Direct energy deposition (DED) process neural network online monitoring system and method based on time sequence data set

Abstract

The invention discloses a time series data set-based direct energy deposition (DED) process neural network online monitoring system and method, and relates to the technical field of additive manufacturing, and the method comprises the steps: obtaining a time series data set in an additive manufacturing process; inputting the time sequence data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model; determining a defect prediction result based on the temperature change trend information; and when the defect prediction result shows that defects will appear, the defect prediction result is input into a preset parameter optimization model, technological parameters output by the preset parameter optimization model are obtained, and real-time additive manufacturing is carried out based on the technological parameters. Compared with an existing mode that the process parameters are adjusted in a warehouse when it is detected that the product has defects, the process state can be accurately controlled, the process parameters are adjusted in advance, the fluctuation degree of the additive manufacturing state is reduced, and the influence of heat accumulation is reduced.

Description

Technical Field

[0001] The present application relates to the field of additive manufacturing technology, and in particular to a neural network online monitoring system and method for a direct energy deposition (DED) process based on a time series data set. Background Art

[0002] Direct Energy Deposition (DED) technology is gradually being applied to additive manufacturing (3D printing) and repair and remanufacturing of metal parts, especially Direct Laser Deposition (DLD), also known as Laser Cladding, which is widely used. However, since this type of process is a very complex metallurgical process with many influencing factors, the process is prone to fluctuations and abnormal mutations, which can cause the process to not proceed smoothly or cause defects. Therefore, it is very necessary to monitor the DED process online. In the DED process, due to the existence of heat accumulation, the molten pool temperature may rise rapidly, the molten pool size may change rapidly, and the control upper limit may be exceeded. The existing method is to monitor the additive manufacturing process in real time, and adjust the process parameters in time when the molten pool temperature rises rapidly or the product has defects. However, it is difficult to quickly change the molten pool temperature and size to a suitable range by hastily adjusting the process parameters. In addition, the adjustment of process parameters with too large a range will also cause fluctuations in the process state. Therefore, how to control the process parameters and improve the efficiency of additive manufacturing has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The main purpose of this application is to provide a direct energy deposition (DED) process neural network online monitoring system and method based on a time series data set, aiming to solve the technical problem of low additive manufacturing efficiency caused by lagging process parameter control during additive manufacturing.

[0004] To achieve the above objectives, the present application proposes a neural network online monitoring method for a direct energy deposition (DED) process based on a time series data set, the method comprising:

[0005] Obtaining time series datasets of additive manufacturing processes;

[0006] Inputting the time series data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model;

[0007] Determine a defect prediction result based on the temperature change trend information;

[0008] When the defect prediction result indicates that a defect is about to occur, the defect prediction result is input into a preset parameter optimization model to obtain process parameters output by the preset parameter optimization model, and real-time additive manufacturing is performed based on the process parameters.

[0009] Optionally, the step of determining a defect prediction result based on the temperature change trend information includes:

[0010] Obtain material information for additive manufacturing;

[0011] Preprocessing the temperature change trend information to obtain target temperature change trend information;

[0012] The target temperature change trend information and the material information are input into a preset defect prediction model to obtain a defect prediction result output by the preset defect prediction model.

[0013] Optionally, the preset defect prediction model includes a convolutional layer and a dense layer:

[0014] The convolution layer includes a one-dimensional convolution layer, a normalization layer, a maximum pooling layer, an activation function, and a global average pooling layer;

[0015] The dense layer is used to infer the relationship between the input value and the target value;

[0016] The preset defect prediction model also includes a dropout layer, and the dropout layer is used to prevent overfitting;

[0017] The preset defect prediction model also includes a dense layer containing a Sigmoid activation function, and the dense layer containing the Sigmoid activation function is used to output a defect prediction result.

[0018] Optionally, when the defect prediction result indicates that a defect is about to occur, the step of inputting the defect prediction result into a preset parameter optimization model to obtain the process parameters output by the preset parameter optimization model includes:

[0019] Determining a defect probability according to the defect prediction result;

[0020] determining whether a defect is about to occur based on the defect probability;

[0021] If it is determined based on the defect probability that a defect will occur, the defect prediction result, the time series data set and the temperature change trend information are input into a preset parameter optimization model to obtain the process parameters output by the preset parameter optimization model.

[0022] Optionally, before the step of obtaining a time series data set in the additive manufacturing process, the method further includes:

[0023] Collecting samples formed by additive manufacturing and production parameters of the samples;

[0024] detecting the sample to obtain defect information of the sample;

[0025] An initial neural network model is trained according to the defect information and the production parameters to obtain a preset temperature trend prediction model.

