WO2023221476A1 - 一种基于机器学习预测船舶t梁焊接机器人作业时间的算法 - Google Patents

一种基于机器学习预测船舶t梁焊接机器人作业时间的算法 Download PDF

Info

Publication number
WO2023221476A1
WO2023221476A1 PCT/CN2022/138652 CN2022138652W WO2023221476A1 WO 2023221476 A1 WO2023221476 A1 WO 2023221476A1 CN 2022138652 W CN2022138652 W CN 2022138652W WO 2023221476 A1 WO2023221476 A1 WO 2023221476A1
Authority
WO
WIPO (PCT)
Prior art keywords
welding
model
robot
operating time
algorithm
Prior art date
Application number
PCT/CN2022/138652
Other languages
English (en)
French (fr)
Inventor
杨润党
董家琛
韦乃琨
孔宁
Original Assignee
上海船舶工艺研究所(中国船舶集团有限公司第十一研究所)
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 上海船舶工艺研究所(中国船舶集团有限公司第十一研究所) filed Critical 上海船舶工艺研究所(中国船舶集团有限公司第十一研究所)
Publication of WO2023221476A1 publication Critical patent/WO2023221476A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning

Definitions

  • the invention relates to the technical field of ship manufacturing, and in particular to an algorithm for predicting the operating time of a ship T-beam welding robot based on machine learning.
  • the T-beam welding robot on the thin plate segmentation line uses advanced offline programming software to plan the welding path and generate welding instructions for robot welding operations.
  • the entire welding process does not require human intervention, but due to some subjective and objective reasons, the welding is interrupted. Or extend, the theoretical operating time of welding time is quite different, the reasons are as follows:
  • the operation time is linearly distributed with the segmental weld meters. That is, the longer the ship segmented T-beam weld meters are, the longer the welding operation time will be. Further analysis , the higher the proportion of welding seam meters, neutral welding seams and through-hole welding seams, the longer the welding operation time.
  • the purpose of the present invention is to provide an algorithm based on machine learning to predict the operating time of a ship T-beam welding robot.
  • An algorithm based on machine learning to predict the operating time of ship T-beam welding robot which is characterized by including the following steps:
  • Step S1 Obtain the weld meters of multiple types of welds and record the data set of their actual operating time;
  • Step S2 Divide the data set into a training set and a test set according to the time dimension, and load the segmented samples (X, Y) of the training set and test set respectively;
  • Step S3 Normalize the weld meters of different weld types
  • Step S4 Use the gradient descent method to solve and build a multi-type weld linear model
  • the multi-type weld linear model is
  • x 1 is the number of flat welding meters
  • w 1 is the weight of the flat welding meters affecting the robot's operating time
  • x 2 is the vertical welding meters
  • w 2 is the weight of the vertical welding meters affecting the robot's operating time
  • x 3 is the welding meter of the patch plate
  • w 3 is the weight of the welding meter of the patch plate that affects the robot's operating time
  • x 4 is the welding meter of the bracket plate
  • w 4 is the weight of the welding meter of the bracket plate that affects the robot's operating time
  • x 5 is the vertical welding meter of the through hole
  • w 5 is the weight of the vertical welding meter of the through hole that affects the robot's operating time
  • b is the linear model constant
  • Step S5 Use the mean square error loss function Evaluate the model to predict the degree of inconsistency between the T-beam robot operating time and the actual operating time;
  • Step S6 Set initial values of hyperparameters and model variables
  • Step S7 Use the gradient descent method to train the model, find the minimum value of the Loss function, and update the algorithm of the weight W;
