CN112069966B - Surface roughness prediction method based on improved LSTM network - Google Patents

Surface roughness prediction method based on improved LSTM network Download PDF

Info

Publication number
CN112069966B
CN112069966B CN202010893197.1A CN202010893197A CN112069966B CN 112069966 B CN112069966 B CN 112069966B CN 202010893197 A CN202010893197 A CN 202010893197A CN 112069966 B CN112069966 B CN 112069966B
Authority
CN
China
Prior art keywords
layer
input
data
activation function
surface roughness
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.)
Active
Application number
CN202010893197.1A
Other languages
Chinese (zh)
Other versions
CN112069966A (en
Inventor
黄玺宁
朱俊江
王君
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Jiliang University
Original Assignee
China Jiliang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China Jiliang University filed Critical China Jiliang University
Priority to CN202010893197.1A priority Critical patent/CN112069966B/en
Publication of CN112069966A publication Critical patent/CN112069966A/en
Application granted granted Critical
Publication of CN112069966B publication Critical patent/CN112069966B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)

Abstract

The application discloses a surface roughness prediction method based on an improved LSTM network, which comprises the following steps: s1, establishing a large data set of vibration signals and corresponding roughness values under different cutting parameters; s2, training the CNN network layer by adopting the large data set in the step S1 to obtain a CNN screening feature set. S3, carrying out manual screening feature on the big data in the step S1 to obtain a manual screening feature set. And S4, training the LSTM network layer by taking the CNN screening feature set in the step S2 and the manual screening feature set in the step S3 as feature data of an input gate to obtain a full feature set. S5, training the full feature set by adopting an FCN network to obtain a prediction model of the vibration signal corresponding to the roughness value. S6, inputting a vibration signal of the roughness to be detected in the prediction model of S5 to obtain a surface roughness value. The prediction model in the application can not filter out the characteristic with high correlation degree after changing the original signal after data preprocessing; the accuracy of part surface roughness prediction is improved.

