Welding penetration status monitoring method and device based on multi-dimensional fusion sensing

Abstract

The present invention provides a method and device for monitoring the weld penetration state based on multidimensional fusion sensing, which relates to the field of welding sensing technology. The method comprises: collecting a real-time image of the weld penetration state and a current real-time one-dimensional signal from the front of the weld pool during welding; inputting the real-time image of the weld penetration state and the real-time one-dimensional signal into a fusion model to obtain the real-time width of the back of the weld pool output by the fusion model; and determining the weld penetration state based on the real-time width of the back of the weld pool; wherein the fusion model is trained based on sample images of the weld penetration state and sample one-dimensional signals from the front of the weld pool as training data, and sample widths of the back of the weld pool as labels. The fusion model establishes an association between feature information of different dimensions and the back width of the weld pool, and can obtain a highly accurate monitoring result of the back width of the weld pool, thereby achieving effective monitoring of the weld penetration state.

Description

Technical Field

[0001] The present invention relates to the field of welding sensing technology, and in particular to a welding penetration state monitoring method and device based on multi-dimensional fusion sensing. Background Art

[0002] Welding sensors play a crucial role in automated welding control. Several types of welding sensors, including arc voltage, vision, sound, ultrasonic, infrared, X-ray, spectral, weld pool oscillation, and multi-dimensional information sensing, are widely used in the welding process. Multi-sensor sensing and fusion methods facilitate the acquisition of multi-dimensional information, and fusion algorithms combine the strengths of each type of information to improve monitoring accuracy.

[0003] Penetration sensing and control are critical and important factors in achieving quality control in the GTAW (Gas Tungsten Arc Weld) process. Since the back width of the weld pool during welding is difficult to observe directly, when using multi-sensor sensing and fusion methods for monitoring, it is difficult to associate the feature information established by different dimensional information with the back width of the weld pool. Therefore, the accuracy of the final monitored back width of the weld pool is still not high, and it is impossible to effectively monitor the welding penetration status. Summary of the Invention

[0004] The present invention provides a welding penetration status monitoring method and device based on multi-dimensional fusion sensing, which is used to solve the problem in the prior art that it is difficult to associate feature information established by information of different dimensions with the penetration width on the back side of the weld.

[0005] The present invention provides a method for monitoring welding penetration status based on multi-dimensional fusion sensing, comprising:

[0006] Collect the real-time image of the penetration status of the weld pool front during welding and the current real-time one-dimensional signal;

[0007] Inputting the real-time image of the penetration state and the real-time one-dimensional signal into a fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model;

[0008] Determining the welding penetration state according to the real-time width of the back side of the weld pool;

[0009] Among them, the fusion model is trained with the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as training data, and the sample width of the back side of the weld pool corresponding to the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as label training.

[0010] According to a welding penetration status monitoring method based on multi-dimensional fusion sensing provided by the present invention, the real-time image of the penetration status of the front side of the weld pool and the sample image of the penetration status are both active visual images collected after mapping the front side information of the weld pool through a laser generator.

[0011] According to a welding penetration status monitoring method based on multi-dimensional fusion sensing provided by the present invention, the real-time image of the penetration status of the front side of the weld pool and the sample image of the penetration status are both passive visual images of the front side of the weld pool collected by a camera.

[0012] According to a welding penetration status monitoring method based on multi-dimensional fusion sensing provided by the present invention, the sample width of the back side of the weld pool is obtained by calibrating and calculating the penetration status image of the back side of the weld pool collected by a camera.

[0013] According to a welding penetration status monitoring method based on multi-dimensional fusion sensing provided by the present invention, the fusion model is composed of a multi-layer perceptron neural network and a convolutional neural network, and at least one fully connected layer is used to fuse and output the output results of the multi-layer perceptron neural network and the output results of the convolutional neural network.

[0014] According to a method for monitoring welding penetration status based on multi-dimensional fusion sensing provided by the present invention, the real-time image of the penetration status and the real-time one-dimensional signal are input into a fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model, including:

[0015] Inputting the real-time one-dimensional signal into a multi-layer perceptron neural network to obtain a first output result;

[0016] Inputting the real-time image of the melt penetration state into a convolutional neural network to obtain a second output result;

[0017] The fully connected layer optimizes the first output result and the second output result to obtain the real-time width of the back side of the weld pool.

