A control method, system and readable storage medium for laser wire feeding process parameters based on machine learning

Abstract

The present invention provides a control method and system for laser wire feeding process parameters based on machine learning, belonging to the field of laser wire feeding. In order to solve the problems that the existing laser wire feeding process requires manual correction of process parameters, has low efficiency, and the manual correction error is relatively large. In the present invention, the bead height and bead width of a single-track deposit obtained by testing different laser powers, scanning speeds, and wire feeding speeds are used, and then a fitting model is obtained by using a BP neural network for the data set, and the model is used for process machining. Using the process parameter data samples, the bead height and bead width of the deposit can be predicted well through the process parameters, and the required process parameters can be inversely predicted through the required bead width and bead height. Therefore, it is possible to reduce the shape deviation of the deposited layer caused by the process personnel's incorrect judgment by observing the molten pool in real time, and it is also possible to realize the mining and analysis of equipment data, and endow the equipment with the ability to process the process intelligently, reducing the professional requirements for the process personnel in the process.

Description

Technical Field

[0001] The present invention relates to the technical field of laser wire feeding, and in particular, to a control method, system and readable storage medium for laser wire feeding process parameters based on machine learning. Background Art

[0002] Laser direct metal deposition (LDMD) belongs to an additive manufacturing method, which can provide a metal cladding layer with good corrosion resistance, good surface finish and precise geometric shape on a substrate, and has been widely used in the fields of aerospace, energy power, metallurgical machinery, etc. However, the process stability of laser wire feeding cladding is one of the important factors restricting its industrial application. Therefore, the control of laser wire feeding process parameters has important significance.

[0003] In direct energy deposition, the laser and the wire metal interact, liquefy and deposit on the substrate material. Due to the complexity of the entire deposition process, it is very difficult to directly construct a very accurate physical model that can accurately show the relationship between process parameters and the bead width and bead height of the deposit.

[0004] The traditional control method is to preset process parameters according to experience, and then observe the state of the deposit and the molten pool to manually adjust the processing parameters. However, correcting process parameters through manual experience is often not the most reasonable, and a technical process engineer needs to observe the whole process during processing, resulting in low efficiency. Summary of the Invention

[0005] The technical problem to be solved by the present invention is:

[0006] To solve the problems that the existing laser wire feeding process requires manual correction of process parameters, has low efficiency, and relatively large manual correction errors.

[0007] The technical solution adopted by the present invention to solve the above technical problems:

[0008] The present invention provides a control method for laser wire feeding process parameters based on machine learning, including the following steps:

[0009] S100. On the same test plate, change the process parameters of the equipment respectively to obtain different single-track deposits, measure the bead height and bead width of the above single-track deposits, and obtain an initial data set. The process parameters include laser power, scanning speed and wire feeding speed;

[0010] S200. Divide the initial data set collected in step S100 into a training set and a test set according to a certain ratio, use the BP neural network to train the training set to obtain a training model, and then use the test set to detect the fitting degree of the training model to obtain the best fitting model;

[0011] In S300, process parameters are imported in batches into the best-fit model obtained in step S200 to obtain data on the bead height and bead width of the corresponding sediment. A database is acquired, and the process parameters of the database are called for processing according to the required shape of the deposited material.

[0012] Further, in step S200, the laser power, scanning speed, and wire feeding speed are set as the input parameters X of the BP neural network, and the bead width and bead height are set as the output Y of the neural network;

[0013] Given the training sample S = {(X1, Y1), (X2, Y2), …, (X n , Y n )}, X i ∈R 3 , Y i ∈R 2 , X is the data of the normalized process parameters, and R 3 is the dimension of the input data, that is, the laser power x1, the scanning speed x2, and the wire feeding speed x3;

[0014] Set the threshold of the j-th neuron in the output layer to be represented by σ j , and the threshold of the m-th neuron in the hidden layer to be represented by τ m . The connection weight between the i-th neuron in the input layer and the m-th neuron in the hidden layer is λ im , and the connection weight between the m-th neuron in the hidden layer and the j-th neuron in the output layer is μ mj ; The input received by the m-th neuron in the hidden layer is:

[0015]

[0016] The input received by the j-th neuron in the output layer is:

[0017]

[0018] where h m represents the value of the m-th neuron in the hidden layer, and q represents a total of q neurons in the hidden layer.

