Molten pool depth prediction method, device and equipment and storage medium

Abstract

The invention discloses a molten pool depth prediction method, device and equipment and a storage medium, and the method comprises the steps: collecting a temperature field distribution image of a molten pool region, and extracting image features in the temperature field distribution image; constructing a molten pool depth prediction model, taking the image features as the input of the prediction model, taking the residual error between the prediction output and the actual measurement as the optimization target of the prediction model, and carrying out iterative training on the prediction model; and the depth of the molten pool is predicted in real time through the trained molten pool depth prediction model. According to the technical scheme provided by the embodiment of the invention, the dependence on expensive detection equipment is eliminated, the technical implementation cost is greatly reduced, non-contact online measurement in a real sense is realized, and zero damage to a printed sample is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of molten pool depth measurement, and in particular to a molten pool depth prediction method, device, equipment and storage medium. Background Art

[0002] In the laser powder bed fusion (L-PBF) additive manufacturing process, melt pool stability is one of the key factors determining the quality of the formed part. Its geometric characteristics (especially the melt pool depth) can intuitively reflect the matching of process parameters and the stability of energy input, and directly affect the density, surface quality and mechanical properties of the formed part.

[0003] The dynamic change of the molten pool depth is closely related to process parameters such as laser power, scanning speed, and powder laying quality. When the parameters are mismatched or there is interference, abnormal fluctuations in the molten pool depth will cause defects such as pores, unfused parts, or keyholes.

[0004] Therefore, developing a prediction method for the melt pool depth can not only evaluate the melt pool stability in real time to optimize process parameters, but also provide a key basis for defect warning and closed-loop control. It is of great significance to improve the quality consistency of formed parts and promote the development of L-PBF technology towards intelligence. Summary of the Invention

[0005] The present invention provides a molten pool depth prediction method, device, equipment and storage medium, which realize non-contact, high-precision real-time prediction of the molten pool depth.

[0006] In a first aspect, the present invention provides a method for predicting molten pool depth, comprising:

[0007] Collecting a temperature field distribution image of the molten pool area and extracting image features from the temperature field distribution image;

[0008] Constructing a molten pool depth prediction model, using the image features as input to the prediction model, using the residual error between the prediction result and the measured depth as an optimization target of the prediction model, and iteratively training the prediction model;

[0009] The molten pool depth is predicted in real time through the trained molten pool depth prediction model; The molten pool depth prediction model is a step-by-step additive model composed of multiple basic models; the objective function of the molten pool depth prediction model is composed of a loss function and a regularization function, and the loss function is composed of the sum of the mean square error between the predicted output of each sample and the actual measurement.

[0010] Furthermore, the process of extracting image features is specifically as follows: using dual-band colorimetric temperature measurement technology to perform online monitoring of the coaxial melt pool and collect the temperature field distribution image of the melt pool area; using the grayscale histogram algorithm to extract statistical features in the temperature field distribution image, and using the scale-invariant feature transformation algorithm to extract local features in the temperature field distribution image; and using the statistical features and the local features together as the image features.

[0011] Furthermore, the loss function is expressed as:

[0012]

[0013] Where n represents the number of data samples, Represents the predicted output in sample i Compared with actual measurement The mean square error between

[0014] The regularization function is expressed as:

[0015]

[0016] Where C is a constant, which represents the sum of the complexities of the first t-1 basic models in the current iteration, j represents the jth node in the tth basic model, T represents the number of nodes in the tth basic model, γ and λ are the penalty strengths of the corresponding terms, represents the weight of the j-th node.

[0017] Furthermore, during the iterative training process, the residual error between the predicted output of the previous model and the label value is corrected by continuously adding new basic models; the label value acquisition process is specifically as follows: the molded part sample corresponding to the temperature field distribution image is directly measured by the metallographic planing method to obtain the label value.

[0018] Furthermore, each basic model in the molten pool depth prediction model is a decision tree model. The decision tree consists of two parts: node weights and the mapping relationship from data samples to nodes, which can be expressed as:

[0019]

[0020] Where x represents the data sample and , represents the node weight and , Represents the mapping relationship between data samples and nodes and , where d is the image feature dimension and T represents the number of nodes in the t-th basic model.

[0021] Furthermore, all samples belonging to the jth node are classified into the same sample set In the basic model newly added at the tth iteration, the weight of the jth node is obtained by taking the sum of the first-order partial derivatives and the sum of the second-order partial derivatives of the loss function of the samples contained in the node.