[0026] Optionally, the process parameters include a laser power adjustment amount:

[0027] The step of performing real-time additive manufacturing based on the process parameters comprises:

[0028] The laser power is adjusted according to the laser power adjustment amount to perform real-time additive manufacturing.

[0029] In addition, to achieve the above purpose, the present application also proposes a direct energy deposition (DED) process neural network online monitoring system based on a time series data set, the system comprising:

[0030] An acquisition module, used to obtain time series data sets during the additive manufacturing process;

[0031] A temperature change trend prediction module, used for inputting the time series data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model;

[0032] A defect prediction result determination module, used to determine a defect prediction result based on the temperature change trend information;

[0033] The additive manufacturing module is used to input the defect prediction result into a preset parameter optimization model when the defect prediction result indicates that a defect is about to occur, obtain the process parameters output by the preset parameter optimization model, and perform real-time additive manufacturing based on the process parameters.

[0034] The present application obtains a time series data set in the additive manufacturing process; inputs the time series data set into a preset temperature trend prediction model to obtain the temperature change trend information output by the preset temperature trend prediction model; determines a defect prediction result based on the temperature change trend information; when the defect prediction result indicates that a defect is about to occur, inputs the defect prediction result into a preset parameter optimization model to obtain the process parameters output by the preset parameter optimization model, and performs real-time additive manufacturing based on the process parameters. Compared with the existing method of hastily adjusting process parameters when defects are detected in a product, the above method of the present application can accurately control the process state, adjust the process parameters in advance, reduce the degree of fluctuation of the additive manufacturing state, and reduce the impact of heat accumulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0037] Figure 1 A schematic diagram of a process flow of a first embodiment of a method for online monitoring of a direct energy deposition (DED) process by a neural network based on a time series data set of the present application;

[0038] Figure 2 An overview diagram of the informer model architecture provided in Example 1 of the method for online monitoring of a direct energy deposition (DED) process by a neural network based on a time series data set of the present application;

[0039] Figure 3 A flow chart of Embodiment 2 of the method for online monitoring of a direct energy deposition (DED) process by a neural network based on a time series data set provided in the present application;

[0040] Figure 4 A schematic diagram of a preset defect prediction model architecture provided in Example 2 of a method for online monitoring of a direct energy deposition (DED) process by a neural network based on a time series data set of the present application;

[0041] Figure 5 A schematic diagram of process parameter prediction provided for Example 2 of a neural network online monitoring method for a direct energy deposition (DED) process based on a time series data set of the present application;

[0042] Figure 6 A schematic diagram of the overall process of online monitoring of a direct energy deposition (DED) process by a neural network based on a time series data set provided in Example 2 of the method for online monitoring of a direct energy deposition (DED) process by a neural network based on a time series data set of the present application;

[0043] Figure 7 This is a schematic diagram of the module structure of a neural network online monitoring system for a direct energy deposition (DED) process based on a time series data set according to an embodiment of the present application;

[0044] Figure 8 Schematic diagram of the equipment structure of the hardware operating environment involved in the direct energy deposition (DED) process neural network online monitoring method based on a time series data set in an embodiment of the present application.

[0045] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0046] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0047] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0048] The main solution of the embodiment of the present application is: obtaining a time series data set in the additive manufacturing process; inputting the time series data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model; determining a defect prediction result based on the temperature change trend information; when the defect prediction result is that a defect is about to occur, inputting the defect prediction result into a preset parameter optimization model to obtain process parameters output by the preset parameter optimization model, and performing real-time additive manufacturing based on the process parameters.

[0049] Compared with the existing method of hastily adjusting process parameters when defects are detected in products, the above-mentioned method of the present application can accurately control the process state, adjust the process parameters in advance, reduce the fluctuation degree of the additive manufacturing state, and reduce the impact of heat accumulation.

[0050] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a process parameter control device, etc. The following takes the process parameter control device as an example to illustrate this embodiment and the following embodiments.

[0051] Based on this, the embodiment of the present application provides a neural network online monitoring method for direct energy deposition DED process based on a time series data set, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the neural network online monitoring method for direct energy deposition (DED) process based on time series data set of the present application.

