on the first welding image channel feature, and the first aggregation feature of the other level is output;

[0033] Based on the first aggregation features of each level, multi-level first aggregation features corresponding to the target circuit board are determined.

[0034] In a preferred selection of the present application, in the above-mentioned circuit board quality analysis method based on artificial intelligence, in each other level other than the first level, using the first semantic aggregation unit, based on the second aggregation feature of the previous level and / or the third aggregation feature of the previous level, the first aggregation feature of the previous level is subjected to a semantic aggregation operation, and the first associated feature corresponding to the other level is output, and the first welding image channel feature is subjected to a semantic aggregation operation based on the first associated feature corresponding to the other level, and the step of outputting the first aggregation feature of the other level includes:

[0035] In each other level except the first level, using the first semantic aggregation unit, determine the average feature of the second aggregation feature of the previous level and the third aggregation feature of the previous level, and determine the first correlation parameter between the average feature and the transposed feature of the first aggregation feature of the previous level, and based on the first correlation parameter, update the first aggregation feature of the previous level, and output the first correlation feature corresponding to the other level;

[0036] Determine a second correlation parameter between the first associated feature corresponding to the other level and the transposed feature of the first welding image channel feature, and based on the second correlation parameter, update the first welding image channel feature to output the first aggregate feature corresponding to the other level.

[0037] In a preferred embodiment of the present application, in the above-mentioned circuit board quality analysis method based on artificial intelligence, the steps of performing semantic mining operations on the circuit board welding image and the welding equipment operation data, respectively, and outputting the circuit board welding image features corresponding to the circuit board welding image and the welding equipment operation data features corresponding to the welding equipment operation data, include:

[0038] Utilizing the semantic mining unit included in the circuit board quality analysis network, semantic mining operations are performed on the circuit board welding image and the welding equipment operation data respectively, and circuit board image mining features corresponding to the circuit board welding image and equipment operation mining features corresponding to the welding equipment operation data are output, wherein the circuit board quality analysis network belongs to a trained neural network, and the semantic mining unit includes two semantic mining branches, and the two semantic mining branches are used to perform semantic mining on the circuit board welding image and welding equipment operation data in different forms of expression respectively;

[0039] Using the first adaptation parameter in the semantic adaptation unit included in the circuit board quality analysis network, an adaptation operation in a feature space is performed on the circuit board image mining feature to form a circuit board welding image feature corresponding to the circuit board welding image;

[0040] The second adaptation parameter in the semantic adaptation unit included in the circuit board quality analysis network is used to perform an adaptation operation in feature space on the equipment operation mining feature to form a welding equipment operation data feature corresponding to the circuit board welding image.

[0041] In a preferred embodiment of the present application, in the above-mentioned circuit board quality analysis method based on artificial intelligence, the step of analyzing the target circuit board quality analysis result corresponding to the target circuit board based at least on the welding characteristics of the multi-layer circuit board includes:

[0042] For each level of the multi-level circuit board soldering features, the fully connected unit included in the circuit board quality analysis network is used to process the circuit board soldering features of the level, and the fully connected features corresponding to the circuit board soldering features of the level are output;

[0043] Utilizing the linear regression unit included in the circuit board quality analysis network, the fully connected features corresponding to at least two levels of circuit board welding features are output and processed respectively to obtain at least two corresponding circuit board quality analysis parameters, and, based on the at least two circuit board quality analysis parameters, the target circuit board quality analysis result corresponding to the target circuit board is determined.

[0044] The present application also provides a circuit board quality analysis device based on artificial intelligence, comprising:

[0045] A welding-related data acquisition module, used to acquire a circuit board welding image collected during the production process of a target circuit board, and to acquire welding equipment operation data of a corresponding target welding equipment during the production process of the target circuit board, wherein the circuit board welding image is at least used to characterize image information of the target circuit board during the welding process of electronic components, and the welding equipment operation data is at least used to characterize operation parameters of the target welding equipment when welding electronic components, including at least welding method, temperature and duration;

[0046] A data semantic mining module, used to perform semantic mining operations on the circuit board welding image and the welding equipment operation data respectively, and output circuit board welding image features corresponding to the circuit board welding image and welding equipment operation data features corresponding to the welding equipment operation data;

[0047] A feature semantic aggregation module, used for performing a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features, and outputting a multi-level circuit board welding feature corresponding to the target circuit board;

[0048] A circuit board quality analysis module is used to analyze a target circuit board quality analysis result corresponding to the target circuit board based at least on the welding characteristics of the multi-layer circuit board, wherein the target circuit board quality analysis result is at least used to characterize that the target circuit board has quality in the welding dimension of electronic components.

[0049] Based on the above, the present application also provides an electronic device, including:

[0050] Memory for storing computer programs;

[0051] A processor connected to the memory is used to execute the computer program stored in the memory to implement the above-mentioned circuit board quality analysis method based on artificial intelligence.

[0052] On the basis of the above, the present application also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is run, each step of the above-mentioned artificial intelligence-based circuit board quality analysis method is executed.

[0053] The circuit board quality analysis method and device, equipment and medium based on artificial intelligence provided by the present application, first, obtain the circuit board welding image and welding equipment operation data of the target circuit board during the production process; secondly, perform semantic mining operations on the circuit board welding image and welding equipment operation data respectively, and output the circuit board welding image features corresponding to the circuit board welding image and the welding equipment operation data features corresponding to the welding equipment operation data; then, perform multi-level semantic aggregation operations on the circuit board welding image features and the welding equipment operation data features, and output multi-level circuit board welding features; finally, at least based on the multi-level circuit board welding features, analyze the target circuit board quality analysis result corresponding to the target circuit board. Based on the above content, since the analysis basis corresponding to the target circuit board quality analysis result is the multi-level circuit board welding feature, and the multi-level circuit board welding feature is obtained by performing multi-level semantic aggregation operations, the multi-level circuit board welding feature includes richer semantic information, including both deep information and shallow information, so that the semantic representation ability is better, and therefore, the reliability of the target circuit board quality analysis result is relatively better. In addition, since the multi-level circuit board welding features not only carry the semantic information of the image information of the target circuit board during the electronic component welding process, but also carry the semantic information of the operating parameters of the target welding equipment when welding electronic components, the semantic representation capability of the multi-level circuit board welding features can be further improved, thereby further improving the reliability of the target circuit board quality analysis results, and further improving the problem of relatively low reliability of circuit board quality analysis in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings.

