A Laser Welding Wire Feeding Control Method for a Robot Servo Control Device

Abstract

The present invention discloses a laser welding wire feeding control method for a robot servo control device, which relates to the field of welding data management. The method comprises the following steps: obtaining operation records of a laser welding wire feeding device and characteristics of a keyhole wall in a unit detection cycle, forming a target cycle group by combining two adjacent unit detection cycles, obtaining function characteristics of operation parameters of the laser welding wire feeding device in the target cycle group, collecting change values of characteristics of the keyhole wall in the target cycle group, obtaining detection records of welds within the target cycle group, collecting detection records corresponding to several target cycle groups to form a detection record set, obtaining image characteristics of the weld surface, calculating surface evaluation values in the weld surface image, collecting detection records of abnormalities inside the weld, calculating internal evaluation values of the weld according to the proportion of abnormal areas inside the weld, extracting characteristics of the surface evaluation values and internal evaluation values of the target cycle group, and establishing a quality evaluation vector for the target cycle group.

Description

Technical Field

[0001] The present invention relates to the field of welding data management, and specifically to a wire feeding control method for laser welding of a robot servo control device. Background Art

[0002] Laser welding is a welding method that uses a laser beam with a high energy density as a heat source and belongs to an efficient and precise welding technology. During the process of laser welding, high-temperature and high-pressure gases are generated inside the molten pool, and these gases will solidify during solidification to form metal pores. If the pore density is too high, keyhole defects will be formed.

[0003] Keyhole defects are a manifestation of substandard welding quality, which will affect the overall quality of welded components. Keyhole defects reduce the effective working cross-section of the weld, resulting in a decrease in the load-bearing capacity of the weld. When subjected to external forces, stress concentration is likely to occur around the keyhole, leading to easier fracture of the weld.

[0004] When the wire feeding speed is too low, the filled liquid metal may not be sufficient to fully wet and fill the weld, resulting in some areas in the weld not being completely melted or having poor fusion. This poor fusion may increase the risk of forming keyhole defects because the unmelted metal areas may form cavities or gaps; when the wire feeding speed is too high, too much liquid metal may compress the laser keyhole, causing the keyhole to be blocked or deformed. In this case, the pressure inside the keyhole may increase, making it easier to form keyhole defects. Summary of the Invention

[0005] The purpose of the present invention is to provide a wire feeding control method for laser welding of a robot servo control device to solve the problems raised in the prior art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] Step S100: Obtain the data characteristics of the operating parameters of the laser welding wire feeding device within a unit detection period, record the combination of one or at least two data characteristics as the first eigenvalue, and collect the change value of the characteristics of the keyhole wall within the target period group, and record it as the second eigenvalue;

[0008] Step S200: Obtain the detection records of the weld within a unit detection period, collect the abnormal records on the surface of the weld and the abnormal records inside the weld in the detection records, and calculate the abnormal evaluation value through the proportion of the abnormal area;

[0009] Step S300: Extract the abnormal evaluation value on the surface of the weld and the abnormal evaluation value inside the weld respectively, and establish a quality evaluation vector within a unit detection period;

[0010] Step S400: Aggregate the first eigenvalue, the second eigenvalue, and the quality evaluation vector of the target period group to form an evaluation sequence corresponding to the target period group. Obtain the evaluation sequences corresponding to several target period groups to form a first evaluation set. Perform clustering analysis on the quality evaluation vectors according to the positions where weld anomalies occur to obtain the clustering cores of each cluster.

[0011] Step S500: Collect the operation records of the laser welding wire feeding device and the change values of the characteristics of the keyhole wall in the current time period. Determine a reference evaluation sequence in the evaluation sequence, perform weighted summation on the quality evaluation vectors of the reference evaluation sequence, and establish a quality prediction vector corresponding to the predicted value.

[0012] Step S600: Calculate the distances between the quality prediction vector and the clustering cores of the clusters respectively, and sort the detection items of the current weld according to the distances.

[0013] Further, step S100 includes:

[0014] Step S101: Set a unit detection period, obtain the operation records of the laser welding wire feeding device and the characteristics of the keyhole wall in the unit detection period. Combine two adjacent unit detection periods to form a target period group, obtain the operation parameters of the laser welding wire feeding device within the target period group, draw an operation function of the operation parameters changing with time, and record the data characteristics of the operation function as the first eigenvalue.