[0052] In this embodiment, the neural network online monitoring method for direct energy deposition (DED) process based on time series data set includes steps S10 to S40:

[0053] Step S10, obtaining a time series data set in the additive manufacturing process;

[0054] It should be noted that the time series data set may include part identity parameters, material parameters, structural parameter data, equipment parameters, process parameters, detection data, and heating time (laser scanning time), which can be expressed in terms of position coordinates and layer sequence numbers. The process parameters may include: laser power, laser spot diameter, laser scanning speed, powder feeding rate, powder carrier gas flow rate, scanning mode, scanning interval, and also include information such as powder particle size and its distribution, and powder spheroidization rate. The time series data set is a data set that characterizes the above data as a data set with a time series as the axis. In order to more intuitively represent the above data, only the data that changes with time in the above data can be characterized as a data set with a time series as the axis, and other data can be recorded statically.

[0055] Step S20, inputting the time series data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model;

[0056] It should be noted that the preset temperature trend prediction model can be a neural network model that is pre-trained with sample data and can predict the future molten pool temperature trend based on a time series data set. The temperature change trend information is the future molten pool temperature trend predicted by the preset temperature trend prediction model, which is characterized by a temperature sequence with time as the axis, and can be displayed in conjunction with the time series data set.

[0057] Furthermore, in order to improve the efficiency of additive manufacturing, the present embodiment uses the Transformer model to train a preset temperature trend prediction model for predicting temperature change trend information, hereinafter referred to as a variant Informer model or Informer model. Figure 2 , Figure 2 The schematic diagram of the informer model architecture overview provided in the first embodiment of the method for online monitoring of the direct energy deposition (DED) process neural network based on a time series data set in this application is that the informer model in this embodiment inherits the attention mechanism in the Transformer. The attention mechanism uses the retrieval vector Q (query vector), key vector K (key vector), and value vector V (value vector) between each position in the sequence to calculate the attention required for each position. This enables the model to effectively grasp the relationship between each position in the sequence. This is also an important reason why the accuracy of the informer model is better than that of the traditional LSTM model. In addition, the Info rmer model uses ProbSparse self-attention as the calculation formula for attention A (Q, K, V), which can be specifically referred to the following formula:

[0058]

[0059] Where d is the dimension of the Key vector. Softmax is a normalized exponential function that converts each element of the input vector into its form as a probability value. Its output is an N×1 vector, x i is the i-th element of the input vector, i = 1, 2, ... N, j = 1, 2, ... N. ProbSparse self-attention only uses the top-ranked search vectors of the density value to form the search matrix It reduces the computational requirements while maintaining the original model effectiveness and makes the temperature trend module smoother in real-time operation.

[0060] The Informer model for temperature trend prediction uses the following MSE loss function for model training:

[0061]

[0062] Among them, MSE is used to characterize the loss function used for model training, Y i is the true probability, is the average prediction probability predicted by the model, and n is the number of samples. This embodiment uses the time series data set collected in real time as the input of the model, and the input is the predicted time series temperature data after being encoded and decoded by the model, that is, the temperature change trend information (future temperature fluctuations and trends). This temperature data will be used for subsequent defect prediction.

[0063] Reference Figure 2 , the encoder input of the Informer model is a string of size (L x ,d model ) of the sequence matrix X en , where L x is the sequence length, d model is the number of features, i.e., the features corresponding to the data in the time series dataset. First, a one-dimensional convolutional layer with a filter size of 3 is used to transform X en The dimension is reduced to a one-dimensional scalar vector u. Then the fixed position embedding is calculated, and the calculation formula is:

[0064]

[0065] Among them, the formula And pos is the position code.

[0066] After the timestamp embedding SE representing the timestamp is obtained through training, the vector X of the input compiler is obtained by substituting it into the following formula: feed .

[0067]

[0068] Here, α is used to balance the size between scalar and embedding. If the input is normalized, α is set to 1.

[0069] After X feed Will be like Figure 2 Through multiple multi-head ProSparse self-attention modules and self-attention purification modules. The self-attention purification module uses a one-dimensional convolutional layer to further reduce the size of the neural network and reduce the difficulty of training while extracting the main attention. At the end, the feature maps in the multi-head self-attention are merged as the output of the encoder. The decoder of the Informer model adopts the decoder structure of the standard transformer model, but it additionally applies generative inference to optimize the prediction.