[0055] Figure 1 A structural block diagram of an electronic device provided in an embodiment of the present application.

[0056] Figure 2 A schematic diagram of the flow chart of the circuit board quality analysis method based on artificial intelligence provided in an embodiment of the present application.

[0057] Figure 3 A schematic diagram of a multi-level semantic aggregation operation provided in an embodiment of the present application.

[0058] Figure 4 A block diagram of an artificial intelligence-based circuit board quality analysis device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0060] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0061] like Figure 1 As shown, an embodiment of the present application provides an electronic device, wherein the electronic device may include a memory, a processor, and a circuit board quality analysis device based on artificial intelligence.

[0062] In detail, the memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, the memory and the processor can be electrically connected through one or more communication buses or signal lines. The circuit board quality analysis device based on artificial intelligence includes at least one software function module stored in the memory in the form of software or firmware. The processor is used to execute an executable computer program stored in the memory, for example, the software function module and computer program included in the circuit board quality analysis device based on artificial intelligence, so as to implement the circuit board quality analysis method based on artificial intelligence provided in the embodiment of the present application.

[0063] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc.

[0064] Furthermore, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0065] Understandably, Figure 1 The structure shown is for illustration only, and the electronic device may also include Figure 1 More or fewer components as shown, or with Figure 1 The different configurations shown, for example, may also include a communication unit for exchanging information with other devices (such as an image acquisition device).

[0066] Combination Figure 2 The embodiment of the present application also provides a circuit board quality analysis method based on artificial intelligence that can be applied to the above electronic device. Among them, the method steps defined in the process related to the circuit board quality analysis method based on artificial intelligence can be implemented by the electronic device.

[0067] The following will Figure 2 The specific process shown is explained in detail.

[0068] Step S110, obtaining a circuit board welding image collected during the production process of the target circuit board, and obtaining welding equipment operation data of a corresponding target welding equipment during the production process of the target circuit board.

[0069] In an embodiment of the present application, the electronic device can obtain a circuit board welding image collected during the production process of the target circuit board, and obtain the welding equipment operation data of the corresponding target welding equipment during the production process of the target circuit board. Among them, the circuit board welding image is at least used to characterize the image information of the target circuit board during the electronic component welding process. For example, the welding process can be monitored by a corresponding image acquisition device to obtain it; in addition, the circuit board welding image can be multiple frames, which can be configured according to actual needs and is not specifically limited here. The welding equipment operation data is at least used to characterize the operating parameters of the target welding equipment when welding electronic components, including at least the welding method, temperature and duration. For example:

[0070] Reflow Soldering: Preheating temperature: gradually heating components and PCB, controlled between 150-180°C; Reflow temperature: the temperature at which the solder melts, usually between 220-250°C; Cooling temperature: the temperature after the solder solidifies, usually room temperature or slightly lower; Heating time: the heating time at different stages, such as the duration of preheating, reflow and cooling stages;

[0071] Atmosphere control: Nitrogen or air: Using nitrogen during reflow can reduce soldering oxidation and improve soldering quality;

[0072] Conveyor speed: The speed at which the PCB passes through the reflow oven, which affects the uniformity of heating;

[0073] Hot air / infrared heating: The heating method of the reflow oven (such as infrared heating or hot air heating) affects the temperature distribution and welding effect;

[0074] Wave Soldering: Solder pool temperature: The solder temperature in the wave soldering pool is usually controlled between 250-270°C; Preheating temperature: The preheating temperature of the PCB board before entering the wave soldering, usually controlled between 100-150°C; Soldering time: The time when the PCB board contacts the solder pool, usually 2-5 seconds. Too long or too short may affect the welding quality; Solder flow: The solder flow control in the wave soldering furnace affects the contact amount of welding; Waveform control: The wave peak height and stability in the solder pool directly affect the welding quality; Atmosphere control: Use nitrogen or air atmosphere to control oxidation to ensure welding quality; Nozzle adjustment: The position and angle of the nozzle can adjust the distribution of solder.

[0075] Step S120, performing semantic mining operations on the circuit board welding image and the welding equipment operation data respectively, and outputting circuit board welding image features corresponding to the circuit board welding image and welding equipment operation data features corresponding to the welding equipment operation data.

[0076] In an embodiment of the present application, after acquiring the circuit board welding image and the welding equipment operation data, the electronic device can perform semantic mining operations on the circuit board welding image and the welding equipment operation data, respectively, and output circuit board welding image features corresponding to the circuit board welding image and welding equipment operation data features corresponding to the welding equipment operation data. In addition, when the circuit board welding image is a multi-frame, the circuit board welding image features corresponding to each frame of the circuit board welding image can be superimposed or averaged to obtain the final circuit board welding image features.

[0077] Step S130, performing a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features, and outputting a multi-level circuit board welding feature corresponding to the target circuit board.

[0078] In an embodiment of the present application, after obtaining the circuit board welding image features and the welding equipment operation data features, the electronic device can perform a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features, and output a multi-level circuit board welding feature corresponding to the target circuit board. In other words, the circuit board welding image features and the welding equipment operation data features can be subjected to a variety of semantic aggregations of different depths, so that the output multi-level circuit board welding features have semantic information of multiple depths, thereby ensuring their semantic representation capabilities.

[0079] Step S140, analyzing a target circuit board quality analysis result corresponding to the target circuit board based at least on the welding characteristics of the multi-layer circuit board.