[0015] Step S102: Represent the three-dimensional morphology characteristics of the keyhole wall by i spatial feature vectors, obtain the i spatial feature vectors of the keyhole wall in the first unit detection period of the target period group, and form the first vector cluster with the spatial feature vectors; Step S103: Obtain the i spatial feature vectors of the keyhole wall in the second unit detection period of the target period group, and form the second vector cluster with the spatial feature vectors.

[0016] Step S104: Record the difference value between the second vector cluster and the first vector cluster as the second eigenvalue.

[0017] Further, step S200 includes:

[0018] Step S201: Obtain an image feature extraction model, and train an identification model for weld surface anomaly features through images of normal weld surfaces and images of abnormal weld surfaces.

[0019] Step S202: Aggregate the detection records with weld surface anomalies into a first group, and identify the images of normal weld surfaces in each detection record in the first group.

[0020] Step S203: Obtain the image corresponding to the weld surface of the k1-th detection record in the first group during the target period group, and record the area of the image as Dk1. Identify the partial image with abnormalities on the weld surface in the image, and record the area of the partial image as dk1;

[0021] Step S204: Calculate the surface evaluation value αk1 of the k1-th detection record in the first group,

[0022] where αk1 = dk1 / Dk1.

[0023] Further, step S200 includes:

[0024] Step S205: Assemble the detection records with normal weld surfaces into the second group, and obtain the abnormal features of the abnormal areas inside the weld. The abnormal features include the corresponding space of the abnormal area or the two-dimensional feature slices of the corresponding space of the abnormal area;

[0025] Step S206: Obtain all the areas corresponding to the weld surface of the k2-th detection record in the second group during the target period group, and record the areas as Vk2. Obtain the partial areas with abnormalities in the k2-th detection record, and record the partial areas as vk2. Among them, when the abnormal feature adopts the corresponding space of the abnormal area, Vk2 represents the volume of the weld inside during the target period group, and vk2 represents the volume of the partial abnormal area. When the abnormal feature adopts two-dimensional feature slices, Vk2 represents the area of the two-dimensional feature slice image during the target period group, and vk2 represents the area of the image corresponding to the partial abnormal area;

[0026] Step S207: Calculate the internal evaluation value βk2 of the k2-th detection record in the second group,

[0027] where βk2 = vk2 / Vk2.

[0028] Further, step S300 includes:

[0029] Step S301: Calculate the surface evaluation value and the internal evaluation value of each detection record in the detection record set. When the weld surface or the inside of the weld in a certain detection record in the detection record set is normal, the corresponding surface evaluation value or internal evaluation value is recorded as 0;

[0030] Step S302: Obtain the surface evaluation value α n and the internal evaluation value β n of the n-th detection record in the detection record set, and calculate the comprehensive evaluation value γ n of the n-th detection record,

[0031] Step S303: Construct the quality evaluation vector e n of the n-th detection record, en (α n ,β n ,γ n ).

[0032] Further, step S400 includes:

[0033] Step S401: Set the first coordinate axis for measuring the first eigenvalue and the second coordinate axis for measuring the second eigenvalue, and establish the management plane of the first evaluation set;

[0034] Step S402: Combine the first eigenvalue and the second eigenvalue of the target cycle group into an eigenvalue pair, mark the position of the eigenvalue pair in the management plane, and correspond the quality evaluation vector of the target cycle group to the position;

[0035] By collecting historical data, the characteristic data of the welding process and the characteristic data of the welding result in the historical data are obtained, and the corresponding relationship between the two types of data is established. When the data in the current welding process is collected, the possible welding results are predicted.

[0036] Further, step S400 includes:

[0037] Step S403: Divide all the quality evaluation vectors in several target cycle groups into three categories: there are abnormalities on the weld surface, there are abnormalities inside the weld, and there are abnormalities on both the weld surface and inside the weld;

[0038] Step S404: Through the clustering algorithm, cluster the three types of quality evaluation vectors respectively according to the coordinates in the vector, and record the clustering cores of the three types as P1, P2, and P3 respectively, where P1 represents the clustering core with abnormalities on the weld surface, P2 represents the clustering core with abnormalities inside the weld, and P3 represents the clustering core with abnormalities on both the weld surface and inside the weld;

[0039] Combined with step S301, when there is only one type of defect in a certain weld, the value of αn or βn is 0. In three-dimensional space, the quality evaluation vector will be in a plane, and the two vectors with only one type of defect are in different planes and have a large separation degree. Therefore, after clustering, it is easy to distinguish the clustering cores of the three situations.