[0070] Generative reasoning allows the decoder to use generative input, specifically extracting L from the original input sequence token Long start symbol sequence X token Then with length L y The zero sequence X0 that serves as a placeholder for the prediction sequence is merged, L y is the length of the final output sequence. For example, assuming that one second is a unit, extract 5 seconds before the target sequence as the starting symbol, and then merge it with a zero sequence with a scalar value of zero but containing timestamp information and a length of 20. The total length of the input sequence is 25, and the output sequence length is still 20. The merged expression of the decoder input sequence is:

[0071]

[0072] in, express After input, Figure 2 This process is repeated twice with residual connections through a masked multi-head ProSparse self-attention module and a multi-head self-attention module and some feed-forward networks omitted in the figure. After that, the value can directly obtain the output sequence after passing through the fully connected layer. The output sequence (i.e., the length of the output sequence) is Figure 2 Shown as the red part.

[0073] Step S30, determining a defect prediction result based on the temperature change trend information;

[0074] It should be noted that the defect prediction result determined based on the temperature change trend information may be determined based on the predicted future temperature in the temperature change trend information to determine whether the additive manufacturing product will have defects. Specifically, the manufacturing temperature range of the additive manufacturing product may be determined based on past experience, and it may be determined whether the predicted future temperature is within the manufacturing temperature range. If not, it is determined that the additive manufacturing product will have defects. Otherwise, it is determined that the additive manufacturing product will not have defects.

[0075] Step S40, when the defect prediction result indicates that a defect is about to occur, the defect prediction result is input into a preset parameter optimization model to obtain process parameters output by the preset parameter optimization model, and real-time additive manufacturing is performed based on the process parameters.

[0076] It should be noted that the preset parameter optimization model can be a neural network model obtained by pre-training with sample data, which can predict the next process parameters according to the temperature change trend information corresponding to the defect prediction result and the time series data set in the additive manufacturing process. The process parameters may include laser power, laser spot diameter, laser scanning speed, molten pool temperature, etc.

[0077] This embodiment obtains a time series data set in the additive manufacturing process; inputs the time series data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model; determines a defect prediction result based on the temperature change trend information; when the defect prediction result indicates that a defect is about to occur, inputs the defect prediction result into a preset parameter optimization model to obtain process parameters output by the preset parameter optimization model, and performs real-time additive manufacturing based on the process parameters. Compared with the existing method of hastily adjusting process parameters when defects are detected in a product, the above method of this embodiment can accurately control the process state, adjust the process parameters in advance, reduce the degree of fluctuation of the additive manufacturing state, and reduce the impact of heat accumulation.

[0078] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 3 , Figure 3 A flow chart of Embodiment 2 of the method for online monitoring of a direct energy deposition (DED) process by a neural network based on a time series data set is provided in this application; the step S30 further includes steps S301 to S303:

[0079] Step S301: Acquire material information for additive manufacturing;

[0080] It should be noted that the material information may include information such as the material name, material melting point, type, etc. corresponding to the product undergoing additive manufacturing.

[0081] Step S302: preprocessing the temperature change trend information to obtain target temperature change trend information;

[0082] It should be noted that the preprocessing of the temperature change trend information may be normalizing and denoising the temperature change trend information to obtain target temperature change trend information.

[0083] Step S303: inputting the target temperature change trend information and the material information into a preset defect prediction model to obtain a defect prediction result output by the preset defect prediction model.

[0084] It should be noted that the preset defect prediction model can be a neural network model pre-trained with sample data for predicting whether an additive manufacturing product will have defects. The defect prediction result includes the probability that the additive manufacturing product will have defects predicted by the preset defect prediction model.

[0085] Furthermore, in order to accurately identify whether an additive manufacturing product is about to have defects, the preset defect prediction model includes a convolutional layer and a dense layer:

[0086] The convolution layer includes a one-dimensional convolution layer, a normalization layer, a maximum pooling layer, an activation function, and a global average pooling layer;

[0087] The dense layer is used to infer the relationship between the input value and the target value;

[0088] The preset defect prediction model also includes a dropout layer, and the dropout layer is used to prevent overfitting;

[0089] The preset defect prediction model also includes a dense layer containing a Sigmoid activation function, and the dense layer containing the Sigmoid activation function is used to output a defect prediction result.

[0090] It should be noted that, please refer to Figure 4 , Figure 4A schematic diagram of the architecture of a preset defect prediction model provided for the second embodiment of the method for online monitoring of a neural network for a direct energy deposition (DED) process based on a time series data set in this application. In the additive manufacturing process, the preset defect prediction model receives the future temperature trend predicted by the preset temperature trend prediction model as input (the temperature change trend information can be spliced ​​with the time series data set as the overall input), and outputs a prediction of whether the future print (i.e., the additive manufacturing product) is defective. If the prediction result is defective, the defect prediction module (a module containing a preset defect prediction model) will provide the parameter optimization module (a module containing a preset parameter optimization model) with the neural network model parameters (i.e., the model parameters of the preset defect prediction model), the current input value (i.e., the temperature change trend information), and the loss function (the model training loss function corresponding to the preset defect prediction model) for subsequent parameter optimization.