[0022] Furthermore, during the iterative training process, a greedy algorithm is used to find the optimal feature splitting point in the t-th basic model. The specific process is as follows: starting from the depth of 0 of the t-th basic model, all available features are enumerated and samples are sorted according to the feature values, the optimal splitting point of the feature is found, and the splitting gain is calculated; after traversing all sample features, the feature and its splitting point when the gain is maximum are recorded, two new nodes are split to the left and right of the splitting point, and each new node is associated with the corresponding sample set; this operation is repeated until the t-th basic model reaches the preset depth or the splitting gain is no longer a positive number, and the training of the t-th basic model is completed; wherein, the splitting gain is expressed as the difference between the objective function before and after the split.

[0023] In a second aspect, the present invention provides a molten pool depth prediction device, comprising:

[0024] A feature extraction module is used to collect a temperature field distribution image of the molten pool area and extract image features from the temperature field distribution image;

[0025] a model building module, configured to build a melt pool depth prediction model, use the image features as input to the prediction model, use the residual error between the prediction output and the actual measurement as an optimization target of the prediction model, and iteratively train the prediction model;

[0026] The real-time prediction module is used to predict the molten pool depth in real time using the trained molten pool depth prediction model; The molten pool depth prediction model is a step-by-step additive model composed of multiple basic models; the objective function of the molten pool depth prediction model is composed of a loss function and a regularization function, and the loss function is composed of the sum of the mean square error between the predicted output of each sample and the actual measurement.

[0027] In a third aspect, an embodiment of the present invention provides an electronic device, the electronic device including:

[0028] at least one processor; and a memory communicatively coupled to the at least one processor;

[0029] The memory stores a computer program that can be executed by at least one processor. The computer program is executed by at least one processor so that the at least one processor can perform the steps of the molten pool depth prediction method of any embodiment of the present invention.

[0030] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a processor to implement the steps of the molten pool depth prediction method of any embodiment of the present invention when executed.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The technical solution in the embodiments of the present invention first collects a temperature field distribution image of the melt pool area and extracts image features from the temperature field distribution image. A melt pool depth prediction model is then constructed and trained using the image features as input. The residual error between the predicted result and the measured depth is used as the optimization target for the prediction model. Using the trained melt pool depth prediction model, the melt pool depth is predicted in real time. This real-time prediction of the melt pool depth eliminates the need for expensive testing equipment, significantly reduces technical implementation costs, and achieves true non-contact online measurement, ensuring zero damage to printed samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 A schematic flow chart of a method for predicting molten pool depth provided by an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of a process for extracting image features provided by an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of the structure of a molten pool depth prediction model provided by an embodiment of the present invention;

[0037] Figure 4 A schematic diagram of a flow chart of a model training process provided by an embodiment of the present invention;

[0038] Figure 5 A schematic diagram of a flow chart of a model optimization process provided by an embodiment of the present invention;

[0039] Figure 6 A schematic structural diagram of a molten pool depth prediction device provided by an embodiment of the present invention;

[0040] Figure 7 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] In related technologies, melt pool depth measurement methods can be mainly divided into two categories: online monitoring and offline analysis:

[0043] 1) Online measurement technology is represented by synchronous high-speed X-ray imaging. This technology uses high-energy X-rays to penetrate the molten pool area and a high-speed camera system to capture the dynamic changes of the molten pool in real time, which can achieve depth measurement with micron-level resolution.

[0044] Although synchronized high-speed X-ray imaging can observe the morphology of the molten pool in real time, its equipment cost is extremely high, it relies on synchrotron radiation light sources or high-power X-ray systems, and is expensive to set up and maintain. In addition, there are many factors that limit its application in industrial scenarios.

[0045] 2) The offline analysis method mainly uses metallographic sample preparation technology, which prepares the molten pool cross-section sample through cutting, polishing, corrosion and other processes, and then combines it with an optical microscope or scanning electron microscope for precise measurement. Although this method destroys the sample, the measurement accuracy can reach the submicron level.

[0046] Offline metallographic sample preparation technology is a destructive detection method that requires cutting, grinding, polishing samples and observing them under a microscope. It is not only time-consuming and labor-intensive, but also cannot be used for real-time feedback in the actual production process.