[0080] In the embodiment of the present application, after obtaining the multi-layer circuit board soldering characteristics, the electronic device can analyze the target circuit board quality analysis result corresponding to the target circuit board based at least on the multi-layer circuit board soldering characteristics, that is, the quality analysis can be performed based only on the multi-layer circuit board soldering characteristics, or on the basis of the multi-layer circuit board soldering characteristics, other related characteristics can be further referred to for analysis. Among them, the target circuit board quality analysis result is at least used to characterize that the target circuit board has quality in the soldering dimension of electronic components.

[0081] Based on the above content, since the analysis basis corresponding to the target circuit board quality analysis result is the multi-level circuit board welding feature, and the multi-level circuit board welding feature is obtained by performing multi-level semantic aggregation operations, the multi-level circuit board welding feature includes richer semantic information, including both deep information and shallow information, so that the semantic representation ability is better, and therefore, the reliability of the target circuit board quality analysis result is relatively better. In addition, since the multi-level circuit board welding feature carries not only the semantic information of the image information of the target circuit board during the electronic component welding process, but also the semantic information of the operating parameters of the target welding equipment when welding electronic components, the semantic representation ability of the multi-level circuit board welding feature can be further improved, thereby further improving the reliability of the target circuit board quality analysis result, thereby improving the problem of relatively low reliability of circuit board quality analysis existing in the prior art.

[0082] It should be noted that for step S120, the specific manner of performing semantic mining operations on the circuit board welding image and the welding equipment operation data is not limited and can be selected according to actual needs. For example, in an alternative implementation, in order to ensure the reliability of the output circuit board welding image features and welding equipment operation data features, so as to facilitate the effective implementation of subsequent semantic aggregation, the above step S120 may further include the following contents:

[0083] First, the semantic mining unit included in the circuit board quality analysis network can be used to perform semantic mining operations on the circuit board welding image and the welding equipment operation data, respectively, and output the circuit board image mining features corresponding to the circuit board welding image and the equipment operation mining features corresponding to the welding equipment operation data, wherein the circuit board quality analysis network belongs to a trained neural network, and the semantic mining unit includes two semantic mining branches, which are used to perform semantic mining on the circuit board welding image and the welding equipment operation data with different expressions, respectively; exemplarily, the first semantic mining branch may include a convolutional neural network, which is used to perform convolution processing on the circuit board welding image and output the corresponding circuit board image mining features; the second semantic mining branch may include word embedding processing, which is used to embed the welding equipment operation data to obtain a corresponding word embedding vector as the corresponding equipment operation mining feature, and the welding equipment operation data is used as "the preheating temperature of reflow soldering is 150 degrees Celsius to 180 degrees Celsius" for corresponding description, and the word embedding vector may include: "reflow soldering": [0.32, -0.25, 0.1, 0.45, -0.03, ];“of”: [0.01, 0.02, 0.01, -0.01, 0.03, 0.02, -0.02,0.00, 0.01, -0.03, 0.01, 0.00, 0.03, 0.01, 0.01, -0.02, 0.02, -0.01, 0.01,...];“warm-up”: [0.41, -0.13, 0.18, 0.33, -0.05, , 0.19, 0.16, -0.08, 0.29, -0.12,0.34, 0.01, 0.27, -0.07, 0.2, 0.22, -0.06, 0.13, 0.23, ...]; "temperature": [0.25,0.17, 0.36, -0.13, 0.27, -0.1, 0.19, 0.03, 0.22, 0.16, 0.1, 0.33, -0.02,0.12, 0.29, -0.18, 0.04, 0.08, -0.09, ...]; "for": [0.05, 0.04, -0.01, 0.02, -0.02, 0.02, 0.01, 0.00, 0.01, 0.03, -0.01, 0.01, 0.02, 0.02, 0.00, 0.01,0.02, 0.01, -0.01, ...]; "150": [0.18, -0.12, 0.07, 0.25, -0.09, 0.3, 0.17, -0.05, 0.22, -0.11, 0.26, 0.15, -0.07, 0.13, 0.2, 0.09, 0.16, 0.28, -0.14,...]; "Celsius": [0.23, 0.11, 0.16, 0.29, -0.06, 0.27, 0.24, 0.05, 0.33, -0.03,0.2, 0.19, 0.14, 0.1, 0.12, 0.03, 0.01, 0.18, 0.05, ...]; "to": [0.12, -0.06,0.14, 0.19, 0.15, 0.09, 0.11, 0.08, -0.01, -0.03, 0.17, 0.1, 0.22, -0.04,0.18, 0.02, 0.16, 0.12, -0.05, ...]; "180": [0.14, 0.09, -0.11, 0.13, 0.18,0.23, 0.26, -0.01, 0.27, 0.16, 0.24, 0.14, 0.16, -0.03, 0.28, 0.11, -0.02,0.03, 0.12, ...]; "Celsius": [0.23, 0.11, 0.16, 0.29, -0.06, 0.27, 0.24, 0.05,0.33, -0.03, 0.2, 0.19, 0.14, 0.1, 0.12, 0.03, 0.01, 0.18, 0.05, ...];.

[0084] Secondly, the first adaptation parameter in the semantic adaptation unit included in the circuit board quality analysis network can be used to perform an adaptation operation in the feature space on the circuit board image mining feature to form a circuit board welding image feature corresponding to the circuit board welding image; for example, the first adaptation parameter can include a first weight matrix and a first bias matrix, so that the first weight matrix and the circuit board image mining feature can be multiplied to obtain a first weight feature, and the first weight feature and the first bias matrix can be added to obtain a circuit board welding image feature;

[0085] Then, the second adaptation parameter in the semantic adaptation unit included in the circuit board quality analysis network can be used to perform feature space adaptation operations on the equipment operation mining features to form welding equipment operation data features corresponding to the circuit board welding image; for example, the second adaptation parameter can include a second weight matrix and a second bias matrix, so that the second weight matrix and the equipment operation mining features can be multiplied to obtain a second weight feature, and the second weight feature and the second bias matrix can be added to obtain the welding equipment operation data features.