  • Step S8 Visualize the results, observe the performance of the Loss function value on the test set data, find out the training times i that minimize the generalization error between the training set Loss function value and the test set Loss function value, retrain the model, and calculate the corresponding With the weight W, a multi-type weld linear model is obtained;
  • Step S9 Predict the operation time of the new ship segment sample data on the model.
  • each segmented T-beam welding operation stage includes flat welding operation, vertical welding operation, repair plate welding operation, toggle plate welding operation and through-hole vertical welding operation, respectively using "HORZ” and "VERTI” , "CL”, “BK” and “VB” indicate that each piece of data in the data set records the actual welding operation time and total actual operation time of a segmented T-beam.
  • step S2 includes slicing the entire period, the past year, the past half year, the past month, and the same period over the years.
  • step S4 includes selecting the number of independent variables and determining the model type
  • the model is a one-dimensional linear model, which is expressed as a line on the two-dimensional coordinate axis;
  • the model is a binary linear model, which is represented by a surface on the three-dimensional coordinate axis;
  • the model is a multivariate linear model, which behaves as a hyperplane.
  • the hyperparameters include the learning rate ⁇ and the number of iterations iter, the initial variable is the weight vector W, and W is initialized to a (w 0 , w 1 , w 2 , w 3 , w 4 , w 5 ) Random value.
  • step S6 includes expressing the Loss function as a vector form
  • X ( X 1 , X is a matrix with n rows and 6 columns;
  • step S8 includes retraining the model, and after calculating the corresponding weight W, denormalizing the weight W.
  • This invention builds a multi-type weld linear model function through the segmented weld sample data characteristics, trains the weight of each weld type in the model that affects the operating time, and predicts the operating time of the new segment in the robot welding station. It can be applied In a variety of robot-based ship manufacturing workshops, such as small group assembly, medium assembly, painting and other workshops that meet the linear or near-linear distribution of operation volume and operation time, it can improve the accuracy of ship manufacturing workshop operation time prediction.
  • Figure 1 is a flow chart of the present invention using the gradient descent method to obtain linear model functions for multiple types of welds
  • Figure 2 is a view of the input sample of the data set of the present invention
  • Figure 3 is a diagram showing the relationship between each welding seam type and operation time according to the present invention.
  • Figure 4 is a diagram of the distribution of training set data according to the present invention.
  • Figure 5 is a diagram of the test set data distribution of the present invention.
  • Figure 6 is a record view of the training loss and test loss of the present invention decreasing with the number of learning times
  • Figure 7 is a visual view of the decrease in training loss and test loss with the number of iterations during the learning process of the present invention
  • Figure 8 is a visual view comparing the training set of the present invention with the actual situation
  • Figure 9 is a visual view comparing the test set training of the present invention with the actual situation.
  • This embodiment discloses an algorithm based on machine learning to predict the operating time of ship T-beam welding robot, as shown in Figure 1, including the following steps:
  • Step S1 obtain the T-beam welding robot offline programming software plan to obtain the weld meters of multiple types of welds, as well as a data set recording its actual operating time, each segmented T-beam welding stage Including various types of welding operations, specifically flat welding, vertical welding, repair plate welding, toggle plate welding, and through-hole vertical welding operations, respectively using "HORZ”, “VERTI”, “CL”, “BK”, " VB” indicates that each piece of data in the data set records the actual welding operation time and total actual operation time of a segmented T-beam.
