Additive manufacturing internal defect detection method and device

Abstract

The invention provides an additive manufacturing internal defect detection method. The additive manufacturing internal defect detection method comprises the following steps: acquiring three-dimensional point cloud data of an additive manufacturing part; generating a three-dimensional model of the additive manufacturing part according to the three-dimensional point cloud data, building a simulation algorithm of the additive manufacturing part through the three-dimensional model, and obtaining a detection direction and a detection route variable of the additive manufacturing part through the simulation algorithm of the additive manufacturing part; the laser ultrasonic system outputs a laser ultrasonic signal through the detection orientation and the detection route variable of the additive manufacturing part, and the laser ultrasonic signal comprises signal graph sequences of different orientations of the additive manufacturing part; building a characteristic selection model of the signal graph sequence, and selecting characteristic information of the signal graph sequence through the characteristic selection model; and building an internal defect detection model based on additive manufacturing, training the internal defect detection model, and introducing the characteristic information of the selected signal graph sequence into the trained internal defect detection model so as to perform internal defect detection on the additive manufacturing part.

Description

Technical Field

[0001] The present invention relates to the technical field of additive manufacturing, and in particular to an additive manufacturing internal defect detection method and an additive manufacturing internal defect detection device. Background Art

[0002] Additive manufacturing technology is a technology that manufactures entities by accumulating materials. In recent years, additive manufacturing technology has made rapid progress and development, and is also considered one of the most promising technologies in aerospace and defense. However, various defects will occur in the process of additive manufacturing, such as agglomeration, spheroidization, internal cracks, etc., which seriously affect the performance of additive manufacturing parts. Therefore, defect detection is particularly important to reduce the hidden dangers of additive manufacturing parts and improve the safety of additive manufacturing parts.

[0003] Traditional defect detection technologies mostly perform defect detection on the surface of additively manufactured parts. There has been no better solution for internal defect detection of additively manufactured parts due to the inability of detection tools to accurately locate the defect position, low sensitivity and slow detection speed. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method and device for detecting internal defects in additive manufacturing, which can accurately locate the internal defects of additively manufactured parts and realize non-contact non-destructive testing during the detection process, with high sensitivity and fast detection speed.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for detecting internal defects in additive manufacturing comprises the following steps: acquiring three-dimensional point cloud data of an additively manufactured part; generating a three-dimensional model of the additively manufactured part according to the three-dimensional point cloud data, building a simulation algorithm for the additively manufactured part through the three-dimensional model, and obtaining the detection position and detection route variables of the additively manufactured part through the simulation algorithm of the additively manufactured part; a laser ultrasonic system outputting a laser ultrasonic signal through the detection position and detection route variables of the additively manufactured part, wherein the laser ultrasonic signal includes a signal graph sequence at different positions of the additively manufactured part; building a feature selection model for the signal graph sequence, and selecting feature information of the signal graph sequence through the feature selection model; building an internal defect detection model based on additive manufacturing, training the internal defect detection model, and introducing the feature information of the selected signal graph sequence into the trained internal defect detection model to perform internal defect detection on the additively manufactured part.

[0007] In one embodiment of the present invention, the feature selection model for building a signal graph sequence specifically includes: collecting a first image and a second image of an image at any position in the signal graph sequence, wherein the first image and the second image are obtained through more than one processing scheme; performing feature selection on the first image, the second image and the remaining images respectively to obtain a first image feature of the first image, a second image feature of the second image and image features of the remaining images, wherein the first image feature is obtained according to a preset first image feature selection algorithm; editing the first image feature through the first image to obtain an edited image of the first image and an editing loss of the first image feature, wherein the editing loss is a feature difference value between the edited image and the first image; obtaining a comparative loss through the second image, the second image feature and the image features of the remaining images, wherein the comparative loss is a feature spacing between the first image feature and the second image feature and a difference value between the first image feature and the image features of the remaining images; updating the first image feature selection algorithm according to the editing loss and the comparative loss to obtain the feature selection model.

[0008] In one embodiment of the present invention, the method of selecting characteristic information of a signal graph sequence through the characteristic selection model specifically includes: obtaining an unreconstructed image in the signal graph sequence, and transmitting the unreconstructed image to the characteristic selection model to obtain source image characteristics of the unreconstructed image; and reconstructing the source image characteristics through image transformation to obtain characteristic information of the unreconstructed image.