[0091] It should be noted that the preset defect prediction model is a neural network model that predicts defects through temperature change trend information and time series data sets. This neural network is used to establish the relationship between temperature and defects and determine whether the predicted temperature is within an appropriate range. Figure 4 As shown in Figure 1, the neural network input is a matrix of (S, n), where S is the sequence length and n is the total number of features. After the input, it first passes through three convolution modules, namely the convolution layer in the figure. The convolution module (see Figure 4 The convolutional layer architecture in () is composed of a one-dimensional convolutional layer, a normalization layer (normalization layer in the model training stage, batch normalization layer in the model use process), maximum pooling, activation function, and global average pooling layer. The one-dimensional convolutional layer can extract the parameter relationship in adjacent time periods. The convolution kernel size f of the convolutional layer is progressively 2, 3, and 4 in the three convolution modules, and the number of convolution kernels is set to 30. Therefore, the input of size (S,n) will become the output of (S-f+1,30) after passing through. The (batch) normalization layer maintains the size of the input and has the function of preventing overfitting and gradient disappearance. The subsequent activation function uses the relu activation function: f(x)=max(0,x). The global average pooling layer can retain the sequence space information while reducing the input space dimension. After entering the global pooling layer, a vector of length 30 will be output. Then through a dense layer ( Figure 4 Not shown, located Figure 4 The convolutional layer 3 and the Dropout layer are used to infer the relationship between the input and the target value, and a Dropout layer is used to prevent overfitting. Finally, a dense layer with a Sigmoid activation function is used to calculate the defect probability σ(x). The Sigmoid function calculation formula is:

[0092]

[0093] Among them, σ(x) is used to represent the defect probability, and x is used to represent the input. The defect probability threshold can be preset to 0.5. If the probability of Sigmoid output is greater than 0.5, it is considered defective, and if it is less than or equal to 0.5, it is considered non-defective.

[0094] The loss function of the preset defect prediction model is the binary cross entropy loss function (Binary Crossentro pyLoss), and the calculation formula is:

[0095]

[0096] Where N is the number of samples, y i is the true label of the ith sample (i.e., binary label 0 or 1), p(y i ) is the output belonging to y i The predicted probability of the label.

[0097] Further, in order to accurately determine the process parameters to be adjusted and improve the additive manufacturing efficiency, the step S40 may include: determining the defect probability according to the defect prediction result;

[0098] determining whether a defect is about to occur based on the defect probability;

[0099] If it is determined based on the defect probability that a defect will occur, the defect prediction result, the time series data set and the temperature change trend information are input into a preset parameter optimization model to obtain the process parameters output by the preset parameter optimization model.

[0100] It should be noted that a defect probability threshold can be preset, for example, 0.5, to determine the defect probability in the defect prediction result. If the defect probability is greater than 0.5, it is classified as a defect that will occur, and if it is less than or equal to 0.5, it is classified as a defect that will not occur for the time being. The process of using the preset parameter optimization model to predict process parameters can be referred to Figure 5 , Figure 5 A schematic diagram of process parameter prediction provided in Example 2 of the direct energy deposition (DED) process neural network online monitoring method based on a time series data set of this application, Figure 5 The power change prediction neural network in is the preset parameter optimization model, and the process parameter output by the preset parameter optimization model is the laser power adjustment amount. Figure 5As shown in the figure, before the start of additive manufacturing, the power change prediction neural network is trained offline using the previous sample data set, and its input is the current temperature, target temperature, heating (laser scanning) time and current laser power. The target temperature is the average temperature value obtained by the defect prediction neural network after the input temperature sequence, that is, the temperature change trend information, is optimized in sequence, and the output is the laser power adjustment amount. In the additive manufacturing process, the binary cross entropy loss function of the neural network in the defect prediction module is used to optimize the input, that is, train the defect prediction neural network to optimize the temperature sequence, that is, the temperature change trend information, and obtain the optimized temperature time series and target temperature. Specifically: ① The defect prediction neural network sets the temperature sequence as a variable and sets other parameters in the neural network as constants. The back propagation algorithm is used to perform gradient descent on the controllable parameters. ② The temperature sequence is gradient descended α times, α is greater than or equal to 1, and the size of α is the same as the learning rate of the gradient descent to determine the fluctuation amplitude of the parameter change during the printing process. ③ Take the average value of the temperature sequence and use it as the target temperature of the power change prediction neural network. Combined with the initial temperature, heating time, and initial laser power, the laser power adjustment amount for optimization is predicted.

