Titanium alloy texture control method, system and device based on hot-cold wire composite

Abstract

The present application belongs to the field of additive manufacturing, and particularly discloses a titanium alloy texture control method, system and device based on hot-cold wire composite, which comprises: obtaining a mechanical property requirement distribution map of different spatial regions of a titanium alloy part numerical model; dividing the part based on the mechanical property requirement distribution map to obtain corresponding subparts of the part; obtaining a first distribution in the subpart, i.e. a grain size distribution, based on the mechanical property requirement, and then determining pulse-related parameters corresponding to the hot wire; obtaining a wire feed speed curve in the subpart; determining a weld bead forming size based on the wire feed speed curve and the pulse-related parameters, slicing the subpart based on the weld bead forming size, and printing each layer of slice based on this, to obtain a printed product by layer-by-layer stacking. The present application realizes the regulation and control of the mechanical properties of the part based on the relationship among the pulse-related parameters-grain size-mechanical properties, so that the mechanical properties of different regions of the titanium alloy part meet the requirements.

Description

Technical Field

[0001] This invention belongs to the field of additive manufacturing, and more specifically, relates to a method, system and device for controlling the texture of titanium alloys based on hot-cold wire composite. Background Technology

[0002] Titanium alloys can reduce aircraft weight and improve thrust-to-weight ratio, and are widely used in the aerospace field. They are commonly used as fuselage structural components such as stabilizer plates, exhaust pipes, ducts, doors, landing gear, tie rods, wing spars, ribs, and stiffeners. Although titanium alloys have excellent properties and wide applications, their difficult machining performance and high manufacturing cost hinder further industrial applications. First, the strong chemical affinity of titanium alloys for cutting tools leads to strong adhesion between chips and tools, reducing the life of carbide tools. This results in slower cutting speeds during titanium alloy milling, affecting machining efficiency. Second, the machining of load-bearing structural components such as frames and beams involves a large amount of material removal, with over 80% of the titanium alloy being removed during machining, resulting in high material waste and high manufacturing costs. Third, titanium alloy milling often relies on high-end imported machine tools, requiring significant initial equipment investment and preventing the formation of a large-scale industrial supply chain.

[0003] Due to the high cost of traditional titanium alloy milling methods, the direct forming of titanium alloys using arc wire additive manufacturing technology has attracted widespread attention in recent years. Arc wire additive manufacturing technology uses an electric arc or other heat source to melt metal wire, depositing it layer by layer to form metal components. Based on a three-dimensional model of the component, it achieves layer-by-layer deposition and effectively ensures the structural dimensions and forming accuracy of the component, representing a near-net-shape forming process. Arc wire additive manufacturing technology can be used to achieve high-efficiency and low-cost manufacturing of large-size titanium alloy aerospace metal components, and has become an important new method for manufacturing titanium alloy metal components, attracting widespread attention both domestically and internationally.

[0004] However, traditional arc-wire additive manufacturing technology can only use a single wire for melting, making it impossible to achieve "composition design" for parts with different stress requirements. In contrast, using a multi-arc, multi-wire approach generates a greater heat input to the molten pool, causing the titanium alloy molten pool to remain in the high-temperature zone for a longer time, resulting in titanium alloy grain growth and impacting mechanical properties. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a method, system and device for controlling the texture of titanium alloy based on hot-cold wire composite. The purpose is to achieve the control of the mechanical properties of the parts based on the relationship between pulse correlation parameters, grain size and mechanical properties, so that the mechanical properties of different regions of the titanium alloy parts meet the requirements.

[0006] To achieve the above objectives, according to a first aspect of the present invention, a method for controlling the texture of titanium alloys based on a hot-cold wire composite is proposed, comprising the following steps:

[0007] S1. Simulate the titanium alloy part model under preset conditions to obtain the distribution map of mechanical performance requirements in different spatial regions of the part model;

[0008] S2. Based on the mechanical performance requirement distribution map, the parts are divided to obtain several sub-parts corresponding to the parts;

[0009] S3. For any sub-part

[0010] The first distribution in the sub-part is obtained based on mechanical performance requirements, and the first distribution represents the grain size distribution in the sub-part of the part.

[0011] Based on the first distribution, the pulse correlation parameters corresponding to the hot wire are determined, and the pulse correlation parameters include at least one of pulse frequency, duty cycle, pulse peak value, and pulse base value;

[0012] Obtain the wire feeding speed curve in the sub-section, the wire feeding speed curve including the hot wire feeding speed curve and the cold wire feeding speed curve;

[0013] S4. Determine the weld bead forming size based on the wire feeding speed curve and the pulse correlation parameters, slice the sub-part based on the weld bead forming size, print each slice based on this, and stack the slices to obtain the printed part.

[0014] As a further preferred embodiment, obtaining a first distribution in the sub-part based on mechanical performance requirements, and determining the pulse-related parameters corresponding to the hot filament based on the first distribution, further includes:

[0015] The mapping relationship between mechanical properties, the first distribution, and impulse correlation parameters is obtained based on a multi-step convolutional neural network.

[0016] The multi-step convolutional neural network comprises two parts: the first part converts the pulse correlation parameters corresponding to the material into a grain size distribution matrix; the second part converts the grain size distribution matrix into the corresponding predicted mechanical properties.

[0017] As a further preferred embodiment, the multi-step convolutional neural network is trained based on the following steps:

[0018] Obtain an initial model of a convolutional neural network; the initial model of the convolutional neural network includes a first part and a second part, wherein the first part includes at least a number of convolutional layers, and the second part includes at least a number of convolutional layers and a fully connected layer.

[0019] Obtain a first training sample set and a second training sample set. The first training sample set includes multiple samples with different grain size distributions under different pulse correlation parameters, and the second training sample set includes multiple measured mechanical property samples with different grain size distributions.

[0020] Multiple rounds of training are performed on the first training sample set and the second training sample set respectively to obtain the first and second parts of the pre-trained initial model.