[0047] In addition to the two direct measurement methods mentioned above, researchers have also developed a variety of statistically significant melt pool depth prediction methods. Finite element analysis-based heat conduction equations can simulate the temperature field evolution during heat source movement and infer the melt pool depth by determining the solid-liquid interface. Deep learning methods, which have emerged in recent years, use neural networks to process data and establish a mapping relationship between the printing process and the melt pool depth. With the development of high-precision sensors and artificial intelligence technologies, melt pool depth measurement is evolving towards multi-information fusion and real-time feedback control.

[0048] While statistical data-based prediction methods avoid direct measurement, they often rely on large amounts of experimental data for model training and parameter calibration. They have high computational costs and limited generalization capabilities, making them difficult to adapt to complex and changing actual production environments.

[0049] In summary, the existing melt pool depth measurement technology still faces problems such as high cost, strong destructiveness or insufficient adaptability. It is urgent to develop a low-cost, non-destructive and highly robust online monitoring method to meet the needs of industrial applications.

[0050] In order to solve at least one of the technical problems existing in the above-mentioned related technologies, the present invention provides a method for predicting the depth of a molten pool. Figure 1 A flow chart of a method for predicting molten pool depth is provided in an embodiment of the present invention. This embodiment is applicable to situations where the molten pool depth is monitored in real time. The method can be executed by a molten pool depth prediction device, which can be implemented in software and / or hardware, and can be configured in an electronic device.

[0051] like Figure 1 As shown in the figure, the implementation process of this method includes: first, collecting the original data, and collecting the temperature field distribution image of the molten pool area through the dual-band colorimetric temperature measurement technology; then performing data preprocessing, and extracting statistical features and local features respectively through the grayscale histogram and scale-invariant feature transformation algorithm; then constructing the XGBoost molten pool depth prediction model and completing the model training; finally, outputting the molten pool depth prediction result based on the currently taken image data.

[0052] The method specifically includes:

[0053] S1, collecting the temperature field distribution image of the molten pool area and extracting the image features in the temperature field distribution image.

[0054] Figure 2 A schematic diagram of a process for extracting image features provided by an example of the present invention is shown in FIG. Figure 2 As shown in the figure, during the laser powder bed fusion (L-PBF) part printing process, a dual-band colorimetric temperature measurement system is used for coaxial online monitoring to collect the temperature field distribution image of the molten pool area. The grayscale histogram and scale-invariant feature transform (SIFT) algorithm are used to extract statistical features and local features respectively, and the statistical features and local features are used together as image features.

[0055] It should be noted that dual-band colorimetric temperature measurement is a non-contact temperature measurement technology that calculates the temperature of a target object by detecting the ratio of its radiation intensity at two different wavelengths. This method, based on Planck's blackbody radiation law, utilizes the ratio of light intensity between two adjacent bands to eliminate the influence of emissivity. It exhibits strong anti-interference capabilities and high measurement accuracy. This technology is often used in conjunction with high-speed cameras to enable real-time temperature distribution monitoring in high-temperature processes. It is widely used for thermal process monitoring in industrial fields such as additive manufacturing and welding.

[0056] In an embodiment of the present invention, the image in the temperature field distribution image will be used as the input of the melt pool depth prediction model to train the melt pool depth prediction model. At the same time, the molded part sample corresponding to the temperature field distribution image will be measured by metallographic planing method and the corresponding melt pool depth data (the distance from the substrate plane to the deepest part of the melt pool) will be obtained as the training label value of the melt pool depth prediction model. The image features and the melt pool depth data will be accurately matched through time alignment and spatial registration technology to ensure the normal training and learning of the model.

[0057] S2, builds a melt pool depth prediction model, uses image features as the input of the prediction model, takes the residual error between the predicted output and the actual measurement as the optimization target of the prediction model, and iteratively trains the prediction model.

[0058] Figure 3 A schematic diagram of the structure of a molten pool depth prediction model provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown in the figure, the melt pool depth prediction model is a step-by-step additive model composed of multiple basic models. During the iterative training process, new basic models are continuously added to correct the residual error between the prediction output of the previous model and the label value. The label value acquisition process is as follows: the molded part sample corresponding to the temperature field distribution image is directly measured by the metallographic planing method to obtain the label value.

[0059] By continuously correcting the residual error between the prediction output of the previous model and the label value, the overall prediction effect is gradually optimized. The melt pool depth prediction model can be expressed as:

[0060]

[0061] in, represents the i-th input data sample, that is, the image features extracted after data preprocessing, It represents the prediction result corresponding to the i-th sample at the t-th iteration. When the number of iterations t is larger, the number of basic models included in the melt pool depth prediction model is larger, and the final prediction output is The closer it gets to the target true value, that is, the measured molten pool depth , but the corresponding model complexity will also increase and the amount of calculation will increase.

