CN116842667A - Method for determining manufacturing feasibility of bent pipe - Google Patents
Method for determining manufacturing feasibility of bent pipe Download PDFInfo
- Publication number
- CN116842667A CN116842667A CN202310749757.XA CN202310749757A CN116842667A CN 116842667 A CN116842667 A CN 116842667A CN 202310749757 A CN202310749757 A CN 202310749757A CN 116842667 A CN116842667 A CN 116842667A
- Authority
- CN
- China
- Prior art keywords
- neural network
- bent pipe
- training
- label value
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 31
- 238000013528 artificial neural network Methods 0.000 claims abstract description 143
- 238000012549 training Methods 0.000 claims abstract description 54
- 238000005452 bending Methods 0.000 claims abstract description 40
- 230000008569 process Effects 0.000 claims abstract description 11
- 230000000694 effects Effects 0.000 claims abstract description 9
- 230000011218 segmentation Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 230000001105 regulatory effect Effects 0.000 claims description 3
- 230000006641 stabilisation Effects 0.000 claims description 2
- 238000011105 stabilization Methods 0.000 claims description 2
- 238000013526 transfer learning Methods 0.000 claims description 2
- 230000005540 biological transmission Effects 0.000 claims 1
- 238000012216 screening Methods 0.000 claims 1
- 230000006870 function Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 7
- 238000002372 labelling Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/18—Network design, e.g. design based on topological or interconnect aspects of utility systems, piping, heating ventilation air conditioning [HVAC] or cabling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/18—Manufacturability analysis or optimisation for manufacturability
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computational Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Algebra (AREA)
- Computing Systems (AREA)
- Fluid Mechanics (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of pipe bending processes, and discloses a method for determining the manufacturing feasibility of a pipe bending, which comprises the following steps: step S1, training data corresponding to a catheter manufacturing constraint condition model is obtained, and a first label value and a second label value are obtained according to the training data; step S2, a first neural network is obtained based on the first label value and the second label value, and each category of the bent pipe and the accuracy of each category are judged according to the first neural network; step S3, training a second neural network according to a prediction result of the first neural network, wherein the structure of the second neural network is identical to that of the first neural network; and S4, feeding back the training effect of the second neural network to the first neural network, and adjusting and determining the feasibility of manufacturing the bent pipe according to the feedback. The invention is used for rapidly judging whether the designed bent pipe can be bent and formed.
Description
Technical Field
The invention relates to the technical field of pipe bending processes, in particular to a method for determining the manufacturing feasibility of a pipe bending, which is used for rapidly judging whether a designed pipe bending can be bent and formed or not.
Background
As an important gas, liquid and other fluid conveying conduit, the aviation conduit can well meet the requirements on light weight, stiffening, complex space layout and the like, and is continuously expanded in depth and breadth of application in the aerospace field.
At present, the typical representative of advanced pipe bending technology is mainly numerical control pipe bending technology, and is an advanced pipe bending technology developed by combining the traditional pipe bending technology, machine tool technology and numerical control technology. The numerical control pipe bending technology can realize the digital control of the metal pipe bending forming process, can well meet the requirements of high precision, high efficiency, high quality and the like of pipe bending forming, and particularly in the technical field of high precision tips of aviation, aerospace and the like, the numerical control pipe bending technical equipment gradually replaces the traditional pipe bending technical equipment, and plays an irreplaceable role in improving the whole manufacturing technical level of the pipe.
However, due to the limitation of the pipe bending machine, the pipe bending often interferes with the pipe bending machine in the manufacturing process, and once the pipe bending often interferes with the pipe bending machine, the pipe bending mold is damaged, and the pipe bending machine is affected. In the prior art, whether the bending can be judged manually by a craftsman or the simulation software is judged, the judgment result is inaccurate due to the fact that the craftsman manually judges depending on experience, and the simulation software has the defects of low judgment speed and low efficiency in the simulation process although the judgment accuracy is high.
Therefore, a method is needed to quickly determine whether the designed bent pipe can be bent.
Disclosure of Invention
The invention aims to provide a method for determining the manufacturing feasibility of an elbow, which is used for rapidly judging whether a designed elbow can be bent and formed or not.