[0019] Further, the discriminant function of the hidden layer and the output layer uses the Sigmoid function For the training set, assuming the current neural network output is Then the mean square error of the BP neural network on the data set is:

[0020]

[0021] Further, the gradient descent algorithm is adopted to adjust the parameters in the negative gradient direction of the target E k , and for the error E k, given the learning rate ξ, after one round of learning:

[0022] Δμ hj = ξf j h m

[0023] Δσ j = -ξf j

[0024] Δλ ih = ξl h x i

[0025] Δτ h = -ξl h

[0026] Where:

[0027] The learning rate ξ controls the step size of each round of learning iteration.

[0028] Furthermore, the value range of ξ is (0, 1).

[0029] A control system for laser wire feeding process parameters based on machine learning, the system has program modules corresponding to the above steps, and executes the steps in the above control method for laser wire feeding process parameters based on machine learning when running.

[0030] A computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of the control method for laser wire feeding process parameters based on machine learning when called by a processor.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] The present invention uses process parameter data samples, and can preferably predict the bead height and bead width of the deposit through process parameters, and inversely predict the required process parameters through the required bead width and bead height. Therefore, it can reduce the shape deviation of the deposited layer caused by the misjudgment of the process personnel observing the molten pool in real time, and can also realize the mining and analysis of equipment data, and endow the equipment with the ability to intelligently process the process, reducing the professional requirements of the process personnel for the process. Description of the Drawings

[0033] Figure 1 It is a flowchart of a control method for laser wire feeding process parameters based on machine learning in an embodiment of the present invention;

[0034] Figure 2 It is a schematic diagram of the principle of the machine learning algorithm in an embodiment of the present invention.

[0035] Description of the Reference Numerals: Detailed implementation manners

[0036] To make the above objects, features and advantages of the present invention more obvious and understandable, the following detailed description will be given to specific embodiments of the present invention with reference to the accompanying drawings.

[0037] Specific implementation manner 1: In combination with Figure 1 and Figure 2 as shown, the present invention provides a control method for laser wire feeding process parameters based on machine learning, including the following steps:

[0038] S100. On the same test plate, change the process parameters of the equipment respectively to obtain different single-track deposits, measure the bead height and bead width of the above single-track deposits, and obtain an initial data set. The process parameters include laser power, scanning speed and wire feeding speed;

[0039] S200. Divide the initial data set collected in step S100 into a training set and a test set according to a certain ratio. Use the BP neural network to train the training set to obtain a training model, and then use the test set to detect the fitting degree of the training model to obtain the best fitting model;

[0040] S300. Batch import the process parameters into the best fitting model obtained in step S200 to obtain the bead height and bead width data of the corresponding deposits, obtain a database, and call the process parameters of the database for processing according to the required deposit shape.

[0041] When using the BP neural network to train the training set, since the process parameters include: laser power, scanning speed, and wire feeding speed parameters are all independent and irrelevant, normalizing these parameters can be used as the input parameter X of the BP neural network. At the same time, the sediment indicators: bead width, bead height can be approximately used as independent and irrelevant variables as the output Y of the neural network. Since the inline relationship between the bead width, bead height and the three process parameters can be trained by the neural network to optimize the parameters, a learning model with higher accuracy can be obtained.

[0042] Given the training sample S = {(X1, Y1), (X2, Y2),..., (X n , Y n )}, X i ∈R 3 , Y i ∈R 2 , X is the data of the normalized process parameters, R 3 is the dimension of the input data, that is, laser power x1, scanning speed x2, wire feeding speed x3. Y represents the sediment index, and R 2 is the dimension of the output data, that is, bead width y1, bead height y2. Figure 2Describes a BP neural network with 3 input neurons, 2 output neurons, and a single hidden layer. If the threshold of the j-th neuron in the output layer is denoted by σ j and the threshold of the m-th neuron in the hidden layer is denoted by τ m and the connection weight between the i-th neuron in the input layer and the m-th neuron in the hidden layer is λ im and the connection weight between the m-th neuron in the hidden layer and the j-th neuron in the output layer is μ mj . Then the input received by the m-th neuron in the hidden layer is:

[0043]

[0044] The input received by the j-th neuron in the output layer is:

[0045]

[0046] where h m represents the value of the m-th neuron in the hidden layer, and q represents the total number of q neurons in the hidden layer;

[0047] If the discriminant function between the hidden layer and the output layer uses the Sigmoid function For the training set, assuming the current neural network output is Then the mean square error of the BP neural network on the data set is:

[0048]

[0049] In this model, a total of (3 + 2 + 1)q + 2 parameters need to be determined. Using the gradient descent algorithm, the parameters are adjusted in the negative gradient direction of the objective E k . For the error E k , given the learning rate ξ, after one round of learning:

[0050] Δμ hj = ξf j h m

[0051] Δσ j = -ξf j

[0052] Δλ ih = ξl h x i

[0053] Δτ h = -ξl h

[0054] where: The learning rate ξ controls the step size of each round of learning iteration and takes values in the range (0, 1).