[0021] Obtain a joint training sample set, which includes multiple measured mechanical performance samples under different pulse correlation parameters;

[0022] Based on the joint training sample set, the first and second parts of the pre-trained initial model are jointly trained to obtain a trained multi-step convolutional neural network model.

[0023] As a further preferred embodiment, obtaining the wire feeding speed curve in the sub-section includes:

[0024] Obtain a second distribution in the sub-part, the second distribution representing the component distribution in the part sub-part;

[0025] Obtain the corresponding components of the hot wire and the cold wire;

[0026] Based on the second distribution and the corresponding components of the hot wire and cold wire, the matching relationship between the hot wire speed and the cold wire speed is determined, and the wire feeding speed curve in the sub-section is determined based on the matching relationship.

[0027] As a further preferred embodiment, determining the matching relationship between the hot wire speed and the cold wire speed based on the second distribution and the corresponding components of the hot wire and the cold wire further includes:

[0028] The speed of the hot wire fluctuates within a first preset threshold range.

[0029] As a further preferred embodiment, the part is divided based on the mechanical performance requirement distribution map, including:

[0030] The first division is performed based on the part forming direction to obtain the first division result;

[0031] Based on the mechanical performance requirement distribution map, the first division result is further divided to obtain part sub-parts.

[0032] As a further preferred embodiment, the first division result is further divided based on the mechanical performance requirement distribution map to obtain part sub-parts, and the method further includes:

[0033] When the size of the sub-part of the component is less than the second preset threshold, it is merged with the nearest neighbor sub-part, or it is merged with the sub-part that has the closest mechanical performance requirements.

[0034] As a further preferred embodiment, determining the pulse correlation parameters corresponding to the hot filament based on the first distribution further includes:

[0035] Determine the threshold value for parameter jump amplitude;

[0036] When the pulse correlation parameter between adjacent sub-parts exceeds the parameter jump amplitude, the arc process parameter is corrected by averaging.

[0037] According to a second aspect of the present invention, a titanium alloy texture control system based on hot-cold wire composite is provided, comprising a simulation module, a part division module, a first distribution acquisition module, a pulse parameter determination module, a wire feeding speed curve acquisition module, and a part forming module, wherein:

[0038] The simulation module is used to simulate the digital model of titanium alloy parts under preset conditions to obtain a distribution map of mechanical property requirements in different spatial regions of the digital model of the parts.

[0039] The parts division module is used to divide the parts based on the mechanical performance requirement distribution map to obtain several sub-parts corresponding to the parts.

[0040] The first distribution acquisition module is used to acquire the first distribution in any sub-part based on mechanical performance requirements. The first distribution represents the grain size distribution in the sub-part of the part.

[0041] The pulse parameter determination module is used to determine the pulse-related parameters corresponding to the hot wire based on the first distribution. The pulse-related parameters include at least one of pulse frequency, duty cycle, pulse peak value, and pulse base value.

[0042] The wire feeding speed curve acquisition module is used to acquire the wire feeding speed curve in the sub-section, which includes the hot wire feeding speed curve and the cold wire feeding speed curve;

[0043] The part forming module is used to determine the weld bead forming size based on the wire feeding speed curve and the pulse correlation parameters, slice the sub-part based on the weld bead forming size, print each slice based on this, and stack the layers to obtain the printed part.

[0044] According to a third aspect of the present invention, a titanium alloy texture control device based on hot-cold wire composite is provided, comprising a processor for executing the above-described titanium alloy texture control method based on hot-cold wire composite.

[0045] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages:

[0046] 1. This invention achieves forced transition of cold wire / hot wire metal through pulse correlation parameters corresponding to the hot wire, and achieves oscillation of liquid metal in the molten pool through pulse correlation parameters, so as to regulate the crystal forming process in the liquid metal (i.e., regulate the grain size), and then realizes the regulation of the mechanical properties of the part based on the coupling relationship between grain size and mechanical properties.

[0047] 2. This invention employs a multi-step convolutional neural network approach, enabling the training of different convolutional layers to be performed in parallel via multiple threads. The training process can also be conducted on different host machines, significantly reducing the time required for the training process.

[0048] 3. This invention combines existing knowledge in the field of metallurgy to obtain a pre-trained model of a multi-step convolutional neural network. Using a pre-trained model allows for the reuse of existing metallurgical knowledge in the multi-step convolutional neural network, significantly reducing the number of training iterations and computational load in subsequent training processes. Furthermore, it reduces the required number of samples for subsequent training, alleviating the cost of sample calibration. After pre-training, the multi-step convolutional neural network is further trained using the jointly trained sample set, resulting in a more accurate model. Attached Figure Description

[0049] Figure 1 This is a system block diagram of a titanium alloy texture control system based on a hot-cold wire composite, as shown in some embodiments of the present invention;

[0050] Figure 2 This is a schematic flowchart illustrating a titanium alloy texture control method based on hot-cold wire composite according to some embodiments of the present invention;

[0051] Figure 3 This is a schematic diagram of the grain size change of titanium alloy based on pulse frequency variation in hot-cold wire composite according to some embodiments of the present invention;

[0052] Figure 4 This is a schematic diagram of the metallographic structure and grain size distribution corresponding to the grain size measurement results when additive manufacturing is performed at different frequencies, as shown in some embodiments of the present invention.

[0053] Figure 5 Images (a) to (j) are the original data and metallographic images of grain size measurement under different pulse effects according to some embodiments of the present invention;

[0054] Figure 6 The diagram shows the results of room temperature tensile tests at different pulse frequencies according to some embodiments of the present invention.

[0055] Figure 7 This is a schematic flowchart illustrating a multi-step convolutional neural network training process according to some embodiments of the present invention;

[0056] Figure 8 This is a schematic diagram of a multi-step convolutional neural network architecture according to some embodiments of the present invention;

[0057] Figure 9 This is a schematic diagram illustrating the data transmission of a convolutional neural network according to some embodiments of the present invention.

