A method for identifying trajectory deviation of an arc additive manufacturing robot based on welding temperature field distribution characteristics

Abstract

This invention provides a method for identifying robot trajectory deviation in arc additive manufacturing based on the distribution characteristics of welding temperature field. It uses an infrared thermal imager to measure the temperature field of the sidewall of the arc additive manufacturing workpiece in real time. A Region of Interest (ROI) is selected outside the molten pool area in the welding temperature field image. An improved LBP algorithm is used to extract the texture features of the ROI temperature field. A dimensionality reduction algorithm is used to convert the high-dimensional temperature field texture features into low-dimensional temperature field texture features. A trained robot trajectory deviation identification model processes the low-dimensional temperature field texture features to output the degree of robot trajectory deviation. This invention achieves online identification of robot trajectory deviation by sensing the temperature field of the sidewall of the arc additive manufacturing workpiece, largely solving the problem of visually monitoring the influence of arc light on the additive manufacturing process, and providing a new method to ensure the quality of arc additive manufacturing workpieces.

Description

Technical Field

[0001] This invention relates to the field of monitoring the arc additive manufacturing process, and in particular to a method for identifying trajectory deviation of an arc additive manufacturing robot based on the characteristics of welding temperature field distribution. Background Technology

[0002] Arc additive manufacturing technology is based on the concept of discrete, additive manufacturing. It uses 3D design software to create a solid model of the workpiece, and uses an electric arc as a heat source to melt metal wire, forming the desired solid workpiece by layer-by-layer deposition along a predetermined path. In the arc additive manufacturing process, the robot's motion trajectory directly determines the forming position of the cladding layer, thus affecting the quality of the formed workpiece. Therefore, researching online monitoring technology for the robot's motion trajectory in arc additive manufacturing, capable of identifying the degree of deviation from the robot's trajectory online, is particularly important.

[0003] During the arc additive manufacturing process, if the robot malfunctions and the welding torch's trajectory deviates, the visual morphological characteristics of the molten pool will differ compared to when the robot's trajectory remains unchanged. Therefore, a robot trajectory deviation recognition method based on molten pool vision can be used to monitor the motion trajectory of the arc additive manufacturing robot online. However, the intense arc light generated during arc additive manufacturing can affect the recognition accuracy of the aforementioned method. Currently, arc additive manufacturing robot trajectory deviation recognition methods based on molten pool vision suffer from drawbacks such as low recognition accuracy and poor stability. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for trajectory offset recognition of arc additive manufacturing robots based on welding temperature field distribution characteristics, the specific technical solution of which is as follows:

[0005] A method for trajectory offset recognition of arc additive manufacturing robots based on welding temperature field distribution characteristics includes the following steps:

[0006] Step 1: Acquire images of the temperature field on the sidewall of the workpiece in arc additive manufacturing: Measure the temperature field on the sidewall of the workpiece using an infrared thermal imager;

[0007] Step 2: Select ROI (Region of Interest): Select the ROI in the heat-affected zone outside the molten pool area in the welding temperature field image;

[0008] Step 3: Extract texture features of the ROI temperature field: Use the improved LBP (Local Binary Pattern) algorithm to extract texture features of the ROI temperature field;

[0009] Step 4: ROI Temperature Field Texture Feature Dimensionality Reduction: Use a dimensionality reduction algorithm to convert the high-dimensional temperature field texture feature vector into a low-dimensional temperature field texture feature vector;

[0010] Step 5: Establish a robot trajectory deviation recognition model: Using low-dimensional temperature field texture features as input, establish a robot trajectory deviation recognition model based on a neural network, and output the degree of deviation of the robot trajectory.

[0011] Furthermore, in step one, the infrared thermal imager is fixed on the welding work platform to collect images of the temperature field of the workpiece sidewall during the arc additive manufacturing process.

[0012] Furthermore, in step two, the size of the ROI is selected as 35 pixels × 90 pixels.

[0013] Furthermore, the improved LBP algorithm in step three is as follows: using a 3×3 square window, calculate the standard deviation S of all windows in the image, and then set the standard deviation S of each window... i Binarization is performed according to a certain threshold. The binarized value is then added to the highest bit of the original LBP binary sequence, effectively improving the traditional 8-bit binary sequence into a 9-bit binary sequence. Finally, the 9-bit binary sequence is converted to decimal, which is the improved LBP value of the center pixel of the window. The comparison threshold for the standard deviation of each window is set to the average of the standard deviations of all windows in the image. The improved LBP calculation formula is as follows:

[0014]

[0015] In the formula, N is the total number of windows in the image, and the improved LBP value ranges from [0, 511].

