CN109472358B - Neural network-based welding process parameter recommendation method and device and robot - Google Patents
Neural network-based welding process parameter recommendation method and device and robot Download PDFInfo
- Publication number
- CN109472358B CN109472358B CN201811207555.8A CN201811207555A CN109472358B CN 109472358 B CN109472358 B CN 109472358B CN 201811207555 A CN201811207555 A CN 201811207555A CN 109472358 B CN109472358 B CN 109472358B
- Authority
- CN
- China
- Prior art keywords
- welding
- neural network
- process parameter
- information
- experience
- 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
Links
Images
Classifications
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
- B23K31/00—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups
- B23K31/12—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups relating to investigating the properties, e.g. the weldability, of materials
- B23K31/125—Weld quality monitoring
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
- B23K31/00—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups
- B23K31/006—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups relating to using of neural networks
Abstract
The invention provides a neural network-based welding process parameter recommendation method, a neural network-based welding process parameter recommendation device and a robot, wherein the method comprises the following steps: acquiring welding experience scene information and welding experience process parameter information, and performing normalization processing to respectively serve as sample input data and sample output data; establishing a neural network, and determining the number of input layer neurons, the number of hidden layers and the number of output layer neurons of the neural network; determining a weight value and a deviation value of a neural network by adopting a back propagation algorithm according to the sample input data and the sample output data, and training the neural network; acquiring welding target scene parameters, inputting the parameters into the trained neural network, and outputting welding target process parameters; the mapping relation between the welding scene parameters and the welding process parameters is learned through the neural network, so that scenes without historical records in a database can be effectively dealt with, and the welding control requirements of changeable welding scenes are met.
Description
Technical Field
The invention relates to the field of artificial intelligence, in particular to a neural network-based welding process parameter recommendation method and device and a robot.
Background
Welding, also known as fusion or welding, is a process and technique for joining metals or other thermoplastics by heating or pressing. Depending on the specific welding process, the welding can be subdivided into other special welding such as gas welding, resistance welding, arc welding, induction welding, laser welding, and the like. Welding is a basic industrial processing mode, and the number of welding workers in the Chinese market is over 500 ten thousand according to conservative estimation at present. With the aggravation of the aging phenomenon of the population and the gradual increase of the labor cost, the problem of difficult labor in the welding field of industrial production, particularly the manufacturing industry, is increasingly serious, and each enterprise is seriously lack of a welder master, particularly an experienced teacher master. The training period of the welder master is long (several years), the difficulty is high, and the gap of the number of the welder workers in China is as high as millions. In order to improve the production efficiency, a factory needs to gradually use a welding robot to replace a worker for welding so as to improve the efficiency and reduce the cost. At present, the field of welding products by using a welding robot, such as welding of mobile phone precision parts, welding of automobile bodies and the like, is increased year by year.
Welding robots are the most common type of industrial robot and are commonly used in the mass production of automotive manufacturing machinery lines for the welding of automotive bodies and other components using welding processes. The welding robot mainly comprises a robot and welding equipment. The robot consists of a robot body and a control cabinet (hardware and software). The welding equipment, taking arc welding and spot welding as examples, is composed of a welding power supply (including a control system thereof), a wire feeder, a welding gun and the like, and further comprises supporting facilities such as shielding gas and the like. The intelligent robot also comprises a sensing system, such as a laser or camera sensor and a control device thereof. The welding robot needs a robot operator to debug before stable operation work, wherein the debugging work comprises the setting work related to a whole set of robot control, process and welding parameters. The robot operator training period is relatively short, but welding related parameters (current, voltage, speed, shielding gas, etc.) in the welding robot cannot be set without knowledge of the welding's basic knowledge.
The setting of initial parameters under a specified welding scene needs to depend on the accumulation of years of welding experience, so that a set of quantitative process kit is urgently needed, scene information (welding base material, thickness and welding type) is input, and welding parameters (current, voltage, speed, shielding gas and the like) are output to be set by an operator so that the welding robot starts to operate effectively and meet the welding quality standard.