[0086] It should be noted that for step S130, the specific method of performing multi-level semantic aggregation operations on the circuit board welding image features and the welding equipment operation data features is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to ensure that the output multi-level circuit board welding features have better semantic representation capabilities, the above-mentioned step S130 can further include step S131, step S132 and step S133, the specific contents of which are described as follows.

[0087] Step S131 , splitting the circuit board welding image feature according to color channels to form a first welding image channel feature, a second welding image channel feature and a third welding image channel feature corresponding to the circuit board welding image feature.

[0088] In the embodiment of the present application, the circuit board welding image feature can be split according to the color channel (it should be noted that in the aforementioned steps, when performing the semantic mining operation, each color channel can also be performed separately) to form the first welding image channel feature, the second welding image channel feature and the third welding image channel feature corresponding to the circuit board welding image feature. In the embodiment, the first welding image channel feature corresponds to the red color channel, the second welding image channel feature corresponds to the green color channel, and the third welding image channel feature corresponds to the blue color channel. In this way, the granularity of the feature can be made smaller to ensure the processing accuracy.

[0089] Step S132, using the first weight, the second weight and the third weight in the image channel fusion unit included in the circuit board quality analysis network, perform weighted sum calculation on the first welding image channel feature, the second welding image channel feature and the third welding image channel feature, and output the weighted welding image channel feature corresponding to the circuit board welding image feature.

[0090] In the embodiment of the present application, after obtaining the features of the three color channels, the first weight, the second weight and the third weight (formed during training) in the image channel fusion unit included in the circuit board quality analysis network can be used to perform weighted sum calculation on the first welding image channel feature, the second welding image channel feature and the third welding image channel feature, and output the weighted welding image channel feature corresponding to the circuit board welding image feature. Among them, the circuit board quality analysis network belongs to a neural network formed by pre-training, and the first weight, the second weight and the third weight belong to the network parameters of the image channel fusion unit, which are formed during training. It should be noted that the semantic information represented by the pixel features of different color channels is different, and the corresponding importance is also different. Therefore, by assigning corresponding weights, different degrees of attention can be paid to features of different importance, and the reliability of its semantics can be guaranteed. The weights are formed during training, so they are actually determined based on training data, and therefore have high reliability.

[0091] Step S133, based on the first welding image channel feature, the second welding image channel feature, the third welding image channel feature and the weighted welding image channel feature, a multi-level semantic aggregation operation is performed on the welding equipment operation data feature to output the multi-level circuit board welding feature corresponding to the target circuit board.

[0092] In the embodiment of the present application, after obtaining the four channel features, the multi-level semantic aggregation operation can be performed on the welding equipment operation data features based on the first welding image channel feature, the second welding image channel feature, the third welding image channel feature and the weighted welding image channel feature to output the multi-level circuit board welding features corresponding to the target circuit board. In other words, the multi-level circuit board welding features not only have multiple semantic information with different depths, but also have semantic information of multiple different channels, so that the richness of the semantic information is further improved.

[0093] It can be understood that in the above-mentioned step S133, the specific method of performing multi-level semantic aggregation operations on the features of various image channels is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to make the multi-level semantic aggregation operation more accurate and ensure that the semantic representation of the multi-level circuit board welding features is more accurate, the above-mentioned step S133 can further include step S133a, step S133b, step S133c, step S133d and step S133e, and the specific content of each step is described as follows.

[0094] Step S133a, using the first semantic aggregation unit included in the circuit board quality analysis network, performs a multi-level semantic aggregation operation on the first welding image channel feature and the welding equipment operation data feature, and outputs a multi-level first aggregation feature corresponding to the target circuit board.

[0095] In the embodiment of the present application, the first semantic aggregation unit included in the circuit board quality analysis network can be used to perform a multi-level semantic aggregation operation on the first welding image channel feature and the welding equipment operation data feature, and output the multi-level first aggregation feature corresponding to the target circuit board. In other words, the semantic information in the welding equipment operation data feature can be fused into the first welding image channel feature, so that the semantics of the output multi-level first aggregation feature is richer.

[0096] Step S133b, using the second semantic aggregation unit included in the circuit board quality analysis network, performs a multi-level semantic aggregation operation on the second welding image channel features and the welding equipment operation data features, and outputs the multi-level second aggregation features corresponding to the target circuit board.

[0097] In the embodiment of the present application, the second semantic aggregation unit included in the circuit board quality analysis network can be used to perform multi-level semantic aggregation operations on the second welding image channel features and the welding equipment operation data features, and output the multi-level second aggregation features corresponding to the target circuit board. In other words, the semantic information in the welding equipment operation data features can be fused into the second welding image channel features, so that the semantics of the output multi-level second aggregation features are richer.

[0098] Step S133c, using the third semantic aggregation unit included in the circuit board quality analysis network, the third welding image channel feature and the welding equipment operation data feature are subjected to multi-level semantic aggregation operations, and the multi-level third aggregation feature corresponding to the target circuit board is output.

[0099] In the embodiment of the present application, the third semantic aggregation unit included in the circuit board quality analysis network can be used to perform multi-level semantic aggregation operations on the third welding image channel feature and the welding equipment operation data feature, and output the multi-level third aggregation feature corresponding to the target circuit board. In other words, the semantic information in the welding equipment operation data feature can be fused into the third welding image channel feature, so that the semantics of the output multi-level third aggregation feature is richer.

[0100] Step S133d, using the fourth semantic aggregation unit included in the circuit board quality analysis network, the weighted welding image channel features and the welding equipment operation data features are subjected to multi-level semantic aggregation operations, and the multi-level fourth aggregation features corresponding to the target circuit board are output.