[0009] In one embodiment of the present invention, the method for detecting internal defects in additive manufacturing further includes: if defects exist, obtaining the orientation and size of the internal defects of the additively manufactured part in the three-dimensional model through the three-dimensional model of the additively manufactured part and a sequence of signal images in different orientations.

[0010] In one embodiment of the present invention, the internal defect detection model based on additive manufacturing is built and the internal defect detection model is trained, specifically including: obtaining an internal defect image data set of the additively manufactured part, and dividing the internal defect image data set into a training set and a test set; presetting a randomly generated neural network model, and performing simulation training on the randomly generated neural network model according to a random maximum likelihood algorithm; inputting the internal defect images in the training set into the randomly generated neural network model in sequence to obtain high-dimensional characteristics of the internal defect images in the training set in multiple dimensions; using the obtained high-dimensional characteristics as the input of the training set, and using the internal defect types mapped by the high-order characteristics as the output of the training set.

[0011] A device for detecting internal defects in additive manufacturing, comprising: a data acquisition unit, the data acquisition unit is used to acquire three-dimensional point cloud data of an additively manufactured part; a model building unit, the model building unit is used to generate a three-dimensional model of the additively manufactured part according to the three-dimensional point cloud data, build a simulation algorithm for the additively manufactured part through the three-dimensional model, and obtain the detection orientation and detection route variables of the additively manufactured part through the simulation algorithm of the additively manufactured part; a signal output unit, the signal output unit is used to output a laser ultrasonic signal according to the detection orientation and detection route variables of the additively manufactured part through a laser ultrasonic system, wherein the laser ultrasonic signal includes a signal graph sequence at different orientations of the additively manufactured part; a feature selection unit, the feature selection unit is used to build a feature selection model of the signal graph sequence, and select feature information of the signal graph sequence through the feature selection model; a defect detection unit, the defect detection unit is used to build an internal defect detection model based on additive manufacturing, train the internal defect detection model, and introduce the feature information of the selected signal graph sequence into the trained internal defect detection model to perform internal defect detection on the additively manufactured part.

[0012] In one embodiment of the present invention, the feature selection unit is also used to: collect the first image and the second image of the image at any position in the signal image sequence, wherein the first image and the second image are obtained through more than one processing scheme; respectively perform feature selection on the first image, the second image and the remaining images to obtain the first image feature of the first image, the second image feature of the second image and the image feature of the remaining images, wherein the first image feature is obtained according to a preset first image feature selection algorithm; edit the first image feature through the first image to obtain the edited image of the first image and the editing loss of the first image feature, wherein the editing loss is the feature difference value between the edited image and the first image; obtain the comparative loss through the second image, the second image feature and the image features of the remaining images, wherein the comparative loss is the feature spacing between the first image feature and the second image feature and the difference value between the first image feature and the image features of the remaining images; update the first image feature selection algorithm according to the editing loss and the comparative loss to obtain the feature selection model.

[0013] In one embodiment of the present invention, the feature selection unit is also used to: obtain an unreconstructed image in a signal image sequence, and transmit the unreconstructed image to the feature selection model to obtain source image characteristics of the unreconstructed image; and reconstruct the source image characteristics through image transformation to obtain characteristic information of the unreconstructed image.

[0014] In one embodiment of the present invention, the device for detecting internal defects in additive manufacturing further includes: a defect acquisition unit, which is used to acquire the orientation and size of the internal defects of the additively manufactured part in the three-dimensional model through the three-dimensional model of the additively manufactured part and a sequence of signal images in different orientations.

[0015] In one embodiment of the present invention, the defect detection unit is also used to: obtain an internal defect image data set of the additively manufactured part, and divide the internal defect image data set into a training set and a test set; preset a randomly generated neural network model, and simulate the training of the randomly generated neural network model according to a random maximum likelihood algorithm; input the internal defect images in the training set into the randomly generated neural network model in sequence to obtain high-dimensional characteristics of the internal defect images in the training set in multiple dimensions; use the obtained high-dimensional characteristics as the input of the training set, and use the internal defect types mapped by the high-order characteristics as the output of the training set.

