A system and method for metal additive manufacturing time prediction based on data fusion

Abstract

The present invention relates to the field of metal additive manufacturing, and in particular to a system and method for predicting metal additive manufacturing time based on data fusion, wherein the system comprises an import extraction module, a processing time prediction module, a preparation time prediction module and a data output module; the import extraction module comprises an import extraction unit and a storage unit, the storage unit is used to store the extracted data, the import extraction unit is used to import training parts or parts to be processed, extract effective feature data of the training parts or parts to be processed, and extract processing parameters of equipment; the processing time prediction module is used to predict the processing time of each layer of the parts to be processed; the preparation time prediction module is used to predict the preparation time of each layer of the parts to be processed; the data output module is used to output the total processing time, total preparation time and overall manufacturing time of the predicted parts. The present invention can quickly and accurately predict the total manufacturing time of the parts to be printed.

Description

Technical Field

[0001] The present invention relates to the field of metal additive manufacturing, and in particular to a system and method for predicting metal additive manufacturing time based on data fusion. Background Art

[0002] Rapid additive manufacturing is a rapid prototyping technology that uses digital models to directly drive the rapid manufacture of three-dimensional physical entities of arbitrary complex shapes. The basic principle of rapid additive manufacturing adopts the manufacturing concept of "layered manufacturing and layer-by-layer stacking", which is different from the traditional "subtractive" manufacturing model and is a green and intelligent "additive" manufacturing.

[0003] Additive manufacturing integrates virtual design and digital manufacturing. Based on the three-dimensional model constructed on the computer, it uses software and numerical control systems to stack special metal materials, non-metallic materials or medical biomaterials layer by layer through extrusion, sintering, melting, photocuring, spraying, etc. to produce physical objects. It can manufacture products in a short time. Compared with traditional machining machine tools and mold manufacturing, it has many advantages such as low cost, short cycle, simple modification and stable size.

[0004] Current additive manufacturing technology generally roughly estimates the manufacturing time of parts based on the volume and height of the parts, and manufactures them according to the estimated time. However, in the actual additive manufacturing process, factors such as the placement of parts, equipment parameters, and time fluctuations of various motion mechanisms will affect the specific time of part manufacturing, resulting in differences between the finished product and the design model. For example, in the metal additive manufacturing process, due to the existence of stress, the processing area needs to be divided into many small areas and irregularly scanned. The time prediction is extremely difficult and the time prediction accuracy is low, resulting in a large difference between the finished product and the design model, a low product qualification rate, and a high production cost. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a system for metal additive manufacturing time prediction based on data fusion, comprising an import and extraction module, a processing time prediction module, a preparation time prediction module and a data output module.

[0006] The import and extraction module includes an import and extraction unit and a storage unit. The storage unit is used to store the extracted data. The import and extraction unit is used to import training parts or parts to be processed, extract valid feature data of training parts or parts to be processed, and extract processing parameters of the equipment.

[0007] The processing time prediction module is used to predict the processing time of each layer of the part to be processed.

[0008] The preparation time prediction module is used to predict the preparation time of each layer of the part to be processed.

[0009] The data output module is used to output the total processing time, total preparation time and overall manufacturing time of the predicted parts.

[0010] According to an embodiment of the present invention, the processing time prediction module includes at least one module training unit, which is used to train the feature data and processing parameters of the currently imported training parts or parts to be processed and obtain the corresponding prediction model, and predict the processing time of each layer of the parts to be processed according to the prediction model.

[0011] According to an embodiment of the present invention, the processing time prediction module includes 2 or more module training units, for example, 2, 3 or 4.

[0012] According to an embodiment of the present invention, different module training units use different methods to train the feature data and processing parameters. The training methods include neural network algorithms and regression algorithms. The regression algorithms include decision tree regression algorithms, SVM regression algorithms, K-nearest neighbor regression algorithms, and random forest regression algorithms.

[0013] According to an embodiment of the present invention, the processing time prediction module includes a first model training unit and a second model training unit.

[0014] According to an embodiment of the present invention, the training method of the first model training unit is a neural network algorithm, which trains the feature data and processing parameters of the currently imported training parts or parts to be processed based on the neural network algorithm to obtain a layer processing time prediction model: the first Model1, the first Model1 is used to predict the processing time T12_n of each layer of the parts to be processed.

[0015] Preferably, the training further includes obtaining the accuracy ACC1 and the mean absolute error MAE1 of the first Model1.