The invention is realized by the following technical scheme: a method for determining the feasibility of manufacturing an elbow, comprising the steps of:
step S1, calibrating training data corresponding to a catheter manufacturing constraint condition model, and acquiring a first label value and a second label value according to the training data;
step S2, a first neural network is obtained based on the first label value, the second label value and the corresponding training data, and the accuracy of each category and each category of the bent pipe is obtained according to the first neural network;
the first neural network is a stepwise training network and is used for other application scenes in a transfer learning mode;
step S3, training a second neural network according to a prediction result of the first neural network, wherein the structure of the second neural network is identical to that of the first neural network;
and S4, feeding back the training effect of the second neural network to the first neural network, and adjusting and determining the feasibility of manufacturing the bent pipe according to the feedback.
In order to better implement the present invention, further, the method for acquiring training data corresponding to the catheter manufacturing constraint condition model in step S1 includes:
acquiring a plurality of pieces of bent pipe data which are judged to be incapable of being bent, extracting axis information and radius information of the plurality of pieces of bent pipe data, and characterizing the axis information and the radius information in a characteristic data mode;
the method for characterizing the data comprises the steps of establishing a first database and a second database;
the data in the first database comprises a first straight line segment length, a bending angle, a second straight line segment length, a bending angle, a third straight line segment length and a radius, and the data in the first database is divided into a first label value and a second label value;
the data of the second database comprises a first straight line segment length, a bending angle, a second straight line segment length, a bending angle, a third straight line segment length and a radius, and the second database does not have any label.
In order to better implement the present invention, further, the step S1 further includes:
the first label value is a manually judged two-class label, and the second label value is the confidence of manual judgment;
the two-class labels are denoted 0 and 1,0 indicating inflexibility and 1 indicating bendability.
In order to better implement the present invention, further, the step S2 includes:
performing string set training according to the database, the first label value and the second label value to obtain a first neural network;
the first neural network is divided into a first part and a second part;
the second part is divided into three branches, namely a binarization image branch, a regression branch and a classification branch;
the first part of the first neural network acquires a bent pipe candidate region in the bent pipe by adopting an interest extraction algorithm, and determines the corresponding position of the bent pipe candidate region in the original catheter based on the ROI alignment module;
and inputting the arranged bent pipe candidate regions into a second part of the first neural network, and respectively passing the arranged bent pipe candidate regions through the full-connection layer of the first neural network in the regression branch and the classification branch of the second part to finally obtain the accuracy of each category and each category of the bent pipe.
In order to better realize the invention, further, in the binary image branches, after passing through the full connection layer, the output result is the binary image of each preset category;
the code of 0 or 1 in each binarized image, 1 in the binarized image represents that the object exists in the pixel point of the category and is an object boundary of the pixel level, 0 in the binarized image represents that the object does not exist in the pixel point of the category, and the final obtained result comprises category information, confidence and the binarized image of the pixel level of each detected object.
In order to better implement the present invention, further, the step S2 includes:
inputting data in a second database based on a first neural network, and generating a prediction tag by forward propagation of the first neural network;
the predictive label is a label value obtained after the first neural network knowledge is processed;
and combining the predicted label obtained based on the first neural network processing with the real label in proportion to realize the training of the second neural network.
In order to better realize the invention, further, after the second neural network finishes training convergence stabilization, the accuracy of each category of the bent pipe is fed back to the first neural network;
finding a bent pipe category with the lowest accuracy or a plurality of bent pipe categories with the accuracy lower than a preset threshold value in feedback, judging whether the accuracy of the bent pipe categories is lower than the preset threshold value in the original first neural network, if so, adding the pictures of the bent pipe categories and point cloud data of the bent pipe categories in an input part, further optimizing the point cloud segmentation network, and if not, adjusting the first neural network
In order to better realize the invention, further, a unidirectional experience-taught network transfer relationship is arranged between the first neural network and the second neural network, the second neural network only passively receives the learned knowledge of the first neural network, the second neural network has no discrimination capability on the learned knowledge, and the effect of the first neural network on the second neural network is not fed back, so that the original learning process of the second neural network cannot be regulated.
Compared with the prior art, the invention has the following advantages:
(1) In the prior art, whether bending can be performed or not is judged by a craftsman manually or by simulation software, the judgment result is inaccurate due to the fact that the craftsman manually judges depending on experience, and the simulation software has the defects of low judgment speed and low efficiency in the simulation process although the judgment accuracy is high.
Drawings
The invention is further described with reference to the following drawings and examples, and all inventive concepts of the invention are to be considered as being disclosed and claimed.