Description

Surface roughness prediction method based on improved LSTM network
Technical Field
The application relates to the field of cutting machining, in particular to a surface roughness prediction method based on an improved LSTM network.
Background
The surface roughness is a main parameter for describing the surface microscopic morphology of the part and measuring the quality of the part, and not only affects the wear resistance, fatigue strength, corrosion resistance, sealing property and stability of the matching of the part, but also affects the surface optical property, electric conduction and heat conduction properties, appearance and the like of the part. The conventional surface roughness measurement method is mainly divided into contact measurement and non-contact measurement. The contact type measurement is limited in application in high-precision detection due to the fact that the measuring tip is easy to abrade and scratch the surface, the non-contact type measurement is sensitive to dirt on the surface of a part, cleaning is needed before measurement, and the roughness measurement efficiency is reduced.
The cutting parameters (spindle speed, feed rate, side draft) during cutting directly affect the surface roughness. At present, the control method of the surface roughness mainly draws out processing parameters according to experience, but the method depends on the working experience of operators, has strong subjectivity and is not beneficial to popularization and use.
Disclosure of Invention
The present application aims to provide a surface roughness prediction method based on an improved LSTM network, which can solve one or more of the above technical problems.
In order to achieve the above purpose, the technical scheme provided by the application is as follows:
the surface roughness prediction method based on the improved LSTM network comprises the following steps:
s1, taking vibration signals generated under different cutting parameter conditions as input, taking roughness values corresponding to the different cutting parameters as label data, and establishing a large data set of the vibration signals and the roughness values corresponding to the different cutting parameters;
s2, training the CNN network layer by adopting the big data set in the step S1 to obtain a CNN screening feature set;
s3, carrying out manual screening characteristics on the big data in the step S1 to obtain a manual screening characteristic set;
the manual screening feature set is mangade_feature= [ a, b, c, d ], wherein a is the mean value of the input signals of the CNN network layer, b is the variance of the input signals of the CNN network layer, c is the peak value of the input signals of the CNN network layer, and d is the kurtosis value of the input signals of the CNN network layer;
s4, training the LSTM network layer by taking the CNN screening feature set in the step S2 and the manual screening feature set in the step S3 as feature data of an input gate to obtain a full feature set so as to ensure the stability of feature extraction;
the LSTM network layer comprises a plurality of unit groups, and each unit group comprises a forgetting gate, an output gate and an input gate;
the calculation of the forgetting gate is shown in formula 1:
f t ＝σ(W f ·[h' t-1 ,x t ]+b f )
(1)
wherein f t For forgetting the door vector, W f Is the parameter to be trained, h' t-1 To improve the cell output at the last instant x t Is the output of the CNN network layer, b f Is a bias term of a forgetting gate, and sigma is an activation function sigmoid;
the input gate calculation is shown in equations 2, 3:
i t ＝σ(W i ·[h' t-1 ,x t ]+b i ) (2)
wherein sigma is an activation function sigmoid and tanh is an activation function tanh, W i 、W C For the parameters to be trained, h' t-1 In order to improve the cell output at the last instant,
b i 、b C as bias term, i t In order to input the gate vector,for the currently entered cell state vector, x t The input vector is the network at the current moment;
the output gate calculation is shown in formulas 4 and 5
o t ＝σ(W o [h' t-1 ,x t ]+b o )
(4)
h t ＝o t *tanh(C t )
(5)
Wherein sigma is an activation function sigmoid, W o For the parameters to be trained, h' t-1 In order to improve the cell output at the last instant,
b o is a bias term, o t To output the gate vector, h t For this time the cell output, x t C is the input vector of the network at the current moment t Is the cell state vector at the current time.
S5, training the full feature set by adopting an FCN network to obtain a prediction model of the vibration signal corresponding to the roughness value.
S6, inputting a vibration signal of the roughness to be detected in the prediction model of S5 to obtain a surface roughness value.
Further: the process in the step S1 is as follows:
s11, taking vibration signals under different milling parameter settings as experimental data; the corresponding workpiece surface roughness value under the milling parameters is used as label data of the model; the data collection corresponding to the big data set is not less than 1000 groups;
s12, preprocessing the original vibration signal in S11, wherein the preprocessing part comprises signal up-sampling, data cutting and digital filtering; preprocessing to obtain a data set XX= [ XX ] 1 ,xx 2 ,…,xx 1000 ]，XF＝[xf 1 ,xf 2 ,…,xf 1000 ]XX is the preprocessed vibration signal data set, XF is the extracted feature set, XX i Data matrix, xf, for a single signal i Is a feature matrix of a single signal.
Further: the CNN network layer in step S2 includes 5 layers, which are respectively:
the number of convolution layer filters in the first layer is 15, the convolution kernel size is 131×131, and the activation function is Relu; pool_size in the maximum pooling layer is 2;
the number of convolution layer filters in the second layer is 17, the convolution kernel size is 73×73, and the activation function is Relu; pool_size in the maximum pooling layer is 2;
the number of convolution layer filters in the third layer is 21, the convolution kernel size is 29 x 29, and the activation function is Relu; pool_size in the maximum pooling layer is 2;
the number of convolution layer filters in the fourth layer is 27, the convolution kernel size is 21 x 21, and the activation function is Relu; pool_size in the maximum pooling layer is 2;
the number of convolution layer filters in the fifth layer is 34, the convolution kernel size is 15×15, and the activation function is Relu; pool_size in the maximum pooling layer is 2.
Further: the node number of the full connection layer in the step S5 is 30, and the activation function is linear; xx for input data i ＝{x 1 ,x 2 ,…,x n Represented by n, where n is the length of a single input signal, x i The ith sampling point value is a single input signal; model parameters during training are: a loss function (mean_squared_error), an optimizer (adam), a learning rate (0.001), and a number of iterations epoch of 300.
The application has the technical effects that:
the roughness prediction method adopted in the application reduces the participation of human experience, avoids the interference caused by introducing human factors, and saves time and labor;
the improved LSTM network layer ensures that corresponding characteristics useful for model prediction are not lost in the process of characteristic extraction in the training process, and the prediction model does not filter the characteristics with high correlation degree after the original signals are changed after data preprocessing; the accuracy of part surface roughness prediction is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application.
In the drawings:
fig. 1 is a schematic of the workflow of the present application.