[0040] Further, step S500 includes:

[0041] Step S501: Combine two adjacent unit detection cycles in the current laser welding process to form the current cycle group, and collect the first eigenvalue m1 of the laser welding wire feeding device and the second eigenvalue m2 of the keyhole wall in the current cycle group;

[0042] Step S502: Combine m1 and m2 to form the target index f(m1, m2).

[0043] Further, step S500 includes:

[0044] Step S503: Represent the index position of the target index in the management plane of the first evaluation set, set a distance threshold r, obtain all evaluation sequences in the first evaluation set whose distance from the index position is less than r, and incorporate the evaluation sequences into the reference sequence set;

[0045] Step S504: Obtain the number of evaluation sequences in the reference sequence set and denote it as n, and obtain the coordinates (q1 j , q2 j ) of the j-th evaluation sequence in the reference sequence set in the management plane;

[0046] Step S505: Calculate the distance lmj between the j-th evaluation sequence and the target index, and calculate the sum of the distances between the n evaluation sequences and the target index and denote it as Ln;

[0047] Step S506: Calculate the weight uj of the j-th evaluation sequence, uj = 1 - (lmj / Ln), and obtain the quality evaluation vector E j ;

[0048] Step S507: Aggregate the weights and quality evaluation vectors of all evaluation sequences in the reference sequence set, and calculate the quality prediction vector H corresponding to the target index,

[0049] Further, step S600 includes:

[0050] Step S601: Calculate the distances between the quality prediction vector and P1, P2, and P3 respectively. Among them, the distance between the quality prediction vector and P1 is denoted as ph1, the distance between the quality prediction vector and P2 is denoted as ph2, and the distance between the quality prediction vector and P3 is denoted as ph3;

[0051] Step S602: Set a threshold ω. When ph3 ≤ ω, prompt the relevant staff to monitor the weld corresponding to the current cycle group;

[0052] Step S603: When ph1 < ph2, preferentially monitor the surface of the weld corresponding to the current cycle group. When ph2 < ph1, preferentially monitor the interior of the weld corresponding to the current cycle group.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows: By collecting the historical operation records of the laser welding wire feeding device and the quality of the corresponding welded products, the relationship between different device operation states and the quality of welded products is summarized. First, a correlation model of the operation of the laser welding wire feeding device, the change of keyhole, and the quality of welded products is established. When the operation data of the laser welding wire feeding device is detected, the possible welding quality is predicted. Further, a correlation model between welding quality and the order of sending for inspection is established, further exploring the relationship between the operation characteristics of the laser welding wire feeding device and the types of specific quality problems generated, and improving the inspection efficiency of product quality problems. Brief Description of the Drawings

[0054] Figure 1 It is a schematic flow chart of a laser welding wire feeding control method for a robot servo control device of the present invention. Detailed Embodiments

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0056] Embodiment: As Figure 1 shown, the present invention provides a technical solution, a laser welding wire feeding control method for a robot servo control device:

[0057] S1: Set a unit detection period, obtain the operation record of the laser welding wire feeding device and the characteristics of the keyhole wall in the unit detection period, form a target period group with two adjacent unit detection periods, obtain the function characteristics of the operation parameters of the laser welding wire feeding device within the target period group, denoted as the first eigenvalue, and collect the change value of the characteristics of the keyhole wall within the target period group, denoted as the second eigenvalue;

[0058] The laser welding wire feeder is a device used to automatically or manually control the wire feeding into the weld during the laser welding process;

[0059] In the embodiment, when obtaining the operation parameters of the laser welding wire feeder, the selectable operation parameter items include: laser power, welding speed, and wire feeding speed;

[0060] In the embodiment, the characteristics of the function with operating parameters varying over time are used as the first eigenvalue, including the timing characteristics of the operating parameters, such as the change trend characteristics of the operating parameters, for example, the increase or decrease of the function, the periodic change region of the function, for example, the change frequency or change period, the statistical characteristics of the operating parameters, such as the mean, variance or standard deviation, such as the waveform characteristics of the function, such as the waveform of the operating parameter varying over time, and the correlation characteristics of the function, for example, the autocorrelation characteristics and cross-correlation characteristics of the function;