[0018] The present invention also provides a welding penetration status monitoring device based on multi-dimensional fusion sensing, comprising:

[0019] Information acquisition module, used to collect real-time images of the penetration status of the weld pool front during welding and the current real-time one-dimensional signal;

[0020] A model execution module is used to input the real-time image of the penetration state and the real-time one-dimensional signal into the fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model;

[0021] A penetration state determination module is used to determine the welding penetration state according to the real-time width of the back side of the weld pool;

[0022] Among them, the fusion model is trained with the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as training data, and the sample width of the back side of the weld pool corresponding to the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as label training.

[0023] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, it implements the welding penetration status monitoring method based on multi-dimensional fusion sensing as described above.

[0024] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for monitoring welding penetration status based on multi-dimensional fusion sensing as described above is implemented.

[0025] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described methods for monitoring welding penetration status based on multi-dimensional fusion sensing.

[0026] The method of the present invention adopts a fusion model that uses sample images of the front penetration state of the weld pool and sample one-dimensional signals as training data and the back width of the weld pool as label training. Through the fusion model, the association between feature information of different dimensions and the back width of the weld pool is established, so that the fusion model is used to identify the real-time collected front penetration state image of the weld pool and the current one-dimensional signal, and a high-precision monitoring result of the back width of the weld pool can be obtained, thereby realizing effective monitoring of the welding penetration state. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 It is a flow chart of the welding penetration state monitoring method based on multi-dimensional fusion sensing provided by the present invention;

[0029] Figure 2 This is a hardware structure diagram of the welding penetration status monitoring method based on multi-dimensional fusion sensing provided by the present invention;

[0030] Figure 3 Schematic diagram of the fusion model in the welding penetration state monitoring method based on multi-dimensional fusion sensing provided by the present invention;

[0031] Figure 4 This is a comparison chart of the actual width of the back of the molten pool when the welding current is 60A and the monitored width output by the fusion model trained by the active visual image and one-dimensional signal using the method of the present invention;

[0032] Figure 5 This is a comparison chart of the actual width of the back of the molten pool when the welding current is 60A and the monitored width output by the fusion model trained by the method of the present invention using passive visual images and one-dimensional signals;

[0033] Figure 6 Schematic diagram of the structure of the welding penetration status monitoring device based on multi-dimensional fusion sensing provided by the present invention;

[0034] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0035] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0036] like Figure 1 As shown, the welding penetration status monitoring method based on multi-dimensional fusion sensing of this embodiment includes:

[0037] Step S110 captures a real-time image of the weld pool's frontal penetration status and the current real-time one-dimensional signal during welding. The one-dimensional signal includes time, arc voltage, welding current, and heat input signals. The real-time one-dimensional signal represents the current time and the arc voltage, welding current, and heat input signals of the welding system when capturing the real-time image of the weld pool's frontal penetration status.

[0038] Step S120 , inputting the real-time image of the penetration state and the real-time one-dimensional signal into a fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model.

[0039] Step S130 determines the weld penetration status based on the real-time width of the back side of the weld pool. Specifically, the weld penetration status of a workpiece is generally classified as incomplete penetration, full penetration, and excessive penetration. Depending on actual conditions, such as workpiece size and material, the width thresholds corresponding to full penetration and excessive penetration vary. For example, a real-time width d of the back side of the weld pool less than 0 indicates incomplete penetration, 0 ≤ d ≤ 5 mm indicates full penetration, and d > 5 mm indicates excessive penetration. 5 mm is the width threshold corresponding to full penetration and excessive penetration. Therefore, the current weld penetration status can be determined by determining the real-time width of the back side of the weld pool.

[0040] Among them, the fusion model is trained with the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as training data, and the sample width of the back side of the weld pool corresponding to the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as label training.

[0041] The method of this embodiment adopts a fusion model that uses sample images of the penetration state of the front side of the weld pool and sample one-dimensional signals as training data, and sample widths of the back side of the weld pool as labels for training. The fusion model establishes an association between feature information of different dimensions and the width of the back side of the weld pool. The fusion model is used to identify the real-time image of the penetration state of the front side of the weld pool and the current real-time one-dimensional signal collected in real time, and can obtain a highly accurate monitoring result of the width of the back side of the weld pool, thereby realizing effective monitoring of the welding penetration state.