[0101] In the embodiment of the present application, the fourth semantic aggregation unit included in the circuit board quality analysis network can be used to perform a multi-level semantic aggregation operation on the fourth welding image channel feature and the welding equipment operation data feature, and output the multi-level fourth aggregated feature corresponding to the target circuit board. In other words, the semantic information in the welding equipment operation data feature can be fused into the fourth welding image channel feature, so that the semantics of the output multi-level fourth aggregated feature is richer.

[0102] Step S133e, superimposing or averaging the multi-level first aggregation feature, the multi-level second aggregation feature, the multi-level third aggregation feature and the multi-level fourth aggregation feature according to corresponding levels, and outputting the multi-level circuit board welding feature corresponding to the target circuit board.

[0103] In an embodiment of the present application, the multi-level first aggregation feature, the multi-level second aggregation feature, the multi-level third aggregation feature and the multi-level fourth aggregation feature can be superimposed or averaged according to the corresponding levels to output the multi-level circuit board welding feature corresponding to the target circuit board. For example, the first aggregation feature of the first level, the second aggregation feature of the first level, the third aggregation feature of the first level, and the fourth aggregation feature of the first level can be superimposed or averaged to obtain the circuit board welding feature of the first level. For another example, the first aggregation feature of the second level, the second aggregation feature of the second level, the third aggregation feature of the second level, and the fourth aggregation feature of the second level can be superimposed or averaged to obtain the circuit board welding feature of the second level. For another example, the first aggregation feature of the third level, the second aggregation feature of the third level, the third aggregation feature of the third level, and the fourth aggregation feature of the third level can be superimposed or averaged to obtain the circuit board welding feature of the third level.

[0104] It can be understood that in the above-mentioned steps S133a, S133b, S133c and S133d, the specific manner of performing the multi-level semantic aggregation operation can be the same. Taking step S133a as an example, in an alternative implementation, in order to make the semantic representation of the features output by the multi-level semantic aggregation operation more reliable, the above-mentioned step S133a can further include step a1, step a2 and step a3. The specific contents of each step are as follows (combined with Figure 3 ).

[0105] Step a1, in the first level, using the first semantic aggregation unit included in the circuit board quality analysis network, based on the welding equipment operation data characteristics, performs semantic aggregation operations on the first welding image channel characteristics, and outputs the first aggregation features of the first level.

[0106] In the embodiment of the present application, at the first level, the first semantic aggregation unit included in the circuit board quality analysis network can be used to perform semantic aggregation operations on the first welding image channel features based on the welding equipment operation data features, and output the first aggregate features of the first level. In addition, based on the same processing method, the second aggregate features of the first level, the third aggregate features of the first level, and the fourth aggregate features of the first level can also be output.

[0107] Step a2, in each other level except the first level, using the first semantic aggregation unit, based on the second aggregation features of the previous level and / or the third aggregation features of the previous level, perform a semantic aggregation operation on the first aggregation features of the previous level, and output the first associated features corresponding to the other levels; and, based on the first associated features corresponding to the other levels, perform a semantic aggregation operation on the first welding image channel features, and output the first aggregation features of the other levels.

[0108] In an embodiment of the present application, after obtaining the first aggregated features of the first level, in each other level other than the first level, the first semantic aggregation unit can be used to perform a semantic aggregation operation on the first aggregated features of the previous level based on the second aggregated features of the previous level and / or the third aggregated features of the previous level, and output the first associated features corresponding to the other level, and, based on the first associated features corresponding to the other level, perform a semantic aggregation operation on the first welding image channel features, and output the first aggregated features of the other level. In other words, a semantic aggregation operation is first performed on the features of the current channel based on the features of other channels, and then a semantic aggregation operation is performed on the first welding image channel features. In this way, the semantic information of the welding equipment operation data features can be fused at different depths. In addition, based on the same processing method, the second aggregated features of other levels, the third aggregated features of other levels, and the fourth aggregated features of other levels can also be output.

[0109] Step a3: determining the multi-level first aggregation features corresponding to the target circuit board based on the first aggregation features of each level.

[0110] In an embodiment of the present application, after obtaining the first aggregation feature of each level, the multi-level first aggregation feature corresponding to the target circuit board can be determined based on the first aggregation feature of each level. Exemplarily, the first aggregation feature of each level can be directly determined as the multi-level first aggregation feature. Alternatively, based on the need to reduce the difficulty of data processing, the first aggregation features of some levels can also be screened out from the first aggregation features of each level, such as screening out the first aggregation features of one level every preset number of levels. In addition, based on the same processing method, multi-level second aggregation features, multi-level third aggregation features, and multi-level fourth aggregation features can also be output.

[0111] It is understandable that in the above step a2, the specific manner of performing the semantic aggregation operation is not limited and can be selected according to actual needs. For example, in an alternative implementation, in order to ensure that the information with associations in different features can get more attention, that is, the important information can be fully mined, the above step a2 can further include the following contents:

[0112] First, in each other level except the first level, the first semantic aggregation unit can be used to determine the mean feature of the second aggregation feature of the previous level and the third aggregation feature of the previous level, and determine the first correlation parameter between the mean feature and the transposed feature of the first aggregation feature of the previous level, and based on the first correlation parameter, update the first aggregation feature of the previous level, and output the first correlation feature corresponding to the other level; for example, currently at the second level, the mean feature of the second aggregation feature of the first level and the third aggregation feature of the first level can be determined first, and then the first correlation parameter between the mean feature and the transposed feature of the first aggregation feature of the first level can be determined, such as multiplying As a result, the first correlation parameter can be multiplied with the first aggregate feature of the first level to obtain the first correlation feature of the second level; based on the same processing method, the second correlation feature of the second level and the third correlation feature of the second level can also be obtained, that is, for the second aggregate feature, corresponding processing is performed based on the average feature of the first aggregate feature and the third aggregate feature, and for the third aggregate feature, corresponding processing is performed based on the average feature of the first aggregate feature and the second aggregate feature; in addition, for the fourth aggregate feature, the first correlation parameter between the fourth aggregate feature and the transposed feature of the fourth aggregate feature can be determined, and the fourth aggregate feature can be updated based on the first correlation parameter to obtain the corresponding fourth correlation feature;