  • Step S2 divide the data set into a training set and a test set according to the time dimension, and slice the entire period, the past year, the past half year, the past month, and the same period over the years, or randomly.
  • Divide the data set into a training set and a test set for example, divide the training set and test set according to 7:3, and load the segmented samples (X, Y) of the training set and test set respectively.
  • Step S3 Normalize the number of weld meters of different weld types. Different types of welds in the data sample have different effects on operating time and weight ratios. In order to improve the learning accuracy, the weld data are The value of is limited to a certain range, so that all types of welds are in the same range and the same order of magnitude, so that the model converges faster and the optimal solution is obtained.
  • Step S4 Select the number of independent variables and determine the model type.
  • the total weld meter as the independent variable to predict the robot operating time, it is a one-dimensional linear model, which is expressed as a line on the two-dimensional coordinate axis;
  • Welding seam prediction of robot operating time is a binary linear model, which is expressed as a surface on a three-dimensional coordinate axis;
  • three or more types of welding seams are used to predict robot operating time, it is a multivariate linear model, which is expressed as a hyperplane.
  • the multi-type weld linear model is
  • x 1 is the number of flat welding meters, w 1 is the weight of the flat welding meters that affects the robot’s operating time;
  • x 2 is the number of vertical welding meters, w 2 is the weight of the vertical welding meters that affects the robot’s operating time;
  • x 3 is the number of meters of patch welding
  • w 3 is the weight of the number of meters of patch welding that affects the robot’s operating time
  • x 4 is the welding meter of the toggle plate
  • w 4 is the weight of the welding meter of the toggle plate that affects the robot's operating time
  • x 5 is the vertical welding meter of the through hole
  • w 5 is the weight of the vertical welding meter of the through hole that affects the robot’s operating time
  • Step S5 Use the mean square error loss function Evaluate the model to predict the degree of inconsistency between the T-beam robot operating time and the actual operating time.
  • Step S6 Set the initial values of hyperparameters and model variables.
  • the hyperparameters include the learning rate ⁇ and the number of iterations iter.
  • the initial value of the model variables refers to the weight vector W, which is initialized to one (w 0 , w 1 , w 2 , w 3 , w 4 ,w 5 ) random values.
  • Step S7 Use the gradient descent method to train the model, specifically, find the minimum value of the mean square error loss function Loss, and express Loss as a vector form,
  • X ( X 1 , X is a matrix with n rows and 6 columns;
  • Step S8 Observe the performance of the Loss function value on the test set data, and find the training times i that minimize the generalization error between the training set Loss function value and the test set Loss function value.
  • the training set Loss function value Marked as Loss_train the test set Loss function value is marked as Loss_test.
  • the test set error begins to rise between 600 and 700 times of training. When the generalization error is minimum, i is about 600. Use this as a benchmark to retrain the model.
  • Step S9 Input the weld data of the new segment, and predict the operation time of the new ship segment sample data on the model based on the multi-type weld linear model.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Manipulator (AREA)
  • Feedback Control In General (AREA)
  • Numerical Control (AREA)