The process kit generally depends on a database for recording welding setting parameters which are successful in the same type of welding scenes in history, foreign companies mostly adopt the mode through decades of technology accumulation, but the application of the mode in the welding scenes in China faces the following problems: if the welding scenes are variable, and if the database only records the parameter combinations in scenes with thicknesses of 3mm and 5mm, the welding scenes cannot be effectively recommended if new scene process packages such as 3.2mm and 4.5mm are encountered in real application, and it is unrealistic to exhaust all possible scenes.
Disclosure of Invention
The invention aims to provide a neural network-based welding process parameter recommendation method, a neural network-based welding process parameter recommendation device and a robot, aiming at the problem that a welding process bag in the prior art cannot meet changeable welding scenes, and meeting the requirements of the changeable welding scenes.
A neural network-based welding process parameter recommendation method comprises the following steps:
acquiring welding experience scene information and welding experience process parameter information, and performing normalization processing to respectively serve as sample input data and sample output data;
establishing a neural network, and determining the number of input layer neurons, the number of hidden layers and the number of output layer neurons of the neural network;
determining a weight value and a deviation value of a neural network by adopting a back propagation algorithm according to the sample input data and the sample output data, and training the neural network;
and acquiring welding target scene parameters, inputting the parameters into the trained neural network, and outputting welding target process parameters.
Further, the welding experience scene information at least comprises a first plate thickness, a second plate thickness and a welding wire diameter;
the welding empirical process parameter information at least comprises a welding current value;
the welding experience scene information and the welding experience process parameter information are in a mapping relation.
Further, the normalization processing is performed on the welding experience scene information and the welding experience process parameter information, and the normalization processing comprises the following steps:
calculating the quantity of each type of welding experience scene information or welding experience process parameter information;
if the number is greater than or equal to a preset number, performing normalization processing by the following formula:
wherein y is a normalized value, and x is welding experience scene information or welding experience worker
The process parameter information is
The average value is obtained, s is a standard deviation, and N is welding experience scene information or the number of welding experience process parameter information;
if the number is less than the preset number, performing normalization processing by the following formula:
y＝x/xmax；
wherein x ismaxIs the maximum value at which x may reasonably occur.
Further, the preset number is greater than 8.
Further, the number of input layer neurons of the neural network is equal to the number of information types in the welding experience scene information;
the number of hidden layers of the neural network is at least one;
the number of output layer neurons is at least one.
Further, the back propagation algorithm employs ADAM, SGD, adagard, or RMSprop.
Further, the activation function of the neural network is Tanh, linear, sigmoid, or relu.
Further, the training period of the neural network is greater than 1000, and the learning rate is 0.01.
A neural network-based welding process parameter recommendation device comprises:
the data processing module is used for acquiring welding experience scene information and welding experience process parameter information, performing normalization processing and respectively serving as sample input data and sample output data;
the neural network establishing module is used for determining the number of input layer neurons, the number of hidden layers and the number of output layer neurons of the neural network;
the training module is used for determining a weight value and a deviation value of the neural network by adopting a back propagation algorithm according to the sample input data and the sample output data, and training the neural network;
and the prediction module is used for acquiring the scene parameters of the welding target, inputting the scene parameters into the trained neural network and outputting the technological parameters of the welding target.
Further, the welding experience scene information acquired by the data processing module at least comprises a first plate thickness, a second plate thickness and a welding wire diameter;
the welding empirical process parameter information at least comprises a welding current value;
the welding experience scene information and the welding experience process parameter information are in a mapping relation.
Further, the data processing module is further configured to:
calculating the quantity of each type of welding experience scene information or welding experience process parameter information;
if the number is greater than or equal to a preset number, performing normalization processing by the following formula:
wherein y is a normalized value, x is welding experience scene information or welding experience process parameter information of the same type, and y is
The average value is obtained, s is a standard deviation, and N is welding experience scene information or the number of welding experience process parameter information;
if the number is less than the preset number, performing normalization processing by the following formula:
y＝x/xmax；
wherein x ismaxIs the maximum value at which x may reasonably occur.
Further, the preset number is greater than 8.
Further, the number of input layer neurons of the neural network is equal to the number of information types in the welding experience scene information;
the number of hidden layers of the neural network is at least one;
the number of output layer neurons is at least one.
Further, the weight and deviation determination module adopts a back propagation algorithm of ADAM, SGD, Adagrad or RMSprop.
Further, the activation function of the neural network is Tanh, linear, sigmoid, or relu.