[0101] Furthermore, in order to improve the efficiency of additive manufacturing, before step S10, the method further includes: collecting samples formed by additive manufacturing and production parameters of the samples;

[0102] detecting the sample to obtain defect information of the sample;

[0103] An initial neural network model is trained according to the defect information and the production parameters to obtain a preset temperature trend prediction model.

[0104] It should be noted that the collection of additively manufactured samples and the production parameters of the samples, and the detection of defect information of the samples can be for parts of typical materials and typical structures, and samples formed by additive manufacturing are collected, the samples are detected, and the locations and types of defects are recorded. The process parameters and adjustment amounts of each position (heating time) of the parts including the defect position during the laser scanning process of the additive manufacturing process, and the molten pool temperature (initial temperature and adjustment target temperature) are traced back to form a time series data set, which is used as a sample for offline training of a preset temperature trend prediction model, a preset defect prediction model and a preset parameter optimization model. The preset parameter optimization model includes a defect prediction neural network and a power change prediction neural network, wherein the defect prediction neural network is a model obtained by training the model parameters of the preset defect prediction model using a binary cross entropy loss function.

[0105] In the specific implementation, we first take zero defects as the goal, and use the sample data set of the previous samples with the same material and the closest structure to train the power change prediction neural network offline. Its input is the initial temperature, the adjusted target temperature, the heating (laser scanning) time and the initial laser power. The initial laser power adjustment amount is determined, and then combined with the initial process parameters, the optimized process parameters are processed to perform additive manufacturing of parts. In the additive manufacturing process of parts, the online monitoring of parameter information such as the molten pool temperature forms a time series data set, which is used as the input of the preset temperature trend prediction model. After the calculation and prediction of each module, the process parameters are adjusted in real time to control the additive manufacturing process. At the same time, the process parameters, molten pool temperature and other data after the laser power adjustment are collected to obtain a new time series data set, which is used as the input of the preset temperature trend prediction model and can also be used for offline training of the preset temperature trend prediction model. After the parts are additively manufactured, they can also be properly processed to analyze and detect their defects. The defective and non-defective positions are traced back to obtain the corresponding process parameters and molten pool temperature information, and the new defect time series data samples are obtained, which can be used for offline training of the preset defect prediction model.

[0106] In the specific implementation, please refer to Figure 6 , Figure 6The overall flow diagram of the online monitoring of the direct energy deposition DED process neural network based on the time series data set provided by the embodiment 2 of the online monitoring method of the direct energy deposition DED process neural network based on the time series data set of the present application; in the process of DED additive manufacturing, the process parameter control device detects the temperature data of the molten pool formed by the continuous laser scanning online and forms a time series data set with the corresponding process parameters and other data, inputs the temperature trend prediction module, and predicts the temperature change trend in real time through the preset temperature trend prediction model in the temperature trend prediction module, which is characterized as a temperature sequence with time as the axis, that is, the temperature change trend information, and the temperature change trend information is pre-processed (normalized, denoising, etc.) and combined with other features (material information of additive manufacturing) and then input into the preset defect prediction model in the defect prediction module, predicting whether a defect is about to occur and inputting the prediction result into the parameter optimization module, if a defect is about to occur, the online parameter optimization module outputs a controllable process parameter adjustment amount, that is, a laser power adjustment amount, sends a control signal to the device, and adjusts the controllable process parameters in time, so that the molten pool temperature is always controlled within the corresponding appropriate temperature range. Then, the new temperature data obtained by the subsequent online detection and other real-time data form a time series data set and input into the temperature trend prediction module to carry out the next round of prediction control cycle. This embodiment has three neural network modules that work in a series loop. The temperature change trend output by the temperature trend prediction module is characterized as a temperature time series data set as the input of the preset defect prediction model, and the defect probability time series data set output by the preset defect prediction model is used as the input of the parameter optimization module. The process parameters optimized by the online parameter optimization module will form a new time series data set together with the corresponding temperature data, and will be used as the input of the temperature trend prediction module for the next round of prediction control cycle.