Abstract

本申请提供一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，包括：获取T梁焊接机器人离线编程软件规划得出多类型焊缝的焊缝米数，以及记录其实际作业时间的数据集；按照时间维度把数据集划分为训练集和测试集；对不同焊缝类型的焊缝米数进行归一化处理；使用梯度下降法求解并构建多类型焊缝线性模型；设置超参数和模型变量初始值；训练模型；结果可视化输出评估模型；预测新的船舶分段样本数据在模型上的作业时间；通过分段焊缝样本数据特征，构建多类型焊缝线性模型函数，训练出模型中各焊缝类型影响作业时间的权值，预测新分段在机器人焊接工位的作业时间，可应用于机器人船舶制造车间，提高船舶制造车间作业时间预测的准确性。

Description

一种基于机器学习预测船舶T梁焊接机器人作业时间的算法 技术领域
本发明涉及船舶制造技术领域，尤其涉及一种基于机器学习预测船舶T梁焊接机器人作业时间的算法。
背景技术
薄板分段流水线上T梁焊接机器人通过先进的离线编程软件，对焊接路径进行规划，生成焊接指令用于机器人焊接作业；理论上焊接全过程无需人员干预，但由于一些主客观原因，导致焊接中断或延长，焊接时间理论作业时间差距较大，存在原因如下：
①机器人操作人员技能水平不达标；
②焊材受潮；
③补板自由边切面不平整；
④焊接路径及焊接参数不匹配；
⑤分段摆放与机器人门架存在较大夹角；
⑥定位焊焊脚尺寸、位置影响机器人寻位和弧压跟踪；
⑦前道装配间隙不达标，一些焊缝无法作业，或模型细节未处理造成动作干涉；
⑧枪头导电性能不佳、清枪不到位有残渣、焊机偶发性故障。
通过对大量实船分段焊接作业时间的记录以及经验估计，作业时间与分段焊缝米数呈线性分布，即船舶分段T梁焊缝米数越长，焊接作业时间越长，进一步分析，焊缝米数中立焊缝、贯穿孔焊缝所占比例越高，焊接作业时间越长。
发明内容
本发明的目的是提供一种基于机器学习预测船舶T梁焊接机器人作业时间的算法。
为了实现上述目的，本发明的技术方案是：
一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，其特征在于，包括以下步骤：
步骤S1、获取多类型焊缝的焊缝米数，记录其实际作业时间的数据集；
步骤S2、按照时间维度把数据集划分为训练集和测试集，分别加载训练集和测试集的分段样本(X,Y)；
步骤S3、对不同焊缝类型的焊缝米数进行归一化处理；
步骤S4、使用梯度下降法求解并构建多类型焊缝线性模型，
多类型焊缝线性模型为
Figure PCTCN2022138652-appb-000001
其中，x 1为平焊米数，w 1为平焊米数影响机器人作业时间的权值，x 2为立焊米数，w 2为立焊米数影响机器人作业时间的权值，x 3为补板焊米数，w 3为补板焊米数影响机器人作业时间的权值，x 4为肘板焊米数，w 4为肘板焊米数影响机器人作业时间的权值，x 5为贯穿孔立焊米数，w 5为贯穿孔立焊米数影响机器人作业时间的权值，b为线性模型常数；
令b＝w 0，x 0＝1，将多类型焊缝线性模型转换为向量形式
Figure PCTCN2022138652-appb-000002
Figure PCTCN2022138652-appb-000003
其中W＝(w 0,w 1,w 2,w 3,w 4,w 5) T,X＝(x 0,x 1,x 2,x 3,x 4,x 5) T,
假设有n个分段样本(X i,y i)，(i＝1,2,…,n),则
Figure PCTCN2022138652-appb-000004
步骤S5、采用均方差损失函数
Figure PCTCN2022138652-appb-000005
评估模型，预测T梁机器人作业时间与实际作业时间不一致程度；
步骤S6、设置超参数和模型变量初始值；
步骤S7、使用梯度下降法训练模型，求Loss函数的最小值，更新权值W的算法；