[0016] Beneficial effects of the present invention:

[0017] The present invention collects three-dimensional point cloud data of additively manufactured parts by detecting laser ultrasonic emission, laser ultrasonic interferometer and pulse excitation laser ultrasound, then generates a three-dimensional model of the additively manufactured part according to the three-dimensional point cloud data, builds a simulation algorithm through the three-dimensional model, and obtains the detection position and detection route variables of the additively manufactured part through the simulation algorithm, and outputs laser ultrasonic signals through the detection position and detection route variables of the additively manufactured part, and selects characteristic information of the signal graph sequence through the laser ultrasonic signal, finally builds an internal defect detection model and trains it, and introduces the selected characteristic information into the trained internal defect detection model to perform internal defect detection on the additively manufactured part, thereby being able to accurately locate the internal defect position of the additively manufactured part, and realize non-contact non-destructive detection during the detection process, with high sensitivity and fast detection speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the internal defect detection method of additive manufacturing according to an embodiment of the present invention;

[0019] Figure 2 Schematic diagram of a block diagram of an internal defect detection device for additive manufacturing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Figure 1 Flow chart of an additive manufacturing internal defect detection method according to an embodiment of the present invention.

[0022] like Figure 1 As shown, the additive manufacturing internal defect detection method according to the embodiment of the present invention comprises the following steps:

[0023] S1, obtaining three-dimensional point cloud data of additively manufactured parts.

[0024] In one embodiment of the present invention, the three-dimensional point cloud data of the additively manufactured part can be acquired by an RGB-D camera, and the three-dimensional coordinates inside the additively manufactured part can be calculated by the three-dimensional point cloud data.

[0025] S2, generating a three-dimensional model of the additively manufactured part according to the three-dimensional point cloud data, building a simulation algorithm for the additively manufactured part through the three-dimensional model, and obtaining the detection orientation and detection route variables of the additively manufactured part through the simulation algorithm for the additively manufactured part.

[0026] Specifically, first, a simulation algorithm for additive manufacturing can be built through a three-dimensional model of additive manufacturing that needs to be detected. By simulating the reaction of lasers with various laser intensities, injection orientations and discrete angles to the refraction of additive manufacturing parts, the corresponding associations of the laser attenuation degree and laser intensity, injection orientation and discrete angles can be obtained. Then, the signal interference to noise ratio, grayscale difference value and recognition of the image generated by the reflected laser ultrasonic signal can be set as the improved target for enhancing the clarity of the image, and the corresponding associations of the laser attenuation degree and laser intensity, injection orientation and discrete angles in different dimensions can be converted into boundary factors of several targets to build an improved algorithm for multiple targets. Secondly, a more comprehensive detection orientation can be obtained by extracting the optimal combination of laser attenuation degree and laser intensity, injection orientation and discrete angles. Finally, since there may be undetectable places and the possibility of repeated detection, and the laser may have weakened intensity inside the additive manufacturing part, the optimal detection orientation and optimal detection route variables of the additive manufacturing part can be obtained by improving the emission position and detection route of the laser emission source.

[0027] S3, the laser ultrasonic system outputs a laser ultrasonic signal through the detection position and detection route variables of the additively manufactured part, wherein the laser ultrasonic signal includes a sequence of signal graphs at different positions of the additively manufactured part.

[0028] In one embodiment of the present invention, by obtaining a preferred detection orientation and detection route variable of an additively manufactured part, a signal graph sequence of laser ultrasonic signal conversion at different orientations can be obtained, thereby enabling the internal structure of the additively manufactured part to be fully captured.

[0029] S4, building a feature selection model for the signal graph sequence, and selecting feature information of the signal graph sequence through the feature selection model.

[0030] In one embodiment of the present invention, building a feature selection model for a signal graph sequence may specifically include: collecting a first image and a second image of images at any position in the signal graph sequence, wherein the first image and the second image are obtained through more than one processing scheme; performing feature selection on the first image, the second image, and the remaining images respectively to obtain a first image feature of the first image, a second image feature of the second image, and image features of the remaining images, wherein the first image feature is obtained according to a preset first image feature selection algorithm; editing the first image feature through the first image to obtain an edited image of the first image and an editing loss of the first image feature, wherein the editing loss is a feature difference value between the edited image and the first image; obtaining a comparative loss from the second image, the second image feature, and the image features of the remaining images, wherein the comparative loss is a feature spacing between the first image feature and the second image feature and a difference value between the first image feature and the image features of the remaining images; updating the first image feature selection algorithm according to the editing loss and the comparative loss to obtain a feature selection model.

[0031] Specifically, two feature selection algorithms with the same structure can be used to select the first picture feature, the second picture feature and the remaining picture features for the first picture, the second picture and the remaining pictures. The detailed extraction process can be performed by selecting the first picture feature through the first picture feature selection algorithm, and selecting the second picture feature and the remaining picture features through the second picture selection algorithm. Among them, the variables of the second picture feature selection algorithm can be obtained by performing moving mean operations on the variables of the first picture selection algorithm.

