Plate-shaped part machining quality detection method, device, equipment, medium and product
By constructing a neural network model, inputting static and dynamic parameters of part characteristics, predicting the quality pass rate and machining error of parts, the problems of slow detection speed and lack of universality in the existing technology are solved, and fast and accurate machining is achieved Quality detection and error prediction.
Patent Information
- Application Number
- CN202411629961.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-15
- Publication Date
- 2025-02-25
AI Technical Summary
The prior art is difficult to quickly and accurately detect the processing quality of waveguide holes in flat plate crack antennas. Especially under the conditions of high dimensional density, the three-coordinate measuring instrument is time-consuming and complex in operation, and the quality prediction model of neural network methods lacks universality. .
The quality pass rate prediction model and feature processing error prediction model are constructed by using neural networks. By inputting the static parameters and dynamic parameters of part features, the quality pass rate and processing error of part features are predicted.
It improves the applicability of the quality prediction model, can quickly and accurately detect the quality pass rate of different parts characteristics of the inspected parts, and realizes accurate prediction of machining errors on the basis of the prediction qualification, improving the prediction accuracy of part quality.
Smart Images
Figure CN119511941A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of parts processing, and in particular to a method, device, equipment, medium and product for detecting the processing quality of plate-shaped parts. Background Art
[0002] With the improvement of technical and tactical requirements, the size density of waveguide holes in flat slot antennas is getting higher and higher, which puts higher and higher requirements on processing quality. The processing quality error requirement has reached less than ±0.012mm. The relevant detection methods mostly use three-coordinate measuring instruments for detection, but three-coordinate detection is time-consuming and complicated to operate, which is difficult to meet the needs of rapid detection.
[0003] Some people have proposed using neural networks to predict processing quality, but this method regards the features in precision thin plate parts as a combination of rectangular, crack and circular hole features, and establishes corresponding quality prediction models for different features. This will result in the established quality prediction model being essentially only applicable to a single type of feature and not universal. Summary of the invention
[0004] The purpose of this application is to provide a method, device, equipment, medium and product for detecting the processing quality of plate-like parts, which can improve the applicability of the quality prediction model and predict the processing errors.
[0005] To achieve the above objectives, this application provides the following solutions:
[0006] In a first aspect, the present application provides a method for detecting the processing quality of a plate-shaped part, comprising:
[0007] Inputting static parameters and dynamic parameters of a part feature of a plate-like part to be inspected into a quality pass rate prediction model to obtain a quality pass rate prediction result of the part feature; the static parameters include design parameters, tool parameters and process parameters, and the dynamic parameters include cutting force parameters; the quality pass rate prediction model is obtained by training a first neural network; the plate-like part to be inspected includes at least one part feature;
[0008] If the quality pass rate prediction result meets the set conditions, the static parameters and dynamic parameters of the part feature are input into a feature processing error prediction model to obtain the processing error of the part feature; the feature processing error prediction model is obtained by training the second neural network.
[0009] Optionally, the design parameters include part material, part dimensions, part feature type and dimensions corresponding to each part feature; the part feature type includes circular hole feature, rectangular feature and crack feature;
[0010] The tool parameters include tool material and tool diameter;
[0011] The process parameters include cutting speed, feed rate and cutting depth when machining the part features;
[0012] The cutting force parameters include cutting force changes in the X direction, cutting force changes in the Y direction, and cutting force changes in the Z direction when machining the part feature.
[0013] Optionally, the cutting force variation in the i direction is represented as a cutting force image in the i direction;
[0014] The process of obtaining the cutting force image in the i direction is:
[0015] Collect the initial time series of cutting force when processing the hth layer in the i-direction of the part feature; the processing in the i-direction is divided into h 0 Segment, 1≤h≤h 0 ; h and h 0 All are integers, i = X, Y, Z;
[0016] Normalizing the sampling time of the initial time series of the cutting force to obtain a first time series of the cutting force after time reconstruction;
[0017] If the time interval between adjacent cutting force sampling points in the first cutting force time series is greater than the set interval value, an interpolation method is used for interpolation; if the time interval between adjacent cutting force sampling points is less than the set interval value, a moving average method is used to process the adjacent cutting force sampling points to obtain a second cutting force time series;
[0018] The second time series of cutting force is converted into a two-dimensional image by using the Gram angle field method to obtain a cutting force image in the i direction.