[0016] Furthermore, the dimensionality reduction algorithm used in step four is principal component analysis, which transforms the 512-dimensional temperature field texture feature vector into a 10-dimensional temperature field texture feature.

[0017] Furthermore, in step five, a backpropagation neural network is used to establish a robot trajectory deviation recognition model. The recognition model has two hidden layers, each with 30 neurons.

[0018] The beneficial effects of this invention are:

[0019] The improved LBP algorithm based on window standard deviation designed in this invention can effectively extract the texture features of the temperature field of the workpiece sidewall. The established robot trajectory deviation recognition model can not only identify whether the robot's trajectory has deviated at the current moment, but also identify the degree of trajectory deviation. It largely solves the problem of the influence of arc light on the online monitoring of trajectory deviation of arc additive manufacturing robots, and provides a necessary strategy for the online monitoring of trajectory deviation of arc additive manufacturing robots. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for identifying trajectory deviation of an arc additive manufacturing robot based on the characteristics of welding temperature field distribution, provided by the present invention.

[0021] Figure 2 This is a schematic diagram of the monitoring system of the present invention.

[0022] Figure 3 This is a schematic diagram of the selection of the ROI of the temperature field on the side wall of the workpiece in the electric arc additive manufacturing of the present invention.

[0023] Figure 4 This is an image showing the effect of the improved LBP algorithm based on window standard deviation in extracting the texture features of the welding temperature field according to the present invention.

[0024] Figure 5 These are grayscale distribution images of the LBP feature maps before and after the improvement of the present invention: (a) grayscale distribution image of the traditional LBP feature map, and (b) grayscale distribution image of the improved LBP feature map.

[0025] Figure 6 This is a comparison diagram between the recognition results of the robot trajectory deviation recognition model of the present invention and the actual results. Detailed Implementation

[0026] The present invention will now be described in further detail with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0027] like Figure 1 The diagram shows a flowchart of a method for identifying trajectory deviation of an arc additive manufacturing robot based on the characteristics of welding temperature field distribution, according to the present invention.

[0028] First, acquire images of the temperature field on the sidewall of the workpiece manufactured by arc additive manufacturing. For example... Figure 2 The diagram shows a robot trajectory deviation monitoring system. The infrared thermal imager is fixed on the welding platform. After the arc additive manufacturing begins, the FPGA module sends a 50Hz signal to trigger the infrared thermal imager to collect the temperature field image of the workpiece sidewall.

[0029] Then, select the ROI. Select the ROI outside the molten pool region in the welding temperature field image, such as... Figure 3 The diagram shows the selection of the ROI on the side wall of the workpiece. The size of the selected ROI is 35 pixels × 90 pixels.

[0030] Extracting texture features of the ROI temperature field. An improved LBP algorithm is used to extract texture features of the ROI temperature field. The improved LBP algorithm is as follows: using a 3×3 square window, calculate the standard deviation S of all windows in the image, and then extract the standard deviation S of each window. iBinarization is performed according to a certain threshold. The binarized value is then added to the highest bit of the original LBP binary sequence, effectively improving the traditional 8-bit binary sequence into a 9-bit binary sequence. Finally, the 9-bit binary sequence is converted to decimal, which is the improved LBP value of the center pixel of the window. The comparison threshold for the standard deviation of each window is set to the average of the standard deviations of all windows in the image. The improved LBP calculation formula is as follows:

[0031]

[0032] In the formula, N is the total number of windows in the image, and the improved LBP value ranges from [0, 511].

[0033] Dimensionality reduction of ROI temperature field texture features. Principal component analysis was used to transform the 512-dimensional temperature field texture feature vector into 10-dimensional temperature field texture features.

[0034] Finally, a robot trajectory deviation recognition model is established. Using low-dimensional temperature field texture features as input, a robot trajectory deviation recognition model is built based on a backpropagation neural network, outputting the degree of deviation in the robot trajectory. The recognition model has two hidden layers, each with 30 neurons.