[0032] Furthermore, after obtaining the first picture characteristic of the first picture, the first picture characteristic can be edited using a decoder to obtain an edited picture of the first picture, and the loss of both the edited picture and the first picture is calculated using a cost function and marked as an editing loss.

[0033] Furthermore, the characteristics obtained by extracting the same image can be arranged as an upper sampling group, and the characteristics obtained by extracting different images can be arranged as a lower sampling group, and the predicted probability and true probability difference operation can be performed on the upper sampling group and the lower sampling group, and marked as a comparative loss. For example, the first image characteristic of the first image and the second image characteristic of the second image can be arranged as an upper sampling group, and the first image characteristic of the first image and each of the remaining images can be arranged as a lower sampling group, thereby obtaining several groups of lower sampling groups. Then, the distance between the upper sampling group and each group of lower sampling groups can be marked as a comparative loss. Finally, the comparative loss can be calculated by the first image characteristic, the second image characteristic and the remaining image characteristics.

[0034] Furthermore, the editing loss and the comparison loss can be added to obtain the comprehensive loss, and the first image feature selection algorithm can be updated according to the comprehensive loss, wherein a constraint standard can be set, and when the comprehensive loss does not meet the constraint standard, the variables of the first image feature selection algorithm can be changed to obtain a first image feature selection algorithm that is better than the first image feature selection algorithm, and the better first image feature selection algorithm is updated, wherein the updating process is as described in the above steps, and is updated until the comprehensive loss meets the constraint standard, and the updated first image feature selection algorithm is used as a feature selection model.

[0035] In one embodiment of the present invention, characteristic information of a signal graph sequence is selected through a characteristic selection model, which may specifically include: obtaining an unreconstructed image in the signal graph sequence, and transmitting the unreconstructed image to the characteristic selection model to obtain source image characteristics of the unreconstructed image; and reconstructing the source image characteristics through image transformation to obtain characteristic information of the unreconstructed image.

[0036] S5, building an internal defect detection model based on additive manufacturing, training the internal defect detection model, and introducing characteristic information of the selected signal graph sequence into the trained internal defect detection model to perform internal defect detection on additively manufactured parts.

[0037] In one embodiment of the present invention, an internal defect detection model based on additive manufacturing is constructed, and the internal defect detection model is trained, which may specifically include: obtaining an internal defect image data set of additively manufactured parts, and dividing the internal defect image data set into a training set and a test set; presetting a randomly generated neural network model, and performing simulation training on the randomly generated neural network model according to a random maximum likelihood algorithm; inputting the internal defect images in the training set into the randomly generated neural network model in sequence to obtain high-dimensional characteristics of the internal defect images in the training set in multiple dimensions; using the obtained high-dimensional characteristics as the input of the training set, and using the internal defect types mapped by the high-dimensional characteristics as the output of the training set.

[0038] Specifically, defect images of different additive manufacturing parts can be obtained from the Internet or additive manufacturing sites, and the defect images should include as many different types of defects as possible, and different defect types are labeled, so that a data set of mapping between images and defect types can be formed. Through the labels on the defect images, the defect image data set can be divided into a standard sampling data set and a defect sampling data set, and the standard sampling data and defect sampling data can be divided into a training set, a test set, and a validation set.

[0039] Furthermore, a randomly generated neural network model can be preset, and the variables in the randomly generated neural network model can be initialized, and the randomly generated neural network model can be simulated and trained according to the random maximum likelihood algorithm, wherein the random maximum likelihood algorithm can operate on the dimension and iterate on the variable by reducing the dimension. After the simulation training of the randomly generated neural network model is completed, the randomly generated neural network model can be verified by the verification set.

[0040] Furthermore, the internal defect images in the training set can be sequentially input into the randomly generated neural network model, and the internal defect images can be selected by the randomly generated neural network model, thereby obtaining high-dimensional characteristics of the defect images in multiple dimensions, wherein the high-dimensional characteristics may include partial characteristics and overall structure of the defect images. Finally, the obtained high-dimensional characteristics can be used as the input of the training set, and the internal defect types mapped by the high-dimensional characteristics can be used as the output of the training set to train the internal defect detection model.