[0107] This embodiment obtains material information for additive manufacturing; pre-processes the temperature change trend information to obtain target temperature change trend information; inputs the target temperature change trend information and the material information into a preset defect prediction model to obtain a defect prediction result output by the preset defect prediction model. This embodiment predicts the upcoming temperature change trend by online detection of a series of data such as temperature, i.e., a time series data set, and identifies whether a defect is about to occur through a preset defect prediction model. When a defect is predicted to occur, the process parameters are pre-controlled in advance through a preset parameter optimization model to improve the quality of DED additive manufacturing of parts and process stability.

[0108] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the present application's method for online monitoring of the direct energy deposition (DED) process using a neural network based on a time series data set. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0109] This application also provides a direct energy deposition DED process neural network online monitoring system based on a time series data set, please refer to Figure 7 The direct energy deposition (DED) process neural network online monitoring system based on time series data set includes:

[0110] An acquisition module 10 is used to acquire a time series data set in an additive manufacturing process;

[0111] The temperature change trend prediction module 20 is used to input the time series data set into a preset temperature trend prediction model to obtain the temperature change trend information output by the preset temperature trend prediction model;

[0112] A defect prediction result determination module 30, used to determine a defect prediction result based on the temperature change trend information;

[0113] The additive manufacturing module 40 is used to input the defect prediction result into a preset parameter optimization model when the defect prediction result indicates that a defect is about to occur, obtain the process parameters output by the preset parameter optimization model, and perform real-time additive manufacturing based on the process parameters.

[0114] This embodiment obtains a time series data set in the additive manufacturing process; inputs the time series data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model; determines a defect prediction result based on the temperature change trend information; when the defect prediction result indicates that a defect is about to occur, inputs the defect prediction result into a preset parameter optimization model to obtain process parameters output by the preset parameter optimization model, and performs real-time additive manufacturing based on the process parameters. Compared with the existing method of hastily adjusting process parameters when defects are detected in a product, the above method of this embodiment can accurately control the process state, adjust the process parameters in advance, reduce the degree of fluctuation of the additive manufacturing state, and reduce the impact of heat accumulation.

[0115] The direct energy deposition DED process neural network online monitoring system based on a time series data set provided by the present application adopts the direct energy deposition DED process neural network online monitoring method based on a time series data set in the above embodiment, which can solve the technical problem that the process parameter control lags during the additive manufacturing process, resulting in low additive manufacturing efficiency. Compared with the prior art, the beneficial effects of the direct energy deposition DED process neural network online monitoring system based on a time series data set provided by the present application are the same as the beneficial effects of the direct energy deposition DED process neural network online monitoring method based on a time series data set provided by the above embodiment, and other technical features of the direct energy deposition DED process neural network online monitoring system based on a time series data set are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0116] The present application provides a direct energy deposition (DED) process neural network online monitoring device based on a time series data set, and the direct energy deposition (DED) process neural network online monitoring device based on a time series data set includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the direct energy deposition (DED) process neural network online monitoring method based on the time series data set in the above-mentioned embodiment one.

[0117] Reference below Figure 8 , which shows a schematic diagram of the structure of a neural network online monitoring device for a direct energy deposition DED process based on a time series data set suitable for implementing an embodiment of the present application. The neural network online monitoring device for a direct energy deposition DED process based on a time series data set in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions: tablet computers), PMPs (Portable Media Players: portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 8 The neural network online monitoring device for the direct energy deposition (DED) process based on a time series data set shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0118] like Figure 8As shown, the direct energy deposition DED process neural network online monitoring device based on the time series data set may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the direct energy deposition DED process neural network online monitoring device based on the time series data set are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the direct energy deposition DED process neural network online monitoring device based on a time series data set to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a direct energy deposition DED process neural network online monitoring device based on a time series data set with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.

[0119] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0120] The direct energy deposition DED process neural network online monitoring device based on a time series data set provided by the present application adopts the direct energy deposition DED process neural network online monitoring method based on a time series data set in the above embodiment, which can solve the technical problem that the process parameter control lags during the additive manufacturing process, resulting in low additive manufacturing efficiency. Compared with the prior art, the beneficial effects of the direct energy deposition DED process neural network online monitoring device based on a time series data set provided by the present application are the same as the beneficial effects of the direct energy deposition DED process neural network online monitoring method based on a time series data set provided by the above embodiment, and the other technical features of the direct energy deposition DED process neural network online monitoring device based on a time series data set are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0121] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0122] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0123] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the direct energy deposition (DED) process neural network online monitoring method based on a time series data set in the above-mentioned embodiment.