Fig. 2 is a schematic diagram of the network operation of the present application.
Detailed Description
The present application will be described in detail below with reference to the drawings and the specific embodiments thereof, wherein the exemplary embodiments and the description are for the purpose of illustrating the application only and are not to be construed as unduly limiting the application.
It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other. The application will be described in detail below with reference to the drawings in connection with embodiments.
As shown in fig. 1, a schematic workflow of the prediction method of the present application is shown. From the drawings, it can be determined that the main function of the LSTM network layer in the application is to keep the characteristic of high correlation and ensure the accuracy of prediction.
The specific process is as follows:
firstly, a triaxial acceleration sensor is used for respectively collecting vibration signals generated during milling under different cutting parameter settings as experimental data.
And after milling, measuring roughness of the processed workpiece surface, and taking the roughness value as label data of the model. And collecting more than 1000 pieces of data, wherein vibration signals and corresponding roughness values measured by different workpieces in each piece of data are taken as a group.
The raw vibration signal collected will be preprocessed, including signal up-sampling, data slicing and digital filtering.
The vibration signal is up-sampled to ensure that its sampling frequency is uniform at 20480Hz. The method comprises the following steps:
to avoid cutting instability at the beginning and end of the process, the first 500 samples and the last 500 samples of each vibration signal were discarded.
The sampling point data of the middle section are filtered by a low-pass filter to remove frequency components higher than 5 kHz.
Obtaining a vibration data set XX= [ XX ] after pretreatment 1 ,xx 2 ,…,xx 1000 ]，XF＝[xf 1 ,xf 2 ,…,xf 1000 ]XX is the preprocessed vibration signal data set, XF is the extracted feature set, XX i Data matrix, xf, for a single signal i Is a feature matrix of a single signal.
The above dataset xx= [ XX ] 1 ,xx 2 ,…,xx 1000 ]、XF＝[xf 1 ,xf 2 ,…,xf 1000 ]For input, the roughness dataset y= [ Y ] 1 ,y 2 ,…,y 1000 ]Is a label (where y i Xx respectively i Corresponding roughness value), pair ofTraining the network model to obtain a prediction model capable of predicting the roughness value;
wherein the prediction model consists of three parts of CNN+LSTM+FCN.
The first part is the CNN network layer (convolutional neural network), which consists of 5 layers:
the number of convolution layer filters in the first layer is 15, the convolution kernel size is 131×131, and the activation function is Relu. Pool_size in the maximum pooling layer is 2; the number of convolution layer filters in the second layer is 17, the convolution kernel size is 73 x 73, and the activation function is Relu. Pool_size in the maximum pooling layer is 2; the number of convolution layer filters in the third layer is 21, the convolution kernel size is 29 x 29, and the activation function is Relu. Pool_size in the maximum pooling layer is 2; the number of convolution layer filters in the fourth layer is 27, the convolution kernel size is 21 x 21, and the activation function is Relu. Pool_size in the maximum pooling layer is 2; the number of convolution layer filters in the fifth layer is 34, the convolution kernel size is 15 x 15, and the activation function is Relu. Pool_size in the maximum pooling layer is 2.
The second part is the LSTM network layer (long and short memory network),
the LSTM network layer is modified, the LSTM network layer comprising a number of cell groups, each cell group consisting of a forget gate, an output gate and a modified input gate.
The calculation of the forgetting gate is shown in formula 1:
f t ＝σ(W f ·[h' t-1 ,x t ]+b f )
(1)
wherein f t For forgetting the door vector, W f Is the parameter to be trained, h' t-1 To improve the cell output at the last instant x t Is the output of CNN layer, b f Is a bias term of a forgetting gate, and sigma is an activation function sigmoid;
the calculation formula of the input gate is improved, as shown in formulas 2 and 3
i t ＝σ(W i ·[h' t-1 ,x t ]+b i )
(2)
Wherein σ is the activation function sigmoid; tanh is the activation function tanh; w (W) i Is a parameter to be trained; w (W) C For the parameters to be trained, h' t-1 To improve the cell output at the last instant, b i Is a bias term; b C As bias term, i t In order to input the gate vector,for the currently entered cell state vector, x t Is the input vector of the network at the current time (i.e., the output of the CNN network layer).
Unlike conventional input gate computation, in the present applicationWherein, mangade_feature= [ a, b, c, d ]]The method comprises the steps of carrying out a first treatment on the surface of the a is the average value of the CNN network layer input signals; b is the variance of the CNN network layer input signal; c is the peak value of the CNN network layer input signal; d is the kurtosis value of the CNN network layer input signal.
In the application, the input signal of the CNN network layer is processed, and then the input signal is combined with the output of the CNN network layer and is added into the input gate of the LSTM network layer together; features with high correlation degree can be effectively reserved, corresponding features which are useful for model prediction are prevented from being lost by a CNN network layer during feature extraction, and accuracy of surface roughness prediction is improved.
The output gate calculation is shown in formulas 4 and 5
o t ＝σ(W o [h' t-1 ,x t ]+b o )
(4)
h t ＝o t *tanh(C t )
(5)
Wherein sigma is an activation function sigmoid, W o For the parameters to be trained, h' t-1 To improve the cell output at the last instant, b o Is a bias term, o t To output the gate vector, h t Unit input for this timeGo out, x t C is the input vector of the network at the current moment (namely the output of CNN) t Is the cell state vector at the current time.
The third part is that the node number of a full connection layer is 30, and the activation function is linear; continuing to train the data after the two network screening;
here, xx is used for input data i ＝{x 1 ,x 2 ,…,x n Represented by n, where n is the single input signal length (i.e., the number of input layer neurons), x i The i-th sample point value (i.e., the value of the i-th neuron) is given to a single input signal.
The model parameters trained at this time were: a loss function (mean_squared_error), an optimizer (adam), a learning rate (0.001), and a number of iterations epoch of 300.
And obtaining a final prediction model after the first part, the second part and the third part, and using the trained final prediction model for predicting the surface roughness of the workpiece.
Prediction of workpiece surface roughness under new milling parameters:
resetting milling parameters to process the workpiece, and collecting vibration signals in the process of processing; processing the signal according to the processing process; and obtaining XX ' and XF ' of the new vibration signals, taking XX ' and XF ' as inputs, and predicting by adopting a final trained prediction model to obtain a corresponding roughness value y ', namely the predicted surface roughness value of the machined workpiece.
The above description is only of the preferred embodiments of the present application and is not intended to limit the present application, but various modifications and variations can be made to the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims (4)