Further, the training period of the neural network is greater than 1000, and the learning rate is 0.01.
The welding robot comprises the neural network-based welding process parameter recommendation device and a welding power supply, wherein the welding target process parameters act on the welding power supply to control the welding robot to perform welding operation.
According to the neural network-based welding process parameter recommendation method, device and robot, the mapping relation between the welding scene parameters and the welding process parameters is learned through the neural network, so that scenes without historical records in a database can be effectively responded, and the welding control requirements of variable welding scenes are met.
Drawings
Fig. 1 is a flowchart of an embodiment of a neural network-based welding process parameter recommendation method provided by the present invention.
Fig. 2 is a schematic structural diagram of an embodiment of a single neuron in the neural network-based welding process parameter recommendation method provided by the invention.
Fig. 3 is a schematic structural diagram of a neural network in an application scenario in the neural network-based welding process parameter recommendation method provided by the invention.
Fig. 4 is a schematic diagram of Linear function definition in the neural network-based welding process parameter recommendation method provided by the invention.
FIG. 5 is a schematic diagram of sigmoid function definition in the neural network-based welding process parameter recommendation method provided by the invention.
FIG. 6 is a schematic diagram of the function definition of relu in the neural network-based welding process parameter recommendation method provided by the present invention.
Fig. 7 is a schematic structural diagram of an embodiment of a neural network-based welding process parameter recommendation device provided by the invention.
Fig. 8 is a schematic structural diagram of an embodiment of a neural network-based welding robot provided by the invention.
Detailed Description
In order to make the objects, technical solutions and effects of the present invention clearer and clearer, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Example one
Referring to fig. 1, the present embodiment provides a neural network-based welding process parameter recommendation method, including:
step S101, obtaining welding experience scene information and welding experience process parameter information, and performing normalization processing to respectively serve as sample input data and sample output data;
step S102, establishing a neural network, and determining the number of input layer neurons, the number of hidden layers and the number of output layer neurons of the neural network;
step S103, determining a weight value and a deviation value of a neural network by adopting a back propagation algorithm according to the sample input data and the sample output data, and training the neural network;
and step S104, acquiring welding target scene parameters, inputting the welding target scene parameters into the trained neural network, and outputting welding target process parameters.
Specifically, step S101 is executed to obtain welding experience scene information and welding experience process parameter information, where the welding experience scene information at least includes a first plate thickness, a second plate thickness, and a welding wire diameter, and the welding experience process parameter information at least includes a welding current value.
In addition, the welding experience scene information may also include information such as whether to open a groove (open 1, not open 0), and the welding experience process parameter information may also include a welding voltage value, a welding speed, and the like.
The welding experience scene information and the welding experience process parameter information are in a mapping relation, for example, in a group of data, the first plate thickness, the second plate thickness and the welding wire diameter correspond to a welding current value, in a neural network, the first plate thickness, the second plate thickness and the welding wire diameter are input data, the welding current value is output data, and the welding experience scene information comprises a plurality of data of the first plate thickness, the second plate thickness and the welding wire diameter.
The welding experience scene information and the welding experience process parameter information come from experience data which are qualified in welding.
Further, it is necessary to convert the welding experience scene information and the welding experience process parameter information into a form that can be understood by machine learning, so that normalization processing is required, which specifically includes:
calculating the quantity of each type of welding experience scene information or welding experience process parameter information;
if the number is greater than or equal to a preset number, performing normalization processing by the following formula:
wherein y is a normalized value, x is welding experience scene information or welding experience process parameter information of the same type, and y is
The average value is obtained, s is a standard deviation, and N is welding experience scene information or the number of welding experience process parameter information;
if the number is less than the preset number, performing normalization processing by the following formula:
y＝x/xmax；
wherein x ismaxIs the maximum value at which x may reasonably occur.
In a preferred embodiment, the predetermined number is greater than 8.
And the normalized welding experience scene information is used as sample input data, and the normalized welding experience process parameter information is used as sample output data.
Further, step S102 is executed to establish a neural network, determine the number of neurons in the input layer of the neural network: the number of input layer neurons of the neural network is equal to the number of information types in the welding experience scene information.
Taking the least types of welding experience scenario information as an example, the welding experience scenario information includes a first plate thickness, a second plate thickness and a welding wire diameter, and therefore the input layer includes three neurons for inputting the first plate thickness, the second plate thickness and the welding wire diameter, respectively.