[0041] According to the additive manufacturing internal defect detection method of an embodiment of the present invention, three-dimensional point cloud data of the additive manufacturing part is collected by detecting laser ultrasonic emission, laser ultrasonic interferometer and pulse excitation laser ultrasound, and then a three-dimensional model of the additive manufacturing part is generated according to the three-dimensional point cloud data, a simulation algorithm is built through the three-dimensional model, and the detection position and detection route variables of the additive manufacturing part are obtained through the simulation algorithm, and the laser ultrasonic signal is output through the detection position and detection route variables of the additive manufacturing part, and the characteristic information of the signal graph sequence is selected through the laser ultrasonic signal, and finally an internal defect detection model is built and trained, and the selected characteristic information is introduced into the trained internal defect detection model to perform internal defect detection on the additive manufacturing part. As a result, the internal defect position of the additive manufacturing part can be accurately located, and non-contact non-destructive detection can be realized during the detection process, with high sensitivity and fast detection speed.

[0042] In order to implement the additive manufacturing internal defect detection method of the above embodiment, the present invention also proposes an additive manufacturing internal defect detection device.

[0043] like Figure 2As shown, the additive manufacturing internal defect detection device of the embodiment of the present invention includes: a data acquisition unit 100, a model building unit 200, a signal output unit 300, a feature selection unit 400 and a defect detection unit 500, wherein the data acquisition unit 100 is used to acquire three-dimensional point cloud data of the additive manufacturing part; the model building unit 200 is used to generate a three-dimensional model of the additive manufacturing part according to the three-dimensional point cloud data, build a simulation algorithm of the additive manufacturing part through the three-dimensional model, and obtain the detection orientation and detection route variables of the additive manufacturing part through the simulation algorithm of the additive manufacturing part; the signal output unit 300 is used to generate a three-dimensional model of the additive manufacturing part according to the three-dimensional point cloud data, build a simulation algorithm of the additive manufacturing part through the three-dimensional model, and obtain the detection orientation and detection route variables of the additive manufacturing part through the simulation algorithm of the additive manufacturing part; the signal output unit 300 is used to generate a three-dimensional model of the additive manufacturing part according to the three-dimensional point cloud data, build a simulation algorithm of the additive manufacturing part according to the three-dimensional model, and obtain the detection orientation and detection route variables of the additive manufacturing part according to ... point cloud data, and obtain the detection orientation The laser ultrasonic system outputs a laser ultrasonic signal through the detection position and detection route variables of the additively manufactured part, wherein the laser ultrasonic signal includes a signal graph sequence at different positions of the additively manufactured part; the characteristic selection unit 400 is used to build a characteristic selection model for the signal graph sequence, and select characteristic information of the signal graph sequence through the characteristic selection model; the defect detection unit 500 is used to build an internal defect detection model based on additive manufacturing, train the internal defect detection model, and introduce the characteristic information of the selected signal graph sequence into the trained internal defect detection model to perform internal defect detection on the additively manufactured part.

[0044] In one embodiment of the present invention, the data acquisition unit 100 may acquire the three-dimensional point cloud data of the additively manufactured part through an RGB-D camera, and calculate the three-dimensional coordinates inside the additively manufactured part through the three-dimensional point cloud data.

[0045] In one embodiment of the present invention, first, the model building unit 200 can build a simulation algorithm for additive manufacturing through a three-dimensional model of additive manufacturing that needs to be detected, and can obtain the corresponding association of the laser attenuation degree and the laser intensity, the injection direction and the different dimensions of the discrete angle by simulating the reaction of the laser with multiple laser intensities, injection directions and discrete angles to the refraction of the additive manufacturing part. Then, the signal interference and noise ratio, grayscale difference value and recognition of the image generated by the reflected laser ultrasonic signal can be set as the improved target for enhancing the clarity of the image, and the corresponding association of the laser attenuation degree and the laser intensity, the injection direction and the different dimensions of the discrete angle can be converted into the boundary factors of several targets to build the improved algorithm of multiple targets. Secondly, a more comprehensive detection direction can be obtained by extracting the optimal combination of the laser attenuation degree and the laser intensity, the injection direction and the discrete angle. Finally, since there may be undetectable places and the possibility of repeated detection, and the laser may have a weakened intensity inside the additive manufacturing part, the optimal detection direction and the optimal detection route variable of the additive manufacturing part can be obtained by improving the emission position and detection route of the laser emission source.

