KR102039054B1 - Apparatus and Method for predicting of welding strength - Google Patents

Abstract

The present invention relates to a welding strength prediction method, comprising: measuring a welded surface of a welded portion of a welded base material through a vision sensor; A shape derivation step of deriving a weld shape using the measured weld surface; A parameter calculating step of calculating a parameter of the welded portion from the welded shape; An input value obtaining step of obtaining the physical property of the base material as an input value; And an intensity prediction step of inputting the parameter and the input value into an artificial neural network and predicting the intensity corresponding to the parameter and the input value.

Description

Apparatus and Method for predicting of welding strength}

The present invention relates to an apparatus and method for predicting welding strength. More particularly, the welding strength predicting the tensile strength and the fatigue strength of the welding portion with high accuracy by inputting the shape of the welding portion measured by the vision sensor into the artificial intelligence network. The present invention relates to a prediction apparatus and a method.

In general, when assembling complex products used in various fields, such as ships, automobiles, industrial equipment, and the like, welding is performed on a large part of each part to fix the respective parts. The above welding work is performed by a welding robot or by an operator. Weld by welding machine.

Since such welding work directly affects the strength of the product, work must be carried out with great care and the welding condition of the weld after welding. In other words, after welding, the strength of the weld should be checked by inspecting the state of the weld, which is a part that substantially joins the two parts.

The inspection method of the welded product is the presence or absence of defects on the inside and outside of the specimen, without changing or destroying the properties and internal structure of materials, device structures, etc. using radiation, ultrasonic waves, electromagnetic waves, fluids, heat, and light. Non-destructive inspection that checks the condition, characteristics, material change, and soundness, and if it is impossible to check with the naked eye of the inspector or if it is difficult to analyze the definite results even with non-destructive inspection equipment, it is cut and analyzed for the defects by analyzing the welding state of the cut part. There are destructive checks that can be expected and, if the defects are presumed, correctable based on them.

In this conventional welding inspection, it is very difficult to predict the welding result, and the management and prediction of the welding quality becomes more complicated, and it takes a lot of time and money to analyze the weld strength, tensile strength, etc. The process was complicated and difficult.

The idea of the present invention was devised in order to meet the above-mentioned requirements, and it is possible to derive the weld shape by measuring the weld surface through a vision sensor, and when inputting the parameter calculated in the weld shape into the artificial neural network, The aim is to predict mechanical properties with a high degree of accuracy. However, these problems are exemplary, and the scope of the present invention is not limited thereby.

Welding strength prediction method according to the idea of the present invention for solving the above problems, measuring the welding surface of the welded portion of the welded base material through the vision sensor; A shape derivation step of deriving a weld shape using the measured weld surface; A parameter calculating step of calculating a parameter of the welded portion from the welded shape; An input value obtaining step of obtaining the physical property of the base material as an input value; And an intensity prediction step of inputting the parameter and the input value into an artificial neural network and predicting the intensity corresponding to the parameter and the input value.

According to the spirit of the present invention, the intensity prediction step comprises: an input step of inputting the parameter and the input value into an artificial neural network; And calculating a tensile strength and a fatigue strength of the welded part through the first hidden layer and the second hidden layer formed of ten nodes, respectively, and outputting the tensile strength and the fatigue strength of the welded part.

According to the spirit of the present invention, the parameters include reinforcement addition, weld bead thickness, actual neck thickness, length, penetration depth, toe angle, welding bead angle, angle angle, reinforcement attachment area, neck area, It may include any one or more of the penetration area and the weld porosity.

According to the spirit of the present invention, the input value may include any one or more of the strength, thickness, plating, texture, top undercut, bottom undercut, and top and bottom gaps of the upper and lower plates of the base material. .

Welding strength prediction apparatus according to the spirit of the present invention for solving the above problems, the measuring unit for measuring the welded surface of the welded portion of the welded base material through the vision sensor; A shape derivation unit for deriving a weld shape using the measured weld surface; A parameter calculator configured to calculate a parameter of the welded part from the welded shape; An input value obtaining unit obtaining the physical property of the base material as an input value; And an intensity estimator configured to input the parameter and the input value to an artificial neural network and predict an intensity corresponding to the parameter and the input value.

