JP7507480B2 - Welding equipment and welding method - Google Patents

Description

The present invention relates to a welding device and a welding method.

For example, as described in Non-Patent Document 1 (Narita Kuniro, "Post-welding treatment and inspection," Journal of the Japan Welding Society, Vol. 50, No. 9, p. 875, 1981), after welding, non-destructive inspection of the weld is performed and defective areas with cracks or other defects are repaired (rewelded). In particular, in large structures, hot cracks are likely to occur in the welds, making rewelding of defective areas essential. Hot cracks are cracks that occur in the welds during welding or during the cooling process after welding.

Kunio Narita, "Post-welding treatment and inspection", Journal of the Japan Welding Society, Vol. 50, No. 9, p. 875, 1981

In order to avoid rewelding of defective areas such as those described above as much as possible, it is necessary to suppress the occurrence of hot cracking during welding. However, there are a wide variety of welding conditions related to the occurrence of hot cracking, and optimizing these conditions is difficult, so the occurrence of hot cracking during welding has not been sufficiently suppressed.

The present invention has been made in consideration of the problems of the conventional technology as described above. More specifically, the present invention provides a welding device and a welding method that can suppress the occurrence of hot cracks.

The welding device of the present invention welds a first member and a second member in a state in which a first end face, which is an end face of the first member in a first direction, and a second end face, which is an end face of the second member in the first direction, face each other. The welding device includes a welding torch that moves on a welded portion where the first member and the second member are welded along a second direction intersecting the first direction, and a sensor that measures the width of a groove in the first direction formed by the first end face and the second end face while moving along the second direction together with the welding torch. The welding device is configured to perform welding according to a first welding condition when the displacement of the groove width is less than a predetermined threshold value, and to perform welding according to a second welding condition different from the first welding condition when the displacement of the groove width is equal to or greater than the threshold value. The displacement of the groove width under the second welding condition is smaller than the displacement of the groove width under the first welding condition.

In the above welding device, the sensor may measure the displacement of the groove width in front of the welding torch.

In the above welding device, the movement speed of the welding torch under the second welding condition may be smaller than the movement speed of the welding torch under the first welding condition.

In the above welding device, the output of the welding torch under the second welding condition may be smaller than the output of the welding torch under the first welding condition.

In the above welding device, the welding torch may weave along the first direction or the second direction under the second welding condition.

The above welding device may further include a storage device and a processor. The storage device may store a trained model. Reinforcement learning may be performed on the trained model. In the reinforcement learning, a first reward may be given when one of a plurality of learning welding conditions is selected. The first reward may increase as a hot crack index, which is an index of the susceptibility of hot cracking in the weld, decreases. In the reinforcement learning, a value function may be updated based on the first reward. The processor may determine the first welding condition based on the value function.

In the above welding device, the hot cracking index when each of the multiple learning welding conditions is selected may be calculated based on finite element analysis.

In the above welding device, the hot cracking index may be at least one of the tensile plastic strain in the solidification brittle temperature range of the weld, the temperature gradient vector in the solidification brittle temperature range, and the displacement of the groove width.

In the above welding device, a second reward may be further provided when one of the multiple learning welding conditions is selected in reinforcement learning. The second reward may increase as the average movement speed of the welding torch increases. In reinforcement learning, the value function may be updated based on the first reward and the second reward.

In the above welding apparatus, the trained model may be subjected to Q-learning.
The above welding device may further include a storage device and a processor. The storage device may store a trained model. When welding conditions are input, the trained model may output a hot crack index that is an index of the likelihood of hot cracking occurring in a welded portion when the welding conditions are applied. The trained model may be subjected to supervised learning using teacher data including the learning welding conditions and the hot crack index when the learning welding conditions are applied. The processor may determine the first welding conditions based on the hot crack index output from the trained model.

The above welding device may further include a storage device and a processor. The storage device may store an evaluation function that outputs an evaluation value of the welding conditions when the welding conditions are input. The evaluation value may include a hot crack index that is an index of the likelihood of hot cracking occurring in the weld when the welding conditions are applied. The processor may determine the first welding conditions by applying a genetic algorithm based on the evaluation value.

The welding method of the present invention is a method for welding a first member and a second member in a state in which a first end face, which is an end face of the first member in a first direction, and a second end face, which is an end face of the second member in the first direction, face each other. The welding method includes a step of welding the first member and the second member by moving a welding torch along a second direction intersecting the first direction, and a step of measuring the width in the first direction of a groove formed by the first end face and the second end face with a sensor moving along the second direction together with the welding torch. The welding of the first member and the second member is performed according to a first welding condition when the width displacement is less than a predetermined threshold value, and according to a second welding condition different from the first welding condition when the width displacement is equal to or greater than the threshold value. The width displacement under the second welding condition is smaller than the width displacement under the first welding condition.

The welding device of the present invention can suppress the occurrence of hot cracks.