[0113] Secondly, the second correlation parameter between the first associated features corresponding to other levels and the transposed features of the first welding image channel features can be determined, and based on the second correlation parameter, the first welding image channel features can be updated to output the first aggregate features corresponding to the other levels; for example, the first associated features of the second level can be multiplied by the transposed features of the first welding image channel features to obtain the second correlation parameter, and then the second correlation parameter can be multiplied by the first welding image channel features to obtain the first aggregate features corresponding to the second level; based on the same processing method, the second aggregate features of the second level and the third aggregate features of the second level can also be obtained.

[0114] It should be noted that the specific method of analyzing the target circuit board quality analysis result corresponding to the target circuit board is not limited and can be selected according to actual needs. For example, in an alternative implementation, in order to ensure the reliability of the target circuit board quality analysis result, the above step S140 can further include the following contents:

[0115] First, for each level of the multi-level circuit board soldering features, the fully connected unit included in the circuit board quality analysis network can be used to process the circuit board soldering features of the level, and output the fully connected features corresponding to the circuit board soldering features of the level;

[0116] Secondly, the linear regression unit (which may include functions such as Sigmoid) included in the circuit board quality analysis network can be used to output the fully connected features corresponding to at least two levels of circuit board welding features respectively, so as to obtain at least two corresponding circuit board quality analysis parameters, and, based on the at least two circuit board quality analysis parameters, determine the target circuit board quality analysis result corresponding to the target circuit board (such as calculating the mean of at least two circuit board quality analysis parameters or selecting a smaller value therebetween as the target circuit board quality analysis result), such as using a 0-1 parameter to characterize the quality, the larger the value, the higher the quality.

[0117] Combination Figure 4 The embodiment of the present application also provides an artificial intelligence-based circuit board quality analysis device that can be applied to the above electronic device. The artificial intelligence-based circuit board quality analysis device can include a welding-related data acquisition module, a data semantic mining module, a feature semantic aggregation module, and a circuit board quality analysis module.

[0118] The welding-related data acquisition module is used to acquire the circuit board welding image collected during the production process of the target circuit board, and acquire the welding equipment operation data of the corresponding target welding equipment during the production process of the target circuit board, wherein the circuit board welding image is at least used to characterize the image information of the target circuit board during the electronic component welding process, and the welding equipment operation data is at least used to characterize the operating parameters of the target welding equipment when welding electronic components, including at least the welding method, temperature and duration. In the embodiment of the present application, the welding-related data acquisition module can be used to perform Figure 2 As shown in step S110, the relevant contents of the welding related data acquisition module can refer to the above description of step S110.

[0119] The data semantic mining module is used to perform semantic mining operations on the circuit board welding image and the welding equipment operation data respectively, and output the circuit board welding image features corresponding to the circuit board welding image and the welding equipment operation data features corresponding to the welding equipment operation data. In the embodiment of the present application, the data semantic mining module can be used to perform Figure 2 As shown in step S120, the relevant contents of the data semantic mining module can refer to the above description of step S120.

[0120] The feature semantic aggregation module is used to perform a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features, and output the multi-level circuit board welding features corresponding to the target circuit board. Figure 2 As shown in step S130, the relevant contents of the feature semantic aggregation module can refer to the above description of step S130.

[0121] The circuit board quality analysis module is used to analyze the target circuit board quality analysis result corresponding to the target circuit board based on at least the welding characteristics of the multi-layer circuit board, wherein the target circuit board quality analysis result is at least used to characterize that the target circuit board has quality in the welding dimension of electronic components. In the embodiment of the present application, the circuit board quality analysis module can be used to perform Figure 2 As shown in step S140, the relevant contents of the circuit board quality analysis module can refer to the above description of step S140.

[0122] In an embodiment of the present application, corresponding to the above-mentioned artificial intelligence-based circuit board quality analysis method applied to the electronic device, a computer-readable storage medium is also provided, in which a computer program is stored, and when the computer program is run, each step of the artificial intelligence-based circuit board quality analysis method is executed.

[0123] Among them, the steps executed when the aforementioned computer program is running will not be described one by one here, and reference can be made to the explanation of the aforementioned circuit board quality analysis method based on artificial intelligence in the previous article.

[0124] In summary, the circuit board quality analysis method and device, equipment and medium based on artificial intelligence provided by the present application, first, obtain the circuit board welding image and welding equipment operation data of the target circuit board during the production process; secondly, perform semantic mining operations on the circuit board welding image and welding equipment operation data respectively, and output the circuit board welding image features corresponding to the circuit board welding image and the welding equipment operation data features corresponding to the welding equipment operation data; then, perform multi-level semantic aggregation operations on the circuit board welding image features and the welding equipment operation data features, and output multi-level circuit board welding features; finally, at least based on the multi-level circuit board welding features, analyze the target circuit board quality analysis result corresponding to the target circuit board. Based on the above content, since the analysis basis corresponding to the target circuit board quality analysis result is the multi-level circuit board welding feature, and the multi-level circuit board welding feature is obtained by performing multi-level semantic aggregation operations, the multi-level circuit board welding feature includes richer semantic information, including both deep information and shallow information, so that the semantic representation ability is better, and therefore, the reliability of the target circuit board quality analysis result is relatively better. In addition, since the multi-level circuit board welding features not only carry the semantic information of the image information of the target circuit board during the electronic component welding process, but also carry the semantic information of the operating parameters of the target welding equipment when welding electronic components, the semantic representation capability of the multi-level circuit board welding features can be further improved, thereby further improving the reliability of the target circuit board quality analysis results, and further improving the problem of relatively low reliability of circuit board quality analysis in the prior art.