[0058] In all the accompanying drawings, the same reference numerals are used to denote the same elements or structures, wherein: 110-Titanium alloy part digital model acquisition module, 120-Simulation module, 130-Part division module, 140-First distribution acquisition module, 150-Pulse parameter determination module, 160-Wire feeding speed curve acquisition module, 170-Part forming module, 310-Slice surface, 320-Sub-part. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0060] Figure 1 This is an exemplary system block diagram of a titanium alloy texture control system based on a hot-cold wire composite, as shown in some embodiments of this specification.

[0061] like Figure 1 As shown, system 100 may include: a titanium alloy part digital model acquisition module 110, a simulation module 120, a part division module 130, a first distribution acquisition module 140, a pulse parameter determination module 150, a wire feeding speed curve acquisition module 160, and a part forming module 170, wherein:

[0062] Titanium alloy parts digital model acquisition module 110 is used to acquire titanium alloy parts digital models;

[0063] The simulation module 120 is used to simulate the digital model of the titanium alloy part under preset conditions to obtain a distribution map of mechanical performance requirements in different spatial regions of the digital model of the part.

[0064] Part division module 130 is used to divide parts based on the mechanical performance requirement distribution map to obtain sub-parts corresponding to several parts;

[0065] The first distribution acquisition module 140 is used to acquire a first distribution in any one of the sub-parts, wherein the first distribution represents the grain size distribution in the part sub-part;

[0066] The pulse parameter determination module 150 is used to determine the pulse-related parameters corresponding to the hot wire based on the first distribution. The pulse-related parameters include at least one of pulse frequency, duty cycle, pulse peak value, and pulse base value.

[0067] The wire feeding speed curve acquisition module 160 is used to acquire the wire feeding speed curve in the sub-part, which includes the hot wire feeding speed curve and the cold wire feeding speed curve.

[0068] The part forming module 170 is used to determine the weld bead forming size based on the wire feeding speed curve and the pulse related parameters, slice the sub-part based on the weld bead forming size, print each slice based on this, and stack the slices to obtain the printed part.

[0069] In some embodiments, the wire feeding speed curve acquisition module 160 is further configured to: acquire a second distribution in the sub-part, the second distribution representing the component distribution in the part sub-part; acquire the corresponding components of the hot wire and the cold wire; determine a matching relationship between the hot wire speed and the cold wire speed based on the second distribution and the corresponding components of the hot wire and the cold wire; and determine the wire feeding speed curve in the sub-part based on the matching relationship. In some embodiments, the hot wire speed fluctuates within a first preset threshold range.

[0070] In some embodiments, a compositional transition region exists at an adjacent location in any two adjacent sub-parts, wherein the accumulated metal composition of the compositional transition region changes gradient between the compositions corresponding to the two sub-parts.

[0071] In some embodiments, the preset conditions include at least one of the following: a preset load cycle change curve, a preset temperature curve, and a preset service acid or alkaline environment.

[0072] In some embodiments, the part division module 130 is further configured to perform a first division based on the part forming direction to obtain a first division result; and to further divide the first division result based on the mechanical performance requirement distribution map to obtain the part sub-parts.

[0073] In some embodiments, the part division module 130 is further configured to merge the part sub-part with the nearest neighbor sub-part when the size of the part sub-part is less than a second preset threshold, or to merge it with the sub-part with the closest mechanical performance requirements.

[0074] In some embodiments, the pulse parameter determination module 150 is further configured to determine a parameter jump amplitude threshold; when the pulse-related parameters between adjacent sub-parts exceed the parameter jump amplitude, the arc process parameters are corrected by averaging.

[0075] It should be understood that the systems and modules described in one or more embodiments of this specification can be implemented in various ways. For example, in some embodiments, the systems and modules can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the methods and systems described above can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The systems and modules described in this specification can be implemented not only by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., but also by software, for example, executed by various types of processors, or by a combination of the aforementioned hardware circuits and software (e.g., firmware).

[0076] It should be noted that the above description of the processing device and its modules is for convenience only and should not be construed as limiting this specification to the scope of the embodiments described. It is understood that those skilled in the art, after understanding the principles of this system, may arbitrarily combine the various modules or construct subsystems connected to other modules without departing from these principles.

[0077] Figure 2 This is a schematic flowchart illustrating a method for controlling the texture of titanium alloys based on a hot-cold wire composite according to some embodiments of this specification. In some embodiments, method 200 may be further executed by system 100.

[0078] Step 210: Obtain the digital model of the titanium alloy part.

[0079] In some embodiments, step 210 may be performed by the titanium alloy part digital model acquisition module 110.

[0080] It is understood that the digital model to be printed can be any part shape that is not subject to printing constraints in this field, and this specification does not impose any restrictions on it.

[0081] Step 220 (i.e. step S1) involves simulating the titanium alloy part model under preset conditions to obtain a distribution map of mechanical performance requirements in different spatial regions of the part model.

[0082] In some embodiments, step 220 may be performed by the simulation module 120.

[0083] like Figure 3 The diagram shown is a typical simulation result of a titanium alloy digital model to be printed. The simulation module 120 can simulate the stress state of the digital model under preset conditions using simulation software such as Ansys, Solidworks, and Deform, to obtain the stress distribution of the digital model. It is understood that simulation techniques are well-known to those skilled in the art, and will not be elaborated upon here.

[0084] In some embodiments, the preset conditions include, but are not limited to, at least one of a preset load cycle variation curve, a preset temperature curve, and a preset service acid / alkali environment. It can be understood that the preset load cycle variation curve refers to the periodic changing tensile and compressive cyclic load variation experienced by the component, such as the alternating stress caused by wind force and vibration during flight climb on an aircraft wing. The preset temperature curve refers to the cyclic temperature change experienced by the component during operation, such as the repeated thermal cycling process experienced by an engine piston rod during operation. The preset service acid / alkali environment refers to the corrosiveness of the environment in which the component operates, such as pH value, biological corrosion, atmospheric corrosion, etc.