[0019] Optionally, the first neural network includes three feature extraction branches with the same structure, the three feature extraction branches are respectively a first feature extraction branch, a second feature extraction branch and a third feature extraction branch; the cutting force image in the X direction is input into the first feature extraction branch, the cutting force image in the Y direction is input into the second feature extraction branch, and the cutting force image in the Z direction is input into the third feature extraction branch;
[0020] The outputs of the three feature extraction branches in the first neural network are weighted averaged and then vector-concatenated with the static parameters to obtain a concatenated vector. The concatenated vector is sequentially subjected to data normalization, a fully connected layer, and a softmax function layer to output a quality pass rate prediction result.
[0021] Optionally, the first feature extraction branch includes multiple feature extraction modules connected in sequence; the structure of each feature extraction module is the same; the feature extraction module includes a convolution layer and a pooling layer connected in sequence, and the convolution layer is a convolution layer integrating a self-attention mechanism; the convolution layer in the feature extraction module is the input, and the pooling layer is the output.
[0022] Optionally, the second neural network includes three feature extraction branches with the same structure, the three feature extraction branches are respectively a fourth feature extraction branch, a fifth feature extraction branch and a sixth feature extraction branch; the cutting force image in the X direction is input into the fourth feature extraction branch, the cutting force image in the Y direction is input into the fifth feature extraction branch, and the cutting force image in the Z direction is input into the sixth feature extraction branch;
[0023] The outputs of the three feature extraction branches in the second neural network are weighted averaged and then vector-concatenated with the static parameters to obtain a concatenated vector. The concatenated vector is sequentially subjected to data normalization, a fully connected layer, a softmax function layer, a Relu function layer, and a logarithmic transformation to output the processing error of the part features.
[0024] In a second aspect, the present application provides a plate-like parts processing quality detection device, the plate-like parts processing quality detection device comprising:
[0025] The quality qualification prediction module is used to input the static parameters and dynamic parameters of a part feature of the plate-like part to be detected into the quality qualification rate prediction model to obtain the quality qualification rate prediction result of the part feature; the static parameters include design parameters, tool parameters and process parameters, and the dynamic parameters include cutting force parameters; the quality qualification rate prediction model is obtained by training the first neural network; the plate-like part to be detected includes at least one part feature;
[0026] The quality error prediction module is used to input the static parameters and dynamic parameters of the part feature into a feature processing error prediction model to obtain the processing error of the part feature if the quality pass rate prediction result meets the set conditions; the feature processing error prediction model is obtained by training the second neural network.
[0027] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-mentioned methods for detecting the processing quality of plate-like parts.
[0028] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the above-mentioned methods for detecting the processing quality of plate-like parts.
[0029] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned methods for detecting the processing quality of plate-like parts.
[0030] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0031] The present application provides a method, device, equipment, medium and product for inspecting the processing quality of plate-like parts, which predicts the quality pass rate of each part to be inspected, and specifically predicts the quality pass rate of part features based on design parameters, tool parameters, process parameters and cutting force parameters, taking into account static parameters and dynamic parameters, and achieving the quality pass rates of different part features of the parts to be inspected, thereby being applicable to parts of various types and sizes, improving the applicability of quality prediction, and furthermore, predicting processing errors after predicting the quality pass, thereby improving the prediction accuracy of part quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0033] Figure 1 A schematic diagram of a process flow of a plate-shaped parts processing quality inspection method provided in one embodiment of the present application;
[0034] Figure 2 A detailed flow chart of a plate-shaped parts processing quality inspection method provided in one embodiment of the present application;
[0035] Figure 3 A schematic diagram of the quality pass rate prediction model structure provided in one embodiment of the present application;
[0036] Figure 4 A schematic diagram of the structure of a feature processing error prediction model provided in one embodiment of the present application;
[0037] Figure 5 A schematic diagram of a plate-shaped component structure provided in one embodiment of the present application;
[0038] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0039] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0040] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0041] The present application provides a method for detecting the processing quality of plate-shaped parts, such as Figure 1 and Figure 2 As shown, the plate-like parts processing quality detection method includes:
[0042] Step 101: Input the static parameters and dynamic parameters of a part feature of a plate-like part to be inspected into a quality pass rate prediction model to obtain a quality pass rate prediction result of the part feature; the static parameters include design parameters, tool parameters and process parameters, and the dynamic parameters include cutting force parameters; the quality pass rate prediction model is obtained by training a first neural network; the plate-like part to be inspected includes at least one part feature.
