Automobile body welding process quality score prediction method and device based on LSTM (Long Short Term Memory) model

A scoring prediction and welding process technology, applied in the field of auto body welding process quality scoring prediction, can solve the problems of lack of intelligent application, high labor cost, poor timeliness, etc., and achieve the effect of improving maintenance accuracy and reducing maintenance time.

Pending Publication Date: 2022-05-31
大唐互联科技(武汉)有限公司 +1
0 Cites 0 Cited by

AI-Extracted Technical Summary

Problems solved by technology

At present, the quality scoring of automobile body welding process mainly relies on the accumulation of manual experience, human judgment and evaluation of welding quality, lack...
View more

Method used

Leaf-wise strategy: in all current leaf nodes, select the node with the largest split benefit to split, so that recursively proceeds, reducing a lot of unnecessary overhead;
Regular term: in the cost function of model, has added regular term, avoids the situation of over-fitting occurring in model;
Set up the relative importance selection model of automobile body welding feature parameter: adopt LightGBM model to carry out relevant importance extraction to welding parameter, select more important parameter to carry out next step prediction, play the effect of simplified model and reduce computational difficulty. LightGBM is an integrated learning algorithm, which is a histogram-based decision tree algorithm. It has the advantages of fast calculation speed, good classification and regression effects, can handle large-scale data, and sup...
View more

Abstract

The invention relates to an automobile body welding process quality score prediction method and device based on an LSTM model, and the method comprises the steps: collecting welding spot quality sample data, including welding spot information parameters, environment parameters, welding equipment parameters and quality score parameters; carrying out training set and test set division on the selected feature parameters, carrying out feature relative importance analysis through an ensemble learning method, and extracting feature parameters with relatively high correlation; and carrying out data analysis and prediction on the extracted characteristic parameter data set through an LSTM deep learning model to obtain a quality scoring result of each welding process of the welding spots. According to the method, an LSTM deep learning prediction model is established, and through the characteristic that the deep neural network memorizes long sequence data, the defects that a traditional prediction model is prone to falling into a local minimum value, low in convergence speed, poor in generalization and the like are overcome, so that accurate prediction of the quality score of the automobile body welding process is achieved.

Application Domain

ForecastingNeural architectures +3

Technology Topic

Deep neural networksLong short term memory +11

Image

  • Automobile body welding process quality score prediction method and device based on LSTM (Long Short Term Memory) model
  • Automobile body welding process quality score prediction method and device based on LSTM (Long Short Term Memory) model
  • Automobile body welding process quality score prediction method and device based on LSTM (Long Short Term Memory) model

Examples

  • Experimental program(2)

Example Embodiment

Table 1: Model parameters of LightGBM
While saving to the solder joint information database, the system will detect that new feature data is generated in the fixed path, and then trigger the LSTM model.
While saving to the solder joint information database, the system will detect that new feature data is generated in the fixed path, and then trigger the LSTM model.
After accepting the data, the solder joint process parameters and information parameter data are stored in the database in the specified format.
Establish the relative importance selection model of automobile body welding characteristic parameter: adopt XGBoost model to weld parameter
[0056] histogram algorithm: the histogram algorithm reduces memory consumption, and when selecting the split feature to calculate the benefit
Leaf-wise strategy: in all current leaf nodes, select the node with the largest split income to split, so
Unilateral Gradient Sampling: Using GOSS can reduce a large number of data instances with only small gradients, so that in computing information
Mutually exclusive feature binding: many mutually exclusive features can be bound as one feature, thus achieving the purpose of dimensionality reduction;
Support category features: The category features supported by LightGBM can accelerate the training speed by 8 times, and the accuracy is one
Support efficient parallelism: do not perform vertical division of data, but save all training data on each machine.
LightGBM uses the greedy strategy and the quadratic optimization algorithm to iteratively allocate and optimize the parameters of each decision tree.
Table 1: Model parameters of LightGBM
[0064]
Preferably, in the process of training and testing LSTM model by described training data set and test data set
[0066]
Represents the minimum value of the sample data set, N" represents the number of samples
training set, validation set and test set. The training set is used to train the model, and the validation set is used to verify the fitting effect of the model and select the
The features are used as the feature data for the next training prediction model. Generally, the data features that account for 80% of the importance are selected, and the features are selected.
[0070] The data after selecting the relative importance of the feature is divided into a training set and a verification set again, and the divided data set is used.
Sparrow search algorithm is a kind of intelligent optimization algorithm, compared with other algorithms such as genetic algorithm, it has stronger global
LSTM is a special Recurrent Neural Network (RNN) that can effectively
Line search, and finally select the appropriate parameters, the number of hidden layer nodes, batch scale and time step of the standard LSTM model are respectively
The quasi-LSTM prediction model is used for performance evaluation, and the calculation formulas are as follows:
[0075]
[0076]
[0077]
[0078] In the formula, i represents the data serial number, yi, represents the actual value and the predicted value, respectively, and N represents the number of test sample sets.
The scoring data is analyzed and compared to obtain the most suitable prediction model.