[0062] Furthermore, the objective function of the melt pool depth prediction model consists of a loss function and a regularization function. The loss function is composed of the sum of the mean square error between the predicted output of each sample and the actual measurement, and the regularization function is composed of the sum of the complexity of each basic model.

[0063] The loss function L can be expressed as:

[0064]

[0065] Where n represents the number of data samples, Represents the predicted output in sample i Compared with actual measurement To prevent the model from overfitting, a regularization term is introduced to constrain the complexity of each basic function.

[0066] The objective function Obj of the molten pool depth prediction model can be expressed as:

[0067]

[0068] During the model iteration process, each time a new base model is added, the previous model is already trained and does not need to be adjusted in subsequent iterations. Therefore, when modifying the objective function, only the newly added base model is trained and constrained. It can be understood that each iteration consists of a previous model and a base model. In the current iteration, the previous t-1 base models serve as the previous models, and the tth base model serves as the subsequent model.

[0069] According to the newly added basic objectives, the objective function is modified It can be expressed as:

[0070]

[0071] It should be noted that the melt pool depth prediction model consists of t basic models, and the sum of the complexity of the first t-1 basic models is It is represented by a constant C.

[0072] To find the optimal , we need to minimize the objective function. We can use Taylor's formula to perform a second-order expansion on the corrected objective function to obtain an approximate value, which can be expressed as:

[0073]

[0074] In the approximate formula, yes The first-order partial derivative of can be expressed as:

[0075]

[0076] In the approximate formula, yes The second-order partial derivative of can be expressed as:

[0077]

[0078] At the tth iteration of training, is a known item, so is a constant and will not affect the function optimization. Therefore, the objective function obtained by removing the correction is All constant terms in the , the simplified objective function can be expressed as:

[0079]

[0080] In the melt pool depth prediction model, the basic model For a decision tree, the decision tree consists of the weights of the nodes And the mapping relationship q from data samples to nodes, which can be expressed as:

[0081]

[0082] in, , , , where d is the sample dimension (image feature dimension) and T represents the number of nodes in the t-th basic model.

[0083] It's important to note that a decision tree is a supervised learning algorithm based on a tree structure. It recursively partitions a dataset into smaller subsets for classification or regression prediction. Its core principle is to gradually partition the data using conditional judgments (nodes) based on feature values, ultimately reaching leaf nodes (decision results). The advantages of decision trees are that the model is intuitive and easy to interpret, requires minimal data preprocessing, and can automatically filter out important features.

[0084] Correspondingly, The complexity of can be characterized as a combination of the number of nodes and the node weight paradigm, which can be expressed as:

[0085]

[0086] Among them, γ and λ are the penalty strengths of the corresponding terms, which can be adjusted empirically in experiments.

[0087] All samples belonging to the jth node are classified into the same sample set Then the simplified objective function can be rewritten to obtain the rewritten objective function, which can be expressed as:

[0088]

[0089] It is understandable that the rewritten objective function is about the node weights The optimal solution of the quadratic function of can be directly obtained through the vertex formula and can be expressed as:

[0090]

[0091] At this point, the objective function can be organized as:

[0092]

[0093] That is, in the basic model newly added at the tth iteration, the weight of the jth node can be obtained by calculating the sum of the first-order partial derivatives and the sum of the second-order partial derivatives of the loss function of the samples contained in the node.

[0094] Figure 4 A flow chart of a model training process provided by an embodiment of the present invention is as follows: Figure 4 As shown, in the actual training process, a greedy algorithm is used to find The optimal feature splitting point in . The specific process is as follows: Starting from the depth of 0, enumerate all available features and sort the samples according to the feature value, find the best splitting point of the feature, and calculate the splitting gain; after traversing all sample features, record the feature and its splitting point when the gain is the largest, split two new nodes on the left and right of the splitting point, and associate the corresponding sample set with each new node; repeat this operation until the The process is completed when the preset depth is reached or the split gain is no longer positive. The split gain is expressed as the difference between the objective functions before and after the split.

[0095] It's important to note that a greedy algorithm is an algorithmic strategy that chooses the optimal (most advantageous) decision at each step, hoping to eventually reach a global optimum by accumulating local optimal solutions. A core feature of this algorithm is that it doesn't backtrack; once a choice is made, it remains unchanged. This makes it computationally efficient and simple to implement, making it a common strategy for solving combinatorial optimization problems such as shortest paths, knapsack problems, and task scheduling.

