JP3200102B2 - Welding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001] The present invention relates to a welding how, particularly, suitable for use in automatic welding which is performed under the new setting conditions,
Soluble guessing welding condition which is not stored in the database
It concerns the method of making contact .

[0002]

2. Description of the Related Art Conventionally, data of a database system is a set of data obtained by actually performing tests and experiments, and those having functions capable of storing, modifying, searching, etc., are generally used in a general database. It is called a system.

However, in the database system using the test results as it is, the man-hour for data collection and the database itself become very large. For example, when application to a welding robot is considered, since the degree of freedom of conditions set before welding is infinite, it is impossible to test all of them and collect the results in a database. In addition, if the database cannot be used for cases other than the case where the test was performed, welding cannot be performed under the new setting conditions, and
It is not enough from the viewpoint of effective use of the database if the database created cannot be used.

[0004] In recent years, in order to digitize expertise and clarify relationships in the field of welding,
"Technique for presenting data to workers by guessing non-existent data" is used, and fuzzy methods, expert systems, calculation of data by calculation formulas, and the like have been proposed.

[0005]

However, in the fuzzy method, the expert system, and the method of calculating data by using the above-mentioned formulas, the above-mentioned methods of digitizing know-how of the welding technique, clarifying the mutual relation of data, determining the formulas, etc. The disadvantage is that specialized and very laborious work is required. In addition, since the numerically determined values are determined with restrictions, they can be applied only under limited setting conditions.

An object of the present invention is to pay attention to such a conventional technique, and it is not stored in a database without spending time for clarifying mutual relations of data and quantifying know-how. It is intended to provide a method of estimating welding conditions and performing welding under these welding conditions.

[0007]

Means for Solving the Problems Conventional problems as described above
In view of the above, the present invention is to determine the optimal heat input and welding speed in advance.
Arc length, welding speed and
Enter the arc length, welding speed and
Neural network that satisfies the relationship between heat and optimal heat input
The work is constructed by learning, and the first new
Using a neural network to determine the maximum
Determine the appropriate heat input and welding speed, and determine the optimum heat input and
Power is determined based on the welding speed and welding speed.
When determining the arc voltage and arc current
The arc current corresponds to multiple plate thicknesses with known arc currents.
Input the values of arc current, arc voltage and power
A second kernel that satisfies the relationship between current, arc voltage and power
The neural network is constructed by learning, and this second
Input the obtained power to a neural network and
An arc current corresponding to the obtained power is obtained, and the obtained arc current is obtained.
Arc voltage in the relationship between the
The welding speed, arc current and arc
This is a welding method in which welding is performed with a welding voltage using welding voltage.

[0008]

[0009]

[0010]

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In the present invention, a neural network is used to estimate data not stored in a database. Therefore, a general neural network will be briefly described first.

A neural network is a function of a neurocomputer, and refers to a processing method for solving various problems using the neurocomputer. Here, the neurocomputer aims at engineeringly realizing the mechanism of the cranial nerves of a living thing, and is composed of many neurons corresponding to the cranial nerves of a living thing. A complex network formed by connecting a large number of these neurons is a neural network. In this neural network, the strength of the connection between these neurons (meaning input data, output data and intermediate processing data) and the presence or absence of the connection change plastically under the influence of the signal itself. A neural network applicable to a signal is sequentially formed by the change, and a learning function is realized.

Therefore, the neurocomputer used here solves a problem not by processing according to a program but by pattern processing based on a given neural network. Therefore, a programming language is used to perform predetermined processing. This is very different from a Neumann-type computer (a so-called general computer) in which, when a program is created and input data is input, a calculation result processed according to the program is output as an answer.

Since the automatically set internal structure has a mechanism for realizing the learning function, the neural network after learning not only presents an output value to the learned input, but also performs learning. Even when a point that has not been input is input, an output value that satisfies the relationship between the data required at the time of learning and the like can be estimated and presented.

Hereinafter, a preferred embodiment of the present invention using this neural network will be described with reference to the drawings.

FIG. 1 shows a case in which the heat input is used as a parameter, welding conditions that are not in the database are estimated using a neural network, and the welding conditions are applied to a TIG welding robot.
This will be described with reference to FIG.

First, in order to be able to output welding conditions not only for limited settings but also for all settings, all input setting conditions are digitized in advance and can be handled by a neural network.