步骤S8、结果可视化，观察Loss函数值在测试集数据上的表现，找出训练集Loss函数值和测试集Loss函数值之间泛化误差最小的训练次数i，重新训练模型，计算得出相应的权值W，得到多类型焊缝线性模型；
步骤S9、预测新的船舶分段样本数据在模型上的作业时间。
进一步地，步骤S1中，每个分段T梁焊接作业阶段包括平焊作业、立焊作 业、补板焊作业、肘板焊作业和贯穿孔立焊作业，分别用“HORZ”、“VERTI”、“CL”、“BK”和“VB”表示，数据集中每条数据均记录一个分段T梁实际焊接作业时间和实际作业总时间。
进一步地，步骤S2包括，对全时段、近一年、近半年、近一个月、历年同期进行切片。
进一步地，步骤S4包括，选取自变量的数量，确定模型类型，
当采用总焊缝米数作为自变量预测机器人作业时间，模型为一元线性模型，表现为二维坐标轴上的一条线；
当采用两种类型焊缝预测机器人作业时间，模型为二元线性模型，表现为三维坐标轴上的一个面；
当采用三类及以上焊缝预测机器人作业时间，模型为多元线性模型，表现为一个超平面。
进一步地，步骤S6中，超参数包括学习率η和迭代次数iter，初始变量为权值向量W，W初始化为一个(w 0,w 1,w 2,w 3,w 4,w 5)的随机值。
进一步地，步骤S6包括，将Loss函数表示为向量形式
Figure PCTCN2022138652-appb-000006
Figure PCTCN2022138652-appb-000007
其中，X＝(X 1,X 2,…,X n) T，Y＝(y 1,y 2,…,y n) T，
Figure PCTCN2022138652-appb-000008
X为一个n行6列的矩阵；
求解
Figure PCTCN2022138652-appb-000009
函数最小值，
对W求偏导
Figure PCTCN2022138652-appb-000010
权值更新算法
Figure PCTCN2022138652-appb-000011
进一步地，步骤S8包括，重新训练模型，计算得出相应的权值W后，对权值W进行反归一化处理。
本发明通过分段焊缝样本数据特征，构建多类型焊缝线性模型函数，训练出 模型中各焊缝类型影响作业时间的权值，预测新分段在机器人焊接工位的作业时间，可应用于多种基于机器人的船舶制造车间，如小组立、中组立、涂装等满足作业物量与作业时间呈线性或接近线性分布的车间，提高船舶制造车间作业时间预测的准确性。
附图说明
图1为本发明使用梯度下降法求多类型焊缝线性模型函数的流程图；
图2为本发明数据集输入样本视图；
图3为本发明各焊缝类型与作业时间的关系图；
图4为本发明训练集数据分布情况图；
图5为本发明测试集数据分布情况图；
图6为本发明训练损失和测试损失随学习次数下降的记录视图；
图7为本发明学习过程中训练损失和测试损失随迭代次数的下降情况可视化视图；
图8为本发明训练集训练完成后与实际对比可视化视图；
图9为本发明测试集训练完成后与实际对比可视化视图。
具体实施方式
下面将结合附图对本发明的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本实施例公开一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，如图1所示，包括以下步骤：
步骤S1、如图2、3所示，获取T梁焊接机器人离线编程软件规划得出多类型焊缝的焊缝米数，以及记录其实际作业时间的数据集，每个分段T梁焊接阶段包括多种焊缝类型的作业，具体为平焊、立焊、补板焊、肘板焊、贯穿孔立焊作业，分别用“HORZ”，“VERTI”，“CL”，“BK”，“VB”表示，数据集中每条数据都记录一个分段T梁实际焊接作业时间和实际作业总时间。
步骤S2、如图4、5所示，按照时间维度把数据集划分为训练集和测试集，对全时段、近一年、近半年、近一个月、历年同期进行切片，也可按随机方式将 数据集划分为训练集和测试集，例如，按7:3划分训练集和测试集，分别加载训练集和测试集的分段样本(X,Y)。