[0042] like Figure 2 As shown in FIG, a hardware support structure of the welding penetration state monitoring method based on multi-dimensional fusion sensing of this embodiment (not limited to this hardware support structure). The hardware support structure mainly includes: welding system, motion system and visual sensing system. Figure 2 As can be seen in the figure, the welding system includes a power supply 201, a welding torch 202, and pure argon shielding gas, which is ejected from the tungsten electrode of the welding torch 202. The motion system, consisting of an industrial robot or walking mechanism, is primarily used to transport the workpiece 203 to be welded. The visual sensing system includes a front camera 204, a rear camera 205, a laser generator 206, and a reflective imaging plane 207. The front camera 204 and rear camera 205 are respectively used to capture images of the front and rear surfaces of the weld pool 208 on the workpiece 203. The front camera 204 has a built-in narrowband filter to filter arc light. The laser emitter 206 emits laser light of a preset frequency band, and the reflective imaging plane 207 is used to illuminate the weld pool 208 and then reflect the image. The welding system, motion system, and visual sensing system all collect and output signals through a data acquisition card in a computer 209, enabling data acquisition, welding parameter control, and motion control during the welding process.

[0043] In this embodiment, the real-time image of the penetration state of the front side of the weld pool and the sample image of the penetration state are active visual images collected after the laser generator maps the front side information of the weld pool. Specifically, Figure 2 The laser generator 206 emits laser light, which illuminates the weld pool 208 and is reflected to the reflection imaging plane 207 to form an active vision image.

[0044] In this embodiment, the real-time image of the penetration state of the front side of the weld pool and the sample image of the penetration state can also be a passive visual image of the front side of the weld pool collected by a camera. Specifically, Figure 2 The image of the front side of the weld pool 208 captured by the central front camera 204 is a passive vision image.

[0045] The sample width of the back side of the weld pool is obtained by calibrating and calculating the penetration state image of the back side of the weld pool collected by the camera. Specifically, Figure 2 The back camera 205 captures a penetration state image of the back side of the weld pool 208. The back side of the weld pool 208 is then calibrated and calculated using an algorithm in the computer 209 to obtain a sample width of the back side of the weld pool. A binary image processing algorithm can be used to calibrate and calculate the penetration state image of the back side of the weld pool 208 to obtain the sample width of the back side of the weld pool.

[0046] like Figure 3 As shown in FIG, the fusion model is composed of a multi-layer perceptron neural network (MLP) and a convolutional neural network (CNN), and at least one fully connected layer is used to fuse the output results of the multi-layer perceptron neural network and the output results of the convolutional neural network. Figure 3 It can be seen that the decision-level fusion method is used to fully connect the output results of MLP and CNN, and finally the recognition result of the fusion model is output. When fusing MLP and CNN during the fusion process, one or more fully connected layers can be used. This embodiment uses one fully connected layer to fuse the output results.

[0047] Before training the fusion model, when collecting a sample image of the front penetration state of the weld pool 208, the current time and the arc voltage, welding current, and heat input signals of the welding system are obtained as sample one-dimensional signals. At the same time, the back camera 205 collects a sample image of the back penetration state of the weld pool 208. The computer 209 calibrates and calculates the back penetration state image of the weld pool 208 to obtain a sample width of the back of the weld pool corresponding to the front penetration state sample image and sample one-dimensional signal. The sample image of the front penetration state of the weld pool, the sample one-dimensional signal, and the sample width of the back of the weld pool are substituted into the fusion model for training.

[0048] During the training of the fusion model, Figure 3 As shown in the figure, the sample one-dimensional signal is input into the MLP, and the sample image of the penetration state on the front of the weld pool is input into the CNN. Multiple iterations are performed, for example: 50 times. After each iteration, the final output result of the fully connected layer is compared with the label, and the error is calculated. During the iteration process, feedback propagation will update the weights in the fully connected layer, thereby automatically optimizing the output results of the above two neural networks and outputting the final monitoring width. The training model corresponding to the number of iterations with the smallest error is selected as the final fusion model.

[0049] Among them, the input sample image can be an active vision image or a passive vision image. The active vision image and the one-dimensional signal are used as samples for model training to obtain a fusion model based on active vision. The passive vision image and the one-dimensional signal are used as samples for model training to obtain a fusion model based on passive vision. Through the error of the test set, it can be found that the active / passive vision fusion model has the smallest error when it is iterated 24 times / 29 times respectively. Therefore, the training model saved after this training is selected as the optimal fusion model, and the respective test set data is verified, and the errors are 0.354mm and 0.525mm respectively. Table 1 below lists the test set error and other information of the optimal active vision fusion model and the optimal passive vision fusion model.

[0050] Table 1 Loss values and errors of different model test sets at welding current 60A

[0051]

[0052] like Figure 4 and Figure 5 As shown in the figure, the comparison results of the monitoring values and the true values of the active vision fusion model and the passive vision fusion model when the welding current is 60A are respectively shown. The results show that to a certain extent, both sets of test data can effectively monitor the back width of the weld pool. Figure 4 As shown in Figure 3, the comparison between these two sets of experiments shows that the active vision fusion model can more accurately monitor the back width of the weld pool.