In a preferred embodiment, the number of hidden layers of the neural network is at least one.
The number of output layer neurons is at least one.
Taking the least type of welding empirical process parameter information as an example, the welding empirical process parameter information includes a welding current value, and the number of neurons in the output layer is 1.
The welding experience scene information is used as the input of the neural network, and each output can build an independent model by using the same input.
If there are multiple types of outputs, each output can be built into an independent model, or multiple neurons can be set in the output layer of a model.
The method provided by the embodiment aims at data with very diversified input types, and a general machine learning method cannot find corresponding characteristic rules from information input with a plurality of different dimensions. Compared with a statistical model, the neural network model is more suitable for a scene with no obvious rules or differences in input characteristics, so that the deep learning neural network is required to be applied to establish the relation between the input scene and the output process parameters.
Referring to fig. 2, the basic units of a neural network are neurons, each of which receives more than 1 input, denoted as a 1-an.
Each received input is multiplied by a weight w1-wn, and summed, and the summed value is added with an offset b to obtain X, and an activation function f is applied to X to obtain the final output t.
Each neuron, except the output layer, serves as an input to the next layer of neurons.
In the embodiment, the simplest application scenario is described, referring to fig. 3, the sample input data includes a first plate thickness, a second plate thickness, and a wire diameter, so that the input layer includes three neurons, the three neurons correspond to the first plate thickness, the second plate thickness, and the wire diameter, and each neuron receives more than one data. The middle layer is a layer and comprises three neurons. The output layer includes a neuron corresponding to the welding current value.
The first layer is an input layer and does not relate to weight values and deviation values, 9 weights are needed in the middle layer of three neurons corresponding to three inputs, the weights are generally expressed in a matrix form, the output layer is only provided with one neuron, and three weights and one deviation are needed corresponding to the three inputs.
Further, step S103 is executed, and the weight values and the bias values are obtained by model automatic learning according to the sample input data and the sample output data through a back propagation algorithm.
Preferably, the back propagation algorithm employs ADAM, SGD, adagard or RMSprop.
The Adam algorithm, namely an Adaptive Moment Estimation method (Adaptive motion Estimation), can calculate the Adaptive learning rate of each parameter. This method not only stores the exponentially decaying averages of the AdaDelta previous squared gradients, but also maintains the exponentially decaying averages of the previous gradients M (t), M (t) being the average at the first instance of the gradient, and V (t) being the non-central variance value at the second instance of the gradient.
And training the neural network according to the sample input data, the weight value and the deviation value, wherein the training period is more than 1000, and the learning rate is 0.01.
Preferably, the activation function of the neural network is Tanh, linear, sigmoid, or relu.
The function definition of Linear is shown in fig. 4, the function definition of sigmoid is shown in fig. 5, and the function definition of relu is shown in fig. 6.
Further, step S104 is executed, and the trained neural network may predict and recommend a required welding target process parameter according to the target scene parameter, for example, input a first plate thickness, a second plate thickness and a welding wire diameter to be welded, and output a welding current value applied to the welding power source.
According to the neural network-based welding process parameter recommendation method provided by the embodiment, the mapping relation between the welding scene parameters and the welding process parameters is learned through the neural network, so that scenes without historical records in a database can be effectively responded, and the welding control requirements of changeable welding scenes are met.
Example two
Referring to fig. 7, the present embodiment provides a neural network-based welding process parameter recommendation device, including:
the data processing module 201 is configured to obtain welding experience scene information and welding experience process parameter information, perform normalization processing, and respectively serve as sample input data and sample output data;
a neural network establishing module 202, configured to determine the number of input layer neurons, the number of hidden layers, and the number of output layer neurons of the neural network;
the training module 203 is configured to determine a weight value and a bias value of a neural network by using a back propagation algorithm according to the sample input data and the sample output data, and train the neural network;
and the prediction module 204 is used for acquiring the welding target scene parameters, inputting the welding target scene parameters into the trained neural network, and outputting the welding target process parameters.
The neural network-based welding process parameter recommendation device provided in this embodiment is an execution subject of the neural network-based welding process parameter recommendation method described in the first embodiment, and the device may be a processor, a central controller, or the like.