Fig. 1 is a schematic structural diagram of a first neural network and a second neural network in a method for determining manufacturing feasibility of bent pipe according to the present invention.
Fig. 2 is a network architecture diagram of a first neural network in a method for determining manufacturing feasibility of bent pipe according to the present invention.
Fig. 3 is a flowchart of a method for determining manufacturing feasibility of bent pipe according to the present invention.
Detailed Description
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it should be understood that the described embodiments are only some embodiments of the present invention, but not all embodiments, and therefore should not be considered as limiting the scope of protection. All other embodiments, which are obtained by a worker of ordinary skill in the art without creative efforts, are within the protection scope of the present invention based on the embodiments of the present invention.
In the description of the present invention, it should be noted that the conduit determined in the present invention is not limited to the aviation conduit.
In addition, in the first neural network architecture diagram of the present invention, ROL ALIGN and ResNet50/101,backbone,Feature Maps,conv,softmax,bbox reg,Proposals,Fully Convolution Nets,head,Mask,FC layers,Coordinates,Category,three branches,RPN are the proprietary English terms of the neural network.
The ROLALIGN module is a regional characteristic aggregation module and can improve the accuracy of the detection model;
ResNet50/101 is ResNet50 and ResNet101, both are residual network models;
backbone is the Backbone network in deep learning networks;
feature Maps are network Feature Maps;
conv is the network switch;
softmax is the regression classifier;
bbox reg, bounding Box Regression, proposals, refers to frame regression;
fully Convolution Nets is a full convolutional network;
the Head is a network for acquiring network output content, and the Head makes predictions by utilizing the features extracted before;
mask is a segmentation Mask;
the FC layer is a fully-connected layer,
coordinates is a coordinate code;
categories are classifications;
threeblanches are three branches;
the RPN is a region generation network, which is called Region Proposal Network in its entirety, and is a network for extracting features.
Example 1:
in the method for determining the manufacturing feasibility of the bent pipe, as shown in fig. 1 and 3, training data corresponding to a constraint condition model of manufacturing a catheter is acquired first, and a first label value and a second label value are acquired according to the training data; obtaining a first neural network based on the first label value and the second label value, and judging whether the boundary frame of the bent pipe and the class of the bent pipe are bendable or not according to the first neural network; then training a second neural network according to the prediction result of the first neural network, wherein the structure of the second neural network is completely the same as that of the first neural network; and finally, the training effect of the second neural network is fed back to the first neural network to determine the feasibility of manufacturing the bent pipe.
The labels (first label and second label) of the training data (hereinafter abbreviated as first training data) of the first neural network are manually calibrated, and thus the sample size is relatively limited (eg.1000). The characteristic proposal of the first neural network is written in the right 1 to add novelty, and the first neural network is a stepwise training network which is modularly designed and can be used for other application scenes in a migration learning mode. Labels (first predictive labels and second predictive labels) of training data (hereinafter abbreviated as second training data) of a second training sample are generated based on a first neural network. The second training data is not manually calibrated, only the characteristic data is obtained by prediction based on the first neural network, and therefore the second training data is called a first prediction label and a second prediction label. The significance of this is mainly to promote the quantity of data (eg.20000) and promote the generalization of the data model.
Example 2:
the present embodiment is further optimized based on embodiment 1, and training data corresponding to the catheter manufacturing constraint condition model is obtained, a plurality of bent pipe data (for example, 1000 items) which are judged to be incapable of bending are obtained, and the axis and radius information thereof are extracted to characterize the data. The database may be built up in the following manner [ first straight segment length, bend angle, second straight segment length, bend angle, third straight segment length …, radius ]. The first label value is a manually determined two-class label (i.e., 0,1. Each represents a bendable, non-bendable) and the second label is a confidence of the manual judgment.
The first neural network must be a network based on artificial labels, that is, a neural network based on experimental selection is necessary to establish a mapping relationship, because the amount of manually labeled data sets is relatively small, but the sample size is very large, so that a neural network similar to a teacher student network can be selected.
The first database is in the presence of tags and the second database is not in any presence of tags.
Other portions of this embodiment are the same as those of embodiment 1, and thus will not be described in detail.
Example 3:
the present embodiment is further optimized based on the above embodiment 1 or 2, and as shown in fig. 2, the training process of the first neural network is based on the database [ the first straight line segment length, the bending angle, the second straight line segment length, the bending angle, the third straight line segment length …, the radius ], the first label value, and the second label, to perform the cluster training, thereby obtaining the first neural network.