步骤S3、对不同焊缝类型的焊缝米数进行归一化处理，数据样本中不同类型的焊缝对作业时间影响和所占权重比是不一样的，为提高学习精度，将焊缝数据的值限定在一定范围内，使所有类型焊缝处于同一个范围、同一个数量级下，使其模型更快收敛，得到最优解。
步骤S4、选取自变量数量，确定模型类型，当采用总焊缝米数作为自变量预测机器人作业时间，是一个一元线性模型，表现为二维坐标轴上的一条线；当采用两种类型焊缝预测机器人作业时间，是一个二元线性模型，表现为三维坐标轴上的一个面；当采用三类及以上焊缝预测机器人作业时间，是一个多元线性模型，表现为一个超平面。
多类型焊缝线性模型为
Figure PCTCN2022138652-appb-000012
x 1为平焊米数，w 1为平焊米数影响机器人作业时间的权值；
x 2为立焊米数，w 2为立焊米数影响机器人作业时间的权值；
x 3为补板焊米数，w 3为补板焊米数影响机器人作业时间的权值；
x 4为肘板焊米数，w 4为肘板焊米数影响机器人作业时间的权值；
x 5为贯穿孔立焊米数，w 5为贯穿孔立焊米数影响机器人作业时间的权值；
b是线性模型常数。
令b＝w 0，x 0＝1，将多类型焊缝线性模型转换为向量形式
Figure PCTCN2022138652-appb-000013
Figure PCTCN2022138652-appb-000014
其中W＝(w 0,w 1,w 2,w 3,w 4,w 5) T,X＝(x 0,x 1,x 2,x 3,x 4,x 5) T,
若假设有n个分段样本(X i,y i)，(i＝1,2,…,n),则
Figure PCTCN2022138652-appb-000015
步骤S5、采用均方差损失函数
Figure PCTCN2022138652-appb-000016
评估模型，预测T梁机器人作业时间与实际作业时间不一致程度。
步骤S6、设置超参数和模型变量初始值，超参数包括学习率η和迭代次数iter，模型变量初始值指权值向量W，初始化为一个(w 0,w 1,w 2,w 3,w 4,w 5)的随机值。
步骤S7、使用梯度下降法训练模型，具体为，求均方差损失函数Loss的最 小值，将Loss表示为向量形式，
Figure PCTCN2022138652-appb-000017
其中，X＝(X 1,X 2,…,X n) T，Y＝(y 1,y 2,…,y n) T，
Figure PCTCN2022138652-appb-000018
X为n行6列的矩阵；
求解
Figure PCTCN2022138652-appb-000019
函数最小值，
对W求偏导
Figure PCTCN2022138652-appb-000020
权值更新算法
Figure PCTCN2022138652-appb-000021
步骤S8、观察Loss函数值在测试集数据上的表现，找出训练集Loss函数值和测试集Loss函数值之间泛化误差最小的训练次数i，如图6所示，训练集Loss函数值标识为Loss_train，测试集Loss函数值标识为Loss_test，在训练600～700次之间测试集误差开始上升，泛化误差最小时i为600左右，以此为基准重新训练模型。
如图7～9所示，进行训练过程和结果的可视化输出，计算得出在泛化误差最小的权值W，对权值W进行反归一化处理，得到较佳的多类型焊缝线性模型，
y＝0.02093x1+0.07075x2+0.21526x3+0.10064x4+0.09198x5+1.63374。
步骤S9、输入新分段的焊缝数据，根据多类型焊缝线性模型预测新的船舶分段样本数据在模型上的作业时间。
最后应说明的是：以上各实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述各实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分或者全部技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的范围。