[0043] Step 102: If the quality pass rate prediction result meets the set conditions, the static parameters and dynamic parameters of the part feature are input into a feature processing error prediction model to obtain the processing error of the part feature; the feature processing error prediction model is obtained by training the second neural network.
[0044] The present application predicts the quality pass rate of each part to be inspected, and specifically predicts the quality pass rate of part features based on design parameters, tool parameters, process parameters and cutting force parameters, taking into account static parameters and dynamic parameters, and adopts the plate-shaped parts processing quality inspection method of the present application to perform quality pass rate inspection and corresponding quality parameter inspection on each part feature of the part to be inspected, thereby being applicable to parts of various types and sizes, improving the applicability of quality prediction, and in addition, performing processing error prediction after predicting the quality pass, thereby improving the prediction accuracy of part quality.
[0045] In an exemplary embodiment, the design parameters include part material, part dimensions, part feature type, and dimensions corresponding to each part feature; the part feature types include circular hole features, rectangular features, and crack features.
[0046] The tool parameters include tool material and tool diameter.
[0047] The process parameters include cutting speed, feed rate and cutting depth when machining the part features.
[0048] The cutting force parameters include cutting force changes in the X direction, cutting force changes in the Y direction, and cutting force changes in the Z direction when machining the part feature.
[0049] In an exemplary embodiment, data collection is performed before the first neural network is trained to obtain a quality pass rate prediction model and the second neural network is trained to obtain a feature processing error prediction model.
[0050] The plate-like parts in this application are precision thin plate parts, and data collection includes: data collection of design parameters, tool parameters, process parameters, cutting force parameters and quality parameters in the actual production process of precision thin plate parts.
[0051] 1) The design parameters are collected by manually counting in the 3D model. The design parameters include the material Mat of the part. com 、Part dimensions (L com ,W com ,H com ), part feature type FeaKind, absolute size of each part feature [(L fea1 ,W fea1 ,H fea1 ), (L fea2 ,W fea2 ,H fea2 ),...,(L feaN ,W feanN ,H feanN )].
[0052] Among them, L com is the length of the part shape, W com H is the width of the part shape. com is the height of the part shape, L fean is the length of the nth part feature, W fean is the width of the nth part feature, H fean is the height of the nth part feature, 1≤n≤N, and N is the number of part features.
[0053] Parts MaterialMat com They include one of the commonly used aluminum alloy materials such as 3A21, 6061, 6063, 7075, etc., and are respectively denoted as (1,0,0,0), (0,1,0,0), (0,0,1,0), and (0,0,0,1).
[0054] In an exemplary embodiment, a part such as Figure 5As shown, the dimensions of the part are 120 mm long, 70 mm wide and 4 mm high.
[0055] Part feature types FeaKind include circular hole features, rectangular features, and crack features, which are coded as (1,0,0), (0,1,0), and (0,0,1) using one-hot encoding. The difference between crack features and rectangular features is that at least one side of the crack feature is less than or equal to 1mm, and the rest are considered rectangular features. Figure 5 The three features inside the part drawing shown are the circular hole feature, the rectangular feature, and the crack feature from left to right, which are recorded as (1,0,0), (0,1,0), and (0,0,1) respectively.
[0056] The absolute size of each feature refers to the design size of each feature in the three directions of length, width and depth. The length and width of the circular hole feature respectively represent the length of the diameter of the two sides of the circular hole feature in any 90° direction, and the height represents the depth of the circular hole feature. Figure 5 In the part drawing shown, the absolute sizes of the three features from left to right are: (8mm, 8mm, 2mm), (10mm, 6mm, 2mm), and (4mm, 1mm, 2mm).