[0055] The data in Table 1 was used to train and test a BP neural network. The final model had an R2 score of 0.842 on the test set and an R2 score of 0.921 on the training set.

[0056] Table 1

[0057]

[0058]

[0059] Specific implementation plan two: A control system for laser wire feeding process parameters based on machine learning according to the present invention. This system has program modules corresponding to the above steps and executes the steps in the control method for laser wire feeding process parameters based on machine learning when running.

[0060] Other combinations and connection relationships in this implementation plan are the same as those in the first specific implementation plan.

[0061] Specific implementation plan three: A computer-readable storage medium according to the present invention. The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of the control method for laser wire feeding process parameters based on machine learning when called by a processor.

[0062] Other combinations and connection relationships in this implementation plan are the same as those in the first specific implementation plan.

[0063] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art of the present invention can make various changes and modifications without departing from the spirit and scope of the present disclosure, and these changes and modifications will all fall within the protection scope of the present invention.

Claims

1. A control method for laser wire feeding process parameters based on machine learning, characterized in that, Including the following steps: S100. On the same test plate, change the process parameters of the equipment respectively to obtain different single-track deposits, measure the bead height and bead width of the above single-track deposits, and obtain an initial data set. The process parameters include laser power, scanning speed, and wire feeding speed; S200. Divide the initial data set collected in step S100 into a training set and a test set according to a certain ratio. Use the BP neural network to train the training set to obtain a training model, and then use the test set to detect the fitting degree of the training model to obtain the best fitting model; Set the laser power, scanning speed, and wire feeding speed as the input parameter X of the BP neural network, and set the bead width and bead height as the output Y of the neural network; Given a training sample \(S =\{(X_1, Y_1),(X_2, Y_2),\cdots,(X n , Y n )\}\), where \(X i \in\mathbb{R} 3 \) and \(Y i \in\mathbb{R} 2 \). Here, \(X\) is the data of normalized process parameters, and \(\mathbb{R} 3 \) is the dimension of the input data, namely the laser power \(x_1\), the scanning speed \(x_2\), and the wire feeding speed \(x_3\). Let the threshold of the j-th neuron in the output layer be denoted by σ j Let the threshold of the m-th neuron in the hidden layer be denoted by τ m Let the connection weight between the i-th neuron in the input layer and the m-th neuron in the hidden layer be λ im Let the connection weight between the m-th neuron in the hidden layer and the j-th neuron in the output layer be μ mj The input received by the m-th neuron in the hidden layer is: The input received by the j-th neuron in the output layer is: Among them, h m represents the value of the m-th neuron in the hidden layer, and q represents the total number of neurons in the hidden layer; S300. Batch import the process parameters into the best fitting model obtained in step S200 to obtain the bead height and bead width data of the corresponding deposit, obtain a database, and call the process parameters of the database for processing according to the required deposit shape.

2. The control method of laser wire feeding process parameters based on machine learning according to claim 1, characterized in that: The discriminant function of the hidden layer and the output layer uses the Sigmoid function For the training set, assume that the current neural network output is Then the mean square error of the BP neural network on the data set is:

3. The control method of laser wire feeding process parameters based on machine learning according to claim 2, characterized in that: Using the gradient descent algorithm, the parameters are adjusted in the negative gradient direction of the target E k For the error E k , given the learning rate ξ, after one round of learning: Δμ hj = ξf j h m Δσ j =-ξf j Δλ ih = ξl h x i Δτ h =-ξl h Wherein: The learning rate ξ controls the step size of each round of learning iteration.

4. The control method of the laser wire feeding process parameters based on machine learning according to claim 3, characterized in that: The value range of ξ is (0, 1).

5. A control system for laser wire feeding process parameters based on machine learning, characterized in that: The system has a program module corresponding to the steps of any one of the above claims 1-4, and executes the steps in the above control method of laser wire feeding process parameters based on machine learning when running.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of the control method of laser wire feeding process parameters based on machine learning according to any one of claims 1-4 when called by a processor.

Images