[0085] In one or more embodiments described herein, for the purpose of simplification, the following are combined with Figure 3 The following is a detailed explanation of the subsequent steps for the titanium alloy digital model to be printed. It is understood that, as... Figure 3 The titanium alloy part to be printed shown is only an example and is not intended to limit the scope of protection of one or more embodiments of this specification.

[0086] Step 230 (i.e. step S2): Based on the mechanical performance requirement distribution map, the parts are divided to obtain several sub-parts corresponding to the parts.

[0087] In some embodiments, step 230 may be performed by the part division module 130.

[0088] like Figure 3The figure shows the stress distribution diagram of the part model obtained after simulating the stress state of the part under preset loads using Ansys simulation software. In the figure, the stress distribution is displayed in different shades, with each 50 MPa increment. Furthermore, the part partitioning module 130 can divide the part into several sub-parts based on the displayed stress distribution diagram. A sub-part can be understood as a part region with roughly the same stress state. In other words, if the stress at a certain point on the simulated part model is within the threshold range of a sub-part, then that point can be identified as belonging to that sub-part.

[0089] In some embodiments, the threshold range of a sub-part can be determined variably by a process technician. For example, the threshold range of each sub-part can be uniform, such as the threshold range of each sub-region being 50 MPa, 30 MPa, 20 MPa, etc. Alternatively, the threshold range of each sub-region can be variably set, such as the threshold range of some sub-regions being 50 MPa, while the threshold range of another sub-region is 20 MPa.

[0090] In one or more embodiments of this specification, a first sub-region may be defined as the stress range of 700 MPa to 720 MPa, a second sub-region may be defined as the stress range of 720 MPa to 750 MPa, a third sub-region may be defined as the stress range of 750 MPa to 800 MPa, and so on.

[0091] In some embodiments, when the printed part is complex, due to the unique nature of arc additive manufacturing, it is difficult to form the part along a single forming direction. In this scenario, the part segmentation module 130 can determine the forming direction of the part based on the normal vector direction of the part, and perform a first segmentation based on the forming direction of the part to obtain part blocks corresponding to different forming directions (i.e., the first segmentation result). For example, for a part such as... Figure 3 As for the part shown, it is clear that the part cannot be formed along one direction, so the first division result can be decomposed into large parts in the first direction and the second direction.

[0092] Furthermore, the part partitioning module 130 can couple the obtained mechanical performance requirement distribution map to further partition the first partitioning result obtained above, obtaining part sub-parts. Specifically, for Figure 3For the part shown, even different parts in the same direction may have different mechanical property requirements. In this case, the part division module 130 can further divide the large part based on the different mechanical property requirements to obtain sub-parts with different mechanical property requirements, and then use different process parameters to print different sub-parts. It can be understood that further dividing the first division result allows different process parameters to be used to print individual large parts (i.e., the first division result) in a one-time molding process, without the need for additional layering and slicing processing.

[0093] In some embodiments, because the sub-parts of the divided part are too small, arc additive manufacturing makes slicing difficult or molding difficult to control. In this case, the sub-parts of the part can be screened based on a second preset threshold, and sub-parts with a volume smaller than the preset second threshold can be merged with other sub-parts. For example, as... Figure 3 The sub-part 320 shown is very small and can be merged with the adjacent parts.

[0094] For ease of explanation, the sub-parts of the component whose size is smaller than the second preset threshold (e.g.) Figure 3 The sub-part 320 shown is referred to as the target sub-part. In some embodiments, when merging target sub-parts, priority is given to merging the target sub-part with the sub-part whose mechanical performance requirements are closest to its target sub-part. In some embodiments, a merging evaluation function P can be established to determine which sub-part (such as any sub-part i) the target sub-part is merged with. Exemplarily, the merging evaluation function P can be calculated using the following formula:

[0095]

[0096] In the above formula, and This represents the hyperparameter, which is between 0 and 1. This represents the distance between the shape centers of the target sub-part and other sub-parts. This represents the degree of difference in mechanical performance requirements between the target sub-part and other sub-parts. Based on the above formula, the minimum value of the merging evaluation function P can be determined for different merging methods, and the corresponding sub-part to be merged with the target sub-part can be determined based on this minimum value.

[0097] In some embodiments, the merging of sub-parts can also be adaptively processed using machine learning.

[0098] Step S3: For any sub-section, obtain the wire feeding speed curve and pulse-related parameters of the sub-section, including steps 240, 250 and 260.

[0099] Step 240: Obtain the first distribution in the sub-part.

[0100] In some embodiments, step 240 may be performed by the first distribution acquisition module 140.

[0101] The first distribution represents the grain size distribution in a sub-part of the part. The first distribution acquisition module 140 can determine the first distribution of grain size corresponding to the sub-part obtained from the mechanical performance requirement distribution map. Because titanium alloys have very low thermal conductivity, their grains tend to grow along the stacking direction during arc additive manufacturing, forming large longitudinal columnar crystals. These longitudinally growing columnar crystals significantly affect the microstructure and properties of titanium alloy arc additive manufacturing. Therefore, in one or more embodiments of this specification, it is necessary to use process adjustment methods (such as adjusting pulse parameters) to break up the columnar crystals of the titanium alloy, thereby controlling the mechanical properties through the control of the columnar crystal size.

[0102] The first distribution acquisition module 140 can obtain the correspondence between grain size distribution and mechanical properties based on the mechanical property database, and determine the grain size distribution (i.e., the first distribution) in the first sub-part based on this.

[0103] Step 250: Determine the pulse correlation parameters corresponding to the hot wire based on the first distribution.

[0104] In some embodiments, step 250 may be performed by the pulse parameter determination module 150.