[0096] The number of iterations and the depth of the decision tree are two key hyperparameters of the melt pool depth prediction model, which need to be repeatedly adjusted based on experimental results until the model performance meets expectations.

[0097] Figure 5 A flow chart of a model optimization process provided by an embodiment of the present invention is as follows: Figure 5 As shown in the figure, after inputting the image features, the number of iterations and the depth of the decision tree are initialized, and the melt pool depth prediction model is trained and its performance is verified. If the performance meets the requirements, the currently trained melt pool depth prediction model is exported as the target model; if the performance does not meet the requirements, the number of iterations and the depth of the decision tree are updated, and the melt pool depth prediction model is retrained and its performance is verified; until the model meets the requirements, it is exported as the target model.

[0098] The technical solution in the embodiment of the present invention addresses the problem of monitoring the molten pool depth in the laser additive manufacturing process and innovatively proposes an intelligent prediction method based on the XGBoost machine learning algorithm and temperature field feature analysis. Compared with the traditional synchronous high-speed X-ray imaging technology, the technical solution in this embodiment completely gets rid of the dependence on expensive detection equipment, greatly reducing the cost of technology implementation; compared with the destructive metallographic sample preparation and detection method, the technical solution in this embodiment realizes true non-contact online measurement, ensuring zero damage to the printed sample. This method uses a dual-band colorimetric temperature measurement system to collect the temperature field distribution image of the molten pool in real time, combined with advanced image feature extraction technology and the powerful prediction ability of the XGBoost algorithm, not only to achieve accurate prediction of the molten pool depth, but also has significant advantages such as fast response speed and strong adaptability. It provides a new solution for intelligent quality monitoring of the additive manufacturing process, which is particularly suitable for large-scale applications in industrial production environments.

[0099] S3, through the trained melt pool depth prediction model, the melt pool depth is predicted in real time.

[0100] The technical solution in the embodiments of the present invention proposes an intelligent melt pool depth prediction method based on the XGBoost (eXtreme GradientBoosting) algorithm and temperature field features. As an advanced ensemble learning algorithm, XGBoost, through its innovative gradient boosting framework and regularization strategy, effectively avoids overfitting while maintaining excellent prediction accuracy, enabling it to demonstrate exceptional performance and generalization capabilities in various machine learning tasks. To address the current technical bottlenecks in melt pool depth measurement—including the high cost of synchronous high-speed X-ray imaging equipment, the destructive nature of metallographic sample preparation, and the insufficient generalization capabilities of traditional prediction methods—this application innovatively combines dual-band colorimetric temperature measurement technology with the XGBoost algorithm to construct a complete intelligent melt pool depth prediction method. This method first uses dual-band colorimetric temperature measurement technology to acquire real-time images of the melt pool temperature field distribution. It then extracts multi-dimensional feature parameters, including grayscale statistical features and scale-invariant features, which are then input into an optimized XGBoost prediction model for depth calculation. This technical solution not only realizes non-contact, high-precision real-time prediction of melt pool depth, but more importantly, it breaks through the limitations of traditional measurement methods in terms of equipment cost and sample destructiveness. At the same time, through the powerful adaptive capabilities of machine learning algorithms, it can flexibly adapt to the prediction needs under different materials and different process parameters, providing a new technical path for intelligent quality monitoring of additive manufacturing.

[0101] Figure 6 A schematic diagram of the structure of a molten pool depth prediction device provided by an embodiment of the present invention is shown in FIG. Figure 6 As shown, the device specifically includes:

[0102] The feature extraction module 100 is used to collect the temperature field distribution image of the molten pool area and extract image features from the temperature field distribution image;

[0103] The model building module 200 is used to build a melt pool depth prediction model, use image features as input to the prediction model, use the residual error between the prediction output and the actual measurement as the optimization target of the prediction model, and iteratively train the prediction model;

[0104] The real-time prediction module 300 is used to predict the molten pool depth in real time using a trained molten pool depth prediction model.

[0105] The technical solution in the embodiment of the present invention first preprocesses the collected coaxial melt pool image, extracts statistical features and local features, and performs regression calculation as the input of the XGBoost model to obtain the melt pool depth prediction result at the corresponding melt pool position, thereby completing the depth prediction process.