Example Embodiment

Table 2: Model parameters of XGBoost
While saving to the solder joint information database, the system will detect that new feature data is generated in the fixed path, and then trigger the LSTM model.
Preferably, in the process of training and testing LSTM model by described training data set and test data set
After accepting the data, the solder joint process parameters and information parameter data are stored in the database in the specified format.
Establish the relative importance selection model of automobile body welding characteristic parameter: adopt XGBoost model to weld parameter
[0095] It should be noted that, herein, the terms "comprising", "comprising" or any other variant thereof are intended to cover non-exclusive
Parallel computing: parallel processing and computing are performed at the model feature level, and each feature gain is computed in a multi-threaded method.
[0087] Pruning process: first establish all subtrees that can be established from the top to the bottom, and then reversely prune from the bottom to the top to avoid
[0088] Column sampling: support column sampling, not only can reduce overfitting, but also reduce the amount of calculation, improve the simulation calculation speed;
[0089] Missing value processing: There are real samples for the value of the feature, and its splitting direction can be automatically learned.
The present invention builds XGBoost feature relative importance selection model, utilizes grid search algorithm to XGBoost model
Table 2: Model parameters of XGBoost
[0092]
In embodiment 2, except that the selected feature parameter relative importance selection model is different from that in embodiment 1, other
Redundancy makes the model trained from data less complex and enhances the robustness of the model. With the performance enhancement of deep learning,

PUM

no PUM

Description & Claims & Application Information

We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.

Similar technology patents

Anti-static rubber raw material stirring device

InactiveCN110405976AReduce maintenance timestir well
Owner:扬州赛思新材料科技有限公司

Device and method for detecting state of card machine internal sensor

InactiveCN104931084Aextended use timeReduce maintenance time
Owner:浪潮(山东)电子信息有限公司

Elastic reset pressure plate for selection piece

InactiveCN101519823AExtended service lifeReduce maintenance time
Owner:冯加林 +1

Detachable skid pressurizing integration system

ActiveCN101858207AReduce maintenance time
Owner:XIAN CHANGQING TECH ENG

Port physical link state detection method and system with external PHY chip

PendingCN113612652AReduce manual interventionReduce maintenance time
Owner:SUZHOU LANGCHAO INTELLIGENT TECH CO LTD

Classification and recommendation of technical efficacy words

  • Reduce maintenance time

System and method for monitoring outdoor display screen by using video image and/or digital photo

ActiveCN102184694AReduce maintenance timeImprove normal and efficient operation
Owner:BEIJING SULIANKE COMM EQUIP

Lead-in cable with a replaceable portion and method

ActiveUS20150020726A1reduce maintenance timeextend lifespan
Owner:SERCEL SAS

Automatic centering and damping brush component for collector ring

ActiveCN105470766AExtended service lifeReduce maintenance time
Owner:成都六三三机电科技有限公司

Configuration file monitoring method of server

ActiveCN110909394Aensure safetyReduce maintenance time
Owner:SUZHOU LANGCHAO INTELLIGENT TECH CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products