[0105] It is understandable that during arc additive manufacturing, the metal in the molten pool is stirred under the action of the arc force, resulting in energy fluctuations and structural fluctuations, which alters the solidification sequence of the liquid metal. Therefore, the frequency of the arc force (i.e., the stirring frequency of the molten pool) and the magnitude of the force (i.e., the stirring amplitude of the molten pool) are not the same when the pulse-related parameters are different, resulting in different grain sizes of the corresponding metal. The pulse-related parameters include at least one of pulse frequency, duty cycle, pulse peak value, and pulse base value. Among them, the pulse frequency can affect the frequency of the arc force, thereby affecting the oscillation frequency of the molten pool under the action of the arc; while the duty cycle can affect the average value of the arc force, thereby affecting the amplitude of the molten pool fluctuation; the difference between the pulse peak value and the pulse base value determines the magnitude of the arc force. The pulse parameter determination module 150 can determine the pulse-related parameters (such as the pulse frequency, duty cycle, pulse peak value, pulse base value, etc. mentioned above) corresponding to the hot wire through the first distribution determined in step 240, thereby realizing the grain size distribution in the solidified metal. It should be noted that among all pulse-related parameters, pulse frequency has the greatest impact on the distribution of grain size. Therefore, in one or more embodiments of this specification, pulse frequency is used as an example to further illustrate the influence of pulse-related parameters on grain size.

[0106] In some embodiments, the pulse parameter determination module 150 can determine a parameter jump amplitude threshold. When the pulse-related parameters between adjacent sub-parts exceed the parameter jump amplitude, the pulse parameter determination module 150 can use averaging to correct the arc process parameters. For example, the pulse frequency in a certain region is 50 Hz, and the pulse frequency in an adjacent region is 80 Hz. If the pulse frequency changes directly from 50 Hz to 80 Hz, arc instability may occur. To avoid this, the pulse parameter determination module 150 can set a parameter jump amplitude threshold (e.g., a threshold of 5 Hz). When the pulse frequency in an adjacent region is greater than 5 Hz, the pulse frequency is gradually changed using averaging, so that the pulse frequency gradually changes from 50 Hz to 80 Hz.

[0107] For information on the effect of pulse frequency on grain size, please refer to [link / reference]. Figure 3 , Figure 4 , Figure 5 The corresponding descriptions will not be repeated here.

[0108] Step 260: Obtain the wire feeding speed curve in the sub-section.

[0109] In some embodiments, step 260 may be performed by the wire feed speed curve acquisition module 160.

[0110] The wire feeding speed curve acquisition module 160 can further determine the wire feeding speed curve in each sub-part based on the pulse correlation parameters corresponding to the hot wire obtained in step 250. The wire feeding speed curve includes the hot wire feeding speed curve and the cold wire feeding speed curve.

[0111] In some embodiments, the metal composition in the sub-parts can be different. In this scenario, the cold wire speed curve and the hot wire speed curve can be matched anisotropically to obtain the stacked metal composition with different components. Specifically, the wire feed speed curve acquisition module 160 can acquire a second distribution in the sub-part, which represents the composition distribution in the part sub-part; based on acquiring the corresponding compositions of the hot and cold wires, the matching relationship between the hot and cold wire speeds is determined, and the wire feed speed curve in the sub-part is determined based on the matching relationship. For example, the room temperature tensile properties of Ti6Al4V, Ti6Al2V, Ti3Al3V, Ti4Al1V, Ti5Al1.5V, Ti6.5Al3.5Mo, and Ti6.5Al2Zr are 850 MPa, 800 MPa, 720 MPa, 750 MPa, 776 MPa, 801 MPa, and 887 MPa, respectively. Furthermore, the hot-cold wire composition determination module 130 acquires the upper and lower limits of the stress distribution of the entire part. If the upper limit of the tensile strength of the entire part is 841 MPa and the lower limit is 724 MPa, then the composition of the entire part can be set as an alloy composed of Ti-xAl-yV, where x represents the content of aluminum alloy in the deposited metal and y represents the content of vanadium alloy in the deposited metal. x and y can be changed in any proportion during the deposition process.

[0112] For specific illustration, we can assume the hot wire composition is Ti-2Al and the cold wire composition is Al-45V. The wire feed speed curve acquisition module 160 can then determine the speed ratio between the hot and cold wires based on the proportion of V in the deposited metal. For example, if the deposited metal is Ti-6Al-4V, the speed ratio between the hot and cold wires can be determined to be 10.25, i.e., the corresponding first ratio is 10.25. Preferably, the cold wire composition can be achieved using aluminum-based flux-cored welding wire to ensure that the alloying elements in the intercalation reach the required proportion.

[0113] Since the hot wire feed speed significantly affects the arc stability in the arc additive manufacturing process, in one or more embodiments of this specification, the hot wire speed should fluctuate within a first preset threshold range (the first preset threshold range may be 10%) to ensure the stability of the printing process.

[0114] Step 270 (i.e. step S4): Determine the weld bead forming size based on the wire feed speed curve and the pulse correlation parameters; slice the sub-part based on the weld bead forming size; print each slice based on this; and stack the slices layer by layer to obtain the printed part.

[0115] In some embodiments, step 270 may be performed by the part printing module 170.

[0116] The part printing module 170 can perform additive manufacturing of the current slice layer based on the process parameter change curve obtained in step 260.

[0117] Figure 3 This is a schematic diagram illustrating the grain size change of a titanium alloy based on pulse frequency variation in some embodiments of this specification.

[0118] When the hot filament is Ti-2Al, the cold filament is Al-45V, and the speed ratio between the hot and cold filaments is 10.25, the composition of the deposited metal is Ti-6Al-4V. The grain size measurement results for additive manufacturing using different frequencies corresponding to the hot filaments are shown in the table below, along with the corresponding metallographic structure and grain size distribution diagrams. Figure 4 As shown, the raw data for grain size measurement are as follows: Figure 5 As shown in (a)~(j).