[0061] Wherein, S1 includes:

[0062] S11: Obtain the operating parameters of the laser welding wire feeding device within the target period group, draw the operating function of the operating parameters varying over time, and record the function characteristics of the operating function as the first eigenvalue;

[0063] S12: Represent the three-dimensional morphology characteristics of the keyhole wall by i spatial feature vectors, obtain the i spatial feature vectors of the keyhole wall in the first unit detection period of the target period group, and form the first vector cluster with the spatial feature vectors; S13: Obtain the i spatial feature vectors of the keyhole wall in the second unit detection period of the target period group, and form the second vector cluster with the spatial feature vectors;

[0064] S14: Record the difference value between the second vector cluster and the first vector cluster as the second eigenvalue;

[0065] In the embodiment, the difference between the two vector clusters can be calculated by comparing the vectors in the vector clusters one by one, or the first vector cluster and the second vector cluster are respectively formed into an i-order matrix, and the second eigenvalue is obtained by comparing the difference values of the two i-order matrices;

[0066] The methods for comparing matrix difference degrees include: Frobenius norm (Frobenius norm, F-norm), Manhattan distance and Euclidean distance.

[0067] S2: Obtain the detection records of the weld within the target period group range, collect the detection records corresponding to several target period groups to form a detection record set, obtain the image characteristics of the weld surface, and calculate the proportion of the abnormal part in the weld surface image to the normal image of the weld surface to obtain the surface evaluation value;

[0068] Wherein, S2 includes:

[0069] S21: Obtain the image feature extraction model, and train the recognition model of the weld surface abnormal features through the normal image of the weld surface and the abnormal image of the weld surface;

[0070] S22: Collect the detection records with abnormal weld surfaces into the first group, and identify the normal images of the weld surfaces in each detection record in the first group;

[0071] S23: Obtain the image corresponding to the weld surface of the k1-th detection record in the first group in the target cycle group, record the area of the image as Dk1, identify the partial image with anomalies on the weld surface in the image, and record the area of the partial image as dk1;

[0072] S24: Calculate the surface evaluation value αk1 of the k1-th detection record in the first group,

[0073] where αk1 = dk1 / Dk1.

[0074] S3: Collect the detection records with anomalies inside the weld, and calculate the internal evaluation value of the weld according to the proportion of the abnormal area inside the weld;

[0075] where S3 includes:

[0076] S31: Collect the detection records with normal weld surfaces into the second group, and obtain the abnormal features of the abnormal area inside the weld. The abnormal features include the corresponding space of the abnormal area or the two-dimensional feature slices of the corresponding space of the abnormal area;

[0077] S32: Obtain all the areas corresponding to the weld surface of the k2-th detection record in the second group in the target cycle group, record the area as Vk2, obtain the partial area with anomalies in the k2-th detection record, and record the partial area as vk2. Among them, when the abnormal feature uses the corresponding space of the abnormal area, Vk2 represents the volume of the weld inside in the target cycle group, and vk2 represents the volume of the partial abnormal area. When the abnormal feature uses two-dimensional feature slices, Vk2 represents the area of the two-dimensional feature slice image in the target cycle group, and vk2 represents the area of the image corresponding to the partial abnormal area;

[0078] S33: Calculate the internal evaluation value βk2 of the k2-th detection record in the second group,

[0079] where βk2 = vk2 / Vk2.

[0080] S4: Extract the features of the surface evaluation value and the internal evaluation value of the target cycle group, and establish the quality evaluation vector of the target cycle group;

[0081] where S4 includes:

[0082] S41: Calculate the surface evaluation value and the internal evaluation value of each detection record in the detection record set. When the weld surface or the weld inside in a certain detection record in the detection record set is normal, the corresponding surface evaluation value or internal evaluation value is recorded as 0;

[0083] S42: Obtain the surface evaluation value αn and the internal evaluation value βn of the nth detection record in the detection record set, and calculate the comprehensive evaluation value γn of the nth detection record.

[0084] S43: Construct the quality evaluation vector en of the nth detection record, en(αn, βn, γn).

[0085] S5: Aggregate the first eigenvalue, the second eigenvalue, and the quality evaluation vector of the target period group to form an evaluation sequence corresponding to the target period group, obtain the evaluation sequences corresponding to several target period groups, and form the first evaluation set.