[0125] In several embodiments provided in the embodiments of the present application, it should be understood that the disclosed device and method can also be implemented in other ways. The device and method embodiments described above are merely schematic, for example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the device, method and computer program product according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0126] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0127] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, electronic device, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code. It should be noted that in this article, the term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0128] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A circuit board quality analysis method based on artificial intelligence, characterized in that: include: Acquire a circuit board welding image collected during the production process of a target circuit board, and acquire welding equipment operation data of a corresponding target welding equipment during the production process of the target circuit board, wherein the circuit board welding image is at least used to characterize image information of the target circuit board during the welding process of electronic components, and the welding equipment operation data is at least used to characterize operation parameters of the target welding equipment when welding electronic components, including at least welding method, temperature and duration; Performing semantic mining operations on the circuit board welding image and the welding equipment operation data respectively, and outputting circuit board welding image features corresponding to the circuit board welding image and welding equipment operation data features corresponding to the welding equipment operation data; Performing a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features, and outputting a multi-level circuit board welding feature corresponding to the target circuit board; At least based on the multi-layer circuit board welding characteristics, analyzing the target circuit board quality analysis result corresponding to the target circuit board, wherein the target circuit board quality analysis result is at least used to characterize that the target circuit board has quality in the welding dimension of electronic components; The step of performing a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features to output the multi-level circuit board welding features corresponding to the target circuit board includes: Splitting the circuit board welding image feature according to the color channel to form a first welding image channel feature, a second welding image channel feature and a third welding image channel feature corresponding to the circuit board welding image feature, wherein the first welding image channel feature corresponds to a red color channel, the second welding image channel feature corresponds to a green color channel, and the third welding image channel feature corresponds to a blue color channel; Using the first weight, the second weight and the third weight in the image channel fusion unit included in the circuit board quality analysis network, the first welding image channel feature, the second welding image channel feature and the third welding image channel feature are weighted summed and calculated, and the weighted welding image channel feature corresponding to the circuit board welding image feature is output, wherein the circuit board quality analysis network is a neural network formed by training in advance, and the first weight, the second weight and the third weight are network parameters of the image channel fusion unit, which are formed during training; Based on the first welding image channel feature, the second welding image channel feature, the third welding image channel feature and the weighted welding image channel feature, a multi-level semantic aggregation operation is performed on the welding equipment operation data feature to output a multi-level circuit board welding feature corresponding to the target circuit board; The steps of performing semantic mining operations on the circuit board welding image and the welding equipment operation data respectively, and outputting circuit board welding image features corresponding to the circuit board welding image and welding equipment operation data features corresponding to the welding equipment operation data, include: Utilizing the semantic mining unit included in the circuit board quality analysis network, semantic mining operations are performed on the circuit board welding image and the welding equipment operation data respectively, and circuit board image mining features corresponding to the circuit board welding image and equipment operation mining features corresponding to the welding equipment operation data are output, wherein the circuit board quality analysis network belongs to a trained neural network, and the semantic mining unit includes two semantic mining branches, and the two semantic mining branches are used to perform semantic mining on the circuit board welding image and welding equipment operation data in different forms of expression respectively; Using the first adaptation parameter in the semantic adaptation unit included in the circuit board quality analysis network, an adaptation operation in a feature space is performed on the circuit board image mining feature to form a circuit board welding image feature corresponding to the circuit board welding image; The second adaptation parameter in the semantic adaptation unit included in the circuit board quality analysis network is used to perform an adaptation operation in feature space on the equipment operation mining feature to form a welding equipment operation data feature corresponding to the circuit board welding image.

2. The circuit board quality analysis method based on artificial intelligence according to claim 1 is characterized in that: The step of performing a multi-level semantic aggregation operation on the welding equipment operation data feature based on the first welding image channel feature, the second welding image channel feature, the third welding image channel feature and the weighted welding image channel feature, and outputting the multi-level circuit board welding feature corresponding to the target circuit board includes: Using a first semantic aggregation unit included in a circuit board quality analysis network, a multi-level semantic aggregation operation is performed on the first welding image channel feature and the welding equipment operation data feature, and a multi-level first aggregation feature corresponding to the target circuit board is output; Using the second semantic aggregation unit included in the circuit board quality analysis network, the second welding image channel feature and the welding equipment operation data feature are subjected to a multi-level semantic aggregation operation, and the multi-level second aggregation feature corresponding to the target circuit board is output; Using the third semantic aggregation unit included in the circuit board quality analysis network, the third welding image channel feature and the welding equipment operation data feature are subjected to a multi-level semantic aggregation operation, and the multi-level third aggregation feature corresponding to the target circuit board is output; Using the fourth semantic aggregation unit included in the circuit board quality analysis network, the weighted welding image channel features and the welding equipment operation data features are subjected to a multi-level semantic aggregation operation, and the multi-level fourth aggregation features corresponding to the target circuit board are output; The multi-level first aggregation feature, the multi-level second aggregation feature, the multi-level third aggregation feature and the multi-level fourth aggregation feature are superimposed or averaged according to corresponding levels, and the multi-level circuit board welding feature corresponding to the target circuit board is output.

3. The circuit board quality analysis method based on artificial intelligence according to claim 2 is characterized in that: The step of using the first semantic aggregation unit included in the circuit board quality analysis network to perform a multi-level semantic aggregation operation on the first welding image channel feature and the welding equipment operation data feature, and outputting the multi-level first aggregation feature corresponding to the target circuit board includes: In the first level, using the first semantic aggregation unit included in the circuit board quality analysis network, based on the welding equipment operation data features, the semantic aggregation operation is performed on the first welding image channel features to output the first aggregation features of the first level; In each other level except the first level, using the first semantic aggregation unit, based on the second aggregation feature of the previous level and / or the third aggregation feature of the previous level, a semantic aggregation operation is performed on the first aggregation feature of the previous level, and the first associated feature corresponding to the other level is output; and based on the first associated feature corresponding to the other level, a semantic aggregation operation is performed on the first welding image channel feature, and the first aggregation feature of the other level is output; Based on the first aggregation features of each level, multi-level first aggregation features corresponding to the target circuit board are determined.