[0119] Table 1. Grain size measurement results of transverse samples

[0120]

[0121] in, Figure 4 The box plot shown illustrates the grain size distribution, with the box range defined as 25%–75% and the range from the lower edge to the upper edge as 5%–95%. The box plot visually reflects the grain size concentration and, more intuitively, the influence of pulse frequency on grain size. Figure 4 It can be seen that when the pulse frequency is 0Hz, the area of ​​the enclosure is relatively large, the distance between the top and bottom edges is relatively large, and the data between 25% and 75% are concentrated in 26×10. -3 ~101×10 -3 mm 2 This indicates that the grain size is relatively dispersed at this point. At a pulse frequency of 50Hz, the box area is small, the distance between the top and bottom edges is small, and 25% to 75% of the data is concentrated in the 14×102 range. -3 ~43×10 - 3 mm 2 The range indicates that the grain size is relatively concentrated, and the grain refinement effect is good. At a pulse frequency of 80Hz, the area of ​​the enclosure and the distance between the top and bottom edges are both large, with 25% to 75% of the data concentrated at 32×10⁻⁶. -3 ~199×10 -3 mm 2 The range indicates that the grain size is relatively dispersed, resulting in poor grain refinement. At a pulse frequency of 110Hz, the box area is small, the distance between the top and bottom edges is small, and the data between 25% and 75% are concentrated around 25×10⁻⁶. -3 ~59×10 -3 mm 2The range indicates that the grain size is relatively concentrated, and the grain refinement effect is good. At a pulse frequency of 160Hz, the area of ​​the enclosure and the distance between its top and bottom edges are both small, with 25% to 75% of the data concentrated at 17×10⁻⁶. -3 ~53×10 -3 mm 2 The range indicates that the grain size is relatively concentrated at this point, and the grain refinement effect is good.

[0122] The reason for this is that at a pulse frequency of 0Hz, the droplet transfer is a short-circuit transfer. At this time, the surface tension of the molten metal and the shear force of the arc are the main driving forces for the flow of molten metal in the molten pool, acting on the entire upper surface of the pool. The droplet transfer is relatively gentle, the flow velocity of the molten metal in the pool is low, and the range of motion is small, resulting in weak stirring within the pool and minimal impact on grain growth, thus leading to large grain sizes. At a pulse frequency of 50Hz, the droplet transfer is a large-droplet transfer. During the time the droplet detaches from the welding wire tip but has not yet entered the molten pool, it is subjected to gravity and electromagnetic forces, further increasing its velocity. Moreover, due to the large mass of the droplet, the impact and stirring effect on the molten pool are significant, resulting in strong stirring and a high flow velocity of the molten metal, leading to fine grains. Subsequently, as the pulse frequency increases, the droplet size decreases, and the impact on the molten pool decreases. Although the number of molten droplets increases, the increase is limited. Due to the insufficient impact force, the stirring effect on the molten pool is small, and the flow velocity of the molten metal in the pool is low, resulting in an increase in the sample grain size. Subsequently, as the pulse frequency continues to increase, the droplet transition mode changes from droplet transition to jet droplet transition. Although the impact force on the molten pool continues to decrease, the droplet drop frequency is extremely high, thus enhancing the stirring effect and the grain refinement effect.

[0123] Furthermore, in order to couple the relationship between pulse frequency, grain size, and mechanical properties, such as Figure 6 The diagram shows the results of room temperature tensile tests at different pulse frequencies. Figure 6It can be seen that the tensile strength and yield strength show the same trend with frequency. Both reach their maximum at a pulse frequency of 50 Hz and their minimum at 80 Hz. The values ​​are not significantly different at pulse frequencies of 110 Hz and 160 Hz. Combining this with the results in Chapter 4, the trends in tensile strength and yield strength are similar to those in grain size variation. Therefore, it is inferred that the mechanical properties of titanium alloy components manufactured by arc additive manufacturing are controlled by grain size. The degree of grain refinement varies depending on the pulse frequency. The grain refinement effect is optimal at 50 Hz, resulting in the best mechanical properties. The grains are largest at 80 Hz, leading to the worst mechanical properties. The average grain size is not significantly different at 110 Hz and 160 Hz, resulting in similar mechanical properties.

[0124] Figure 7 This is a schematic flowchart illustrating an exemplary multi-step convolutional neural network training process according to some embodiments of this specification.

[0125] Step 710: Obtain the initial model of the convolutional neural network.

[0126] The initial model of a convolutional neural network includes a first part and a second part, wherein the first and second parts of the initial model include at least a number of convolutional layers; the structure of the convolutional neural network can be found in [reference needed]. Figure 8 As shown, it will not be elaborated further here.

[0127] The initial model of a convolutional neural network includes initialized model parameters and a complete model structure. In some embodiments, the initial model can be an untrained convolutional neural network or a convolutional neural network that has not been fully trained. Each layer of the initial model can be set with initial parameters, which can be continuously adjusted during training until training is complete.

[0128] In some embodiments, a pre-trained model of a multi-step convolutional neural network can be obtained based on statistical methods combined with existing knowledge in the field of metallurgy. Specifically, a small number of specific samples can be selected statistically using a quadratic general rotation combination experiment or an orthogonal experiment, and then a pre-trained model of a multi-step convolutional neural network with lower accuracy can be obtained based on this small number of specific samples. It is understood that the knowledge in this field can qualitatively explain the correspondence between pulse correlation parameters, grain size, and mechanical properties, that is, it can reflect the convergence direction of model training. By adopting the above-mentioned method of obtaining the pre-trained model, on the one hand, existing metallurgical knowledge can be reused in multi-step convolutional neural networks, greatly reducing the number of training iterations and computational load in the subsequent training process; on the other hand, it can reduce the number of samples required in the subsequent training process and alleviate the cost of sample calibration.

[0129] Step 720: Obtain the first training sample set and the second training sample set.

[0130] The first training sample set includes multiple samples with different grain size distributions under different pulse correlation parameters, and the second training sample set includes multiple measured mechanical property samples with different grain size distributions. In some embodiments, the samples with different grain size distributions under different pulse correlation parameters and the measured mechanical property samples with different grain size distributions can be manually labeled. In some alternative embodiments, both the samples with different grain size distributions under different pulse correlation parameters and the measured mechanical property samples with different grain size distributions can be labeled with experimentally measured values. For example, Figure 5 The method shown is used for labeling.