Claims (7)

  1. 一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，其中，包括以下步骤：
    步骤S1、获取多类型焊缝的焊缝米数，记录其实际作业时间的数据集；
    步骤S2、按照时间维度把数据集划分为训练集和测试集，分别加载训练集和测试集的分段样本(X,Y)；
    步骤S3、对不同焊缝类型的焊缝米数进行归一化处理；
    步骤S4、使用梯度下降法求解并构建多类型焊缝线性模型，
    多类型焊缝线性模型为
    Figure PCTCN2022138652-appb-100001
    其中，x 1为平焊米数，w 1为平焊米数影响机器人作业时间的权值，x 2为立焊米数，w 2为立焊米数影响机器人作业时间的权值，x 3为补板焊米数，w 3为补板焊米数影响机器人作业时间的权值，x 4为肘板焊米数，w 4为肘板焊米数影响机器人作业时间的权值，x 5为贯穿孔立焊米数，w 5为贯穿孔立焊米数影响机器人作业时间的权值，b为线性模型常数；
    令b＝w 0，x 0＝1，将多类型焊缝线性模型转换为向量形式
    Figure PCTCN2022138652-appb-100002
    Figure PCTCN2022138652-appb-100003
    其中W＝(w 0,w 1,w 2,w 3,w 4,w 5) T,X＝(x 0,x 1,x 2,x 3,x 4,x 5) T,
    假设有n个分段样本(X i,y i)，(i＝1,2,…,n),则
    Figure PCTCN2022138652-appb-100004
    步骤S5、采用均方差损失函数
    Figure PCTCN2022138652-appb-100005
    评估模型，预测T梁机器人作业时间与实际作业时间不一致程度；
    步骤S6、设置超参数和模型变量初始值；
    步骤S7、使用梯度下降法训练模型，求Loss函数的最小值，更新权值W的算法；
    步骤S8、结果可视化，观察Loss函数值在测试集数据上的表现，找出训练集Loss函数值和测试集Loss函数值之间泛化误差最小的训练次数i，重新训练模型，计 算得出相应的权值W，得到多类型焊缝线性模型；
    步骤S9、预测新的船舶分段样本数据在模型上的作业时间。
  2. 根据权利要求1所述的一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，其中，步骤S1中，每个分段T梁焊接作业阶段包括平焊作业、立焊作业、补板焊作业、肘板焊作业和贯穿孔立焊作业，分别用“HORZ”、“VERTI”、“CL”、“BK”和“VB”表示，数据集中每条数据均记录一个分段T梁实际焊接作业时间和实际作业总时间。
  3. 根据权利要求1所述的一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，其特征在于，步骤S2包括，对全时段、近一年、近半年、近一个月、历年同期进行切片。
  4. 根据权利要求1所述的一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，其中，步骤S4包括，选取自变量的数量，确定模型类型，
    当采用总焊缝米数作为自变量预测机器人作业时间，模型为一元线性模型，表现为二维坐标轴上的一条线；
    当采用两种类型焊缝预测机器人作业时间，模型为二元线性模型，表现为三维坐标轴上的一个面；
    当采用三类及以上焊缝预测机器人作业时间，模型为多元线性模型，表现为一个超平面。
  5. 根据权利要求1所述的一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，其中，步骤S6中，超参数包括学习率η和迭代次数iter，初始变量为权值向量W，
    W初始化为一个(w 0,w 1,w 2,w 3,w 4,w 5)的随机值。
  6. 根据权利要求5所述的一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，其中，步骤S7包括，将Loss函数表示为向量形式
    Figure PCTCN2022138652-appb-100006
    Figure PCTCN2022138652-appb-100007
    其中，X＝(X 1,X 2,…,X n) T，Y＝(y 1,y 2,…,y n) T，
    Figure PCTCN2022138652-appb-100008
    X为一个n行6列的矩阵；
    求解
    Figure PCTCN2022138652-appb-100009
    函数最小值，
    对W求偏导
    Figure PCTCN2022138652-appb-100010
    更新算法权值
    Figure PCTCN2022138652-appb-100011
  7. 根据权利要求6所述的一种基于机器学习预测船舶T梁焊接机器人作业时间的算法，其中，步骤S8包括，重新训练模型，计算得出相应的权值W后，对权值W进行反归一化处理。
PCT/CN2022/138652 2022-05-19 2022-12-13 一种基于机器学习预测船舶t梁焊接机器人作业时间的算法 WO2023221476A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210545352.X 2022-05-19
CN202210545352.XA CN115270592A (zh) 2022-05-19 2022-05-19 一种基于机器学习预测船舶t梁焊接机器人作业时间的算法