According to the spirit of the present invention, the intensity predicting unit, an input unit for inputting the parameter and the input value to the artificial neural network; And a calculation unit for calculating the tensile strength and the fatigue strength of the welded portion through the artificial neural network having the first hidden layer and the second hidden layer formed of ten nodes, respectively; and an output unit for outputting the tensile strength and the fatigue strength of the welded portion. can do.

According to the welding strength prediction apparatus and method according to the present invention, it is possible to predict the weld shape only by measuring the weld surface using a vision sensor, and various parameters can be calculated in the weld shape, so that the process time and cost in inspecting the weld position Can be reduced.

In addition, the strength of the artificial neural network is used as a method of predicting the strength, and it is not necessary to check the results through multiple experiments and set new process variables every time the welded material is changed, and have high prediction accuracy with minimal input data. It can reduce the process time and cost, maximize the cost and time-to-time welding quality, and the welded material without data has the effect of welding prediction. The scope of the present invention is not limited by these effects.

FIG. 1 is a flowchart illustrating a welding strength prediction method using the welding strength prediction apparatus of FIG. 1.
2 is a flow chart showing a welding strength prediction method according to another embodiment of the present invention.
3 is a view showing a welding strength prediction apparatus according to an embodiment of the present invention.
4 is a perspective view showing a measurement step according to the welding strength prediction method of the present invention.
5 is a view showing the shape of the welded portion of the shape derivation step according to an embodiment of the present invention.
6 is a cross-sectional view showing the parameters of the welded portion in the parameter calculation step according to an embodiment of the present invention.
7 is a view showing the calculation of the artificial neural network of the intensity prediction step according to an embodiment of the present invention.
8 is a view showing a node value for the strength prediction experiment according to an embodiment of the present invention.
9 is a view showing a tensile strength prediction experiment example according to an embodiment of the present invention.
10 is a graph showing the results of tensile strength prediction experiments according to an embodiment of the present invention.
11 is a graph showing the results of fatigue strength prediction experiments according to an embodiment of the present invention.

Hereinafter, various exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

The embodiments of the present invention are provided to more fully explain the present invention to those skilled in the art, and the following examples can be modified in various other forms, and the scope of the present invention is It is not limited to an Example. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. In addition, the thickness or size of each layer in the drawings is exaggerated for convenience and clarity of description.

Embodiments of the present invention will now be described with reference to the drawings, which schematically illustrate ideal embodiments of the present invention. In the figures, for example, variations in the shape shown may be expected, depending on manufacturing techniques and / or tolerances. Accordingly, embodiments of the inventive concept should not be construed as limited to the specific shapes of the regions shown herein, but should include, for example, changes in shape resulting from manufacturing.

1 and 2 is a flow chart showing a welding strength prediction method according to the invention, Figure 3 is a view showing a welding strength prediction apparatus according to an embodiment of the present invention, Figures 4 to 7 is an embodiment of the present invention A diagram showing each step according to.

As shown in Figure 1, the welding strength prediction method of the present invention, the measurement step (S10), the shape derivation step (S20), the parameter calculation step (S30), the input value acquisition step (S40) and the strength prediction step (S50) ) May be included.

In the measuring step S10, the welded surface of the welded base material is measured by the vision sensor 3. For example, the welded surface of the welded base material is measured by the vision sensor 3. It may include a measuring unit 10.

More specifically, as shown in FIGS. 1 and 4, in the measuring step S10, after welding to the base material, the welding part of the base material may be measured by the vision sensor 3 formed in the measuring unit 10. Can be. In this case, in order to measure the welded portion with the vision sensor 3, the welded portion may be accurately measured on the welded surface of the welded portion of the base material using the vision sensor 3 under the vision sensor 3. have.

In general, a laser vision sensor is used as the vision sensor 3. The laser vision sensor device typically detects a welding line by photographing an image of a laser irradiation device (laser diode) that emits a laser and an irradiated laser on a welded object. The camera may be configured to accurately measure the weld surface.