[0057] The absolute size of each part feature is segmented according to the size range. The segmentation is based on four segments: (0,1], (1,5], (5,10], and (10,+∞], which are recorded as (1,0,0,0), (0,1,0,0), (0,0,1,0), and (0,0,0,1). In this way, the sizes of different part features can be encoded as: [(L' fea1 ,W f ' ea1 ,H' fea1 ), (L' fea2 ,W f ' ea2 ,H' fea2 ),...,(L' fean ,W f ' ean ,H' fean )].like Figure 5 As shown in the figure, the dimensions of the parts can be encoded as [(0,0,1,0), (0,0,1,0), (0,1,0,0)], [(0,0,0,1), (0,0,1,0), (0,1,0,0)], [(0,1,0,0), (1,0,0,0), (0,0,0,1)]. Encoding different part features can effectively process various types of non-numeric data and unify the dimensions. The purpose of encoding tool parameters later is the same.
[0058] 2) The tool material in the tool parameters is recorded as Mattool-sn , the tool diameter is recorded as D tool-sn , s is the number of the part to be tested, n is the number of the part feature, and the tool parameters are manually recorded when compiling the NC program. For example, the tool diameter D used in the processing process tool =1mm.
[0059] The tool materials include high-speed steel tools, carbide tools, ceramic tools, and diamond tools, which are denoted as (1,0,0,0), (0,1,0,0), (0,0,1,0), and (0,0,0,1), respectively. Figure 5 The tool material for machining the middle part is high-speed steel tool, denoted as (1,0,0,0).
[0060] 3) The process parameters include the three major elements of cutting speed, feed rate and cutting depth when machining each feature, which are denoted as v c-sn 、f 0-sn and a p-sn , statistical records are made manually when compiling NC programs.
[0061] According to the orthogonal test method, Figure 5 The test was carried out on the parts in the machine, where the cutting speeds were 25000rpm, 27500rpm, 30000rpm, 32500rpm, and 35000rpm, the feed rates were 2600, 2800, 3000, 3200, and 3400, and the cutting depths ranged from 0.01mm, 0.03mm, 0.05mm, 0.07mm, and 0.1mm. A total of 75 parts were processed.
[0062] 4) Cutting force parameters refer to the cutting force variation curves in the three directions of X, Y, and Z when machining each feature of the part. The cutting force parameters are collected by using a cutting force collector to collect the cutting force variation curves of each feature in sections, which are recorded as The sampling frequency of the cutting force is set between 2000 Hz and 5000 Hz. For example, s = 1, 2, ..., 75, n = 1, 2, 3, and the sampling frequency of the cutting force is set to 4000 Hz.
[0063] 5) Quality parameters include machining errors in the length direction, width direction and height direction of the part.
[0064] The machining error of each part feature is collected by using a three-coordinate measuring machine to form a quality parameter data set: (δ Lfea1-s ,δ Wfea1-s ,δ Hfea1-s ), (δ Lfea2-s ,δ Wfea2-s ,δHfea2-s ),...,(δ LfeaN-s ,δ WfeaN-s ,δ HfeaN-s ), where δ Lfean-s is the machining error of the nth part feature of the sth part in the length direction, δ Wfean-s is the machining error of the nth part feature of the sth part in the width direction, δ Hfean-s is the machining error of the nth part feature of the sth part in the height direction.
[0065] The machining errors of rectangular features and crack features refer to the differences between the actual values and the absolute sizes after machining in the length, width and depth directions of the rectangle / crack, respectively. The machining error of circular hole features refers to the difference between the actual values and the absolute sizes of the length and depth of the diameter of any cross 90° direction of the circular hole feature. The quality parameters are collected by measuring the actual sizes of different features one by one through a three-coordinate measuring machine and calculating them.
[0066] The design parameters, tool parameters, and process parameters of a single part feature are fused to form a static parameter data set of a single feature, which is recorded as:
[0067] P arsta-sn =(Mat com ,FeaKind,Mat tool ,D tool ,v c-sn ,f 0-sn ,a p-sn ,L com ,W com ,H com ,L' fea1 ,W f ' ea1 ,H' fea1 ), for example, P arsta-sn =[(1,0,0,0),FeaKind,(1,0,0,0),1,v c-sn ,f 0-sn ,a p-sn ,(120,70,4),(L' fean-s ,W f ' ean-s ,H' fean-s )].
[0068] Among them, P arsta-sn It is the static parameter of the nth part feature of the sth part.