[0086] Among them, S5 includes:

[0087] S51: Set the first coordinate axis for measuring the first eigenvalue and the second coordinate axis for measuring the second eigenvalue, and establish the management plane of the first evaluation set.

[0088] S52: Combine the first eigenvalue and the second eigenvalue of the target period group to form an eigenvalue pair, mark the position of the eigenvalue pair in the management plane, and correspond the quality evaluation vector of the target period group to the position.

[0089] S6: Obtain all the evaluation vectors in several target period groups, and respectively cluster the evaluation vectors according to the presence of abnormalities on the weld surface, the presence of abnormalities inside the weld, and the presence of abnormalities on both the weld surface and inside the weld, and obtain the clustering cores of the three clusters.

[0090] Among them, S6 includes:

[0091] S61: Divide all the evaluation vectors in several target period groups into three categories: the presence of abnormalities on the weld surface, the presence of abnormalities inside the weld, and the presence of abnormalities on both the weld surface and inside the weld.

[0092] S62: Through the clustering algorithm, cluster the three categories of evaluation vectors respectively according to the coordinates in the vector, and record the clustering cores of the three categories as P1, P2, and P3 respectively, where P1 represents the clustering core of the presence of abnormalities on the weld surface, P2 represents the clustering core of the presence of abnormalities inside the weld, and P3 represents the clustering core of the presence of abnormalities on both the weld surface and inside the weld.

[0093] S7: Collect the operating parameters of the current laser welding wire feeding device and the characteristics of the keyhole wall, form the current period group with the current two adjacent unit detection periods, obtain the first eigenvalue and the second eigenvalue in the current period group, and form the target index.

[0094] Among them, S7 includes:

[0095] S71: Form the current cycle group from two adjacent unit detection cycles during the current laser welding process, and collect the first eigenvalue m1 of the wire feeding device for laser welding and the second eigenvalue m2 of the keyhole wall in the current cycle group;

[0096] S72: Combine m1 and m2 to form the target index f(m1, m2).

[0097] S8: Map the target index into the first evaluation set, obtain the evaluation sequence within the distance threshold range from the target index, calculate the weights of the target index and the evaluation sequence according to the distance, and perform a weighted operation on the quality evaluation vectors in the evaluation sequence to obtain the quality prediction vector;

[0098] Among them, S8 includes:

[0099] S81: Represent the index position of the target index in the management plane of the first evaluation set, set the distance threshold r, obtain all evaluation sequences in the first evaluation set whose distances from the index position are less than r, and incorporate the evaluation sequences into the reference sequence set;

[0100] S82: Obtain the number of evaluation sequences in the reference sequence set and denote it as n, and obtain the coordinates (q1j, q2j) of the j-th evaluation sequence in the reference sequence set in the management plane;

[0101] S83: Calculate the distance lmj between the j-th evaluation sequence and the target index, Calculate the sum of the distances between the n evaluation sequences and the target index and denote it as Ln;

[0102] S84: Calculate the weight uj of the j-th evaluation sequence, uj = 1 - (lmj / Ln), and obtain the evaluation vector Ej in the j-th evaluation sequence;

[0103] S85: Aggregate the evaluation vectors of the weights of all evaluation sequences in the reference sequence set, and calculate the quality prediction vector H corresponding to the target index,

[0104] S9: Calculate the distances between the quality prediction vector and the clustering cores of the three clusters respectively, and sort the detection items of the current weld seam according to the distances.

[0105] Among them, S9 includes:

[0106] S91: Calculate the distances between the quality prediction vector and P1, P2, and P3 respectively. Among them, the distance between the quality prediction vector and P1 is denoted as ph1, the distance between the quality prediction vector and P2 is denoted as ph2, and the distance between the quality prediction vector and P3 is denoted as ph3;

[0107] S92: Set the threshold ω. When ph3 ≤ ω, prompt the relevant staff to monitor the weld seam corresponding to the current cycle group;

[0108] S93: When ph1 < ph2, preferentially monitor the surface of the weld corresponding to the current cycle group; when ph2 < ph1, preferentially monitor the interior of the weld corresponding to the current cycle group.