Other portions of this embodiment are the same as those of embodiment 1 or 2 described above, and thus will not be described again.
Example 4:
the present embodiment is further optimized based on any one of the above embodiments 1 to 3, in order to more accurately reflect whether the bent pipe is bendable, a two-stage algorithm is combined into a first neural network, a first part of the first neural network adopts an interest extraction algorithm to obtain a bent pipe candidate region in a bent pipe, and a corresponding position of the bent pipe candidate region in an original pipe is determined based on an ROI alignment module.
Based on the above, the arranged bent pipe candidate region is input into a second part of the first neural network, and the second part is divided into three branches, namely a binary image branch, a regression branch and a classification branch, wherein the binary image branch is the Mask branch in fig. 1. In the regression branch and the classification branch, the candidate areas pass through the full connection layer respectively to finally obtain the boundary frame of the target object and the class whether the target object is bendable, wherein the bendable is 1 and the non-bendable is 0.
In the binary image branches, after dimension expansion and full connection layer, the output result is the binary image of each preset category, each binary image is a code of 0 or 1, and 1 represents that the category has a target object at the pixel point and is an object boundary at the pixel level. The final result includes class information, confidence and pixel-level binarized images for each detected object.
The first neural network comprises three branches, so that its loss function is also composed of three parts, namely a classification loss function, a detection loss function and a segmentation error loss function.
L＝L cls +L box +L mask ；
Wherein L is a loss function of the first neural network, L cls To classify the loss function, L box To detect the loss function, L mask Is a segmentation error loss function.
Classification branching uses cross entropy, calculates the loss for each region of interest's resulting class, and averages the number of ROIs. Loss of bounding box portion is directed to accuracy of ROI position.
Assuming that the size of the input image of the first neural network is m×n, and k curved segments are output and detected, the resulting dimension is m×n×k+k×2, where m×n×k is a binary image of k curved segments, and k×2 is confidence and category information of k curved tubes. When the binarized images of the k detected bent pipes are synthesized, each pixel point in m x n can be processed by taking the confidence coefficient as a standard, and the bending section number (namely the serial number in k) with the Mask value of 1 and the highest confidence coefficient is selected as the labeling result of the pixel point. After integration, the dimension of the result output by the first neural network is m×n×1+k×1, where m×n×1 is the number of the detected bent pipe, the value is between 1 and k, and k×1 is the category information of each detected bent pipe.
Therefore, training of the first neural network is achieved, and the first neural network can achieve prediction of the first label value and the second label based on the training set.
Other portions of this embodiment are the same as any of embodiments 1 to 3 described above, and thus will not be described again.
Example 5:
in this embodiment, further optimization is performed based on any one of the above embodiments 1 to 4, in the training process of the second neural network, because the number of manually labeled labels is limited, the labeling is performed by adopting the manual label mode, although the accurate model training result can be obtained, the cost of manually labeling is too high, and the number of labels can be limited. Therefore, training of the second neural network needs to be achieved based on the prediction results of the first neural network.
First, a plurality of bent pipe data (for example, 1000 items) are acquired, the axis and radius information thereof are extracted, and the characteristic data are characterized, for example, a second database [ first straight line segment length, bending angle, second straight line segment length, bending angle, third straight line segment length …, radius ] can be established in a following manner. No tag is present in the second database.
The feed forward (first neural network forward propagation) generates predictive labels based on the data input in the second database by the first neural network. It will be appreciated that the predicted tag is not an actual tag, but rather a tag value that results from the first neural network knowledge processing.
The structure of the second neural network is identical to that of the first neural network, the prediction labels obtained based on the processing of the first neural network and the real labels are combined according to the proportion of 9:1, and the training of the second neural network can be realized according to other proportions, and the training is not repeated here.
It can be appreciated that the training of the larger data volume model can be achieved by adopting the second neural network mode, and the knowledge inheritance of the first neural network is achieved. The second neural network obtained by training is more universal.
Other portions of this embodiment are the same as any of embodiments 1 to 4 described above, and thus will not be described again.