[0053] In this embodiment, the multidimensional information fusion model (MLP+CNN) uses two neural network deep learning mechanisms to automatically extract features from different data types. This eliminates the need to develop a corresponding feature extraction algorithm for each image or one-dimensional signal. It also achieves decision-level fusion of different data types, thus resolving the challenges of extracting features from different data types and merging them after extraction. Furthermore, the fusion of multidimensional information achieves higher monitoring accuracy than single-sensor methods, providing support for subsequent online control of the welding process.

[0054] After the fusion model training is completed, the fusion model can be used to monitor the back width of the weld pool online. At this time, the back camera is no longer needed to collect images of the back of the weld pool. Figure 3 The fusion model, step S120 includes:

[0055] The real-time one-dimensional signal is input into a multi-layer perceptron neural network to obtain a first output result, where the first output result is the real-time width of the back side of the weld pool output by the MLP.

[0056] The real-time image of the penetration state is input into a convolutional neural network to obtain a second output result, which is the real-time width of the back side of the weld pool output by the CNN.

[0057] The fully connected layer optimizes the first and second output results to obtain the real-time width of the back side of the weld pool. It should be noted that the optimization here is performed using the optimization method determined by the trained fusion model.

[0058] The welding penetration state monitoring device based on multi-dimensional fusion sensing provided by the present invention is described below. The welding penetration state monitoring device based on multi-dimensional fusion sensing described below and the welding penetration state monitoring method based on multi-dimensional fusion sensing described above can be referenced to each other.

[0059] like Figure 6 As shown, the present invention is a welding penetration state monitoring device based on multi-dimensional fusion sensing, comprising:

[0060] The information acquisition module 610 is used to acquire a real-time image of the penetration state of the front side of the weld pool and the current real-time one-dimensional signal during the welding process.

[0061] The model execution module 620 is used to input the real-time image of the penetration state and the real-time one-dimensional signal into the fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model.

[0062] The penetration state determination module 630 is used to determine the welding penetration state according to the real-time width of the back side of the weld pool.

[0063] Among them, the fusion model is trained with the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as training data, and the sample width of the back side of the weld pool corresponding to the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as label training.

[0064] The device of this embodiment adopts a fusion model that uses sample images of the penetration state of the front side of the weld pool and sample one-dimensional signals as training data, and sample widths of the back side of the weld pool as labels for training. The fusion model establishes an association between feature information of different dimensions and the penetration width of the back side of the weld. The fusion model is used to identify the real-time image of the penetration state of the front side of the weld pool and the current real-time one-dimensional signal collected in real time, and can obtain a highly accurate monitoring result of the width of the back side of the weld pool, thereby realizing effective monitoring of the welding penetration state.

[0065] Optionally, the fusion model is composed of a multi-layer perceptron neural network and a convolutional neural network, and at least one fully connected layer is used to fuse and output the output results of the multi-layer perceptron neural network and the output results of the convolutional neural network.

[0066] Optionally, the model execution module 620 includes:

[0067] The first model execution module is used to input the real-time one-dimensional signal into a multi-layer perceptron neural network to obtain a first output result.

[0068] The second model execution module is used to input the real-time image of the melt penetration state into the convolutional neural network to obtain a second output result.

[0069] A fully connected output module is used for the fully connected layer to optimize the first output result and the second output result to obtain the real-time width of the back side of the weld pool.

[0070] It should be noted that the welding penetration status monitoring method and device based on multi-dimensional fusion sensing of the present invention are not only applicable to GTAW, but also to welding methods such as gas metal arc welding (GMAW), plasma welding (PAW), and laser welding (LW), which can use visual mapping of the molten pool surface characteristics and arc characteristics, and can monitor the welding penetration status in real time during the welding process of the above welding methods.

[0071] Figure 7 The present invention provides a schematic diagram of the physical structure of an electronic device, which may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740. The processor 710, the communication interface 720, and the memory 730 communicate with each other via the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute a method for monitoring welding penetration status based on multi-dimensional fusion sensing, which includes:

[0072] Collect the real-time image of the penetration status of the weld pool front during welding and the current real-time one-dimensional signal.

[0073] The real-time image of the penetration state and the real-time one-dimensional signal are input into a fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model.

[0074] Among them, the fusion model is trained with the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as training data, and the sample width of the back side of the weld pool corresponding to the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as label training.