[0109] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

Claims

1. A laser welding wire feeding control method for a robot servo control device, characterized in that: The method includes the following steps: Step S100: Obtain the data characteristics of the operating parameters of the laser welding wire feeding device within a unit detection period, denote the combination of one or at least two data characteristics as the first eigenvalue, and collect the change value of the characteristics of the keyhole wall within the target period group, denoted as the second eigenvalue; Step S200: Obtain the detection records of the weld seam within a unit detection period, collect the abnormal records on the surface of the weld seam and the abnormal records inside the weld seam in the detection records, and calculate the abnormal evaluation value through the proportion of the abnormal area; Step S300: Extract the abnormal evaluation value on the surface of the weld seam and the abnormal evaluation value inside the weld seam respectively, and establish a quality evaluation vector within a unit detection period; Step S400: Aggregate the first eigenvalue, the second eigenvalue and the quality evaluation vector of the target period group to form an evaluation sequence corresponding to the target period group, obtain a first evaluation set composed of evaluation sequences corresponding to several target period groups, and perform clustering analysis on the quality evaluation vector according to the position where the weld seam abnormality appears to obtain the clustering core of each clustering; Step S500: Collect the operating records of the laser welding wire feeding device and the change value of the characteristics of the keyhole wall in the current time period, determine a reference evaluation sequence in the evaluation sequence, perform weighted summation on the quality evaluation vector of the reference evaluation sequence, and establish a quality prediction vector corresponding to the predicted value; Step S600: Calculate the distances between the quality prediction vector and the clustering cores of the clusters respectively, and sort the detection items of the current weld seam according to the distances.

2. A laser welding wire feeding control method for a robot servo control device according to claim 1, characterized in that: Step S100 includes: Step S101: Set a unit detection period, obtain the operating records of the laser welding wire feeding device and the characteristics of the keyhole wall in the unit detection period, form a target period group with two adjacent unit detection periods, obtain the operating parameters of the laser welding wire feeding device within the target period group, draw an operating function of the change of the operating parameters with time, and denote the data characteristics of the operating function as the first eigenvalue; Step S102: Represent the three-dimensional morphological characteristics of the keyhole wall through i spatial feature vectors, obtain the i spatial feature vectors of the keyhole wall in the first unit detection period in the target period group, and form the first vector cluster with the spatial feature vectors; Step S103: Obtain the i spatial feature vectors of the keyhole wall in the second unit detection period in the target period group, and form the second vector cluster with the spatial feature vectors; Step S104: Denote the difference value between the second vector cluster and the first vector cluster as the second eigenvalue.

3. A laser welding wire feeding control method for a robot servo control device according to claim 2, characterized in that: The method for calculating the abnormal evaluation value on the surface of the weld seam in Step S200 includes: Step S201: Obtain an image feature extraction model, and train an identification model for abnormal features on the surface of the weld seam through the images of the normal surface of the weld seam and the abnormal images of the surface of the weld seam; Step S202: Aggregate the detection records with abnormalities on the surface of the weld seam into the first group, and identify the images of the normal surface of the weld seam in each detection record in the first group; Step S203: Obtain the image corresponding to the weld surface of the k1-th detection record in the first group in the target cycle group, and record the area of the image as Dk1. Identify the partial image with abnormalities on the weld surface in the image, and record the area of the partial image as dk1; Step S204: Calculate the surface evaluation value αk1 of the k1-th detection record in the first group, where αk1 = dk1 / Dk1.

4. A laser welding wire feeding control method for a robot servo control device according to claim 3, characterized in that: The method for calculating the abnormality evaluation value inside the weld in Step S200 includes: Step S205: Collect the detection records with normal weld surfaces into the second group, and obtain the abnormal features of the abnormal areas inside the weld. The abnormal features include the corresponding space of the abnormal area or the two-dimensional feature slices of the corresponding space of the abnormal area; Step S206: Obtain all the areas corresponding to the weld surface of the k2-th detection record in the second group in the target cycle group, and record the area as Vk2. Obtain the partial area with abnormalities in the k2-th detection record, and record the partial area as vk2. Among them, when the abnormal feature adopts the corresponding space of the abnormal area, Vk2 represents the volume of the weld inside in the target cycle group, and vk2 represents the volume of the partial abnormal area. When the abnormal feature adopts two-dimensional feature slices, Vk2 represents the area of the two-dimensional feature slice image in the target cycle group, and vk2 represents the area of the image corresponding to the partial abnormal area; Step S207: Calculate the internal evaluation value βk2 of the k2-th detection record in the second group, where βk2 = vk2 / Vk2.