[0046] In one embodiment of the present invention, the signal output unit 300 can obtain a signal graph sequence of laser ultrasonic signal conversion at different orientations by obtaining a preferred detection orientation and detection route variable of the additively manufactured part, thereby being able to fully capture the internal structure of the additively manufactured part.

[0047] In one embodiment of the present invention, building a feature selection model for a signal graph sequence may specifically include: collecting a first image and a second image of images at any position in the signal graph sequence, wherein the first image and the second image are obtained through more than one processing scheme; performing feature selection on the first image, the second image, and the remaining images respectively to obtain a first image feature of the first image, a second image feature of the second image, and image features of the remaining images, wherein the first image feature is obtained according to a preset first image feature selection algorithm; editing the first image feature through the first image to obtain an edited image of the first image and an editing loss of the first image feature, wherein the editing loss is a feature difference value between the edited image and the first image; obtaining a comparative loss from the second image, the second image feature, and the image features of the remaining images, wherein the comparative loss is a feature spacing between the first image feature and the second image feature and a difference value between the first image feature and the image features of the remaining images; updating the first image feature selection algorithm according to the editing loss and the comparative loss to obtain a feature selection model.

[0048] Specifically, two feature selection algorithms with the same structure can be used to select the first picture feature, the second picture feature and the remaining picture features for the first picture, the second picture and the remaining pictures. The detailed extraction process can be performed by selecting the first picture feature through the first picture feature selection algorithm, and selecting the second picture feature and the remaining picture features through the second picture selection algorithm. Among them, the variables of the second picture feature selection algorithm can be obtained by performing moving mean operations on the variables of the first picture selection algorithm.

[0049] Furthermore, after obtaining the first picture characteristic of the first picture, the first picture characteristic can be edited using a decoder to obtain an edited picture of the first picture, and the loss of both the edited picture and the first picture is calculated using a cost function and marked as an editing loss.

[0050] Furthermore, the characteristics obtained by extracting the same image can be arranged as an upper sampling group, and the characteristics obtained by extracting different images can be arranged as a lower sampling group, and the predicted probability and true probability difference operation can be performed on the upper sampling group and the lower sampling group, and marked as a comparative loss. For example, the first image characteristic of the first image and the second image characteristic of the second image can be arranged as an upper sampling group, and the first image characteristic of the first image and each of the remaining images can be arranged as a lower sampling group, thereby obtaining several groups of lower sampling groups. Then, the distance between the upper sampling group and each group of lower sampling groups can be marked as a comparative loss. Finally, the comparative loss can be calculated by the first image characteristic, the second image characteristic and the remaining image characteristics.

[0051] Furthermore, the editing loss and the comparison loss can be added to obtain the comprehensive loss, and the first image feature selection algorithm can be updated according to the comprehensive loss, wherein a constraint standard can be set, and when the comprehensive loss does not meet the constraint standard, the variables of the first image feature selection algorithm can be changed to obtain a first image feature selection algorithm that is better than the first image feature selection algorithm, and the better first image feature selection algorithm is updated, wherein the updating process is as described in the above steps, and is updated until the comprehensive loss meets the constraint standard, and the updated first image feature selection algorithm is used as a feature selection model.

[0052] In one embodiment of the present invention, the feature selection unit 400 can also be used to: obtain unreconstructed images in the signal image sequence, and transmit the unreconstructed images to the feature selection model to obtain source image characteristics of the unreconstructed images; reconstruct the source image characteristics through image transformation to obtain characteristic information of the unreconstructed images.

[0053] In one embodiment of the present invention, the defect detection unit 500 can also be used to: obtain an internal defect image data set of an additively manufactured part, and divide the internal defect image data set into a training set and a test set; preset a randomly generated neural network model, and simulate the training of the randomly generated neural network model according to a random maximum likelihood algorithm; input the internal defect images in the training set into the randomly generated neural network model in sequence to obtain high-dimensional characteristics of the internal defect images in the training set in multiple dimensions; use the obtained high-dimensional characteristics as the input of the training set, and use the internal defect types mapped by the high-dimensional characteristics as the output of the training set.

[0054] Specifically, defect images of different additive manufacturing parts can be obtained from the Internet or additive manufacturing sites, and the defect images should include as many different types of defects as possible, and different defect types are labeled, so that a data set of mapping between images and defect types can be formed. Through the labels on the defect images, the defect image data set can be divided into a standard sampling data set and a defect sampling data set, and the standard sampling data and defect sampling data can be divided into a training set, a test set, and a validation set.