1. The surface roughness prediction method based on the improved LSTM network is characterized by comprising the following steps of: the method comprises the following steps:
s1, taking vibration signals generated under different cutting parameter conditions as input, taking roughness values corresponding to the different cutting parameters as label data, and establishing a large data set of the vibration signals and the roughness values corresponding to the different cutting parameters;
s2, training the CNN network layer by adopting the big data set in the step S1 to obtain a CNN screening feature set;
s3, carrying out manual screening characteristics on the big data in the step S1 to obtain a manual screening characteristic set;
the manual screening feature set is mangade_feature= [ a, b, c, d ], wherein a is the mean value of the input signals of the CNN network layer, b is the variance of the input signals of the CNN network layer, c is the peak value of the input signals of the CNN network layer, and d is the kurtosis value of the input signals of the CNN network layer;
s4, training the LSTM network layer by taking the CNN screening feature set in the step S2 and the manual screening feature set in the step S3 as feature data of an input gate to obtain a full feature set so as to ensure the stability of feature extraction;
the LSTM network layer comprises a plurality of unit groups, and each unit group comprises a forgetting gate, an output gate and an input gate;
the calculation of the forgetting gate is shown in formula 1:
f t ＝σ(W f ·[h' t-1 ,x t ]+b f )(1)
wherein f t For forgetting the door vector, W f Is the parameter to be trained, h' t-1 To improve the cell output at the last instant x t Is the output of the CNN network layer, b f Is a bias term of a forgetting gate, and sigma is an activation function sigmoid;
the input gate calculation is shown in equations 2, 3:
i t ＝σ(W i ·[h' t-1 ,x t ]+b i )(2)
wherein sigma is an activation function sigmoid and tanh is an activation function tanh, W i 、W C For the parameters to be trained, h' t-1 In order to improve the cell output at the last instant,
b i 、b C as bias term, i t In order to input the gate vector,for the currently entered cell state vector, x t The input vector is the network at the current moment;
the output gate calculation is shown in formulas 4 and 5
o t ＝σ(W o [h' t-1 ,x t ]+b o )(4)
h t ＝o t *tanh(C t )(5)
Wherein sigma is an activation function sigmoid, W o For the parameters to be trained, h' t-1 In order to improve the cell output at the last instant,
b o is a bias term, o t To output the gate vector, h t For this time the cell output, x t C is the input vector of the network at the current moment t A cell state vector at the current moment;
s5, training the full feature set by adopting an FCN network to obtain a prediction model corresponding to the vibration signal and the roughness value;
s6, inputting a vibration signal of the roughness to be detected in the prediction model of S5 to obtain a surface roughness value.
2. The improved LSTM network-based surface roughness prediction method of claim 1, wherein: the process in the step S1 is as follows:
s11, taking vibration signals under different milling parameter settings as experimental data; the corresponding workpiece surface roughness value under the milling parameters is used as label data of the model; the data collection corresponding to the big data set is not less than 1000 groups;
s12 to S11, preprocessing an original vibration signal, wherein the preprocessing part comprises signal up-sampling, data cutting and digital filtering; preprocessing to obtain a data set XX= [ XX ] 1 ,xx 2 ,…,xx 1000 ]，XF＝[xf 1 ,xf 2 ,…,xf 1000 ]XX is the preprocessed vibration signal data set, XF is the extracted feature set, XX i Data matrix, xf, for a single signal i Is a feature matrix of a single signal.
3. The improved LSTM network-based surface roughness prediction method of claim 1, wherein: the CNN network layer in step S2 includes 5 layers, which are respectively:
the number of convolution layer filters in the first layer is 15, the convolution kernel size is 131×131, and the activation function is Relu; pool_size in the maximum pooling layer is 2;
the number of convolution layer filters in the second layer is 17, the convolution kernel size is 73×73, and the activation function is Relu; pool_size in the maximum pooling layer is 2;
the number of convolution layer filters in the third layer is 21, the convolution kernel size is 29 x 29, and the activation function is Relu; pool_size in the maximum pooling layer is 2;
the number of convolution layer filters in the fourth layer is 27, the convolution kernel size is 21 x 21, and the activation function is Relu; pool_size in the maximum pooling layer is 2;
the number of convolution layer filters in the fifth layer is 34, the convolution kernel size is 15×15, and the activation function is Relu; pool_size in the maximum pooling layer is 2.
4. The improved LSTM network-based surface roughness prediction method of claim 1, wherein: the node number of the full connection layer in the step S5 is 30, and the activation function is linear; xx for input data i ＝{x 1 ,x 2 ,…,x n Represented by n, where n is the length of a single input signal, x i The ith sampling point value is a single input signal; model parameters during training are: the loss function is mean_squared_error, the optimizer is adam, the learning rate is 0.001, and the iteration number epoch is 300.
CN202010893197.1A 2020-08-31 2020-08-31 Surface roughness prediction method based on improved LSTM network Active CN112069966B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010893197.1A CN112069966B (en) 2020-08-31 2020-08-31 Surface roughness prediction method based on improved LSTM network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010893197.1A CN112069966B (en) 2020-08-31 2020-08-31 Surface roughness prediction method based on improved LSTM network