[0016] According to an embodiment of the present invention, the training method of the second model training unit is a regression algorithm, which trains the feature data and processing parameters of the currently imported training parts or parts to be processed based on the regression algorithm to obtain a layer processing time prediction model: the second Model2, and the second Model2 is used to predict the processing time T22_n of each layer of the parts to be processed.

[0017] Preferably, the training further includes obtaining the accuracy ACC2 and the mean absolute error MAE2 of the second Model2.

[0018] According to an embodiment of the present invention, the training method of the second model training unit includes at least one of a decision tree regression algorithm, a SVM regression algorithm, a K-nearest neighbor regression algorithm and a random forest regression algorithm, for example, a random forest regression algorithm.

[0019] According to an embodiment of the present invention, the preparation time prediction module includes a third model training unit, which is used to train the characteristic data and processing parameters of the currently imported training parts or parts to be processed based on a neural network algorithm to obtain a layer preparation time prediction model: a third Model3, and the third Model3 is used to predict the preparation time T32_n of each layer of the parts to be processed.

[0020] According to an embodiment of the present invention, the system also includes a model fusion module, which is used to fuse the processing time of each layer of the part to be processed predicted by the first Model 1 and the processing time of each layer of the part to be processed predicted by the second Model 2 to obtain a fused processing time for each layer.

[0021] According to an embodiment of the present invention, the fusion module calculates the fusion processing time of each layer using Formula 1:

[0022]

[0023] According to an embodiment of the present invention, the system also includes a data output module, which is used to predict the total processing time, total preparation time and overall manufacturing time of the part based on the predicted fusion processing time and preparation time of each layer.

[0024] According to an implementation scheme of the present invention, the effective feature data of the part include but are not limited to the total number of layers of the part, the area corresponding to each layer of the structure, the outer contour circumference, the inner contour circumference, the number of outer contours, the overlap rate, the partition size, the maximum value of the X-direction coordinate of the current layer of the part, the minimum value of the X-direction coordinate of the current layer of the part, the maximum value of the Y-direction coordinate of the current layer of the part, the minimum value of the Y-direction coordinate of the current layer of the part, and other data.

[0025] According to an embodiment of the present invention, the processing parameters of the equipment include but are not limited to the light spot corresponding to the manufacture of each layer of structure, laser off delay, laser on delay, jump delay, marking delay, inflection point delay, scanning speed, jump speed, randomness, contour speed, filling angle and other data.

[0026] According to an embodiment of the present invention, the storage unit is used to store effective feature data and processing parameters corresponding to the number of part layers.

[0027] According to an embodiment of the present invention, the system also includes a feature selection module, which is used to extract effective feature data in each layer and / or feature data in processing parameters whose correlation with processing time is arranged in an order ≥N, where N ≥ 15, preferably N ≥ 10, for example, N is 8, 10 or 12.

[0028] According to an embodiment of the present invention, the feature selection module is also used to extract the effective feature data in each layer and / or feature data in the processing parameters whose correlation with the preparation time is arranged in an order ≥N, where N ≥ 15, preferably N ≥ 10, for example, N is 5, 8, 10 or 15.

[0029] According to an embodiment of the present invention, the total processing time is TR2_total=∑TR2_n, the total preparation time is T32_total=∑T32_n, and the overall manufacturing total time is Time_total=TR2_total+T32_total.

[0030] According to an embodiment of the present invention, the training part is a processed part, and the storage module stores data information of all processed parts.

[0031] The present invention also provides a prediction method based on the above prediction system, comprising the following steps:

[0032] Step 1: Import the training parts and the corresponding additive manufacturing equipment, extract the effective feature data of the parts, and extract the processing parameters of the equipment.

[0033] Step 2: Input the effective feature data and processing parameters of the training parts into the processing time prediction module and the preparation time prediction module to obtain the prediction model.

[0034] Step 3: Input the effective feature data and processing parameters of the part to be trained into the prediction model to obtain the predicted processing time and preparation time.

[0035] According to an embodiment of the present invention, inputting effective feature data and processing parameters into a processing time prediction module to obtain processing time comprises the following steps:

[0036] The effective feature data and processing parameters are respectively input into the first model training unit and the second model training unit for training to obtain the first prediction model and the second prediction model. The effective feature data and processing parameters of the parts to be processed are input into the first prediction model and the second prediction model to obtain the first processing time T12_n and the second processing time T22_n.