[0069] In an exemplary embodiment, the cutting force variation in the i-direction is a cutting force image in the i-direction, that is, the cutting force time series data is processed.
[0070] The process of obtaining the cutting force image in the i direction is:
[0071] The real-time cutting forces in the X, Y and Z directions of different part features are obtained The data is processed smoothly, abnormal values and redundant values in the acquisition process are eliminated, and missing values are supplemented. The cutting force after data processing is recorded as: Where s represents the number of the part and n represents the number of the part feature.
[0072] Cutting force segmentation: The real-time cutting force in the three directions of X, Y, and Z after data stabilization is divided into h segments, where h 0 =np.math.ceil(H com / H fean ), thus obtaining the cutting force variation curve when machining the h layer in the i direction of the nth feature of the sth part and The available time series can be expressed as: in, The mth sampling time t is when processing the hth layer of the feature i of the nth part of the sth part ijh,m-sn The cutting force, M is the number of sampling points, 0≤m≤M, t k-sn =t ijh,0-sn ,t ijh,1-sn ,...,t ijh,M-sn Represents the sampling time series when machining the h layers in the i direction of the nth feature of the sth part.
[0073] Collect the initial time series of cutting force when machining the hth layer in the i direction of the part feature The processing in the i direction is divided into h 0 Segment, 1≤h≤h 0 ; h and h 0 All are integers, i = X, Y, Z. Specifically include:
[0074] The sampling time of the initial time series of the cutting force is normalized to obtain the first time series of the cutting force after time reconstruction, which specifically includes: normalizing the sampling time according to the formula Normalize to time, t k is the kth sampling time point, T k t k Normalized time, T t It represents the total processing time of each layer.
[0075] Cutting force variation curve Reconstruct according to the normalized time to obtain the cutting force variation curve when processing each feature, each direction, and each layer in the time range (0,1), that is:
[0076]
[0077] T ijh,k-s =T ijh,0-s ,T ijh,1-s ,...,T ijh,M-s It represents the normalized sampling time series when processing the h layers in the i direction of the n-th feature of the s-th part.
[0078] For each feature, each direction, and each layer, the cutting force variation curve is calculated within the time range (0, 1) according to the sampling point time interval p (p = (max (T ijh,k-s )-min(T ijh,k-s )) / 2) is reconstructed to form the time series of the cutting force variation curve of the hth layer in the i-th direction of the nth feature at the same time interval Right now:
[0079] If the time interval between adjacent cutting force sampling points in the first cutting force time series is greater than the set interval value, the interpolation method is used for interpolation; if the time interval between adjacent cutting force sampling points is less than the set interval value, the cutting force sampling points within the set interval value are discarded to obtain the second cutting force time series. That is to say, if the time interval between the sampling points of the original cutting force change curve in (0,1) is greater than p, the interpolation method is used for filling; if the time interval between the sampling points of the original cutting force change curve in (0,1) is less than p, the moving average method is used to process the over-sampling points, that is, the adjacent cutting force sampling points.
[0080] The second time series of cutting force is converted into a two-dimensional image using the Gram angle field method to obtain a cutting force image in the i direction. In this way, the cutting force signal of any part feature can be converted into 3×h 0 images, a total of 3×h images are obtained for each part 0 ×N images, for example, h 0 =2, N=3.
[0081] Classification of quality data. According to the design dimensional error requirements as the dividing line, the measured quality parameter data sets of each feature are divided into four categories: A, B, C, and D, which are recorded as (1,0,0,0), (0,1,0,0), (0,0,1,0), and (0,0,0,1), respectively, to form a quality label data set. Among them, Category A means that the machining errors of the long and wide sides of the part features are both within the qualified range, and all Category A data sets are recorded as qualified quality parameter data sets. Category B means that only the machining error of the long side of the preset feature is within the qualified range, and the machining error of the wide side of the feature is outside the qualified range. Category C means that only the machining error of the wide side of the preset feature is within the qualified range, and the machining error of the long side of the feature is outside the qualified range. Category D means that the machining errors of the long and wide sides of the preset feature are both outside the qualified range.
[0082] The quality pass rate prediction result is Class A, Class B, Class C or Class D, the setting condition is that the quality pass rate prediction result is Class A, or the setting condition is that the quality pass rate prediction result is Class A or Class B.