As such, the shape of the weld can be predicted only by measuring the surface of the weld using the vision sensor, and various parameters can be calculated in the shape of the weld, thereby reducing the process time and cost in inspecting the weld.

As shown in FIG. 1, FIG. 3, and FIG. 5, the shape derivation step (S20) is a step of deriving a weld shape using the measured weld surface, and derives a weld shape using the measured weld surface. The shape derivation unit 20 may be included.

The weld shape may be formed in a specific shape by melting a portion of the upper plate 1 and the lower plate 2 of the base material, and in order to calculate the strength of the weld, the weld shape is a very important factor.

In the shape derivation step S20, the welding part surface measured by the vision sensor 3 in the measuring step S20 is specified, and the welding part between the upper plate 1 and the lower plate 2 of the base material is measured by the welding part surface. The shape can be derived.

In the shape deriving step S20, the welded surface measured by the vision sensor 3 in the measuring step S10 may be input to an artificial neural network. For example, the data of the weld surface is input to the artificial neural network, and the data of the weld surface is output similar to the data that the artificial neural network has learned, and compared with the data of the weld surface. At this time, if the data similarity is a predetermined value or more, the corresponding data is outputted. If the data similarity is less than the predetermined value, the searching for similar data is repeated again.

As shown in Fig. 1, Fig. 3 and Fig. 6, the parameter calculating step (S30) is a step of calculating a parameter of the welded portion from the welded shape, and calculating a parameter of calculating the parameter of the welded portion from the welded shape. It may include a portion (30).

The parameter calculating step (S30) is to calculate numerical values for calculating the weld strength in the shape of the welded portion derived in the shape derivation step (S20), the molten joined of the upper plate 1 and the lower plate 2 of the base material The numerical value of the welded area can be calculated.

The parameters include reinforcement addition (L1), weld bead thickness (L2), actual neck thickness (L3), square lengths (L4, L5), penetration depth (L6), toe angle (θ1), welding at the weld. It may include any one or more of the bead angle (θ2), rectangular angle (θ3), reinforcement addition area (A1), neck area (A2), penetration area (A3) and weld porosity (N24).

More specifically, as shown in FIG. 6, the parameter is a reinforcement of the melting part representing the thickness of the convex part of the convex shape in which the upper plate 1 and the lower plate 2 of the base material are melted and joined in the cross section of the welding part. Addition L1, welding bead thickness L2 representing the thickness of the joint, actual neck thickness L3 representing the thickness of the neck formed by joining the upper plate 1 and the lower plate 2, and the melting length of the upper plate 1 And lengths L4 and L5 representing the melting length of the lower plate 2, penetration depth L6 representing the molten depth of the lower plate 2, tow angle angle θ1 representing the outer angle of the convex shape that is melted and joined, Welding bead angle (θ2) representing the inner angle of the weld bead, rectangular angle angle (θ3) representing the melted angle of the upper plate 1, reinforcement addition area A1 representing the cross-sectional area of the reinforcement addition L1, and actual thickness (L3) Melted area of the lower plate (A2), indicating the cross-sectional area of the neck, including The penetration area A3 representing the cross-sectional area of the depth and the weld porosity N24 measured at the weld surface may be included.

As shown in FIGS. 1 and 3, the input value obtaining step S40 is a step of acquiring a physical property of the base material as an input value, and an input value obtaining part 40 obtaining the physical property of the base material as an input value. ) May be included.

In the input value obtaining step S40, the physical properties of the upper plate 1 and the lower plate 2 of the base material may be input, or the physical properties of the base material may be extracted and imported through a plurality of welding data.

The input value is the strength (N1, N5), thickness (N2, N6), plating (N3, N7), structure (N4, N8), upper plate undercut (N21) of the upper plate 1 and lower plate 2 of the base material ), A lower plate undercut N22, and a gap N23 of the upper plate 1 and the lower plate 2 may be included.