[0037] According to an embodiment of the present invention, inputting effective feature data and processing parameters into a preparation time prediction module to obtain preparation time comprises the following steps:

[0038] The effective feature data and processing parameters are respectively input into the third model training unit to obtain the third prediction model, and the effective feature data and processing parameters of the part to be processed are input into the third prediction model to obtain the preparation time T32_n.

[0039] According to an embodiment of the present invention, before step 2, the following steps are also included: using a feature selection module to select valid feature data in each layer and / or feature data in the processing parameters whose correlation with the processing time is arranged in an order ≥ N as input data.

[0040] According to an embodiment of the present invention, before step 2, the following steps are also included: using a feature selection module to select valid feature data in each layer and / or feature data in processing parameters whose correlation with preparation time is arranged in an order ≥ N as input data.

[0041] According to the implementation scheme of the present invention, after step 2 and before step 3, the following steps are also included: using the feature selection module to select the valid feature data in each layer and / or feature data in the processing parameters whose correlation order with the processing time and preparation time is ≥N as input data.

[0042] According to an embodiment of the present invention, after step 3, the following step is also included: integrating the predicted processing time and preparation time to obtain the total processing time.

[0043] Preferably, the predicted total processing time of each layer is TR2_n, and the total processing time is calculated by TR2_n according to Formula 1:

[0044]

[0045] Preferably, the total processing time is TR2_total=∑TR2_n.

[0046] Preferably, the total preparation time is T32_total=∑T32_n, and the overall manufacturing total time is Time_total=TR2_total+T32_total.

[0047] A method for additive manufacturing using the above method comprises the following steps:

[0048] Parts are processed according to the predicted processing time of each layer of the part to be processed, the predicted preparation time of each layer of the part to be processed, the total processing time and the total preparation time.

[0049] According to an embodiment of the present invention, after the part processing is completed, the effective feature data of the part and the actual processing parameters are stored in the storage unit.

[0050] According to the implementation scheme of the present invention, before processing parts, it also includes arranging a suitable placement form according to the structure of the parts to be processed, and inputting the current placement form and equipment parameters into the prediction model according to the placement form.

[0051] The present invention also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the method described is implemented.

[0052] The present invention also provides a device, comprising the storage medium and a processor, wherein the processor implements the steps of the method when executing the computer program on the storage medium.

[0053] Beneficial Effects

[0054] 1. The metal additive manufacturing time prediction method based on data fusion of the present invention establishes an adaptive prediction method, classifies the feature data, uses the characteristics of the parts to be processed and the processing parameters of the equipment as the input of the prediction model, and predicts the processing time and preparation time of each layer of the parts to be processed, so as to quickly and accurately predict the total manufacturing time of the parts to be printed.

[0055] 2. The method of the present invention can quickly predict the processing time and preparation time before the parts are processed. If there are requirements for the manufacturing time, the parameters of the equipment can be modified, and the predicted manufacturing time can be checked at any time using this method to achieve a balance between manufacturing quality and manufacturing time.

[0056] 3. The method of the present invention predicts the layer processing time and the layer preparation time separately, which is convenient for optimizing the processing parameters of the equipment and improving the utilization rate of the equipment.

[0057] 4. The method of the present invention adopts two layer processing time prediction models, and fuses the two prediction models according to the errors of the models to obtain a more accurate and stable fusion model, thereby improving the accuracy of time prediction.

[0058] 5. With the use of additive manufacturing equipment, the method of the present invention can convert the parts that can be actually processed by the present invention into training part data, continuously optimize the existing time prediction model, and thus further improve the accuracy of the prediction; the method of the present invention is not limited by the type of additive manufacturing equipment and can be extended to different equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic flow chart of a method for predicting metal additive manufacturing time based on data fusion according to an embodiment of the present invention;

[0060] Figure 2 It is a structural schematic diagram of a metal additive manufacturing time prediction system based on data fusion according to an embodiment of the present invention;

[0061] Figure 3 Schematic diagram of a training part in Example 2 of the present invention;

[0062] Figure 4 This is a schematic diagram of the model to be predicted in Example 2 of the present invention. DETAILED DESCRIPTION

[0063] The system and method of the present invention will be further described in detail below in conjunction with specific embodiments. It should be understood that the following embodiments are only exemplary descriptions and explanations of the present invention and should not be construed as limiting the scope of protection of the present invention. All technologies implemented based on the above content of the present invention are included in the scope of protection that the present invention is intended to protect.

[0064] Example 1

[0065] like Figure 1 As shown, a metal additive manufacturing time prediction method based on data fusion includes the following steps:

[0066] Step 1: Import the training part and the processing parameters of the additive manufacturing equipment as the first parameters.