[0055] Furthermore, a randomly generated neural network model can be preset, and the variables in the randomly generated neural network model can be initialized, and the randomly generated neural network model can be simulated and trained according to the random maximum likelihood algorithm, wherein the random maximum likelihood algorithm can operate on the dimension and iterate on the variable by reducing the dimension. After the simulation training of the randomly generated neural network model is completed, the randomly generated neural network model can be verified by the verification set.

[0056] Furthermore, the internal defect images in the training set can be sequentially input into the randomly generated neural network model, and the internal defect images can be selected by the randomly generated neural network model, thereby obtaining high-dimensional characteristics of the defect images in multiple dimensions, wherein the high-dimensional characteristics may include partial characteristics and overall structure of the defect images. Finally, the obtained high-dimensional characteristics can be used as the input of the training set, and the internal defect types mapped by the high-dimensional characteristics can be used as the output of the training set to train the internal defect detection model.

[0057] To summarize, the present invention collects three-dimensional point cloud data of additively manufactured parts through detection of laser ultrasonic emission, laser ultrasonic interferometer and pulse excitation laser ultrasound, and then generates a three-dimensional model of the additively manufactured part according to the three-dimensional point cloud data, builds a simulation algorithm through the three-dimensional model, and obtains the detection position and detection route variables of the additively manufactured part through the simulation algorithm, and outputs laser ultrasonic signals through the detection position and detection route variables of the additively manufactured part, and selects characteristic information of the signal graph sequence through the laser ultrasonic signal, and finally builds an internal defect detection model and trains it, and introduces the selected characteristic information into the trained internal defect detection model to perform internal defect detection on the additively manufactured part, thereby being able to accurately locate the internal defect position of the additively manufactured part, and realize non-contact non-destructive testing during the detection process, with high sensitivity and fast detection speed.

[0058] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. "Multiple" means two or more, unless otherwise clearly and specifically defined.

[0059] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0060] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0061] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0062] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0063] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0064] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0065] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0066] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0067] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for detecting internal defects in additive manufacturing, characterized in that: The following steps are involved: Obtain 3D point cloud data of additively manufactured parts; Generating a three-dimensional model of the additively manufactured part according to the three-dimensional point cloud data, building a simulation algorithm of the additively manufactured part through the three-dimensional model, and obtaining detection position and detection route variables of the additively manufactured part through the simulation algorithm of the additively manufactured part; The laser ultrasonic system outputs a laser ultrasonic signal through the detection position and detection route variables of the additively manufactured part, wherein the laser ultrasonic signal includes a sequence of signal graphs at different positions of the additively manufactured part; Building a feature selection model for a signal graph sequence, and selecting feature information of the signal graph sequence through the feature selection model; An internal defect detection model based on additive manufacturing is built, the internal defect detection model is trained, and characteristic information of the selected signal graph sequence is introduced into the trained internal defect detection model to perform internal defect detection on the additively manufactured part.

2. The additive manufacturing internal defect detection method according to claim 1, characterized in that: The feature selection model for building a signal graph sequence specifically includes: Collecting a first image and a second image of an image at any position in the signal image sequence, wherein the first image and the second image are obtained through more than one processing scheme; Selecting characteristics of the first picture, the second picture, and the remaining pictures respectively to obtain a first picture characteristic of the first picture, a second picture characteristic of the second picture, and picture characteristics of the remaining pictures, wherein the first picture characteristic is obtained according to a preset first picture characteristic selection algorithm; Editing the first picture characteristic through the first picture to obtain an edited picture of the first picture and an editing loss of the first picture characteristic, wherein the editing loss is a characteristic difference value between the edited picture and the first picture; Obtaining a comparative loss through the second picture, the second picture characteristic, and the picture characteristics of the remaining pictures, wherein the comparative loss is a characteristic interval between the first picture characteristic and the second picture characteristic and a difference value between the first picture characteristic and the picture characteristics of the remaining pictures; The first image feature selection algorithm is updated according to the editing loss and the comparison loss to obtain the feature selection model.

3. The additive manufacturing internal defect detection method according to claim 2, characterized in that: The said selecting characteristic information of the signal graph sequence by the said characteristic selection model specifically includes: Acquire an unreconstructed picture in the signal image sequence, and transmit the unreconstructed picture to the feature selection model to obtain source picture features of the unreconstructed picture; The source image characteristics are reconstructed through image transformation to obtain characteristic information of the unreconstructed image.