[0083] In an exemplary embodiment, Figure 3 As shown, the first neural network includes three feature extraction branches with the same structure, which are respectively a first feature extraction branch, a second feature extraction branch and a third feature extraction branch; the cutting force image in the X direction is input into the first feature extraction branch, the cutting force image in the Y direction is input into the second feature extraction branch, and the cutting force image in the Z direction is input into the third feature extraction branch.
[0084] The outputs of the three feature extraction branches in the first neural network are weighted averaged and then vector-concatenated with the static parameters to obtain a concatenated vector. The concatenated vector is sequentially subjected to data normalization, a fully connected layer, and a softmax function layer to output a quality pass rate prediction result.
[0085] The first feature extraction branch includes multiple feature extraction modules connected in sequence; the structure of each feature extraction module is the same; the feature extraction module includes a convolution layer and a pooling layer connected in sequence, and the convolution layer is a convolution layer integrating a self-attention mechanism; the convolution layer in the feature extraction module is the input, and the pooling layer is the output.
[0086] In an exemplary embodiment, Figure 4As shown, the second neural network includes three feature extraction branches with the same structure, which are the fourth feature extraction branch, the fifth feature extraction branch and the sixth feature extraction branch; the cutting force image in the X direction is input into the fourth feature extraction branch, the cutting force image in the Y direction is input into the fifth feature extraction branch, and the cutting force image in the Z direction is input into the sixth feature extraction branch. The fourth feature extraction branch, the fifth feature extraction branch and the sixth feature extraction branch have the same structure as the first feature extraction branch.
[0087] The outputs of the three feature extraction branches in the second neural network are weighted averaged and then vector-concatenated with the static parameters to obtain a concatenated vector. The concatenated vector is sequentially subjected to data normalization, a fully connected layer, a softmax function layer, a Relu function layer, and a logarithmic (log) transformation to output the processing error of the part feature.
[0088] Data splicing. The obtained static parameter dataset, dynamic parameter dataset, quality parameter dataset and quality label dataset are spliced separately to form four types of datasets, including static parameter-quality label dataset, dynamic parameter-quality label dataset, static parameter-qualified quality parameter dataset, and dynamic parameter-qualified quality parameter dataset.
[0089] Quality prediction model construction and training: Use static parameter-quality label data set and dynamic parameter-quality label data set to build a quality pass rate prediction model, and use static parameter-qualified quality parameter data set and dynamic parameter-qualified quality parameter data set to build a feature processing error prediction model. The quality pass rate prediction model is used to predict whether the processing errors of different features are qualified, and then the feature processing error prediction model is used to predict the specific error values of different features under the premise of qualification.
[0090] Construction and training of quality pass rate prediction model, including:
[0091] 1) Data imbalance processing. The obtained data set is processed by mixed sampling method. The minority class samples in the original data set are oversampled using the synthetic minority oversampling technique (SMOTE) algorithm; then the TomekLink connection sample pairs are found and cleaned up, and the noise samples generated by the SMOTE algorithm are removed to complete the mixed sampling.
[0092] Data normalization. Normalize the static parameters and quality label data. Use the Min-max normalization method to normalize the data, perform linear transformation on the static data, and map the static data to the interval [0,1].
[0093] 2) Construction of quality pass rate prediction model. The dynamic parameters and static parameters of a single feature (single part feature) are used as the input of the model, and the quality label data is used as the output to build a quality pass rate prediction model.
[0094] 21) h of the single characteristic cutting force signal in the X, Y, and Z directions are respectively 0 The images are weighted averaged according to the pixels to obtain an image covering all the features of the cutting force signal in the single feature X, Y, and Z directions, respectively, denoted as P ix , P iy , P iz .
[0095] 22) P ix , P iy , P iz Feature extraction is performed separately. Feature extraction is achieved through multiple layers of convolutional layers and pooling layers that integrate self-attention mechanisms. The number of layers may include 5 to 10 layers. The feature vectors after feature extraction are expressed as
[0096] 23) Perform weighted averaging to obtain the dynamic characteristic vector of a single characteristic cutting force And converted into the form of column vector.
[0097] 24) The static parameter P of a single feature ars Convert the vector form into column vector form, recorded as
[0098] 25) The dynamic parameter vector and static parameter vector of a single feature are concatenated in columns to form a cutting force feature vector of a single feature.