[0106] Figure 7 Schematic diagram of the structure of an electronic device for implementing the method for predicting the depth of the molten pool according to an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0107] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0108] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0109] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any other suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the melt pool depth prediction method.

[0110] In some embodiments, the melt pool depth prediction method can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the melt pool depth prediction method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute the melt pool depth prediction method in any other suitable manner (e.g., via firmware).

[0111] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0112] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0113] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0114] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0115] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0116] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0117] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0118] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting molten pool depth, characterized in that: include: Collecting a temperature field distribution image of the molten pool area and extracting image features from the temperature field distribution image; Constructing a molten pool depth prediction model, using the image features as input to the prediction model, using the residual error between the prediction output and the actual measurement as an optimization target of the prediction model, and iteratively training the prediction model; The molten pool depth is predicted in real time through the trained molten pool depth prediction model; The molten pool depth prediction model is a step-by-step additive model composed of multiple basic models; the objective function of the molten pool depth prediction model is composed of a loss function and a regularization function, and the loss function is composed of the sum of the mean square error between the predicted output of each sample and the actual measurement.

2. The method according to claim 1, characterized in that The specific process of extracting image features is as follows: using dual-band colorimetric temperature measurement technology to perform online monitoring of the coaxial molten pool and collect the temperature field distribution image of the molten pool area; using the grayscale histogram algorithm to extract the statistical features in the temperature field distribution image, and using the scale-invariant feature transformation algorithm to extract the local features in the temperature field distribution image; and using the statistical features and the local features together as the image features.

3. The method according to claim 1, characterized in that The loss function is expressed as: ； Where n represents the number of data samples, Represents the predicted output in sample i Compared with actual measurement The mean square error between The regularization function is expressed as: ； Where C is a constant, which represents the sum of the complexities of the first t-1 basic models in the current iteration, j represents the jth node in the tth basic model, T represents the number of nodes in the tth basic model, γ and λ are the penalty strengths of the corresponding terms, represents the weight of the j-th node.

4. The method according to claim 1, wherein During the iterative training process, the residual error between the predicted output of the previous model and the label value is corrected by continuously adding new basic models; the label value acquisition process is specifically as follows: the molded part sample corresponding to the temperature field distribution image is directly measured by the metallographic planing method to obtain the label value.

5. The method according to claim 1, wherein Each basic model in the molten pool depth prediction model is a decision tree model. The decision tree consists of two parts: node weight and the mapping relationship from data samples to nodes, which can be expressed as: ； Where x represents the data sample and , represents the node weight and , Represents the mapping relationship between data samples and nodes and , where d is the image feature dimension and T represents the number of nodes in the t-th basic model.

6. The method according to claim 5, characterized in that All samples belonging to the jth node are classified into the same sample set In the basic model newly added at the t-th iteration, the weight of the j-th node is obtained by taking the sum of the first-order partial derivatives and the sum of the second-order partial derivatives of the loss function of the samples contained in the node.

7. The method according to claim 6, characterized in that During the iterative training process, a greedy algorithm is used to find the optimal feature splitting point in the t-th basic model. The specific process is as follows: starting from the depth of 0 of the t-th basic model, all available features are enumerated and samples are sorted according to the feature values, the optimal splitting point of the feature is found, and the splitting gain is calculated; after traversing all sample features, the feature and its splitting point with the maximum gain are recorded, two new nodes are split to the left and right of the splitting point, and each new node is associated with the corresponding sample set; this operation is repeated until the t-th basic model reaches the preset depth or the splitting gain is no longer a positive number, and the training of the t-th basic model is completed; wherein, the splitting gain is expressed as the difference between the objective function before and after the split.

8. A device for predicting molten pool depth, characterized in that: The device is configured to implement the method according to any one of claims 1 to 7, and includes: A feature extraction module is used to collect a temperature field distribution image of the molten pool area and extract image features from the temperature field distribution image; a model building module, configured to build a melt pool depth prediction model, use the image features as input to the prediction model, use the residual error between the prediction output and the actual measurement as an optimization target of the prediction model, and iteratively train the prediction model; The real-time prediction module is used to predict the molten pool depth in real time using the trained molten pool depth prediction model; The molten pool depth prediction model is a step-by-step additive model composed of multiple basic models; the objective function of the molten pool depth prediction model is composed of a loss function and a regularization function, and the loss function is composed of the sum of the mean square error between the predicted output of each sample and the actual measurement.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the steps of the molten pool depth prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the steps of the molten pool depth prediction method according to any one of claims 1 to 7 when executed.

Images