Publications (2)

Publication Number Publication Date
CN112069966A CN112069966A (en) 2020-12-11
CN112069966B true CN112069966B (en) 2023-11-28

Family

ID=73664765

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010893197.1A Active CN112069966B (en) 2020-08-31 2020-08-31 Surface roughness prediction method based on improved LSTM network

Country Status (1)

Country Link
CN (1) CN112069966B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112884081B (en) * 2021-03-31 2023-07-14 福州大学 On-line intelligent predicting method for quality of workpieces produced in small batches of multiple varieties
CN113592064B (en) * 2021-07-05 2024-02-09 中国科学院上海光学精密机械研究所 Ring polishing process parameter prediction method, system, application, terminal and medium
CN113899809B (en) * 2021-08-20 2024-02-27 中海石油技术检测有限公司 In-pipeline detector positioning method based on CNN classification and RNN prediction
CN114897028B (en) * 2022-05-27 2023-04-07 清华大学 Numerical control machining surface roughness prediction method and device
CN114756977B (en) * 2022-06-16 2022-10-25 成都飞机工业（集团）有限责任公司 Method, device and equipment for predicting boring cutter yield of intersection hole of airplane and storage medium
CN117495211B (en) * 2024-01-03 2024-03-19 东北大学 Industrial master machining workpiece quality prediction method based on self-adaptive period discovery