[0099] 26) The cutting force characteristic vector Normalize the data.
[0100] 27) The fully connected layer and softmax function layer are used to predict the quality label data. The grid search method and 5-fold cross validation method are used during the training process.
[0101] 3) Quality pass rate prediction model training: During the training process, a grid search method and a 5-fold cross validation method are used to obtain a quality pass rate prediction model with accuracy that meets the requirements.
[0102] Construction and training of qualified feature processing error prediction model. The dynamic parameters and static parameters of a single feature are used as the input of the model, and the qualified quality parameters are used as the output to construct the feature processing error prediction model.
[0103] Construction and training of feature processing error prediction model: The dynamic parameters and static parameters of a single feature are used as the input of the model, and the qualified quality parameters are used as the output to construct the feature processing error prediction model.
[0104] The process of constructing the feature processing error prediction model is the same as the process of constructing the quality pass rate prediction model.
[0105] The fully connected layer and ReLU function layer are used to predict the normalized qualified quality parameters.
[0106] The predicted results are log transformed to finally achieve regression prediction of qualified quality parameters.
[0107] The feature processing error prediction model training is completed to obtain a feature processing error prediction model with accuracy that meets the requirements.
[0108] According to the above steps, the present application realizes the construction of a quality prediction model for precision thin plate parts, which can realize the prediction of the quality pass rate of different features on precision thin plate parts, and at the same time, based on the predicted pass rate, realize the prediction of the actual processing errors of different features.
[0109] This application divides the various parameters involved in the processing of precision thin plate parts into dynamic parameters and static parameters, where static parameters include design, process, and tool, and dynamic parameters are cutting force parameters in the processing process, and the cutting force parameters are transformed from time series signals to image data through smooth processing, stratification, time normalization, reconstruction, etc. This application constructs a quality pass rate prediction model and a feature processing error prediction model for different features by combining a convolutional neural network with an attention mechanism. It can predict the processing pass rate of different features of thin plate parts, and can also predict the specific processing error of the feature on the basis of the predicted pass, thereby realizing the quality prediction of different processing features on thin plate parts. The prediction method is reasonable and feasible, with high prediction accuracy.
[0110] Based on the same inventive concept, the embodiment of the present application also provides a plate-like part processing quality detection device for implementing the plate-like part processing quality detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the plate-like part processing quality detection device provided below can refer to the limitations of the plate-like part processing quality detection method above, and will not be repeated here.
[0111] In an exemplary embodiment, the present application provides a plate-shaped parts processing quality detection device comprising:
[0112] The quality qualification prediction module is used to input the static parameters and dynamic parameters of a part feature of a plate-like part to be detected into a quality qualification rate prediction model to obtain a quality qualification rate prediction result of the part feature; the static parameters include design parameters, tool parameters and process parameters, and the dynamic parameters include cutting force parameters; the quality qualification rate prediction model is obtained by training a first neural network; the plate-like part to be detected includes at least one part feature.
[0113] The quality error prediction module is used to input the static parameters and dynamic parameters of the part feature into a feature processing error prediction model to obtain the processing error of the part feature if the quality pass rate prediction result meets the set conditions; the feature processing error prediction model is obtained by training the second neural network.
[0114] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store plate-like parts processing quality detection data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a plate-like parts processing quality detection method is implemented.
[0115] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.
[0116] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.
[0117] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.
[0118] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0119] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0120] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a data processing logic of a programmable logic device, etc., but is not limited thereto.
[0121] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0122] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A method for inspecting the processing quality of plate-shaped parts, characterized in that: The plate-like parts processing quality detection method comprises: Inputting static parameters and dynamic parameters of a part feature of a plate-like part to be inspected into a quality pass rate prediction model to obtain a quality pass rate prediction result of the part feature; the static parameters include design parameters, tool parameters and process parameters, and the dynamic parameters include cutting force parameters; the quality pass rate prediction model is obtained by training a first neural network; the plate-like part to be inspected includes at least one part feature; If the quality pass rate prediction result meets the set conditions, the static parameters and dynamic parameters of the part feature are input into a feature processing error prediction model to obtain the processing error of the part feature; the feature processing error prediction model is obtained by training the second neural network.