[0067] The first feature data that affects the processing time of each layer in the training parts is extracted and saved by layer, and the first parameter that affects the processing time of each layer in the processing parameters of the additive manufacturing equipment is extracted and saved by layer.

[0068] The training parts are parts that have been prepared according to the processing parameters of the additive manufacturing equipment.

[0069] Step 2: Use the first feature data, the first parameter and the actual processing time of each layer as input data and input them into the layer processing time prediction model based on the neural network for training to obtain the layer processing time prediction model Model1 (hereinafter referred to as the first Model), and at the same time obtain the accuracy ACC1 and the mean absolute error MAE1 of the first Model.

[0070] Step 3: Use the first feature data, the first parameter and the actual processing time of each layer as input data and input them into the layer processing time prediction model based on Random Forest Regressor. The trained layer processing time prediction model Model2 (hereinafter referred to as the second Model2) is obtained. At the same time, the accuracy ACC2 and the mean absolute error MAE2 of the second Model2 are obtained.

[0071] Step 4: The first characteristic data, the first parameter and the actual preparation time of each layer are used as input data and input into the layer preparation time prediction model based on the neural network, and the layer preparation time prediction model Model 3 (hereinafter referred to as the third Model 3) is obtained by training.

[0072] Step 5: Import the part 2 to be processed, extract the second characteristic data that affects the processing time in the part 2 to be processed and save it by layer, and extract the second parameter that affects the processing time in the processing parameters of the additive manufacturing equipment and save it by layer.

[0073] Step 6: Input the second characteristic data and the second parameter of each layer as input data into the first Model 1 to obtain the first predicted layer processing time T12_n.

[0074] The second characteristic data and the second parameter of each layer are taken as input data and input into the second Model 2 to obtain the second predicted layer processing time T22_n.

[0075] According to the mean absolute error between the first Model1 and the second Model2, the layer processing time predicted by the two models is fused to obtain the fused time prediction result TR2_n. The fusion is calculated according to formula 1:

[0076]

[0077] Step 7: Input the second characteristic data and the second parameter as input data into the third Model 3 to obtain the predicted layer preparation time T32_n.

[0078] Step 8: Based on the predicted processing time TR2_n and preparation time T32_n of each layer, predict the total processing time, total preparation time and overall manufacturing time of part 2.

[0079] The total processing time is the sum of all predicted layer processing times, and the calculation formula is:

[0080] TR2_total=∑TR2_n

[0081] The total preparation time is the sum of all predicted layer preparation times, calculated as:

[0082] T32_total=∑T32_n

[0083] The total manufacturing time of the parts is:

[0084] Time_total=TR2_total+T32_total.

[0085] Among them, in step 1, extracting the first characteristic data of the training part 1 and extracting the first parameter affecting the processing time from the processing parameters of the additive manufacturing equipment specifically include the following steps:

[0086] S11: Based on the experience of additive manufacturing, feature data that may affect the processing time in training part 1 is collected as the first feature data. The first feature data includes but is not limited to the total number of layers of the part, the area corresponding to each layer of the structure, the outer contour circumference, the inner contour circumference, the number of outer contours, the overlap rate, the partition size, the maximum value of the X-direction coordinate of the current layer of the part, the minimum value of the X-direction coordinate of the current layer of the part, the maximum value of the Y-direction coordinate of the current layer of the part, the minimum value of the Y-direction coordinate of the current layer of the part, and other data. Except for the data showing the overall structure such as the total number of layers of the part, other data are saved by layer.

[0087] S12: Collect parameters in the additive manufacturing equipment that may affect the processing time as the first parameter. The first parameter includes but is not limited to the data of the light spot, laser off delay, laser on delay, jump delay, mark delay, inflection point delay, scanning speed, jump speed, randomness, contour speed, filling angle, etc. corresponding to the manufacturing of each layer of structure. The first parameter is saved by layer.

[0088] S13: The data saved by layer are matched one by one with the actually recorded processing time and preparation time of each layer to obtain a time prediction training set.

[0089] S14: Input the time prediction training set into the hybrid module of the multi-category feature extraction (ReliefF) algorithm and the mutual information (MutualInformation) algorithm for feature selection. The hybrid module obtains the correlation coefficient of each feature through the ReliefF algorithm and the MutualInformation algorithm, removes the features with a correlation coefficient less than the threshold, the preferred threshold is 0.6, and sorts the features according to the size of the correlation coefficient. Extract the 10 feature data with the highest correlation with the processing time of each layer as the second feature data, extract the 10 category feature data with the highest correlation with the preparation time of each layer as the second parameter, and remove other data.