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004255535A (en) * 2003-02-27 2004-09-16 Gumma Prefecture Method of diagnosing processing state in cutting processing
CN110355608A (en) * 2019-07-18 2019-10-22 浙江大学 Based on the tool abrasion prediction technique from attention mechanism and deep learning
CN111177911A (en) * 2019-12-24 2020-05-19 大连理工大学 Part surface roughness online prediction method based on SDAE-DBN algorithm
CN111366123A (en) * 2020-03-06 2020-07-03 大连理工大学 Part surface roughness and cutter wear prediction method based on multi-task learning

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106897714B (en) * 2017-03-23 2020-01-14 北京大学深圳研究生院 Video motion detection method based on convolutional neural network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004255535A (en) * 2003-02-27 2004-09-16 Gumma Prefecture Method of diagnosing processing state in cutting processing
CN110355608A (en) * 2019-07-18 2019-10-22 浙江大学 Based on the tool abrasion prediction technique from attention mechanism and deep learning
CN111177911A (en) * 2019-12-24 2020-05-19 大连理工大学 Part surface roughness online prediction method based on SDAE-DBN algorithm
CN111366123A (en) * 2020-03-06 2020-07-03 大连理工大学 Part surface roughness and cutter wear prediction method based on multi-task learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于神经网络的铣削大理石表面粗糙度预测模型;苏晓云;汪建新;辛李霞;表面技术;第46卷(第8期);274-279 *

Also Published As

Publication number Publication date
CN112069966A (en) 2020-12-11

Similar Documents

Publication Publication Date Title
CN112069966B (en) Surface roughness prediction method based on improved LSTM network
CN111177911B (en) Part surface roughness online prediction method based on SDAE-DBN algorithm
CN111136509B (en) Cutter state monitoring method based on ELM-SDAE algorithm
Wang et al. On line tool wear monitoring based on auto associative neural network
CN114619292B (en) Milling cutter wear monitoring method based on fusion of wavelet denoising and attention mechanism with GRU network
CN111832432B (en) Cutter wear real-time prediction method based on wavelet packet decomposition and deep learning
CN111633467B (en) Cutter wear state monitoring method based on one-dimensional depth convolution automatic encoder
CN111475921A (en) Tool residual life prediction method based on edge calculation and L STM network
CN110263474A (en) A kind of cutter life real-time predicting method of numerically-controlled machine tool
CN110561192A (en) Deep hole boring cutter state monitoring method based on stacking self-encoder
Niu et al. Remaining useful life prediction of machining tools by 1D-CNN LSTM network
CN111783544B (en) Method for building diamond milling grinding head state monitoring system for processing ceramic mobile phone backboard
CN107622276B (en) Deep learning training method based on combination of robot simulation and physical sampling
CN113752089A (en) Cutter state monitoring method based on singularity Leersian index
CN113126564B (en) Digital twin driven numerical control milling cutter abrasion on-line monitoring method
CN113780208A (en) Cutter wear state prediction method and device based on anti-migration learning
CN111650891A (en) Five-axis precise small gantry numerical control machining center with constant-force adaptive control method
CN113554621A (en) Tool wear state identification system and method based on wavelet scale map and depth migration learning
CN114297912A (en) Tool wear prediction method based on deep learning
CN115741235A (en) Wear prediction and health management method based on five-axis machining center cutter
CN114462480A (en) Multi-source sensor rolling mill fault diagnosis method based on non-equilibrium data set
CN113609672B (en) Digital twin system co-evolution method based on incremental model
CN108956783B (en) HDP-HSMM-based grinding sound grinding wheel passivation state detection method
Wang et al. Research on tool remaining life prediction method based on CNN-LSTM-PSO
CN112720071A (en) Cutter real-time state monitoring index construction method based on intelligent fusion of multi-energy domain signals

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
GR01 Patent grant
GR01 Patent grant