[0035] This invention is based on single-pass multi-layer electric arc additive manufacturing. It monitors the offset of the cladding layer during the electric arc additive manufacturing process. The entire electric arc additive manufacturing process is carried out according to the idea of ​​reciprocating stacking. The welding current is set to 160A, the welding speed is set to 8mm / s, the welding length is set to 12cm, the height of the welding torch lifting for each layer is set to 2.6mm, and the waiting time between layers is set to 30s. After the deposition of six cladding layers, the seventh layer begins to alter the welding torch's trajectory. This invention employs five methods to complete the deposition of the seventh cladding layer: the first method involves a welding torch ignition point offset of -2.4mm, with the torch movement direction parallel to the previous deposition direction; the second method involves a welding torch ignition point offset of -1.2mm, with the torch movement direction parallel to the previous deposition direction; the third method involves no offset at the welding torch ignition point, with the torch movement direction parallel to the previous deposition direction, indicating normal additive manufacturing; the fourth method involves a welding torch ignition point offset of +1.2mm, with the torch movement direction parallel to the previous deposition direction; and the fifth method involves a welding torch ignition point offset of +2.4mm, with the torch movement direction parallel to the previous deposition direction. Figure 4 The image shown is an example of the effect of extracting welding temperature field texture features using an improved LBP algorithm based on window standard deviation. Figure 5To improve the grayscale distribution of the LBP feature maps before and after, the output of the robot trajectory offset recognition model uses 1 to represent a robot trajectory offset of -2.4mm, 2 to represent a robot trajectory offset of -1.2mm, 3 to represent a robot trajectory no offset, 4 to represent a robot trajectory offset of +1.2mm, and 5 to represent a robot trajectory offset of +2.4mm. 750 data points were collected for each offset state, resulting in a total of 3750 data points. 1500 data points were used as the training set, and the remaining 2250 data points were used as the test set. Figure 6 The figure shown is a comparison between the recognition results of the robot trajectory offset recognition model and the actual results. The overall recognition accuracy of the robot trajectory offset recognition model exceeds 98%, demonstrating that the trajectory offset recognition method for arc additive manufacturing robots based on welding temperature field distribution characteristics of this invention has high recognition accuracy.

Claims

1. A method for identifying robot trajectory offsets for arc additive manufacturing based on welding temperature field distribution characteristics, the method comprising: Includes the following steps: ​ Step 1: Acquire images of the temperature field on the sidewall of the workpiece in arc additive manufacturing: Measure the temperature field on the sidewall of the workpiece using an infrared thermal imager; Step 2: Select ROI: Select the ROI outside the molten pool area in the welding temperature field image; Step 3: Extract texture features of the ROI temperature field: Use the improved LBP algorithm to extract texture features of the ROI temperature field; In step three, the improved LBP algorithm is as follows: using a 3×3 square window, calculate the standard deviation of all windows in the image. The standard deviation of each window Binarization is performed according to a certain threshold, and the binarized value is added to the highest bit of the original LBP binary sequence, which improves the traditional 8-bit binary sequence into a 9-bit binary sequence. Finally, the 9-bit binary sequence is converted into decimal, which is the improved LBP value of the center pixel of the window. The comparison threshold for the standard deviation of each window is set as the average of the standard deviations of all windows in the image. The improved LBP calculation formula is as follows: （1）； wherein is the total number of windows in the image, the range of the improved LBP value is ; Step 4: ROI Temperature Field Texture Feature Dimensionality Reduction: Use a dimensionality reduction algorithm to convert the high-dimensional temperature field texture feature vector into a low-dimensional temperature field texture feature vector; Step 5: Establish a robot trajectory deviation recognition model: Using low-dimensional temperature field texture features as input, establish a robot trajectory deviation recognition model based on a neural network, and output the degree of deviation of the robot trajectory.

2. The method of claim 1, wherein: In step one, an infrared thermal imager is fixed on the welding work platform to collect images of the temperature field on the sidewall of the workpiece during the arc additive manufacturing process.

3. The method for trajectory offset recognition of an arc additive manufacturing robot based on welding temperature field distribution characteristics according to claim 1, characterized in that: In step two, the ROI size is selected as 35 pixels × 90 pixels.

4. The method for trajectory offset recognition of an arc additive manufacturing robot based on welding temperature field distribution characteristics according to claim 1, characterized in that: In step four, the dimensionality reduction algorithm used is principal component analysis, which converts the 512-dimensional temperature field texture feature vector into a 10-dimensional temperature field texture feature.

5. The method of claim 1, wherein: In step five, a robot trajectory deviation recognition model is established using a backpropagation neural network. The recognition model has two hidden layers, each with 30 neurons.

Images