4. The circuit board quality analysis method based on artificial intelligence according to claim 3 is characterized in that: The step of using the first semantic aggregation unit to perform a semantic aggregation operation on the first aggregation feature of the previous level based on the second aggregation feature of the previous level and / or the third aggregation feature of the previous level in each other level other than the first level, and outputting the first associated feature corresponding to the other level, and performing a semantic aggregation operation on the first welding image channel feature based on the first associated feature corresponding to the other level, and outputting the first aggregation feature of the other level, comprises: In each other level except the first level, using the first semantic aggregation unit, determine the average feature of the second aggregation feature of the previous level and the third aggregation feature of the previous level, and determine the first correlation parameter between the average feature and the transposed feature of the first aggregation feature of the previous level, and based on the first correlation parameter, update the first aggregation feature of the previous level, and output the first correlation feature corresponding to the other level; Determine a second correlation parameter between the first associated feature corresponding to the other level and the transposed feature of the first welding image channel feature, and based on the second correlation parameter, update the first welding image channel feature to output the first aggregate feature corresponding to the other level.

5. The circuit board quality analysis method based on artificial intelligence according to any one of claims 1 to 4, characterized in that: The step of analyzing a target circuit board quality analysis result corresponding to the target circuit board based at least on the welding characteristics of the multi-layer circuit board comprises: For each level of the multi-level circuit board soldering features, the fully connected unit included in the circuit board quality analysis network is used to process the circuit board soldering features of the level, and the fully connected features corresponding to the circuit board soldering features of the level are output; Utilizing the linear regression unit included in the circuit board quality analysis network, the fully connected features corresponding to at least two levels of circuit board welding features are output and processed respectively to obtain at least two corresponding circuit board quality analysis parameters, and, based on the at least two circuit board quality analysis parameters, the target circuit board quality analysis result corresponding to the target circuit board is determined.

6. A circuit board quality analysis device based on artificial intelligence, characterized in that: include: A welding-related data acquisition module, used to acquire a circuit board welding image collected during the production process of a target circuit board, and to acquire welding equipment operation data of a corresponding target welding equipment during the production process of the target circuit board, wherein the circuit board welding image is at least used to characterize image information of the target circuit board during the welding process of electronic components, and the welding equipment operation data is at least used to characterize operation parameters of the target welding equipment when welding electronic components, including at least welding method, temperature and duration; A data semantic mining module, used to perform semantic mining operations on the circuit board welding image and the welding equipment operation data respectively, and output circuit board welding image features corresponding to the circuit board welding image and welding equipment operation data features corresponding to the welding equipment operation data; A feature semantic aggregation module, used for performing a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features, and outputting a multi-level circuit board welding feature corresponding to the target circuit board; A circuit board quality analysis module, used to analyze a target circuit board quality analysis result corresponding to the target circuit board based at least on the welding characteristics of the multi-layer circuit board, wherein the target circuit board quality analysis result is used to at least characterize that the target circuit board has quality in the welding dimension of electronic components; The step of performing a multi-level semantic aggregation operation on the circuit board welding image features and the welding equipment operation data features to output the multi-level circuit board welding features corresponding to the target circuit board includes: Splitting the circuit board welding image feature according to the color channel to form a first welding image channel feature, a second welding image channel feature and a third welding image channel feature corresponding to the circuit board welding image feature, wherein the first welding image channel feature corresponds to a red color channel, the second welding image channel feature corresponds to a green color channel, and the third welding image channel feature corresponds to a blue color channel; Using the first weight, the second weight and the third weight in the image channel fusion unit included in the circuit board quality analysis network, the first welding image channel feature, the second welding image channel feature and the third welding image channel feature are weighted summed and calculated, and the weighted welding image channel feature corresponding to the circuit board welding image feature is output, wherein the circuit board quality analysis network is a neural network formed by training in advance, and the first weight, the second weight and the third weight are network parameters of the image channel fusion unit, which are formed during training; Based on the first welding image channel feature, the second welding image channel feature, the third welding image channel feature and the weighted welding image channel feature, a multi-level semantic aggregation operation is performed on the welding equipment operation data feature to output a multi-level circuit board welding feature corresponding to the target circuit board; The performing of semantic mining operations on the circuit board welding image and the welding equipment operation data respectively, and outputting circuit board welding image features corresponding to the circuit board welding image and welding equipment operation data features corresponding to the welding equipment operation data, comprises: Utilizing the semantic mining unit included in the circuit board quality analysis network, semantic mining operations are performed on the circuit board welding image and the welding equipment operation data respectively, and circuit board image mining features corresponding to the circuit board welding image and equipment operation mining features corresponding to the welding equipment operation data are output, wherein the circuit board quality analysis network belongs to a trained neural network, and the semantic mining unit includes two semantic mining branches, and the two semantic mining branches are used to perform semantic mining on the circuit board welding image and welding equipment operation data in different forms of expression respectively; Using the first adaptation parameter in the semantic adaptation unit included in the circuit board quality analysis network, an adaptation operation in a feature space is performed on the circuit board image mining feature to form a circuit board welding image feature corresponding to the circuit board welding image; The second adaptation parameter in the semantic adaptation unit included in the circuit board quality analysis network is used to perform an adaptation operation in feature space on the equipment operation mining feature to form a welding equipment operation data feature corresponding to the circuit board welding image.

7. An electronic device, characterized in that: include: Memory for storing computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the circuit board quality analysis method based on artificial intelligence as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is run, the circuit board quality analysis method based on artificial intelligence according to any one of claims 1 to 5 is executed.

Images