Example 6:
the embodiment is further optimized based on any one of the above embodiments 1 to 5, as shown in fig. 1, a unidirectional empirical teaching network is arranged between the first neural network and the second neural network, the second neural network only passively receives the learned knowledge of the first neural network, and has no discrimination capability on the learned knowledge, and the effect of the first neural network on the second neural network is not fed back, so that the original learning process of the first neural network cannot be regulated, so that the invention further provides an interactive network with interaction and regulation effects.
Errors generated in the knowledge inheritance of the first neural network to the second neural network are concentrated in three aspects of calibration errors, first neural network reasoning errors and boundary labeling errors. Here we mainly focus on the situation that the boundary error is not marked and mismarked, because if we need to perform the correction work of whether to bend, we keep the information of the single bent tube during the knowledge transfer, the point cloud semantic result obtained after the transfer should be n×1+k1, n×1 is the number of each point in the detected k bent segments, and k×1 is the category of the k detected bent segments. The correction firstly carries out class II clustering on the point cloud of each detected bent pipe, and aims to remove the mislabeled miscellaneous points around the bent section. Because the result of the convolutional neural network is obtained through a mature two-dimensional image example segmentation network and has certain credibility, the correct part in the curved section point cloud is considered to occupy a larger proportion. After clustering, the labels of the impurity clusters with a small number are removed, and the situation of partial misclassification is optimized. And then, marking the point cloud which originally belongs to the bending section but is not regenerated in the body by using a regrowth method for each detected bending section, thereby improving the unclassified condition at the boundary.
Other portions of this embodiment are the same as any of embodiments 1 to 5 described above, and thus will not be described again.
Example 7:
the present embodiment is further optimized based on any one of the above embodiments 1 to 6, in the feedback of the interaction network, the training effect of the second neural network should be fed back to the first neural network to improve the knowledge inheritance quality thereof, that is, the training situation of the original first neural network is improved according to the network quality of the second neural network. The feedback form and the adjustment direction are quite large, for example, the loss or the accuracy of the training of the second neural network can be fed back to the first neural network, and the first neural network can adjust the aspects of input, network structure, output form and the like according to the network quality of the second neural network. The class accuracy of the second neural network training can be used for feedback to adjust the class proportion of the input data of the first neural network, so that the training process adopts a gradient descent method, an Adam algorithm and other conventional neural network algorithms, and the description is omitted.
It should be noted that, the feedback here is not started simultaneously with the training of the second neural network, but after the second neural network converges and stabilizes, the accuracy of each category of the bent tube in the space is fed back to the first neural network. The type with the lowest accuracy or the types with the accuracy lower than a certain threshold value are found in the feedback, and the preset threshold value can be any number between 0 and 1, and can be 0.9 or other values.
Firstly, judging whether the accuracy of the categories is lower in the original first neural network, if so, adding the corresponding category of pictures, and performing fine tuning optimization on the first neural network. If the accuracy is lower in the second neural network only and higher in the first neural network, the images of the several categories and the point cloud data are added in the input part, and the point cloud segmentation network is further optimized. Because of the problem of the mismatch of the ROI alignment module (that is, the accurate data of the straight line segment length, the bending angle, etc. cannot be extracted), the addition of the point cloud data and the point cloud segmentation network can be understood as a reference line in geometry mathematics so as to improve the identification accuracy of the ROI alignment module.
Other portions of this embodiment are the same as any of embodiments 1 to 6 described above, and thus will not be described again.
The foregoing description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, and any simple modification and equivalent variation of the above embodiment according to the technical matter of the present invention falls within the scope of the present invention.
Claims (8)
1. A method for determining the feasibility of manufacturing an elbow, comprising the steps of:
step S1, calibrating training data corresponding to a catheter manufacturing constraint condition model, and acquiring a first label value and a second label value according to the training data;
step S2, a first neural network is obtained based on the first label value, the second label value and the corresponding training data, and the accuracy of each category and each category of the bent pipe is obtained according to the first neural network;
the first neural network is a stepwise training network and is used for other application scenes in a transfer learning mode;
step S3, training a second neural network according to a prediction result of the first neural network, wherein the structure of the second neural network is identical to that of the first neural network;
and S4, feeding back the training effect of the second neural network to the first neural network, and adjusting and determining the feasibility of manufacturing the bent pipe according to the feedback.