[0090] Through the feature extraction algorithm, feature vectors with strong correlation are extracted, which reduces the misjudgment when selecting features based on experience, eliminates useless feature data, reduces the complexity of the model, and obtains higher prediction accuracy with less data.

[0091] The first feature data, the first parameter and the actual processing time of each layer are taken as input data and input into the layer processing time prediction model based on the neural network for training to obtain the layer processing time prediction model Model1 (hereinafter referred to as the first Model), and at the same time, the accuracy ACC1 and the mean absolute error MAE1 of the first Model are obtained.

[0092] In step 2, according to the nature of the input data, the hidden layer is designed to be two layers, plus the input layer and the output layer, and the number of layers of the layer processing time prediction model based on the neural network is determined to be 4 layers, and the output is 1, and the output is the predicted processing time. The optimal threshold of the model is automatically calculated through the combination of longicorn beetle whiskers and particle swarm algorithm, the hyperparameters of the model are determined, and the radial basis neural network model is constructed. After a limited number of trainings, the first prediction model Model 1 is obtained.

[0093] Among them, the optimal threshold of the model is automatically calculated by combining the beetle whiskers and particle swarm algorithm, and the hyperparameters of the model are determined, including: taking the 20 input feature data (such as the first feature data, the first parameter) as the dimension of a particle for parameter optimization, and the accuracy of the processing time as the fitness function. Each particle updates its own optimization process according to the BAS algorithm, and updates the optimal value position of the group through POS, thereby continuously approaching the optimal value.

[0094] The first Model 1 selects Adam as the optimizer and LeakyReLU as the activation function. The ratio of the training set, the validation set, and the test set is 6:2:2. In the embodiment, the number of hidden layer nodes of the neural network layer of the first Model 1 is determined according to the following empirical formula:

[0095]

[0096] Among them, h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is the adjustment constant.

[0097] This embodiment uses a combination of longicorn beetle whiskers and particle swarm algorithms to automatically determine the threshold and hyperparameters of the model, which is beneficial to improving the prediction accuracy of the model; generally, additive manufacturing has a large number of layers, and dividing 20% ​​as a validation set is beneficial to improving the prediction accuracy of the model.

[0098] In the step 3, the 10 category feature data with the highest correlation with the processing time of each layer extracted in step 1 are input into at least two regression algorithms for training respectively, and the regression algorithms include at least two of a decision tree regression algorithm, a SVM regression algorithm, a K-nearest neighbor regression algorithm, and a random forest regression algorithm. A comprehensive comparison is made on the prediction accuracy and mean square error of each regression algorithm, and the random forest regression algorithm with the smallest mean square error is determined as the final algorithm to be adopted. The second Model 2 is calculated according to the random forest regression algorithm.

[0099] Using a variety of regression algorithms for training separately to determine the algorithm model suitable for additive manufacturing time prediction is beneficial to improving the prediction accuracy of the model.

[0100] In step 4, since the prediction accuracy of the preparation time of each layer is relatively high in the additive manufacturing process, the number of layers of the layer preparation time prediction model based on the neural network is determined to be 3, and the output is 1, that is, the predicted layer preparation time. The layer preparation time prediction only uses a single neural network model to predict, which reduces the calculation amount and complexity of the entire prediction method, but still guarantees a high prediction accuracy.

[0101] In step 5, the method of extracting the feature data of the part to be processed is the same as the method of extracting the training part data in step 1, but the processing time and preparation time of each layer of the part to be processed are not actually recorded.

[0102] In formula 1 of step 6, according to the prediction accuracy of the two models (the first Model 1 and the second Model 2), a higher weight factor is given to the model with higher accuracy and a lower weight factor is given to the model with lower accuracy, which has a higher prediction accuracy than directly averaging the predicted times of the two models.

[0103] In step 8, the characteristic data for predicting the layer preparation time is not completely the same as the characteristic data for predicting the layer processing time. By separately predicting the layer preparation time and the layer processing time, it is convenient to optimize the processing parameters of the equipment and improve the utilization rate of the equipment.

[0104] With the use of the equipment, the parts that are actually processed have the data of the processing time and preparation time of each layer. Therefore, these processing time and preparation time data can be used as part data for training models. Continuously updating the existing time prediction model can further improve the accuracy of the prediction. With the use of additive manufacturing equipment, the accuracy of the time prediction model will become higher and higher.