The parameter calculation step (S30) and the input value acquisition step (S40), the strength (N1, N5), thickness (N2, N6), plating (N3, N7) of the upper plate 1 and the lower plate 2 of the base material, Structure (N4, N8), reinforcement addition of molten part (L1), weld bead thickness (L2), actual neck thickness (L3), square length (L4, L5), penetration depth (L6), toe angle (θ1), welding Bead angle (θ2), angle angle (θ3), reinforcement area (A1), neck area (A2), penetration area (A3), top plate undercut (N21), bottom plate undercut (N22), top plate (1) and bottom plate ( 24 may include 24 node values represented by a gap N23 and a weld porosity N24.

The node value may be included in an input layer input to the artificial neural network in the strength prediction step (S50).

As shown in FIG. 1, FIG. 3, and FIG. 6, the intensity prediction step (S50) is a step of inputting the parameter and the input value into an artificial neural network and predicting the intensity corresponding to the parameter and the input value. The apparatus may include an intensity estimator 50 for inputting the parameter and the input value to an artificial neural network and predicting an intensity corresponding to the parameter and the input value.

For example, the node value is input to the artificial neural network, and the inputted node value outputs data most similar to the data learned by the artificial neural network, and is compared with the data of the node value. In this case, when the data similarity is greater than or equal to a certain value, the intensity value corresponding to the corresponding data is output. When the data similarity is less than or equal to a certain value, the searching for similar data is repeated.

The data learned by the artificial neural network may be learned using a plurality of previously stored welding data, wherein the plurality of previously stored welding data may be existing welding data obtained and stored by experiment in advance.

Artificial neural network (ANN) is a kind of function that connects input and output by connecting a virtual neural device, perceptron, which is a model of a neuron, a neuron of the human body, into a network. Perceptron's complex network makes it relatively simple to reproduce or predict complex nonlinear phenomena.

The prediction method using the artificial neural network is largely divided into a learning stage and a production stage. In the learning phase, a series of input / output relationships are supplied, and thus a functional relationship is identified in the artificial neural network. After the output variable is calculated from the given input variable, the weight of each function is adjusted by the learning algorithm according to the actual output value and the error, and the learning continues. When the error falls within the allowable range, the learning is terminated and the artificial neural network can store the functional relationship between the current input and output.

As shown in FIGS. 2 and 7, the intensity prediction step S50 may include an input step S51, an operation step S52, and an output step S53.

The input step S51 may include inputting the parameter and the input value to the artificial neural network, and may include an input unit 51 for inputting the parameter and the input value to the artificial neural network.

The input step S51 is a step of inputting the 24 node values calculated in the parameter calculating step S30 and the input value obtaining step S40 into the artificial neural network.

The calculating step (S52) is a step of calculating the tensile strength and the fatigue strength of the weld through the first hidden layer (S521) and the second hidden layer (S522) formed of 10 nodes, respectively, the first hidden layer (S521) and the first 2 may include a calculation unit 52 for calculating the tensile strength and the fatigue strength of the weld through the artificial neural network having a hidden layer (S522).

In the calculation step S52, the 24 nodes input in the input step S51 are calculated through the first hidden layer S521 including 10 nodes to calculate a first calculation value, and the first calculation value is 10. The second operation value may be calculated through the second hidden layer S522 including the two nodes, and the second operation value may be extracted as one node.

Output step (S53), the step of outputting the tensile strength and fatigue strength of the weld, it may include an output unit 53 for outputting the tensile strength and fatigue strength of the weld.

In the outputting step S53, one node extracted as the quadratic operation value may be output as tensile strength or fatigue strength.

As shown in FIG. 3, the display unit 60 may further include a display unit 60 capable of displaying the output tensile strength and the fatigue strength to a user.

8 and 9 are diagrams showing a node value and an experimental example for the strength prediction experiment according to an embodiment of the present invention.

The number of the nodes is 24, the strength (N1, N5), thickness (N2, N6), plating (N3, N7) of the upper plate (1) and lower plate (2) of the base material from the first node value to the node number 24 Structure (N4, N8), reinforcement addition (L1, N9) of melting part, weld bead thickness (L2, N10), actual neck thickness (L3, N11), square (L4, N12) of upper plate (1), lower plate (2) ) L5, N13, penetration depth (L6, N14), toe angle (θ1, N15), weld bead angle (θ2, N16), square angle (θ3, N17), reinforcement area (A1, N18) ), Neck area (A2, N19), penetration area (A3, N20), top plate undercut (N21), bottom plate undercut (N22), gap (N23) of top plate (1) and bottom plate (2), top plate undercut N21, a lower plate undercut N22, a gap N23 between the upper plate 1 and the lower plate 2, and a weld porosity N24.