2. The method for determining the feasibility of manufacturing an elbow according to claim 1, wherein the method for acquiring training data corresponding to the model of the constraint condition for manufacturing a conduit in step S1 comprises:
acquiring a plurality of pieces of bent pipe data which are judged to be incapable of being bent, extracting axis information and radius information of the plurality of pieces of bent pipe data, and characterizing the axis information and the radius information in a characteristic data mode;
the method for characterizing the data comprises the steps of establishing a first database and a second database;
the data in the first database comprises a first straight line segment length, a bending angle, a second straight line segment length, a bending angle, a third straight line segment length and a radius, and the data in the first database is divided into a first label value and a second label value;
the data of the second database comprises a first straight line segment length, a bending angle, a second straight line segment length, a bending angle, a third straight line segment length and a radius, and the second database does not have any label.
3. The method for determining the manufacturing feasibility of the bent pipe according to claim 1, wherein said step S1 further comprises:
the first label value is a manually judged two-class label, and the second label value is the confidence of manual judgment;
the two-class labels are denoted 0 and 1,0 indicating inflexibility and 1 indicating bendability.
4. The method for determining the manufacturing feasibility of the bent pipe according to claim 1, wherein said step S2 comprises:
performing string set training according to the database, the first label value and the second label value to obtain a first neural network;
the first neural network is divided into a first part and a second part;
the second part is divided into three branches, namely a binarization image branch, a regression branch and a classification branch;
the first part of the first neural network acquires a bent pipe candidate region in the bent pipe by adopting an interest extraction algorithm, and determines the corresponding position of the bent pipe candidate region in the original catheter based on the ROI alignment module;
and inputting the arranged bent pipe candidate regions into a second part of the first neural network, and respectively passing the arranged bent pipe candidate regions through the full-connection layer of the first neural network in the regression branch and the classification branch of the second part to finally obtain the accuracy of each category and each category of the bent pipe.
5. The method for determining the manufacturing feasibility of the bent pipe according to claim 4, comprising:
in the binary image branches, after passing through the full connection layer, outputting a result which is a binary image of each preset category;
the code of 0 or 1 in each binarized image, 1 in the binarized image represents that the object exists in the pixel point of the category and is an object boundary of the pixel level, 0 in the binarized image represents that the object does not exist in the pixel point of the category, and the final obtained result comprises category information, confidence and the binarized image of the pixel level of each detected object.
6. The method for determining the manufacturing feasibility of the bent pipe according to claim 1, wherein said step S3 comprises:
inputting data in a second database based on a first neural network, and generating a prediction tag by forward propagation of the first neural network;
the predictive label is a label value obtained after the first neural network knowledge is processed;
and combining the predicted label obtained based on the first neural network processing with the real label in proportion to realize the training of the second neural network.
7. The method for determining the manufacturing feasibility of the bent pipe according to claim 1, wherein said step S4 comprises:
after the second neural network finishes training convergence stabilization, feeding the accuracy of each category of the bent pipe back to the first neural network;
finding one bent pipe type with the lowest accuracy or a plurality of bent pipe types with the accuracy lower than a preset threshold in feedback, judging whether the accuracy of the bent pipe types is lower than the preset threshold in the original first neural network, if so, adding the pictures of the bent pipe types and point cloud data of the bent pipe types in an input part, further optimizing the point cloud segmentation network, and if not, adjusting the first neural network.