[0105] In actual applications, before starting additive manufacturing, it is necessary to arrange a suitable placement pattern according to the structure of the parts. The software automatically runs steps 6, 7, 8, and 9 according to the placement pattern to obtain the predicted total processing time, total preparation time, and total manufacturing time of the parts under the current placement pattern and equipment parameters, which is convenient for workers to operate and judge.

[0106] like Figure 2 As shown, the present invention also provides a metal additive manufacturing time prediction system based on data fusion, including an import and extraction module, a processing time prediction module, a preparation time prediction module and a data output module.

[0107] The import and extraction module is used to import training parts or parts to be processed, extract the effective feature data of the parts, and extract the processing parameters of the equipment.

[0108] The processing time prediction module is used to predict the processing time of each layer of the part to be processed.

[0109] The preparation time prediction module is used to predict the preparation time of each layer of the part to be processed.

[0110] The data output module is used to output the total processing time, total preparation time and overall manufacturing time of the predicted parts.

[0111] The processing time prediction module includes a first model training module and a second model training module.

[0112] The first model training module is used to train the characteristic data and processing parameters of the currently imported training parts or the parts to be processed based on the neural network algorithm, obtain the layer processing time prediction model Model1, and obtain the accuracy and mean absolute error of the layer processing time prediction model; the first Model1 is used to predict the processing time of each layer of the parts to be processed.

[0113] The second model training module is used to train the feature data and processing parameters of the currently imported training parts or parts to be processed based on the random forest algorithm, obtain the layer processing time prediction model Model2, and obtain the accuracy and mean absolute error of the layer processing time prediction model; predict the processing time of each layer of the parts to be processed based on the second Model2.

[0114] The third model training module is used to train the feature data and processing parameters of the currently imported training parts or parts to be processed based on the neural network algorithm, obtain the layer preparation time prediction model Model3, and predict the preparation time of each layer of the parts to be processed according to the third Model3.

[0115] The system in this embodiment also includes a model fusion module, which is used to fuse the processing time of each layer of the part to be processed predicted by the first Model 1 and the processing time of each layer of the part to be processed predicted by the second Model 2 to obtain a fused processing time of each layer.

[0116] The data output module is used to predict the total processing time, total preparation time and overall manufacturing time of the parts based on the predicted fusion processing time and preparation time of each layer.

[0117] The system of the present invention extracts effective feature data of parts and processing parameters of equipment by importing training parts or parts to be processed; obtains layer processing time by using a multi-model fusion method; improves prediction accuracy by separately predicting layer processing time and layer preparation time; finally, predicts the total processing time, total preparation time and overall manufacturing time of parts based on the predicted processing time and preparation time of each layer; as the equipment is used, the data of parts actually processed can be converted into part data for training models, and the existing time prediction model is continuously updated, thereby further improving the prediction accuracy of the system.

[0118] Example 2

[0119] Taking selective laser melting metal additive manufacturing as an example, the training part is a bracelet, such as Figure 3 , the maximum size is 100.123mm×84.744mm×30.000mm; the part to be predicted is the Eiffel Tower, such as Figure 4 , the maximum size is 54.662mm×54.662mm×120.944mm.

[0120] The specific implementation steps are as follows:

[0121] Extract valid feature data of the bracelet parts, where the total number of parts layers is 1500, and extract the first feature data and the first parameter of each layer in turn. For example, the first characteristic data of the first layer are as follows: cross-sectional area is 101.789mm2, outer contour circumference is 251.083mm, inner contour circumference is 248.495mm, outer contour number is 1, overlap rate is 25%, partition size is 5mm, maximum value of X-direction coordinate of the current layer of the part is 43.251mm, minimum value of X-direction coordinate of the current layer of the part is -43.194mm, maximum value of Y-direction coordinate of the current layer of the part is 36.573mm, minimum value of Y-direction coordinate of the current layer of the part is -36.573mm; the first parameters of the first layer are as follows: spot 0.12mm, laser off delay 240us, laser on delay 180us, jump delay 1500us, mark delay 1500us, inflection point delay 50us, scanning speed 1m / s, jump speed 3m / s, contour speed 1m / s, filling angle 90; the scanning time corresponding to the first layer is 8.522s, and the preparation time is 6.233s. The above 1500 layers of data are imported into the prediction model for training to obtain the first Model, and the accuracy ACC1 of the first Model is 0.941678, and the mean absolute error MAE1 is 1.065; to the second Model2, the accuracy ACC2 of the second Model2 is 0.932943, and the mean absolute error MAE2 is 1.744; and the third Model3 is obtained.

[0122] Extract the effective feature data of the Eiffel Tower, the part to be predicted, where the total number of layers of the part is 6045, and extract the second feature data and second parameters of each layer in turn. The second characteristic data of the first layer are as follows: cross-sectional area is 564.918mm2, outer contour circumference is 190.144mm, inner contour circumference is 0mm, outer contour number is 4, overlap rate is 25%, partition size is 5mm, maximum value of X-direction coordinate of the current layer of the part is 27.331mm, minimum value of X-direction coordinate of the current layer of the part is -27.331mm, maximum value of Y-direction coordinate of the current layer of the part is 27.331mm, minimum value of Y-direction coordinate of the current layer of the part is -27.331mm; the second parameters of the first layer are as follows: spot 0.12mm, laser off delay 240us, laser on delay 180us, jump delay 1500us, mark delay 1500us, inflection point delay 40us, scanning speed 1m / s, jump speed 3m / s, contour speed 1m / s, filling angle 90.

[0123] Importing the above layer data into Model1, the layer processing prediction time is 17.289s; importing the above layer data into Model2, the layer processing prediction time is 17.223s; according to formula 1, the fused layer processing time is 17.263s; importing the above layer data into Model3, the layer preparation prediction time is 6.323s. During actual processing, the actual layer processing time is measured to be 17.296, and the layer preparation time is 6.288; it can be seen that the fusion method is more accurate than directly averaging the prediction time of Model1 and Model2. By analogy, the processing time and preparation time of each layer are predicted, and finally a more accurate overall manufacturing time of the part is obtained.

[0124] The above is an explanation of the embodiments of the present invention. However, the present invention is not limited to the above embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A system for metal additive manufacturing time prediction based on data fusion, characterized in that: It includes an import and extraction module, a processing time prediction module, a preparation time prediction module and a data output module; The import and extraction module includes an import and extraction unit and a storage unit, wherein the storage unit is used to store the extracted data, and the import and extraction unit is used to import the training parts or the parts to be processed, extract the effective feature data of the training parts or the parts to be processed, and extract the processing parameters of the equipment; The processing time prediction module is used to predict the processing time of each layer of the part to be processed; The preparation time prediction module is used to predict the preparation time of each layer of the part to be processed; The data output module is used to output the total processing time, total preparation time and overall total manufacturing time of the predicted parts; The processing time prediction module includes a first model training unit and a second model training unit; The training method of the first model training unit is a neural network algorithm, which trains the feature data and processing parameters of the currently imported training parts or the parts to be processed based on the neural network algorithm to obtain a layer processing time prediction model: a first Model1, the first Model1 is used to predict the processing time T12_n of each layer of the parts to be processed; The training method of the second model training unit is a regression algorithm, which trains the feature data and processing parameters of the currently imported training part or the part to be processed based on the regression algorithm to obtain a layer processing time prediction model: a second Model2, and the second Model2 is used to predict the processing time T22_n of each layer of the part to be processed; The system further includes a model fusion module, which is used to fuse the processing time of each layer of the to-be-processed part predicted by the first Model 1 and the processing time of each layer of the to-be-processed part predicted by the second Model 2 to obtain a fused processing time of each layer; the fusion module calculates the fused processing time of each layer using Formula 1: Among them, MAE1 is the mean absolute error of the first model, and MAE2 is the mean absolute error of the second model.

2. The system for metal additive manufacturing time prediction based on data fusion according to claim 1, characterized in that: The preparation time prediction module includes a third model training unit, which is used to train the feature data and processing parameters of the currently imported training parts or parts to be processed based on a neural network algorithm to obtain a layer preparation time prediction model: the third Model3, and the third Model3 is used to predict the preparation time T32_n of each layer of the parts to be processed.

3. The system for metal additive manufacturing time prediction based on data fusion according to claim 1, characterized in that: The system also includes a data output module, which is used to predict the total processing time, total preparation time and overall manufacturing time of the part based on the predicted fusion processing time and preparation time of each layer.

4. The system for metal additive manufacturing time prediction based on data fusion according to any one of claims 1 to 3, characterized in that: The storage unit is used to store effective feature data and processing parameters according to the number of part layers.

5. The system for metal additive manufacturing time prediction based on data fusion according to any one of claims 1 to 3, characterized in that: The effective feature data of the part include the total number of part layers, the area corresponding to each layer structure, the outer contour circumference, the inner contour circumference, the number of outer contours, the overlap rate, the partition size, the maximum value of the X-direction coordinate of the current layer of the part, the minimum value of the X-direction coordinate of the current layer of the part, the maximum value of the Y-direction coordinate of the current layer of the part, and the minimum value of the Y-direction coordinate of the current layer of the part.

6. The system for metal additive manufacturing time prediction based on data fusion according to any one of claims 1 to 3, characterized in that: The processing parameters of the equipment include but are not limited to the light spot corresponding to each layer structure, laser off delay, laser on delay, jump delay, mark delay, inflection point delay, scanning speed, jump speed, randomness, contour speed, and filling angle.

7. The system for metal additive manufacturing time prediction based on data fusion according to any one of claims 1 to 3, characterized in that: The system further comprises a feature selection module, which is used to extract effective feature data in each layer and / or feature data in the processing parameters whose correlation with the processing time is arranged in an order ≥ N.

8. The system for metal additive manufacturing time prediction based on data fusion according to claim 7, characterized in that: The feature selection module is also used to extract effective feature data in each layer and / or feature data in the processing parameters whose order of relevance to the preparation time is ≥N.

9. The system for metal additive manufacturing time prediction based on data fusion according to claim 2, characterized in that: The total processing time is TR2_total=∑TR2_n, the total preparation time is T32_total=∑T32_n, and the overall manufacturing time is Time_total=TR2_total+T32_total.

10. A prediction method based on the system for predicting metal additive manufacturing time according to any one of claims 1 to 9, comprising the following steps: Step 1: Import the training parts and the corresponding additive manufacturing equipment, extract the effective feature data of the parts, and extract the processing parameters of the equipment; Step 2: Input the effective feature data and processing parameters of the training parts into the processing time prediction module and the preparation time prediction module to obtain the prediction model; Step 3: Input the effective feature data and processing parameters of the part to be trained into the prediction model to obtain the predicted processing time and preparation time.

11. The prediction method according to claim 10, characterized in that: Inputting effective feature data and processing parameters into the processing time prediction module to obtain the processing time includes the following steps: The effective feature data and processing parameters are respectively input into the first model training unit and the second model training unit for training to obtain the first prediction model and the second prediction model. The effective feature data and processing parameters of the parts to be processed are input into the first prediction model and the second prediction model to obtain the first processing time T12_n and the second processing time T22_n.

12. The prediction method according to claim 10, characterized in that: Inputting effective feature data and processing parameters into the preparation time prediction module to obtain the preparation time includes the following steps: The effective feature data and processing parameters are respectively input into the third model training unit to obtain the third prediction model, and the effective feature data and processing parameters of the part to be processed are input into the third prediction model to obtain the preparation time T32_n.

13. The prediction method according to claim 10, characterized in that: Before step 2, the following steps are also included: using a feature selection module to select effective feature data in each layer and / or feature data in the processing parameters whose correlation with the processing time is arranged in an order ≥ N as input data.

14. The prediction method according to claim 10, characterized in that: Before step 2, the following steps are also included: using a feature selection module to select effective feature data in each layer and / or feature data in the processing parameters whose order of relevance to the preparation time is ≥ N as input data.

15. The prediction method according to any one of claims 10 to 14, characterized in that: After step 2 and before step 3, the following steps are also included: using the feature selection module to select valid feature data in each layer and / or feature data in the processing parameters whose correlation order with the processing time and preparation time is ≥N as input data.

16. A method for additive manufacturing using the prediction method according to any one of claims 10 to 15, characterized in that: The following steps are involved: Parts are processed according to the predicted processing time of each layer of the part to be processed, the predicted preparation time of each layer of the part to be processed, the total processing time and the total preparation time.

17. An additive manufacturing method according to claim 16, characterized in that: After the part processing is completed, the effective feature data of the part and the actual processing parameters are stored in the storage unit.

18. An additive manufacturing method according to claim 16, characterized in that: Before processing parts, it also includes arranging a suitable placement configuration according to the structure of the parts to be processed, and inputting the current placement configuration and equipment parameters into the prediction model according to the placement configuration.

19. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the prediction method described in any one of claims 10 to 15 or the additive manufacturing method described in any one of claims 16 to 18 is implemented.

20. A device, characterized in that: The method comprises the storage medium and the processor as described in claim 19, wherein the processor implements the steps of the prediction method as described in any one of claims 10 to 15 or the additive manufacturing method as described in any one of claims 16 to 18 when executing the computer program on the storage medium.

Images