2. The plate-like parts processing quality inspection method according to claim 1 is characterized in that: The design parameters include part material, part dimensions, part feature type and the dimensions corresponding to each part feature; the part feature type includes circular hole feature, rectangular feature and crack feature; The tool parameters include tool material and tool diameter; The process parameters include cutting speed, feed rate and cutting depth when machining the part features; The cutting force parameters include cutting force changes in the X direction, cutting force changes in the Y direction, and cutting force changes in the Z direction when machining the part feature.
3. The plate-like parts processing quality inspection method according to claim 2, characterized in that: The cutting force variation in the i direction is represented as the cutting force image in the i direction; The process of obtaining the cutting force image in the i direction is: The initial time series of the cutting force when processing the hth layer in the i-direction of the part feature is collected; the processing in the i-direction is divided into h0 segments, 1≤h≤h0; h and h0 are both integers, i=X,Y,Z; Normalizing the sampling time of the initial time series of the cutting force to obtain a first time series of the cutting force after time reconstruction; If the time interval between adjacent cutting force sampling points in the first cutting force time series is greater than the set interval value, an interpolation method is used for interpolation; if the time interval between adjacent cutting force sampling points is less than the set interval value, a moving average method is used to process the adjacent cutting force sampling points to obtain a second cutting force time series; The second time series of cutting force is converted into a two-dimensional image by using the Gram angle field method to obtain a cutting force image in the i direction.
4. The plate-like parts processing quality inspection method according to claim 1, characterized in that: The first neural network includes three feature extraction branches with the same structure, which are respectively a first feature extraction branch, a second feature extraction branch and a third feature extraction branch; the cutting force image in the X direction is input into the first feature extraction branch, the cutting force image in the Y direction is input into the second feature extraction branch, and the cutting force image in the Z direction is input into the third feature extraction branch; The outputs of the three feature extraction branches in the first neural network are weighted averaged and then vector-concatenated with the static parameters to obtain a concatenated vector. The concatenated vector is sequentially subjected to data normalization, a fully connected layer, and a softmax function layer to output a quality pass rate prediction result.
5. The plate-like parts processing quality inspection method according to claim 4, characterized in that: The first feature extraction branch includes multiple feature extraction modules connected in sequence; the structure of each feature extraction module is the same; the feature extraction module includes a convolution layer and a pooling layer connected in sequence, and the convolution layer is a convolution layer integrating a self-attention mechanism; the convolution layer in the feature extraction module is the input, and the pooling layer is the output.
6. The plate-like parts processing quality inspection method according to claim 1, characterized in that: The second neural network includes three feature extraction branches with the same structure, the three feature extraction branches are respectively a fourth feature extraction branch, a fifth feature extraction branch and a sixth feature extraction branch; the cutting force image in the X direction is input into the fourth feature extraction branch, the cutting force image in the Y direction is input into the fifth feature extraction branch, and the cutting force image in the Z direction is input into the sixth feature extraction branch; The outputs of the three feature extraction branches in the second neural network are weighted averaged and then vector-concatenated with the static parameters to obtain a concatenated vector. The concatenated vector is sequentially subjected to data normalization, a fully connected layer, a softmax function layer, a Relu function layer, and a logarithmic transformation to output the processing error of the part features.
7. A plate-like parts processing quality inspection device, characterized in that: The plate-like parts processing quality detection device comprises: The quality qualification prediction module is used to input the static parameters and dynamic parameters of a part feature of the plate-like part to be detected into the quality qualification rate prediction model to obtain the quality qualification rate prediction result of the part feature; the static parameters include design parameters, tool parameters and process parameters, and the dynamic parameters include cutting force parameters; the quality qualification rate prediction model is obtained by training the first neural network; the plate-like part to be detected includes at least one part feature; The quality error prediction module is used to input the static parameters and dynamic parameters of the part feature into a feature processing error prediction model to obtain the processing error of the part feature if the quality pass rate prediction result meets the set conditions; the feature processing error prediction model is obtained by training the second neural network.
8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the plate-like parts processing quality detection method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the plate-like parts processing quality detection method described in any one of claims 1 to 6 is implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the plate-like parts processing quality detection method described in any one of claims 1 to 6 is implemented.