Publications (1)

Publication Number Publication Date
WO2023221476A1 true WO2023221476A1 (zh) 2023-11-23

Family

ID=83759312

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/138652 WO2023221476A1 (zh) 2022-05-19 2022-12-13 一种基于机器学习预测船舶t梁焊接机器人作业时间的算法

Country Status (2)

Country Link
CN (1) CN115270592A (zh)
WO (1) WO2023221476A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117733439A (zh) * 2024-01-19 2024-03-22 西咸新区大熊星座智能科技有限公司 一种具有焊缝成形质量实时预测功能的焊接机器人

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115270592A (zh) * 2022-05-19 2022-11-01 上海船舶工艺研究所(中国船舶工业集团公司第十一研究所) 一种基于机器学习预测船舶t梁焊接机器人作业时间的算法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110023610A1 (en) * 2009-08-03 2011-02-03 Georgia Tech Research Corporation Methods and systems for classifying the type and severity of defects in welds
CN110837959A (zh) * 2019-10-31 2020-02-25 上海船舶工艺研究所(中国船舶工业集团公司第十一研究所) 一种基于焊接物量和作业方式衡定焊接派工工时的方法
CN112364560A (zh) * 2020-10-19 2021-02-12 武汉理工大学 矿山凿岩装备作业工时智能预测方法
CN113706116A (zh) * 2021-09-01 2021-11-26 上海外高桥造船有限公司 船舶焊接工时确定方法、装置、计算机设备和存储介质
CN115270592A (zh) * 2022-05-19 2022-11-01 上海船舶工艺研究所(中国船舶工业集团公司第十一研究所) 一种基于机器学习预测船舶t梁焊接机器人作业时间的算法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110023610A1 (en) * 2009-08-03 2011-02-03 Georgia Tech Research Corporation Methods and systems for classifying the type and severity of defects in welds
CN110837959A (zh) * 2019-10-31 2020-02-25 上海船舶工艺研究所(中国船舶工业集团公司第十一研究所) 一种基于焊接物量和作业方式衡定焊接派工工时的方法
CN112364560A (zh) * 2020-10-19 2021-02-12 武汉理工大学 矿山凿岩装备作业工时智能预测方法
CN113706116A (zh) * 2021-09-01 2021-11-26 上海外高桥造船有限公司 船舶焊接工时确定方法、装置、计算机设备和存储介质
CN115270592A (zh) * 2022-05-19 2022-11-01 上海船舶工艺研究所(中国船舶工业集团公司第十一研究所) 一种基于机器学习预测船舶t梁焊接机器人作业时间的算法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117733439A (zh) * 2024-01-19 2024-03-22 西咸新区大熊星座智能科技有限公司 一种具有焊缝成形质量实时预测功能的焊接机器人
CN117733439B (zh) * 2024-01-19 2024-05-28 西咸新区大熊星座智能科技有限公司 一种具有焊缝成形质量实时预测功能的焊接机器人

Also Published As

Publication number Publication date
CN115270592A (zh) 2022-11-01

Similar Documents

Publication Publication Date Title
WO2023221476A1 (zh) 一种基于机器学习预测船舶t梁焊接机器人作业时间的算法
CN106886213B (zh) 一种基于核相似度支持向量数据描述的间歇过程故障检测方法
JP5068637B2 (ja) 製造プロセスにおける操業と品質の関連解析装置、解析方法、プログラム、及びコンピュータ読み取り可能な記録媒体
CN106217128B (zh) 基于大数据挖掘的变工况下刀具磨损状态预测方法
JP6404909B2 (ja) 技術的な系の出力量のモデルを算出する方法
CN107506941A (zh) 一种基于大数据技术的建筑施工企业信用评价方法和系统
Wang et al. Flood control operations based on the theory of variable fuzzy sets
Klochkov et al. Information model-based forecasting of technological process state
CN101893884A (zh) 密炼机的橡胶混炼过程中质量指标数据的软测量方法
CN110991495B (zh) 生产制造过程中产品质量预测方法、系统、介质及设备
CN107464017A (zh) 基于带时间差分贝叶斯网络的自适应软测量预测方法
CN109241493A (zh) 基于马尔可夫随机场和em算法的关键性能指标软测量方法
Budak et al. Accuracy improvement of point data reduction with sampling-based methods by Fuzzy logic-based decision-making
Tran et al. A robust surrogate data assimilation approach to real-time forecasting using polynomial chaos expansion
CN110705841A (zh) 一种基于改进模糊层次分析法的化工生产安全评估方法
CN112597435B (zh) 基于设备监理的火电设备质量数据处理方法及装置
Syafei et al. Machine Learning Model Using Extreme Gradient Boosting (XGBoost) Feature Importance and Light Gradient Boosting Machine (LightGBM) to Improve Accurate Prediction of Bankruptcy
Guo et al. Data mining and application of ship impact spectrum acceleration based on PNN neural network
JP4299508B2 (ja) 製造プロセスにおける操業と品質の関連分析装置、関連分析方法及びコンピュータ読み取り可能な記憶媒体
KR20190136770A (ko) Ｇｍａ 용접에서의 비드 형상 예측 방법 및 장치
CN110795891B (zh) 一种耙吸式挖泥船操作参数的学习方法
CN109325241B (zh) 基于一致性计算的翻译机器人优选方法及其计算机系统
CN116292057A (zh) 一种可变速抽水蓄能机组的控制方法、装置、设备及介质
CN111522240B (zh) 四旋翼飞行器mtmlp-arx模型、辨识方法、系统及存储介质
CN107480647A (zh) 基于归纳式一致性异常检测的异常行为实时检测方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22942496

Country of ref document: EP

Kind code of ref document: A1