[0131] Step 730: Perform multiple rounds of training on the first part and the second part of the initial model based on the first training sample set to obtain the pre-trained first and second parts of the initial model.

[0132] The initial model can perform forward propagation based on one or a series of samples from the first training sample set to obtain the predicted shape value H1 of the intersection region, and based on the labels of the first training samples. Constructing the loss function = Then, backpropagation is performed to obtain the correction values ​​(or gradients) of the model parameters of each layer, including multiple matrix elements (such as gradient elements), which correspond one-to-one with the model parameters. Each gradient element reflects the direction (increase or decrease) and amount of correction of the parameters.

[0133] For ease of understanding, Figure 9 The model shown is an example to illustrate one round of training. It includes three convolutional layers, with a total of six convolutional kernels. The operations at each convolutional kernel are similar to those at kernel 6, and the forward propagation process of the convolutional neural network can be described using the following two formulas:

[0134] (1)

[0135] (2)

[0136] in, The activation function representing the convolution kernel Input data, This represents the output of the convolution kernel. For the convolution kernel of the model's output layer, It can be the model's prediction result on training samples or the object to be predicted; the subscript n or m represents the index of the convolution kernel. The set of indices representing the sequence numbers of the convolution kernels preceding kernel n, in _____. Figure 9 For example, convolution kernel 4 receives the outputs of convolution kernel 1, convolution kernel 2, and convolution kernel 3. . This represents the weights that map convolution kernel m to convolution kernel n. is the constant term corresponding to the convolution kernel n. Where, as well as The model parameters that make up the convolutional neural network model can be obtained through training.

[0137] Through forward propagation, the feature data of the training samples can be processed layer by layer through each convolutional layer of the convolutional neural network model to obtain the prediction results.

[0138] The backpropagation algorithm compares the prediction results of a specific training sample with the labeled data to determine the update magnitude of each weight in the network. In other words, the backpropagation algorithm is used to determine how the loss function changes relative to each weight (also known as the gradient or error derivative), denoted as... .

[0139] by Figure 9 Taking an exemplary neural network model as an example, firstly, the gradient of the loss function value relative to the output of convolution kernel 6 can be calculated. When the loss function is the mean squared error loss function hour, ,in For the predicted results, This is the labeled data. Subsequently, the weights of the loss function relative to convolution kernel 6 and convolution kernel 5 can be calculated using the chain rule. The gradient of the output and the gradient of the loss function value relative to the output of convolution kernel 5. :

[0140] (3)

[0141] = (4)

[0142] (5)

[0143] By analogy, the gradient of the loss function value with respect to each weight can be calculated one by one.

[0144] Given the above process, it is possible to base the loss function value on... Backpropagation of gradients is performed until the gradient of the loss function value relative to each element in the initial output matrix is ​​calculated. The model is then updated based on these gradients, thus completing one round of model updates. This algorithm is used for multiple iterations until the model converges or its performance metrics meet a threshold requirement, resulting in the first and second parts of the trained convolutional neural network.

[0145] In some embodiments, the decision to proceed to the next iteration or to determine the final trained model can be based on the difference in performance and the sample labels. The criteria for this decision may include whether the preset number of iterations has been reached, whether the updated model meets a preset performance threshold, or whether a termination instruction has been received. If it is determined that the next iteration is necessary, it can be performed based on the updated model from the current iteration. If it is determined that the next iteration is not necessary, the updated model obtained during the current iteration can be used as the final trained model.

[0146] Step 740: Obtain the joint training sample set.

[0147] The joint training sample set includes measured mechanical performance samples under various pulse-correlation parameters. Its main function is to further train the multi-step convolutional neural network (CNN) after it has been pre-trained, resulting in a more accurate model.

[0148] Step 750: Based on the joint training sample set, perform joint training on the first and second parts of the pre-trained initial model to obtain a trained multi-step convolutional neural network model.

[0149] The first and second parts of the pre-trained initial model can be concatenated and then forward propagated through joint training samples to obtain the predicted value H, based on the labels of the joint training samples. Constructing the loss function = Then, backpropagation is performed to obtain the corrected values ​​(or gradients) of the model parameters at each layer. These values ​​include multiple matrix elements (such as gradient elements), each corresponding one-to-one with a model parameter. Each gradient element reflects the direction (increase or decrease) and amount of parameter correction. For more information on forward and backward propagation, please refer to [link to documentation / reference]. Figure 9 The corresponding description will not be repeated here.

[0150] If the performance of the resulting multi-step convolutional neural network model is found to be poor or the convergence consistency is not good during joint training, steps 730 and 740 can be returned to perform more precise training of the first and second parts of the initial model. This will improve the performance of the pre-trained model before proceeding to the joint training in step 750.

[0151] It is understandable that by using a multi-step convolutional neural network, the training process of different convolutional layers can be carried out in parallel using multiple threads, and the training process can also be performed on different host machines, which greatly reduces the time required for the training process.

[0152] In some embodiments, the third part of the multi-step convolutional neural network trained in step 340 can be trained and used independently. In this scenario, the third part of the multi-step convolutional neural network can be pre-trained and determined, and then the first and second parts of the multi-step convolutional neural network are trained in reverse. For example, when the multi-step convolutional neural network is jointly trained, the third part is not updated at all, and the reverse operation result of the model can be directly passed to the first and second parts of the model, thereby achieving joint training to obtain a more accurate multi-step convolutional neural network model.

[0153] Computer storage media may contain a propagated data signal containing computer program code, for example, on baseband or as part of a carrier wave. This propagated signal may take various forms, including electromagnetic, optical, and suitable combinations thereof. Computer storage media can be any computer-readable medium other than a computer-readable storage medium, which can be connected to an instruction execution system, apparatus, or device to enable communication, propagation, or transmission of a program for use. The program code located on the computer storage medium can be propagated through any suitable medium, including radio, cable, fiber optic cable, RF, or similar media, or any combination of the above media.