Here, the heat input Q (J / cm) is the amount of heat applied from the outside to the welded portion, and means the electric energy generated per unit length (1 cm) of welding in arc welding. This heat input Q depends on the arc voltage E
(V), arc current I (A), welding speed V (cm / mi)
n) is given by the following equation. Q = 60 * E * I / V [J / cm] (1) Therefore, from the above equation, when the heat input Q is constant, it can be said that the welding speed V and the arc current I are in a proportional relationship (see FIG. 1). ). In addition, the relationship between the plate thickness T and the heat input Q generally indicates that as the plate thickness T increases, the optimal heat input Q also increases (see FIG. 2).

In order to estimate the welding conditions using the heat input Q, first, the heat input Q of the welding condition to be learned is determined from the known arc current I, welding speed V, and arc voltage E from the equation (1). Keep it. Next, the arc length, the welding speed V, the optimum value of the heat input Q, and the like are input for the four types of plate thicknesses for which the heat input Q is known by performing the above calculation, and the neural network shown in FIG. Use to learn. That is, a neural network that satisfies the relationship between the arc length, the welding speed V, and the optimal heat input Q is constructed by learning, and the optimal heat input Q for a plate thickness for which there is no data using the learned neural network. (See FIG. 4), welding speed V (see FIG. 5), and the like.

In this way, an optimum heat input Q for a new set condition is obtained, but an arc current I is necessary for performing an actual welding operation. However, since the welding speed V has already been learned and recalled, the arc current I can be determined. That is, since the obtained optimal heat input Q can be realized by various combinations of the arc current I and the welding speed V, the welding speed V is fixed to a value reminiscent of the welding speed V, and the arc current I is one. Only to be found. Therefore, (1)
From the equation, the following equation is obtained.

E * I = Q * V / 60 [W] (2) P = E * I [W] (3) P is electric power and is obtained from the above equations (2) and (3). Can be In this case, since the electric power P is obtained but the arc voltage E and the arc current I are not obtained, both the arc voltage E and the arc current I must be obtained from one electric power P. Since this solution is innumerable, this problem is again solved using a neural network (see FIG. 6). That is, before
A neural network that satisfies the relationship between the arc current I, the arc voltage E and the power P is learned by inputting a point where the relationship between the arc current I, the arc voltage E and the power P is known for the four types of plate thicknesses. Complete the network. An arc current I (see FIG. 7) is obtained by inputting a power value P to be decomposed into an arc voltage E and an arc current I into the neural network and recalling the same. Further, since the arc current I is obtained, the arc voltage E is easily obtained from the equation (3).

In this manner, the heat input Q and the welding speed V are determined by the neural network of FIG. 3, and the arc voltage E and the arc current I for realizing the heat input Q are obtained by the neural network of FIG. The welding can be performed based on these welding conditions.

In the welding performed under the welding conditions obtained by the above estimation, the appearance rate of the back bead is as shown in FIG. 8 by using the parameter of heat input Q. Although this is not shown here, the appearance rate of the back bead is partially reduced compared to the case where the arc current I and the welding speed V are separately treated first, and the arc current I and the welding speed V are reduced. This indicates that the adverse effects of independent handling are avoided and reliability is improved.

Next, another embodiment will be described with reference to FIGS. Here, when recalling welding conditions using a neural network, the problem of how to distinguish jigs that exist for each product is dealt with. The result of the simulation of the neural network shown in FIG. 9 is shown in FIG. The horizontal axis is the plate thickness T of the workpiece, the vertical axis is the optimal heat input Q for welding the plate thickness, and the jig A in the figure is Y = 6 * X * X, that is, Q = 6 * T *. Assuming that welding conditions exist on the curve of T, the jig B is similarly Q = 2 *
It is assumed that welding conditions exist on the curve of T * T. This assumption is made for the sake of convenience in order to make it easier to compare the results of the jig C to be recalled, and it goes without saying that actual welding data is used for the jigs A and B. .

Here, the data known in advance include eight points obtained by welding plate thicknesses of 2, 4, 6, and 8 mm with jig A and jig B, and a result obtained by welding plate thickness of 2 mm with jig C. Is a total of 9 points. Here, it is assumed that the jig C is Q = 4 * T * T.

First, the data of each of the four points of the jigs A and B are once learned and guessed by a neural network to obtain eight points of each (a total of 16 points). A total of 17 data points are learned by the neural network. The result of recalling the heat input Q of the jig C having a plate thickness of 3 to 9 mm is a value close to a curve (Q = 4 * T * T) assumed as shown in FIG. 10 (black circles in FIG. 10). showed that. In other words, this means that if there is welding data of one type of plate thickness when the jig is changed, welding conditions for other plate thicknesses can be estimated.

In this embodiment, when a certain jig is used, the optimum heat input Q for a certain plate thickness is obtained. However, when welding is actually performed, the optimum heat input Q is obtained as described in the previous embodiment. It is necessary to obtain the arc current I and the welding speed V from the heat input Q obtained.

Although this embodiment is directed to a jig, the same can be applied to a change in material and joint shape.

As described above, since the neural network is used as a tool in order to estimate welding conditions for which no data is accumulated in the TIG welding database system, what items affect the welding results and to what extent. The welding conditions can be estimated without spending time clarifying the interrelationship of data such as whether or not and quantifying know-how. Therefore, knowledge of a skilled person is not required for estimating welding conditions.

Further, since the heat input Q is used as a parameter without directly using the welding speed V and the arc current I, the reliability of the estimated value of the welding condition can be improved.

Further, in addition to general welding conditions such as welding speed V, arc current I and arc voltage E, a neural network can be used even when other welding conditions such as a jig type and a joint shape are added. By repeating learning and guessing, it is possible to process even cases that are not stored in the database.

[0032]

The present invention can be understood from the above description of the embodiments.
As described above, in the present invention, the arc length, the welding speed V, and the
Neural network that satisfies the relationship between
And the arc currents I and A
Second neural network that satisfies the relationship between the peak voltage E and the power P
The first network is constructed by learning the first network.
Ral network allows proper heat input Q and thickness
And welding speed V, and the obtained heat input Q and welding speed
The power P is obtained based on the degree V, and the power P
The arc current I is calculated by the second neural network.
The obtained arc current I and the obtained power P
Arc voltage E in relation to
Welding speed V, arc current I and arc voltage E
Since welding is performed as
It is easy to find welding conditions for plate thickness
Yes, welding can always be performed under appropriate welding conditions
It is.

[0033]

[Brief description of the drawings]

FIG. 1 is a graph showing a relationship between an arc current and a welding speed when heat input is constant.

FIG. 2 is a graph showing a relationship between a plate thickness and an optimal heat input.

FIG. 3 is an explanatory diagram showing the structure of a neural network that learns heat input, welding speed, and arc length for various plate thicknesses by learning heat input, welding speed, and arc length.

FIG. 4 is a graph showing heat input values recalled for various plate thicknesses using a neural network.

FIG. 5 is a graph showing welding speeds recalled for various plate thicknesses using a neural network.

FIG. 6 is an explanatory diagram showing a structure of a neural network for obtaining an arc current and an arc voltage from electric power.

FIG. 7 is a graph showing arc currents recalled for various thicknesses using a neural network.

FIG. 8 is a graph showing the appearance rate of back beads in welding performed based on welding conditions obtained by a neural network using heat input as a parameter.

FIG. 9 is an explanatory diagram showing a structure of a neural network for recalling heat input to a jig.

10 is a graph showing the relationship between the plate thickness and the heat input of the jig recalled by the neural network shown in FIG.

[Explanation of symbols]

 Q Heat input T Plate thickness (Welding conditions) V Welding speed (Welding conditions) I Arc current (Welding conditions) E Arc voltage (Welding conditions) P Electric power

Continuation of front page (56) References JP-A-2-70384 (JP, A) JP-A-3-135497 (JP, A) JP-A-62-290685 (JP, A) JP-A-4-55060 (JP) JP-A-63-252671 (JP, A) JP-A-4-143075 (JP, A) JP-A-4-77219 (JP, A) JP-A-2-153476 (JP, A) (58) Field surveyed (Int.Cl. ⁷ , DB name) B23K 9/095 G05B 13/02

Claims (1)

(57) [Claims] 1. Optimal heat input (Q) and welding speed in advance
(V) corresponds to a plurality of plate thicknesses (T) for which
Enter the values for the work length, welding speed (V) and optimal heat input (Q).
Forced arc length, welding speed (V) and optimal heat input
First neural network satisfying the relationship of (Q)
Is constructed by learning, and the learned first neural network is constructed.
Optimum input for sheet thickness without data using network
The heat (Q) and the welding speed (V) are obtained, and the obtained optimum
Calculate electric power (P) based on heat input (Q) and welding speed (V)
Therefore, the arc voltage (E) suitable for the obtained power (P)
When determining the arc current (I), an appropriate arc current
For multiple sheet thicknesses (T) whose flow (I) is known
Arc current (I), arc voltage (E) and power (P)
Enter the values and enter the arc current (I) and arc voltage (E)
Neural network that satisfies the relationship between power and power (P)
The second neural network is constructed by learning the network.
Input the calculated power (P) into the network
Arc current (I) corresponding to the electric power (P)
The relationship between the determined arc current (I) and the determined power (P)
The arc voltage (E) was determined in the section
Contact speed (V), arc current (I) and arc voltage (E)
The welding method characterized by performing welding with welding conditions
Law.