[0075] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, 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. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0076] On the other hand, the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the welding penetration state monitoring method based on multi-dimensional fusion sensing provided by the above methods, which includes:

[0077] Collect the real-time image of the penetration status of the weld pool front during welding and the current real-time one-dimensional signal.

[0078] The real-time image of the penetration state and the real-time one-dimensional signal are input into a fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model.

[0079] Among them, the fusion model is trained with the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as training data, and the sample width of the back side of the weld pool corresponding to the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as label training.

[0080] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for monitoring welding penetration status based on multi-dimensional fusion sensing provided by the above methods is implemented. The method includes:

[0081] Collect the real-time image of the penetration status of the weld pool front during welding and the current real-time one-dimensional signal.

[0082] The real-time image of the penetration state and the real-time one-dimensional signal are input into a fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model.

[0083] Among them, the fusion model is trained with the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as training data, and the sample width of the back side of the weld pool corresponding to the sample image of the penetration state on the front side of the weld pool and the sample one-dimensional signal as label training.

[0084] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0085] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A welding penetration status monitoring method based on multi-dimensional fusion sensing, characterized in that: include: Collect the real-time image of the penetration status of the weld pool front during welding and the current real-time one-dimensional signal; Inputting the real-time image of the penetration state and the real-time one-dimensional signal into a fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model; Determining the welding penetration state according to the real-time width of the back side of the weld pool; The fusion model is trained using sample images and one-dimensional signals of the front side of the weld pool as training data, and sample widths of the back side of the weld pool corresponding to the sample images and one-dimensional signals of the front side of the weld pool as labels; The fusion model is composed of a multi-layer perceptron neural network and a convolutional neural network, and uses at least one fully connected layer to fuse the output results of the multi-layer perceptron neural network and the output results of the convolutional neural network for output; Inputting the real-time image of the penetration state and the real-time one-dimensional signal into the fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model includes: Inputting the real-time one-dimensional signal into a multi-layer perceptron neural network to obtain a first output result; Inputting the real-time image of the melt penetration state into a convolutional neural network to obtain a second output result; The fully connected layer optimizes the first output result and the second output result to obtain the real-time width of the back side of the weld pool.

2. The method for monitoring welding penetration status based on multi-dimensional fusion sensing according to claim 1, characterized in that: The real-time image of the penetration state and the sample image of the penetration state are both active visual images collected after the front information of the weld pool is mapped by a laser generator.

3. The method for monitoring welding penetration status based on multi-dimensional fusion sensing according to claim 1, characterized in that: The real-time image of the penetration state of the front side of the weld pool and the sample image of the penetration state are both passive visual images of the front side of the weld pool collected by a camera.

4. The method for monitoring welding penetration status based on multi-dimensional fusion sensing according to claim 1, characterized in that: The sample width of the back side of the weld pool is obtained by calibrating and calculating the penetration state image of the back side of the weld pool collected by the camera.

5. A welding penetration status monitoring device based on multi-dimensional fusion sensing, characterized in that: include: Information acquisition module, used to collect real-time images of the penetration status of the weld pool front during welding and the current real-time one-dimensional signal; A model execution module is used to input the real-time image of the penetration state and the real-time one-dimensional signal into the fusion model to obtain the real-time width of the back side of the weld pool output by the fusion model; A penetration state determination module is used to determine the welding penetration state according to the real-time width of the back side of the weld pool; The fusion model is trained using sample images and one-dimensional signals of the front side of the weld pool as training data, and sample widths of the back side of the weld pool corresponding to the sample images and one-dimensional signals of the front side of the weld pool as labels; The fusion model is composed of a multi-layer perceptron neural network and a convolutional neural network, and uses at least one fully connected layer to fuse the output results of the multi-layer perceptron neural network and the output results of the convolutional neural network for output; The model execution module includes: A first model execution module is configured to input the real-time one-dimensional signal into a multilayer perceptron neural network to obtain a first output result; A second model execution module is used to input the real-time image of the melt penetration state into a convolutional neural network to obtain a second output result; A fully connected output module is used for the fully connected layer to optimize the first output result and the second output result to obtain the real-time width of the back side of the weld pool.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the welding penetration status monitoring method based on multi-dimensional fusion sensing as described in any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for monitoring welding penetration status based on multi-dimensional fusion sensing as described in any one of claims 1 to 4 is implemented.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for monitoring welding penetration status based on multi-dimensional fusion sensing as described in any one of claims 1 to 4 is implemented.

Images