[0154] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of range in some embodiments of this specification are approximate values, in specific embodiments, such values ​​are set as accurately as feasible.

[0155] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for controlling the texture of titanium alloys based on hot-cold wire composite, characterized in that, Includes the following steps: S1. Simulate the titanium alloy part model under preset conditions to obtain the distribution map of mechanical performance requirements in different spatial regions of the part model; S2. Based on the mechanical performance requirement distribution map, the parts are divided to obtain several sub-parts corresponding to the parts; S3. For any sub-part The first distribution in the sub-part is obtained based on mechanical performance requirements, and the first distribution represents the grain size distribution in the sub-part of the part. Based on the first distribution, the pulse correlation parameters corresponding to the hot wire are determined, and the pulse correlation parameters include at least one of pulse frequency, duty cycle, pulse peak value, and pulse base value; Obtain the wire feeding speed curve in the sub-section, the wire feeding speed curve including the hot wire feeding speed curve and the cold wire feeding speed curve; Obtain the wire feed speed curve in the sub-section, including: Obtain a second distribution in the sub-part, the second distribution representing the component distribution in the part sub-part; obtain the corresponding components of the hot wire and the cold wire; determine the matching relationship between the hot wire speed and the cold wire speed based on the second distribution and the corresponding components of the hot wire and the cold wire; determine the wire feeding speed curve in the sub-part based on the matching relationship. S4. Determine the weld bead forming size based on the wire feeding speed curve and the pulse correlation parameters. Slice the sub-parts based on the weld bead forming size. Print each slice based on this, and stack the slices layer by layer to obtain the printed part.

2. The titanium alloy texture control method based on hot-cold wire composite as described in claim 1, characterized in that, Based on mechanical performance requirements, the first distribution in the sub-part is obtained; based on the first distribution, the pulse correlation parameters corresponding to the hot filament are determined; and the process further includes: The mapping relationship between mechanical properties, the first distribution, and impulse correlation parameters is obtained based on a multi-step convolutional neural network. The multi-step convolutional neural network comprises two parts: the first part converts the pulse correlation parameters corresponding to the material into a grain size distribution matrix; the second part converts the grain size distribution matrix into the corresponding predicted mechanical properties.

3. The titanium alloy texture control method based on hot-cold wire composite as described in claim 2, characterized in that, The multi-step convolutional neural network is trained based on the following steps: Obtain an initial model of a convolutional neural network; the initial model of the convolutional neural network includes a first part and a second part, wherein the first part includes at least a number of convolutional layers, and the second part includes at least a number of convolutional layers and a fully connected layer. Obtain a first training sample set and a second training sample set. The first training sample set includes multiple samples with different grain size distributions under different pulse correlation parameters, and the second training sample set includes multiple measured mechanical property samples with different grain size distributions. Multiple rounds of training are performed on the first training sample set and the second training sample set respectively to obtain the first and second parts of the pre-trained initial model. Obtain a joint training sample set, which includes multiple measured mechanical performance samples under different pulse correlation parameters; Based on the joint training sample set, the first and second parts of the pre-trained initial model are jointly trained to obtain a trained multi-step convolutional neural network model.

4. The titanium alloy texture control method based on hot-cold wire composite as described in claim 1, characterized in that, Determining the matching relationship between the hot wire speed and the cold wire speed based on the second distribution and the corresponding components of the hot wire and cold wire further includes: The speed of the hot wire fluctuates within a first preset threshold range.

5. The titanium alloy texture control method based on hot-cold wire composite as described in claim 1, characterized in that, Parts are divided based on the aforementioned mechanical performance requirement distribution diagram, including: The first division is performed based on the part forming direction to obtain the first division result; Based on the mechanical performance requirement distribution map, the first division result is further divided to obtain part sub-parts.

6. The titanium alloy texture control method based on hot-cold wire composite as described in claim 5, characterized in that, The first division result is further divided based on the mechanical performance requirement distribution map to obtain part sub-parts, which also include: When the size of the sub-part of the component is less than the second preset threshold, it is merged with the nearest neighbor sub-part, or it is merged with the sub-part that has the closest mechanical performance requirements.

7. The titanium alloy texture control method based on hot-cold wire composite as described in claim 1, characterized in that, Based on the first distribution, the pulse correlation parameters corresponding to the hot filament are determined, and the method further includes: Determine the threshold value for parameter jump amplitude; When the pulse correlation parameter between adjacent sub-parts exceeds the parameter jump amplitude, the pulse correlation parameter is corrected by averaging.

8. A titanium alloy texture control system based on hot-cold wire composite for implementing the titanium alloy texture control method based on hot-cold wire composite as described in any one of claims 1-7, characterized in that, It includes a simulation module, a part division module, a first distribution acquisition module, a pulse parameter determination module, a wire feed speed curve acquisition module, and a part forming module, wherein: The simulation module is used to simulate the digital model of titanium alloy parts under preset conditions to obtain a distribution map of mechanical property requirements in different spatial regions of the digital model of the parts. The parts division module is used to divide the parts based on the mechanical performance requirement distribution map to obtain several sub-parts corresponding to the parts. The first distribution acquisition module is used to acquire the first distribution in any sub-part based on mechanical performance requirements. The first distribution represents the grain size distribution in the sub-part of the part. The pulse parameter determination module is used to determine the pulse-related parameters corresponding to the hot wire based on the first distribution. The pulse-related parameters include at least one of pulse frequency, duty cycle, pulse peak value, and pulse base value. The wire feeding speed curve acquisition module is used to acquire the wire feeding speed curve in the sub-section, which includes the hot wire feeding speed curve and the cold wire feeding speed curve; The part forming module is used to determine the weld bead forming size based on the wire feeding speed curve and the pulse correlation parameters, slice the sub-part based on the weld bead forming size, print each slice based on this, and stack the layers to obtain the printed part.

9. A titanium alloy texture control device based on hot-cold wire composite, characterized in that, Includes a processor for executing the titanium alloy texture control method based on hot-cold wire composite as described in any one of claims 1-7.

Images