5. A laser welding wire feeding control method for a robot servo control device according to claim 4, characterized in that: Step S300 includes: Step S301: Calculate the surface evaluation value and the internal evaluation value of each detection record in the detection record set. When the weld surface or the inside of the weld in a certain detection record in the detection record set is normal, the corresponding surface evaluation value or internal evaluation value is recorded as 0; Step S302: Obtain the surface evaluation value α of the nth detection record in the detection record set n and the internal evaluation value β n , and calculate the comprehensive evaluation value γ of the nth detection record n , Step S303: Construct the quality evaluation vector e of the nth detection record n , e n (α n , β n , γ n ).

6. A laser welding wire feeding control method for a robot servo control device according to claim 5, characterized in that: The method for constructing the evaluation sequence in Step S400 includes: Step S401: Set the first coordinate axis for measuring the first eigenvalue and the second coordinate axis for measuring the second eigenvalue, and establish the management plane of the first evaluation set; Step S402: Combine the first eigenvalue and the second eigenvalue of the target cycle group into an eigenvalue pair, mark the position of the eigenvalue pair in the management plane, and correspond the quality evaluation vector of the target cycle group to the position.

7. A wire feeding control method for laser welding of a robot servo control device according to claim 6, characterized in that: The method for cluster analysis in Step S400 includes: Step S403: Divide all the quality evaluation vectors in several target cycle groups into three categories: there are abnormalities on the weld surface, there are abnormalities inside the weld, and there are abnormalities on both the weld surface and inside the weld; Step S404: Through the clustering algorithm, cluster the three categories of quality evaluation vectors respectively according to the coordinates in the vector, and record the clustering cores of the three categories as P1, P2, and P3 respectively. Among them, P1 represents the clustering core with abnormalities on the weld surface, P2 represents the clustering core with abnormalities inside the weld, and P3 represents the clustering core with abnormalities on both the weld surface and inside the weld.

8. A laser welding wire feeding control method for a robot servo control device according to claim 7, characterized in that: The processing method for the current time period in Step S500 includes: Step S501: Form a current cycle group from two adjacent unit detection cycles during the current laser welding process, and collect the first eigenvalue m1 of the wire feeding device for laser welding and the second eigenvalue m2 of the keyhole wall in the current cycle group; Step S502: Combine m1 and m2 to form a target index f(m1, m2).

9. A laser welding wire feeding control method for a robot servo control device according to claim 8, characterized in that: The method for establishing the quality prediction vector corresponding to the predicted value in Step S500 includes: Step S503: Represent the index position of the target index in the management plane of the first evaluation set, set a distance threshold r, obtain all evaluation sequences in the first evaluation set whose distance from the index position is less than r, and incorporate the evaluation sequences into the reference sequence set; Step S504: Obtain the number of evaluation sequences in the reference sequence set, denoted as n, and obtain the coordinates (q1 j , q2 j ) of the j-th evaluation sequence in the reference sequence set in the management plane; Step S505: Calculate the distance lmj between the j-th evaluation sequence and the target index. Calculate the sum of the distances between the n evaluation sequences and the target index, denoted as Ln. Step S506: Calculate the weight uj of the j-th evaluation sequence, uj = 1 - (lmj / Ln), and obtain the quality evaluation vector E in the j-th evaluation sequence j ; Step S507: Aggregate the weights and quality evaluation vectors of all evaluation sequences in the reference sequence set, and calculate the quality prediction vector H corresponding to the target index.

10. A method for controlling wire feeding in laser welding of a robot servo control device according to claim 9, characterized in that: Step S600 includes: Step S601: Calculate the distances between the quality prediction vector and P1, P2, and P3 respectively. Among them, the distance between the quality prediction vector and P1 is denoted as ph1, the distance between the quality prediction vector and P2 is denoted as ph2, and the distance between the quality prediction vector and P3 is denoted as ph3; Step S602: Set a threshold ω. When ph3 ≤ ω, prompt the relevant staff to monitor the weld corresponding to the current cycle group; Step S603: When ph1 < ph2, preferentially monitor the surface of the weld corresponding to the current cycle group. When ph2 < ph1, preferentially monitor the interior of the weld corresponding to the current cycle group.

Images