Further, the welding experience scene information acquired by the data processing module 201 at least includes a first plate thickness, a second plate thickness and a welding wire diameter;
the welding empirical process parameter information at least comprises a welding current value;
the welding experience scene information and the welding experience process parameter information are in a mapping relation.
In addition, the welding experience scene information may also include information such as whether to open a groove (open 1, not open 0), and the welding experience process parameter information may also include a welding voltage value, a welding speed, and the like.
The welding experience scene information and the welding experience process parameter information are in a mapping relation, for example, in a group of data, the first plate thickness, the second plate thickness and the welding wire diameter correspond to a welding current value, in a neural network, the first plate thickness, the second plate thickness and the welding wire diameter are input data, the welding current value is output data, and the welding experience scene information comprises a plurality of data of the first plate thickness, the second plate thickness and the welding wire diameter.
The welding experience scene information and the welding experience process parameter information come from experience data which are qualified in welding.
Further, the welding experience scene information and the welding experience process parameter information need to be converted into a form which can be understood by machine learning, so that normalization processing needs to be performed, and the data processing module is further used for:
calculating the quantity of each type of welding experience scene information or welding experience process parameter information;
if the number is greater than or equal to a preset number, performing normalization processing by the following formula:
wherein y is a normalized value, x is welding experience scene information or welding experience process parameter information of the same type, and y is
The average value is obtained, s is a standard deviation, and N is welding experience scene information or the number of welding experience process parameter information;
if the number is less than the preset number, performing normalization processing by the following formula:
y＝x/xmax；
wherein x ismaxIs the maximum value at which x may reasonably occur.
Further, the preset number is greater than 8.
And the normalized welding experience scene information is used as sample input data, and the normalized welding experience process parameter information is used as sample output data.
Further, the number of input layer neurons of the neural network is equal to the number of information types in the welding experience scene information;
the number of hidden layers of the neural network is at least one;
the number of output layer neurons being at least one
Taking the least type of welding empirical process parameter information as an example, the welding empirical process parameter information includes a welding current value, and the number of neurons in the output layer is 1.
The welding experience scene information is used as the input of the neural network, and each output can build an independent model by using the same input.
If there are multiple types of outputs, each output can be built into an independent model, or multiple neurons can be set in the output layer of a model.
Further, the back propagation algorithm employed by the training module 203 is to employ ADAM, SGD, adagard, or RMSprop.
Further, the activation function of the neural network is Tanh, linear, sigmoid, or relu.
The function definition of Linear is shown in fig. 4, the function definition of sigmoid is shown in fig. 5, and the function definition of relu is shown in fig. 6.
Further, the training period of the neural network is greater than 1000, and the learning rate is 0.01.
According to the neural network-based welding process parameter recommendation device provided by the embodiment, the mapping relation between the welding scene parameters and the welding process parameters is learned through the neural network, so that scenes without historical records in a database can be effectively responded, and the welding control requirements of changeable welding scenes are met.
EXAMPLE III
Referring to fig. 8, the present embodiment provides a welding robot, which includes the neural network-based welding process parameter recommendation device 301, and further includes a welding power source 302, where the welding target process parameter acts on the welding power source to control the welding robot to perform a welding operation.
For the structure and the working principle of the neural network-based welding process parameter recommendation device 301, please refer to the first embodiment and the second embodiment, which will not be described herein again.
The welding robot provided by the embodiment learns the mapping relation between the welding scene parameters and the welding process parameters through the neural network, can effectively deal with scenes without historical records in a database, and meets the welding control requirement of changeable welding scenes.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
Claims (13)
1. A neural network-based welding process parameter recommendation method is characterized by comprising the following steps:
acquiring welding experience scene information and welding experience process parameter information, and performing normalization processing to respectively serve as sample input data and sample output data;
establishing a neural network, and determining the number of input layer neurons, the number of hidden layers and the number of output layer neurons of the neural network;
determining a weight value and a deviation value of a neural network by adopting a back propagation algorithm according to the sample input data and the sample output data, and training the neural network;
acquiring welding target scene parameters, inputting the parameters into the trained neural network, and outputting welding target process parameters;
the welding experience scene information at least comprises a first plate thickness, a second plate thickness and a welding wire diameter;
the welding empirical process parameter information at least comprises a welding current value;
the welding experience scene information and the welding experience process parameter information are in a mapping relation;
the method for normalizing the welding experience scene information and the welding experience process parameter information comprises the following steps:
calculating the quantity of each type of welding experience scene information or welding experience process parameter information;
if the number is greater than or equal to a preset number, performing normalization processing by the following formula:
wherein y is normalizedThe transformed value x is the information of a kind of welding experience scene or welding experience process parameter,
the average value is obtained, s is a standard deviation, and N is welding experience scene information or the number of welding experience process parameter information;
if the number is less than the preset number, performing normalization processing by the following formula:
y＝x/xmax；
wherein x ismaxIs the maximum value at which x may reasonably occur.