4. The additive manufacturing internal defect detection method according to claim 3, characterized in that: Also includes: If there is a defect, the position and size of the internal defect of the additively manufactured part in the three-dimensional model are obtained through the three-dimensional model of the additively manufactured part and the signal image sequence in different positions.

5. The additive manufacturing internal defect detection method according to claim 4, characterized in that: The building of the internal defect detection model based on additive manufacturing and training of the internal defect detection model specifically include: Acquire an internal defect image dataset of the additively manufactured part, and divide the internal defect image dataset into a training set and a test set; Presetting a randomly generated neural network model, and performing simulation training on the randomly generated neural network model according to a random maximum likelihood algorithm; The internal defect images in the training set are sequentially input into the randomly generated neural network model to obtain high-dimensional characteristics of the internal defect images in the training set in multiple dimensions; The obtained high-dimensional characteristics are used as the input of the training set, and the internal defect types mapped by the high-dimensional characteristics are used as the output of the training set.

6. An additive manufacturing internal defect detection device, characterized in that: include: A data acquisition unit, the data acquisition unit is used to acquire three-dimensional point cloud data of the additively manufactured part; A model building unit, the model building unit is used to generate a three-dimensional model of the additively manufactured part according to the three-dimensional point cloud data, build a simulation algorithm of the additively manufactured part through the three-dimensional model, and obtain a detection position and a detection route variable of the additively manufactured part through the simulation algorithm of the additively manufactured part; A signal output unit, the signal output unit is used to output a laser ultrasonic signal according to a detection position and a detection route variable of the laser ultrasonic system through the additively manufactured part, wherein the laser ultrasonic signal includes a sequence of signal graphs at different positions of the additively manufactured part; A feature selection unit, the feature selection unit is used to build a feature selection model for the signal graph sequence, and select feature information of the signal graph sequence through the feature selection model; A defect detection unit, wherein the defect detection unit is used to build an internal defect detection model based on additive manufacturing, train the internal defect detection model, and introduce characteristic information of the selected signal image sequence into the trained internal defect detection model to perform internal defect detection on the additively manufactured part.

7. The additive manufacturing internal defect detection device according to claim 6, characterized in that: The feature selection unit is also used for: Collecting a first image and a second image of an image at any position in the signal image sequence, wherein the first image and the second image are obtained through more than one processing scheme; Selecting characteristics of the first picture, the second picture, and the remaining pictures respectively to obtain a first picture characteristic of the first picture, a second picture characteristic of the second picture, and picture characteristics of the remaining pictures, wherein the first picture characteristic is obtained according to a preset first picture characteristic selection algorithm; Editing the first picture characteristic through the first picture to obtain an edited picture of the first picture and an editing loss of the first picture characteristic, wherein the editing loss is a characteristic difference value between the edited picture and the first picture; Obtaining a comparative loss through the second picture, the second picture characteristic, and the picture characteristics of the remaining pictures, wherein the comparative loss is a characteristic interval between the first picture characteristic and the second picture characteristic and a difference value between the first picture characteristic and the picture characteristics of the remaining pictures; The first image feature selection algorithm is updated according to the editing loss and the comparison loss to obtain the feature selection model.

8. The additive manufacturing internal defect detection device according to claim 7, characterized in that: The feature selection unit is also used for: Acquire an unreconstructed picture in the signal image sequence, and transmit the unreconstructed picture to the feature selection model to obtain source picture features of the unreconstructed picture; The source image characteristics are reconstructed through image transformation to obtain characteristic information of the unreconstructed image.

9. The additive manufacturing internal defect detection device according to claim 8, characterized in that: Also includes: A defect acquisition unit is used to acquire the position and size of the internal defects of the additively manufactured part in the three-dimensional model through the three-dimensional model of the additively manufactured part and the signal image sequences in different positions.

10. The additive manufacturing internal defect detection device according to claim 9, characterized in that: The defect detection unit is also used for: Acquire an internal defect image dataset of the additively manufactured part, and divide the internal defect image dataset into a training set and a test set; Presetting a randomly generated neural network model, and performing simulation training on the randomly generated neural network model according to a random maximum likelihood algorithm; The internal defect images in the training set are sequentially input into the randomly generated neural network model to obtain high-dimensional characteristics of the internal defect images in the training set in multiple dimensions; The obtained high-dimensional characteristics are used as the input of the training set, and the internal defect types mapped by the high-dimensional characteristics are used as the output of the training set.

Images