The node value may be input to the artificial neural network to predict tensile strength and weld strength.

FIG. 10 is a graph showing the results of predicted tensile strength and experimental tensile strength as tensile strength prediction results according to the welding strength prediction method according to the experimental example of FIG. 9.

SGAFH 590FB 2.3 mm was used as the experimental material. The experiment used overlapped joint shape. Welding conditions were DC, CMT, welding speed 3, 5, 7, 9 m / min, welding speed 60, 80 cm / mn, gap 0, 0.1, 0.2 , 0.5, 1.0 mm were used. CTWD was fixed at 15 mm and working angle at 45 °, and 90% Ar + 10% CO ₂ gas mixture was used.

The average error was 19MPa, which was about 4.0%, and the coefficient of determination (R ² ) was 0.8199, indicating that the average tensile strength and the experimental tensile strength were in line with each other.

In particular, the tensile strengths of 580 MPa to 630 MPa showed that most of the predicted tensile strengths and the experimental tensile strengths were matched, and the predicted rate was high at 580 MPa to 630 MPa.

FIG. 11 is a graph illustrating fatigue strength prediction results according to the example of FIG. 9.

As a fatigue strength prediction result according to the welding strength prediction method according to an embodiment of the present invention, SGAFH 590FB 2.3 mm was used as the experimental material. The experiment used overlapped joint shape, and welding conditions were CMT, welding speed 5 m / min, welding speed 80 cm / min, and gap 0 mm. The CTWD was fixed at 15 mm and the working angle was 45˚. The 90% Ar + 10% CO ₂ gas mixture was used, and the fatigue life according to the load was shown in cycles.

As shown in FIG. 11, the load is shown in a graph inversely proportional to the fatigue life, and the fatigue strength graph shown in the experiment and the fatigue strength graph shown in the prediction are almost identical.

In the graph of FIG. 10, the experimental fatigue life and the predicted fatigue life were compared in a table at 31MPa, 47MPa, 62MPa, 92MPa, 122MPa, 183MPa, 244MPa, 305MPa, and 366MPa.

In terms of the logarithm of each load, the experimental fatigue life and the predicted fatigue life error was 2.6% or less at 2.6%, 1.8%, 2.6% and 2.4%, respectively.

The error mean of the real value is 13.3% and the error mean of the common logarithm is 1.3%.

As a result of experiment by changing welding condition, the average of error of experimental fatigue life and predicted fatigue life is shown in [Table 1].

The material used for the experiment was SGAFH 590FB 2.3 mm material, welding power was CMT, wire feeding speed was 5, 7, 9 m / min, gap 0, 0.1, 0.2, 0.5, welding speed was 60cm / min. CTWD 15mm, welding working angle was fixed at 45˚, 90% Ar + 10% CO ₂ mixed gas was used.

CMT (m / min) Gap (mm) Real
% Error Logarithm error mean (%) 5 m / min 0.1 mm 6.4 0.7 5 m / min 0.2 mm 4.4 0.4 5 m / min 0.5 mm 4.7 0.4 5 m / min 1.0 mm 1.7 0.1 7 m / min 0.1 mm 5.7 0.5 7 m / min 0.2 mm 4.8 0.5 7 m / min 0.5 mm 7.8 0.8 7 m / min 1.0 mm 2.4 0.2 9 m / min 0 mm 8.2 0.7 9 m / min 0.1 mm 3.7 0.3 9 m / min 0.2 mm 2.1 0.1 9 m / min 0.5 mm 2.5 0.2 9 m / min 1.0 mm 1.7 0.2

As shown in the experimental results, as the gap increases, the mean of the real number error is significantly lower, and the mean of the common logarithmic error is very low, at a minimum of 0.2%.