2. The neural network-based welding process parameter recommendation method of claim 1, wherein the preset number is greater than 8.
3. The neural network-based welding process parameter recommendation method of claim 1, wherein the number of input layer neurons of the neural network is equal to the number of information types in the welding experience scenario information;
the number of hidden layers of the neural network is at least one;
the number of output layer neurons is at least one.
4. The neural network-based welding process parameter recommendation method of claim 1, wherein the back propagation algorithm employs ADAM, SGD, adagard, or RMSprop.
5. The neural network-based welding process parameter recommendation method of claim 1, wherein the activation function of the neural network is Tanh, linear, sigmoid, or relu.
6. The neural network-based welding process parameter recommendation method of claim 1, wherein a training period of the neural network is greater than 1000 and a learning rate is 0.01.
7. A neural network-based welding process parameter recommendation device is characterized by comprising:
the data processing module is used for acquiring welding experience scene information and welding experience process parameter information, performing normalization processing and respectively serving as sample input data and sample output data;
the neural network establishing module is used for determining the number of input layer neurons, the number of hidden layers and the number of output layer neurons of the neural network;
the training module is used for determining a weight value and a deviation value of the neural network by adopting a back propagation algorithm according to the sample input data and the sample output data, and training the neural network;
the prediction module is used for acquiring welding target scene parameters, inputting the welding target scene parameters into the trained neural network and outputting welding target process parameters;
the welding experience scene information acquired by the data processing module at least comprises a first plate thickness, a second plate thickness and a welding wire diameter;
the welding empirical process parameter information at least comprises a welding current value;
the welding experience scene information and the welding experience process parameter information are in a mapping relation;
the data processing module is further configured to:
calculating the quantity of each type of welding experience scene information or welding experience process parameter information;
if the number is greater than or equal to a preset number, performing normalization processing by the following formula:
wherein y is a normalized value, x is welding characteristic scene information or welding characteristic process parameter information,
the average value is obtained, s is a standard deviation, and N is welding experience scene information or the number of welding experience process parameter information;
if the number is less than the preset number, performing normalization processing by the following formula:
y＝x/xmax；
wherein x ismaxIs the maximum value at which x may reasonably occur.
8. The neural network-based welding process parameter recommendation device of claim 7, wherein the preset number is greater than 8.
9. The neural network-based welding process parameter recommendation device of claim 7, wherein the number of input layer neurons of the neural network is equal to the number of information types in the welding experience scenario information;
the number of hidden layers of the neural network is at least one;
the number of output layer neurons is at least one.
10. The neural network-based welding process parameter recommendation device of claim 7, wherein the weight and deviation determination module employs a back propagation algorithm that employs ADAM, SGD, Adagrad, or RMSprop.
11. The neural network-based welding process parameter recommendation device of claim 7, wherein the activation function of the neural network is Tanh, linear, sigmoid, or relu.
12. The neural network-based welding process parameter recommendation device of claim 7, wherein a training period of the neural network is greater than 1000 and a learning rate is 0.01.