FIG. 1 is a block diagram of a welding device 100. FIG. 2 is a plan view of a first member 110 and a second member 120 in abutted state against each other. 3 is a cross-sectional view taken along line III-III in FIG. 2. 1 is a schematic high-temperature ductility curve of a solidification brittle temperature region between a liquidus temperature and a solidus temperature. 1 is a schematic graph showing the relationship between tensile plastic strain and the amount of displacement of width W in the solidification brittle temperature region of a weld.

Details of the embodiment will be described below with reference to the drawings. Here, the same or corresponding parts will be given the same reference symbols, and redundant explanations will not be repeated.

First Embodiment
A welding device according to a first embodiment (hereinafter referred to as "welding device 100") will be described below.

<Configuration of welding device 100>
Fig. 1 is a block diagram of a welding device 100. As shown in Fig. 1, the welding device 100 includes a welding torch 10, a drive mechanism 20, a power supply device 30, a sensor 40, and a controller 50.

The welding device 100 welds a first member 110 and a second member 120. FIG. 2 is a plan view of the first member 110 and the second member 120 in a butted state. FIG. 3 is a cross-sectional view taken along line III-III in FIG. 2. As shown in FIGS. 2 and 3, the first member 110 and the second member 120 are disposed adjacent to each other in the first direction DR1.

The first member 110 has an end surface 110a and an end surface 110b. The end surface 110a and the end surface 110b are the end surfaces of the first member 110 in the first direction DR1. The second member 120 has an end surface 120a and an end surface 120b. The end surface 120a and the end surface 120b are the end surfaces of the second member 120 in the first direction DR1. The first member 110 and the second member 120 are arranged such that the end surface 110a and the end surface 120b face each other in the first direction DR1.

The first member 110 has a first main surface 110c and a second main surface 110d. The first main surface 110c and the second main surface 110d are end surfaces of the first member 110 in the thickness direction. The second member 120 has a first main surface 120c and a second main surface 120d. The first main surface 120c and the second main surface 120d are end surfaces of the second member 120 in the thickness direction.

The end faces 110a and 120a form a groove. In the example shown in Figures 2 and 3, the groove has a V-shaped groove shape. That is, the end face 110a is inclined so that the distance from the end face 110b increases from the first main surface 110c side toward the second main surface 110d side, and the end face 120a is inclined so that the distance from the end face 120b increases from the first main surface 120c side toward the second main surface 120d side.

However, the shape of the groove is not limited to this. The groove formed by the end face 110a and the end face 120a may be an I-shaped groove. In other words, the end face 110a and the end face 120a do not have to be inclined.

The width of the groove formed by end face 110a and end face 120a in the first direction DR1 is defined as width W. Width W is the maximum distance between end face 110a and end face 120a in the first direction DR1. In the example shown in Figures 2 and 3, width W is the distance between the end of end face 110a on the first main surface 110c side and the end of end face 120a on the first main surface 120c side.

The first member 110 and the second member 120 are, for example, flat plate-shaped members. The first member 110 and the second member 120 are, for example, made of steel. The first member 110 and the second member 120 are, for example, rolled steel plate for welded structures (SM material) as specified in the JIS standard (JIS G 3106:2008).

The first member 110 and the second member 120 are formed, for example, from the same material. The first member 110 and the second member 120 may be formed from different materials. The first member 110 and the second member 120 do not have to be made of steel. The first member 110 and the second member 120 may be made of a non-ferrous alloy, for example, an aluminum (Al) alloy.

The welding torch 10 is, for example, a welding torch for arc welding. However, the welding torch 10 is not limited to this. The welding torch 10 may be a welding torch for laser welding. The welding torch 10 is moved along the second direction DR2 by the drive mechanism 20. More specifically, the welding torch 10 moves from one side in the second direction DR2 to the other side in the second direction DR2 on the portion (weld portion) where the first member 110 and the second member 120 are welded. The welding torch 10 welds the first member 110 and the second member 120 from the first main surface 110c and the first main surface 120c side. The second direction DR2 is a direction that intersects (preferably perpendicular to) the first direction DR1.

Preferably, the drive mechanism 20 can oscillate the welding torch 10 along at least one of the first direction DR1 and the second direction DR2. From another perspective, the drive mechanism 20 can weave the welding torch 10.

When a current is supplied from the power supply unit 30 to the welding torch 10, an arc is generated in the welding torch 10, and the first member 110 and the second member 120 are welded by the arc. More specifically, the first member 110 and the second member 120 are welded by forming a weld metal 130 between the first member 110 and the second member 120.

The weld metal 130 is formed by mixing the first member 110 and the second member 120 melted by the welding torch 10 (if welding material is used, the molten welding material is also mixed) and cooling. The output of the welding torch 10 is adjusted by changing the current supplied from the power supply unit 30.

The sensor 40 measures the displacement of the width W. The sensor 40 is, for example, a displacement sensor such as an optical displacement sensor. However, the sensor 40 is not limited to this. The sensor 40 is, for example, an imaging element, and the displacement of the width W may be calculated by image processing based on an image captured by the imaging element.