As described above, according to the welding strength method and apparatus according to the present invention, a weld shape can be calculated by measuring a weld surface using an artificial neural network as a method of predicting strength, and using various parameters of the calculated weld shape, Whenever the material to be welded changes, it is not necessary to repeat the results through multiple experiments and to set new process variables, and to have high prediction accuracy with minimal input data, thereby reducing process time and cost, and reducing costs and time. Contrast welding has the effect of maximizing the quality. The scope of the present invention is not limited by these effects.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

1: top plate
2: lower plate
3: vision
10: measuring unit
20: shape derivation unit
30: parameter calculation unit
40: input value acquisition unit
50: strength prediction unit
51: input unit
52: calculator
53: output unit
60 display unit
A1: reinforcement area
A2: neck area
A3: penetration area
L1: reinforcement added
L2: Weld Bead Thickness
L3: Actual Neck
L4: The square of the top plate
L5: lower square
L6: penetration depth
θ1: Toe part angle
θ2: welding bead angle
θ3: Angular angle

Claims (6)

A measurement step of measuring a welded surface of the welded base material through the vision sensor;
A shape derivation step of deriving a weld shape using the measured weld surface;
A parameter calculating step of calculating a parameter of the welded portion from the welded shape;
An input value obtaining step of obtaining the physical property of the base material as an input value; And
An intensity prediction step of inputting the parameter and the input value into an artificial neural network and predicting an intensity corresponding to the parameter and the input value;
Including,
The shape derivation step,
Weld surface data inputting the data of the weld surface measured by the vision sensor into the artificial neural network;
Weld shape data output step of comparing the data of the weld surface with the data learned in the artificial neural network to find the data having a similarity or more than a predetermined value to derive the weld shape between the upper plate and the lower plate of the base material to output the weld shape;
Including,
The parameter calculating step,
Calculating the numerical value of the welded portion which is the inside of the welded surface where the upper and lower plates of the base metal melted and joined in the welded shape data output step are joined;
The parameter is
Reinforcement addition of the molten part, welding bead thickness, actual neck thickness, length, penetration depth, toe angle, welding bead angle, angle of angle, reinforcement addition area, neck area, penetration area And weld porosity at least one of the weld strength prediction methods. The method of claim 1,
The intensity prediction step,
An input step of inputting the parameter and the input value into an artificial neural network;
Computing step of calculating the tensile strength and the fatigue strength of the welded portion through the first and second hidden layer formed of ten nodes, respectively:
An output step of outputting the tensile strength and the fatigue strength of the welded part;
Including, welding strength prediction method. delete The method of claim 1,
The input value is
Weld strength prediction method comprising any one or more of the strength, thickness, plating, texture, top plate undercut, bottom plate undercut and the top plate and the bottom plate of the base material and the bottom plate. A measuring unit measuring a welded surface of the welded base metal by using a vision sensor;
A shape derivation unit for deriving a weld shape using the measured weld surface;
A parameter calculator configured to calculate a parameter of the welded part from the welded shape;
An input value obtaining unit obtaining the physical property of the base material as an input value; And
An intensity estimator for inputting the parameter and the input value to an artificial neural network and predicting an intensity corresponding to the parameter and the input value;
Including,
The shape derivation unit,
A weld surface data input unit configured to input data of the weld surface measured by the vision sensor to the artificial neural network;
A welder shape data output unit for comparing the data of the weld surface with the data learned in the artificial neural network to find data having a similarity or more than a predetermined value, and to derive and output the shape of the welder;
Including,
The parameter calculator,
The upper and lower plates of the base material derived from the weld shape data output unit are melted to calculate a numerical value of the welded joints,
The parameter is
Reinforcement addition of the molten part, welding bead thickness, actual neck thickness, length, penetration depth, toe angle, welding bead angle, angle of angle, reinforcement addition area, neck area, penetration area And at least one of weld porosity. The method of claim 5,
The strength prediction unit,
An input unit configured to input the parameter and the input value to an artificial neural network;
Computation unit for calculating the tensile strength and fatigue strength of the welded portion through the artificial neural network having a first hidden layer and a second hidden layer formed of ten nodes, respectively:
An output unit for outputting tensile strength and fatigue strength of the welded part;
To include, welding strength prediction apparatus.

Images