13. A welding robot comprising the neural network-based welding process parameter recommendation device of any one of claims 7-12, and a welding power supply, wherein the welding target process parameter acts on the welding power supply to control the welding robot to perform a welding operation.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811207555.8A CN109472358B (en) | 2018-10-17 | 2018-10-17 | Neural network-based welding process parameter recommendation method and device and robot |
| PCT/CN2018/113613 WO2020077694A1 (en) | 2018-10-17 | 2018-11-02 | Neural network-based welding process parameter recommendation method and device, and robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811207555.8A CN109472358B (en) | 2018-10-17 | 2018-10-17 | Neural network-based welding process parameter recommendation method and device and robot |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109472358A CN109472358A (en) | 2019-03-15 |
| CN109472358B true CN109472358B (en) | 2021-10-19 |
Family
ID=65663993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811207555.8A Active CN109472358B (en) | 2018-10-17 | 2018-10-17 | Neural network-based welding process parameter recommendation method and device and robot |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109472358B (en) |
| WO (1) | WO2020077694A1 (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109940622B (en) * | 2019-04-28 | 2020-10-27 | 西安交通大学 | Non-sensing collision detection method for robot mechanical arm based on motor current |
| JP7332144B2 (en) * | 2019-06-13 | 2023-08-23 | 株式会社ブイ・テクノロジー | LASER MODIFICATION METHOD, LASER MODIFICATION DEVICE |
| CN111069736A (en) * | 2019-12-27 | 2020-04-28 | 唐山松下产业机器有限公司 | Storage medium, welding equipment, welding abnormity detection method and device |
| CN111914473A (en) * | 2020-06-24 | 2020-11-10 | 浙江吉利汽车研究院有限公司 | Welding parameter determination method and device for resistance spot welding, electronic equipment and storage medium |
| CN112231855A (en) * | 2020-09-14 | 2021-01-15 | 中国船舶重工集团公司第七一六研究所 | Pipe welding process automatic generation method and system based on machine learning |
| CN113779330B (en) * | 2021-09-13 | 2024-01-26 | 河海大学常州校区 | Ultrasonic detection process parameter intelligent computing system based on similarity measurement |
| CN114415616B (en) * | 2022-01-20 | 2023-09-29 | 成都熊谷加世电器有限公司 | Pipeline welding process planning method |
| CN114202262B (en) * | 2022-02-21 | 2022-07-01 | 德州联合拓普复合材料科技有限公司 | Prepreg process improvement method and system based on neural network and storage medium |
| CN114888500A (en) * | 2022-03-09 | 2022-08-12 | 柳州铁道职业技术学院 | Automatic welding robot for railway shaping combination cabinet wiring |
| CN115858940B (en) * | 2023-02-14 | 2023-05-16 | 滨州市博恒工程管理服务有限公司 | Steel structure welding process quality management recommendation method based on big data processing |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101882176A (en) * | 2010-06-08 | 2010-11-10 | 南京航空航天大学 | Interface system method of mechanical property prediction of welding joint based on computer technology |
| CN101908083A (en) * | 2010-05-13 | 2010-12-08 | 上海市特种设备监督检验技术研究院 | Artificial neural network-based Q345 welding joint mechanical property prediction method |
| CN108182427A (en) * | 2018-01-30 | 2018-06-19 | 电子科技大学 | A kind of face identification method based on deep learning model and transfer learning |
Family Cites Families (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090200281A1 (en) * | 2008-02-08 | 2009-08-13 | Gm Global Technology Operations, Inc. | Welding power supply with neural network controls |
| US9445739B1 (en) * | 2010-02-03 | 2016-09-20 | Hrl Laboratories, Llc | Systems, methods, and apparatus for neuro-robotic goal selection |
| US20120091185A1 (en) * | 2010-10-18 | 2012-04-19 | Georgia Tech Research Corporation | In-process weld geometry methods & systems |
| CN102999676A (en) * | 2012-12-24 | 2013-03-27 | 湖南大学 | Process optimization method of steel/aluminum laser welding brazing |
| CN103551709B (en) * | 2013-11-16 | 2015-04-22 | 沈阳工业大学 | Welding joint forming detection device and method for A-TIG welding |
| CN105224984B (en) * | 2014-05-31 | 2018-03-13 | 华为技术有限公司 | A kind of data category recognition methods and device based on deep neural network |
| CN105279291B (en) * | 2014-06-17 | 2018-08-17 | 武汉科技大学 | A kind of bead-welding technology parameter optimization method towards batch machining |
| CN105618897A (en) * | 2014-10-27 | 2016-06-01 | 西安扩力机电科技有限公司 | Remote control method for underwater welding |