8. The method for determining the manufacturing feasibility of the bent pipe according to claim 1, comprising:
the first neural network and the second neural network are in a network transmission relation taught by unidirectional experience, the second neural network only passively receives knowledge learned by the first neural network, the second neural network has no screening capability on the learned knowledge, and the first neural network does not receive feedback on the effect of the second neural network, so that the original learning process of the second neural network cannot be regulated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310749757.XA CN116842667A (en) | 2023-06-25 | 2023-06-25 | Method for determining manufacturing feasibility of bent pipe |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310749757.XA CN116842667A (en) | 2023-06-25 | 2023-06-25 | Method for determining manufacturing feasibility of bent pipe |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116842667A true CN116842667A (en) | 2023-10-03 |
Family
ID=88164586
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310749757.XA Pending CN116842667A (en) | 2023-06-25 | 2023-06-25 | Method for determining manufacturing feasibility of bent pipe |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116842667A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583342A (en) * | 2018-11-21 | 2019-04-05 | 重庆邮电大学 | Human face in-vivo detection method based on transfer learning |
| CN109886245A (en) * | 2019-03-02 | 2019-06-14 | 山东大学 | A kind of pedestrian detection recognition methods based on deep learning cascade neural network |
| CN110009628A (en) * | 2019-04-12 | 2019-07-12 | 南京大学 | A kind of automatic testing method for polymorphic target in continuous two dimensional image |
| US20200320685A1 (en) * | 2017-10-02 | 2020-10-08 | Promaton Holding B.V. | Automated classification and taxonomy of 3d teeth data using deep learning methods |
| CN112990211A (en) * | 2021-01-29 | 2021-06-18 | 华为技术有限公司 | Neural network training method, image processing method and device |
| CN114821063A (en) * | 2022-05-11 | 2022-07-29 | 北京百度网讯科技有限公司 | Semantic segmentation model generation method and device and image processing method |
| CN116229053A (en) * | 2022-12-07 | 2023-06-06 | 安徽唯嵩光电科技有限公司 | Dynamic adjustment method for red date machine striker plate based on neural network |
-
2023
- 2023-06-25 CN CN202310749757.XA patent/CN116842667A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200320685A1 (en) * | 2017-10-02 | 2020-10-08 | Promaton Holding B.V. | Automated classification and taxonomy of 3d teeth data using deep learning methods |
| CN109583342A (en) * | 2018-11-21 | 2019-04-05 | 重庆邮电大学 | Human face in-vivo detection method based on transfer learning |
| CN109886245A (en) * | 2019-03-02 | 2019-06-14 | 山东大学 | A kind of pedestrian detection recognition methods based on deep learning cascade neural network |
| CN110009628A (en) * | 2019-04-12 | 2019-07-12 | 南京大学 | A kind of automatic testing method for polymorphic target in continuous two dimensional image |
| CN112990211A (en) * | 2021-01-29 | 2021-06-18 | 华为技术有限公司 | Neural network training method, image processing method and device |
| CN114821063A (en) * | 2022-05-11 | 2022-07-29 | 北京百度网讯科技有限公司 | Semantic segmentation model generation method and device and image processing method |
| CN116229053A (en) * | 2022-12-07 | 2023-06-06 | 安徽唯嵩光电科技有限公司 | Dynamic adjustment method for red date machine striker plate based on neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299274B (en) | Natural scene text detection method based on full convolution neural network | |
| WO2022127454A1 (en) | Method and device for training cutout model and for cutout, equipment, and storage medium | |
| CN112528963A (en) | Intelligent arithmetic question reading system based on MixNet-YOLOv3 and convolutional recurrent neural network CRNN | |
| CN112836650B (en) | Semantic analysis method and system for quality inspection report scanning image table | |
| CN110648310A (en) | Weak supervision casting defect identification method based on attention mechanism | |
| CN115880704B (en) | Automatic cataloging method, system, equipment and storage medium for cases | |
| CN114693942A (en) | Multimode fault understanding and auxiliary labeling method for intelligent operation and maintenance of instruments and meters | |
| CN112541491A (en) | End-to-end text detection and identification method based on image character region perception | |
| CN114972952B (en) | Model lightweight-based industrial part defect identification method | |
| CN111597943B (en) | Table structure identification method based on graph neural network | |
| CN113076950A (en) | Image data automatic labeling method and system based on deep reinforcement learning | |
| CN110377282B (en) | Method for generating Web code based on UI generating countermeasure and convolutional neural network | |
| CN115862045A (en) | Case automatic identification method, system, equipment and storage medium based on image-text identification technology | |
| CN112884135B (en) | Data annotation correction method based on frame regression | |
| CN111222546A (en) | Multi-scale fusion food image classification model training and image classification method | |
| CN112215301B (en) | Image straight line detection method based on convolutional neural network | |
| CN111105423B (en) | Deep learning-based kidney segmentation method in CT image | |
| CN116385466B (en) | Method and system for dividing targets in image based on boundary box weak annotation | |
| CN116075820A (en) | Method, non-transitory computer readable storage medium and apparatus for searching image database | |
| CN116842667A (en) | Method for determining manufacturing feasibility of bent pipe | |
| CN111612802A (en) | Re-optimization training method based on existing image semantic segmentation model and application | |
| CN116912872A (en) | Drawing identification method, device, equipment and readable storage medium | |
| CN115984894A (en) | 2D drawing feature identification method, system, device and medium | |
| CN115511061A (en) | Knowledge distillation method based on YOLOv5 model | |
| CN115331052A (en) | Garbage data labeling system and method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |