JPH05177353A - Arc sensor for automatic welding - Google Patents

Abstract

PURPOSE:To obtain a correction value at the coordinate system of welding control in accordance with the potional deviation by detecting the deviation of the target position from an arc current waveform when any welding condition is applied without preparing the programming of arithmetic processing at every welding condition. CONSTITUTION:The arc sensor for automatic welding is provided with a current waveform storage device 5 for storing the waveform of an arc current of one period of weaving, a neural network system 6 for detecting the deviation of the target position from inference operation based on the data of the stored waveform of the storage device 5 and a fuzzy inference device 7 for calculating the correction value of the deviation of the target position at the coordinate system of welding control from the execution of an inference rule based on the detected result of the neural network system 6 and the welding conditions such as the welding direction and the welding posture.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an arc sensor for automatic welding, which detects a deviation of the intended position of automatic welding.

[0002]

2. Description of the Related Art Conventionally, when performing fillet welding of members (joints) 1 and 2 as welding works as shown in FIG. 2 by a welding robot, for example, the welding line 1 is controlled based on the control of the welding machine and the arm of the robot. The welding torch 3 moves due to the movement in the direction and the weaving movement of the solid line w based on the weaving pattern, and the core wire 4 is projected from the torch 3 at a constant speed, and welding proceeds along the welding line l.

At this time, in order to maintain the optimum welding state, it is necessary to control the welding aiming position so that the center of the weaving is along the welding line 1 and the length of the core wire 4 is appropriate. An arc sensor for automatic welding is used to control the aiming position, and this sensor detects the deviation of the aiming position from the arc current flowing between the core wire 4 and the welding work during welding.

When the detection result is obtained in the coordinate system (X, Y, Z) of the welding work of FIG. 2, the detection result is obtained in the coordinate system of the welding robot, that is, the coordinate system of the welding control shown in FIG. x, y, z), and the arm position of the welding robot is corrected according to the result after the conversion.

[0005]

In the case of the conventional automatic arc sensor for welding, the deviation of the aimed position from the specific numerical calculation processing based on the detected value of the arc current is used for the welding control coordinate system (x, y, z). Therefore, a huge amount of calculation processing program corresponding to each welding condition is required to detect the correction value and the correction value is not detected correctly even if the welding condition changes even a little. The present invention provides a highly versatile arc sensor for automatic welding, which can accurately detect the deviation of the aiming position under various welding conditions in a coordinate system for welding control without preparing enormous programming of arithmetic processing. With the goal.

[0006]

In order to achieve the above object, in the arc sensor for automatic welding of the present invention, a current waveform storage device for storing the waveform of the arc current of one cycle of weaving, and this storage device. Neural network device that detects the deviation of the target position from the inference operation based on the data of the stored waveform of the robot, and the execution of the inference rule based on the detection result of this device and the welding conditions such as the welding direction and the welding posture, and the coordinate system for the welding control. And a fuzzy inference device that calculates a correction value for the deviation of the target position.

[0007]

In the case of the arc sensor for automatic welding of the present invention configured as described above, the neural network device infers and detects the deviation of the aimed position based on the arc current waveform data of one cycle of the current waveform storage device. To do. Further, based on the detection result and the welding condition, the fuzzy inference device infers and obtains the correction value of the deviation of the aimed position in the welding control coordinate system.

Then, in order to obtain the correction value in the welding control system by detecting the deviation of the aimed position from the inference operation of the neural network device and the execution of the inference rule of the fuzzy inference device, the calculation process for each welding condition of the conventional sensor is performed. No need for enormous programming, etc. Moreover, based on the learning in advance of the neural network device, the deviation of the target position can be accurately detected under any welding condition, and in addition, the learning in advance can be performed by fuzzy reasoning. Since the correction value according to the deviation detection result is obtained without using the coordinate system for welding control, a highly versatile arc sensor with high versatility can be provided.

[0009]

EXAMPLE One example will be described with reference to FIG. In FIG. 1, 5 is a rewritable current waveform storage device composed of a RAM or the like, 6 is a neural network device of, for example, a three-layer feedforward configuration, and 7 is a fuzzy inference device. When applied to the fillet welding of FIG. 2, the arc current between the core wire 4 and the welding work is constantly detected by a welding machine (not shown), and the arc current detection signal Si of this welding machine is stored in the storage device 5. Supplied. Further, as the weaving information, for example, timing information Pw at the start point and the end point of each cycle of the weaving is supplied to the storage device 5.

Then, the storage device 5 stores the timing information Pw.
The detection signal Si of each cycle is taken in according to
The latest waveform data of the average waveform I of the detection signal Si of three cycles including is stored rewritably as the data of the arc current waveform of one cycle. Further, the waveform stored in the storage device 5, that is, the data of the latest average waveform I is read out to the network device 6, and by this device 6, the deviation of the weaving center toward the member 1 or 2 and the tendency of the length of the core wire 4 to become short. Is inferred as a deviation of the intended position of welding, and is obtained in the coordinate system (X, Y, Z) of the welding work.

By the way, if the center of the weaving coincides with the welding line 1 and is precisely positioned at the corner of the members 1 and 2, the center crosses the welding line 1 at 1/2 cycle, and the waveform I Has a symmetric waveform that has a minimum value at 1/2 cycle, but if the center shifts toward the member 1 or 2, the timing at which the center crosses the welding line 1 is delayed or advanced from the 1/2 cycle, and the waveform I Becomes asymmetric with respect to the point of 1/2 cycle.

The arc current is the welding torch 3 of the core wire 4.
If the average value of the waveform I when the length of the core wire 4 is appropriate is a standard value and the average length of the core wire 4 changes as the standard value, the average of the waveform I is a standard value. It changes to smaller and larger. Therefore, the deviation of the weaving center toward the member 1 or 2 can be inferred from the symmetry deviation of the waveform I, and the length and shortness of the core wire 4 can be inferred from the average of the waveform I.

Then, in order to be able to infer the deviation of the aimed position with desired accuracy under various welding conditions, the network device 6 has in advance a typical welding condition for each typical welding condition, such as by back propagation learning. Correct inference results when the weaving center and the length of the core wire 4 are appropriate and improper are learned, and the deviations of the weaving center members 1 and 2 are detected to detect the center deviation determination signals S1 and S2. It is set to output each of them, detect whether the core wire 4 is too long or too short, and output each of the identification signals S3 and S4 of the core wire length failure.

Based on this learning and setting, the network device 6 after learning automatically presents the current inference operation based on the waveform I without giving programming of individual welding conditions or numerical operation processing which matches the conditions. The tendency of deviation of the target position is accurately detected in the coordinate system (X, Y, Z) of the welding work, and the discrimination signals S1, S2, S3 corresponding to the detection result are detected.
Alternatively, S4 is output. Then, the respective discrimination signals S1 to S4 are supplied to the inference device 7 as detection signals of the welding state. In addition, as welding condition information, the current welding direction, welding position,
The groove shape data Da, Db, Dc are supplied to the inference device 7.

By the way, the inference device 7 uses, as an inference rule, an "if-then" format for obtaining a correction value in the coordinate system (x, y, z) of the welding control for the detection signal of the welding state under typical welding conditions. Rules are described in advance. Then, based on the discrimination signals S1 to S4 and the data Da to Dc given from the network device 6, the inference device 7 executes an inference rule and detects without applying programming of coordinate conversion that matches individual welding conditions. The correction values Δx, Δy, Δz in the coordinate system (x, y, z) corresponding to the deviation of the aimed position are obtained.

Then, the arm position of the welding robot is adjusted by a unit amount in the x-, y-, and z-axis directions by the correction values Δx, Δy, Δz, and the aimed position is corrected and maintained at an appropriate position by repeating this adjustment. In the above embodiment, in order to prevent transient fluctuations, the data of the arc current waveform for one cycle was obtained from the average of three cycles of the detection signal Si, but N (=
Needless to say, it may be obtained from the average of (integer of 1, 2, 3, ...) Period. Further, the configurations and the like of the respective devices 5 to 7 are not limited to the embodiments. And, of course, it can be applied to various automatic welding.

[0017]

Since the present invention is configured as described above, it has the following effects. The neural network device 6 infers and detects the deviation of the aim position based on the data of the arc current waveform of one cycle stored in the current waveform storage device 5, and based on the detection result and the welding condition, the fuzzy inference device 7 Since the correction value of the deviation of the target position in the coordinate system for welding control is inferred and obtained, prior learning of the neural network device 6 without preparing enormous programming of arithmetic processing for each welding condition of the conventional sensor. Based on the above, the target position deviation is accurately detected under any welding condition, and the correction value according to the deviation detection result is obtained in the welding control coordinate system without performing prior learning by fuzzy reasoning. It is possible to provide a high-performance arc sensor with high versatility.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of an arc sensor for automatic welding according to the present invention.

FIG. 2 is a welding explanatory view of an example of automatic welding.

[Explanation of symbols]

 5 Current waveform storage device 6 Neural network device 7 Fuzzy inference device

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Office reference number FI technical display location G05B 19/403 M 9064-3H V 9064-3H

Claims (1)

[Claims] 1. An automatic welding arc sensor for detecting a deviation of a welding aim position from an arc current when welding along a welding line by moving in the direction of the welding line and weaving, and the arc of one cycle of weaving. A current waveform storage device that stores a waveform of a current, a neural network device that detects the deviation of the target position from an inference operation based on the data of the storage waveform of the storage device, a detection result of the network device, a welding direction, and a welding posture An arc sensor for automatic welding, comprising: a fuzzy inference device that calculates a correction value for the deviation of the target position in the coordinate system for welding control from execution of an inference rule based on welding conditions such as.