| CN105787940A (en) * | 2016-02-29 | 2016-07-20 | 长安大学 | High-frequency resistance straight seam welding quality state online detection method |
| CN105891215B (en) * | 2016-03-31 | 2019-01-29 | 浙江工业大学 | Welding visible detection method and device based on convolutional neural networks |
| CN105938563A (en) * | 2016-04-14 | 2016-09-14 | 北京工业大学 | Weld surface defect identification method based on image texture |
| CN108037190A (en) * | 2017-11-03 | 2018-05-15 | 大连理工大学 | A kind of high temperature furnace pipe damage rank intelligence assessment method based on neutral net |
| CN108500498B (en) * | 2018-03-26 | 2020-05-19 | 华中科技大学 | Weld joint forming quality monitoring method |
| CN108544146A (en) * | 2018-04-28 | 2018-09-18 | 辽宁机电职业技术学院 | A kind of one side welding device and its control method |
| CN108615075B (en) * | 2018-05-09 | 2021-09-28 | 深圳市微埃智能科技有限公司 | Automatic parameter adjusting method |
-
2018
- 2018-10-17 CN CN201811207555.8A patent/CN109472358B/en active Active
- 2018-11-02 WO PCT/CN2018/113613 patent/WO2020077694A1/en active Application Filing
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101908083A (en) * | 2010-05-13 | 2010-12-08 | 上海市特种设备监督检验技术研究院 | Artificial neural network-based Q345 welding joint mechanical property prediction method |
| CN101882176A (en) * | 2010-06-08 | 2010-11-10 | 南京航空航天大学 | Interface system method of mechanical property prediction of welding joint based on computer technology |
| CN108182427A (en) * | 2018-01-30 | 2018-06-19 | 电子科技大学 | A kind of face identification method based on deep learning model and transfer learning |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020077694A1 (en) | 2020-04-23 |
| CN109472358A (en) | 2019-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109472358B (en) | Neural network-based welding process parameter recommendation method and device and robot | |
| CN109409520B (en) | Welding process parameter recommendation method and device based on transfer learning and robot | |
| US10254750B2 (en) | Machining machine system which determines acceptance/rejection of workpieces | |
| US10502253B2 (en) | Machine learning device for learning assembly operation and component assembly system | |
| JP6453805B2 (en) | Production system for setting judgment values for variables related to product abnormalities | |
| Sadrfaridpour et al. | Modeling and control of trust in human-robot collaborative manufacturing | |
| Kim et al. | An investigation into an intelligent system for predicting bead geometry in GMA welding process | |
| Atkeson | Using locally weighted regression for robot learning | |
| CN112650187B (en) | Workshop scheduling method, device and system | |
| Pilani et al. | A hybrid intelligent systems approach for die design in sheet metal forming | |
| CN111199072B (en) | All-aluminum vehicle body riveting system based on online process library self-learning and implementation method thereof | |
| Weigelt et al. | Conceptual design of an intelligent ultrasonic crimping process using machine learning algorithms | |
| US11185990B2 (en) | Method for learning and embodying human facial expression by robot | |
| CN109934385A (en) | Goods amount prediction technique and system based on length Memory Neural Networks | |
| CN110880044A (en) | Markov chain-based load prediction method | |
| Hamedi et al. | Optimizing spot welding parameters in a sheet metal assembly by neural networks and genetic algorithm | |
| Vision | Improved automatic quality inspections through the integration of state-of-the-art machine vision and collaborative robots | |
| Srinivasan et al. | Prediction of spring-back and bend force in air bending of electro-galvanised steel sheets using artificial neural networks | |
| JP2005078545A (en) | Method and device for adjusting process model | |
| CN111914473A (en) | Welding parameter determination method and device for resistance spot welding, electronic equipment and storage medium | |
| Grilo et al. | Computer Vision in Industrial Automation and Mobile Robots | |
| Bekar et al. | Using Adaptive Neuro-Fuzzy Inference System, Artificial Neural Network and Response Surface Method to Optimize Overall Equipment Effectiveness for An Automotive Supplier Company. | |
| KR102574092B1 (en) | Method, device and system for determining mold abnormalities using artificial intelligence-based temperature and vibration sensor data | |
| Hwang et al. | A neurorobotics approach to investigating the emergence of communication in robots | |
| Barot et al. | Machine learning algorithm comparison for IoT enabled Submerged Arc Welding Process monitoring |
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 |