[0124] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0125] The above-mentioned computer-readable storage medium can be included in the direct energy deposition (DED) process neural network online monitoring device based on a time series data set; or it can exist independently without being assembled into the direct energy deposition (DED) process neural network online monitoring device based on a time series data set.

[0126] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the direct energy deposition DED process neural network online monitoring device based on the time series data set, the direct energy deposition DED process neural network online monitoring device based on the time series data set executes the above-mentioned direct energy deposition DED process neural network online monitoring method based on the time series data set.

[0127] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0128] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0129] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0130] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned method for online monitoring of a neural network for a direct energy deposition (DED) process based on a time series data set, and can solve the technical problem that the process parameter control lags during the additive manufacturing process, resulting in low additive manufacturing efficiency. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the method for online monitoring of a neural network for a direct energy deposition (DED) process based on a time series data set provided in the above-mentioned embodiment, and will not be repeated here.

[0131] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for online monitoring of a direct energy deposition (DED) process by a neural network based on a time series data set.

[0132] The computer program product provided by the present application can solve the technical problem that the process parameter control lags in the additive manufacturing process, resulting in low additive manufacturing efficiency. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the direct energy deposition DED process neural network online monitoring method based on the time series data set provided by the above embodiment, and will not be repeated here.

[0133] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A neural network online monitoring method for direct energy deposition (DED) process based on time series data set, characterized in that: The method comprises the following steps: Obtaining time series datasets of additive manufacturing processes; Inputting the time series data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model; Determine a defect prediction result based on the temperature change trend information; When the defect prediction result indicates that a defect is about to occur, the defect prediction result is input into a preset parameter optimization model to obtain process parameters output by the preset parameter optimization model, and real-time additive manufacturing is performed based on the process parameters.

2. The method according to claim 1, characterized in that The step of determining the defect prediction result based on the temperature change trend information comprises: Obtain material information for additive manufacturing; Preprocessing the temperature change trend information to obtain target temperature change trend information; The target temperature change trend information and the material information are input into a preset defect prediction model to obtain a defect prediction result output by the preset defect prediction model.

3. The method according to claim 2, characterized in that The preset defect prediction model includes a convolutional layer and a dense layer: The convolution layer includes a one-dimensional convolution layer, a normalization layer, a maximum pooling layer, an activation function, and a global average pooling layer; The dense layer is used to infer the relationship between the input value and the target value; The preset defect prediction model also includes a dropout layer, and the dropout layer is used to prevent overfitting; The preset defect prediction model also includes a dense layer containing a Sigmoid activation function, and the dense layer containing the Sigmoid activation function is used to output a defect prediction result.

4. The method according to claim 1, characterized in that When the defect prediction result indicates that a defect is about to occur, the step of inputting the defect prediction result into a preset parameter optimization model to obtain the process parameters output by the preset parameter optimization model comprises: Determining a defect probability according to the defect prediction result; determining whether a defect is about to occur based on the defect probability; If it is determined based on the defect probability that a defect will occur, the defect prediction result, the time series data set and the temperature change trend information are input into a preset parameter optimization model to obtain the process parameters output by the preset parameter optimization model.

5. The method according to any one of claims 1 to 4, characterized in that: Before the step of obtaining the time series data set in the additive manufacturing process, the method further includes: Collecting samples formed by additive manufacturing and production parameters of the samples; detecting the sample to obtain defect information of the sample; An initial neural network model is trained according to the defect information and the production parameters to obtain a preset temperature trend prediction model.

6. The method according to any one of claims 1 to 4, characterized in that: The process parameters include laser power adjustment: The step of performing real-time additive manufacturing based on the process parameters comprises: The laser power is adjusted according to the laser power adjustment amount to perform real-time additive manufacturing.

7. A neural network online monitoring system for direct energy deposition (DED) process based on time series data set, characterized in that: The system comprises: An acquisition module, used to obtain time series data sets during the additive manufacturing process; A temperature change trend prediction module, used for inputting the time series data set into a preset temperature trend prediction model to obtain temperature change trend information output by the preset temperature trend prediction model; A defect prediction result determination module, used to determine a defect prediction result based on the temperature change trend information; The additive manufacturing module is used to input the defect prediction result into a preset parameter optimization model when the defect prediction result indicates that a defect is about to occur, obtain the process parameters output by the preset parameter optimization model, and perform real-time additive manufacturing based on the process parameters.

Images