The sensor 40 is moved along the second direction DR2 together with the welding torch 10, for example, by the drive mechanism 20. Preferably, the sensor 40 measures the displacement of the width W forward of the welding torch 10 (on the other side of the welding torch 10 in the second direction DR2). However, the sensor 40 may also measure the displacement of the width W rearward of the welding torch 10 (on one side of the welding torch 10 in the second direction DR2).

The controller 50 is, for example, a personal computer. However, the controller 50 is not limited to this. The controller 50 transmits a control signal to the drive mechanism 20 via an input/output interface (not shown), and transmits a control signal to the power supply device 30 via the input/output interface. The controller 50 also receives an output signal indicating the displacement of the width W output from the sensor 40 via the input/output interface.

The controller 50 has a storage device 51 and a CPU (Central Processing Unit) 52. A predetermined threshold value is stored in the storage device 51. The CPU 52 compares the magnitude relationship between this threshold value and the displacement of the width W measured by the sensor 40.

Based on the result of this comparison, the controller 50 (CPU 52) changes the welding conditions of the welding device 100. More specifically, if the displacement of the width W is less than the above threshold, the welding device 100 performs welding according to the first welding conditions. On the other hand, if the displacement of the width W is equal to or greater than the above threshold, the welding device 100 performs welding according to the second welding conditions. The second welding conditions are welding conditions different from the first welding conditions. According to the second welding conditions, the displacement of the width W is smaller than that of the first welding conditions.

The first welding condition and the second welding condition, for example, differ in the movement speed of the welding torch 10. In this case, the movement speed of the welding torch 10 under the second welding condition is lower than the movement speed of the welding torch 10 under the first welding condition. The controller 50 changes the movement speed of the welding torch 10 by, for example, changing the control signal sent to the drive mechanism 20 when the displacement of the width W is less than the above threshold value and when the displacement of the width W is equal to or greater than the above threshold value.

The output of the welding torch 10 (i.e., the amount of heat input from the welding torch 10 to the welded portion) may be different between the first and second welding conditions. In this case, the output of the welding torch 10 under the second welding condition is smaller than the output of the welding torch 10 under the first welding condition. The controller 50 changes the output of the welding torch 10, for example, by changing the control signal sent to the power supply 30 when the displacement of the width W is less than the threshold value and when the displacement of the width W is equal to or greater than the threshold value.

The first welding condition and the second welding condition may differ in whether or not the welding torch 10 weaves. In this case, the welding torch 10 weaves under the second welding condition, whereas the welding torch 10 does not weave under the first welding condition. The controller 50 switches between whether or not the welding torch 10 weaves by, for example, changing the control signal sent to the drive mechanism 20 when the displacement of the width W is less than the threshold value and when the displacement of the width W is equal to or greater than the threshold value.

The difference between the first and second welding conditions may be something other than the difference in the operation of the welding torch 10. In short, the second welding conditions are not particularly limited as long as they are welding conditions determined so that the displacement of the width W is small.

2, the first member 110 and the second member 120 can be divided into n regions (n: a natural number equal to or greater than 2) along the second direction DR2. These regions are referred to as regions R ₁ to R _n . The first welding conditions when the welding torch 10 passes through region R _k (k: a natural number equal to or greater than 2 and equal to or less than n) may be different from the first welding conditions when the welding torch 10 passes through region R _l (l: a natural number equal to or greater than 2 and equal to or less than n, different from k). That is, the first welding conditions may change as the welding torch 10 moves along the second direction DR2.

Note that if the displacement of the width W again becomes less than the threshold value after switching from the first welding conditions to the second welding conditions, the welding conditions may be switched back from the second welding conditions to the first welding conditions.

<Effects of the welding device 100>
Heat input from the welding torch 10 to the welded portion forms a molten pool immediately below the welding torch 10. The molten pool is composed of the first member 110 and the second member 120 that are melted and mixed (if a welding material is used, the molten pool also contains the molten welding material).

A brittle temperature region (BTR) is formed adjacent to the rear end of the molten pool (one end in the second direction DR2). The brittle temperature region is a region in which the solid phase and the liquid phase coexist. In other words, the brittle temperature region is a region whose temperature exceeds the solidus temperature of the base material (first member 110 and second member 120) and is below the liquidus temperature of the base material.

Fig. 4 is a schematic high-temperature ductility curve of the solidification brittle temperature region between the liquidus temperature and the solidus temperature. In Fig. 4, the horizontal axis represents the temperature in the solidification brittle temperature region, and the vertical axis represents the tensile plastic strain applied in the solidification brittle temperature region. The high-temperature ductility curve (solid line) in Fig. 4 indicates the limit strain at which hot cracking occurs at a specified temperature. Furthermore, T _L and T _S in Fig. 4 indicate the liquidus temperature and solidus temperature, respectively.

As shown in Figure 4, in the solidification brittle temperature region, tensile plastic strain increases as the molten metal solidifies (i.e., as the temperature drops) (see the dotted line in Figure 4). If this dotted line passes through the region above the high-temperature ductility curve (see the hatched region in Figure 4), hot cracking will occur in the solidification brittle temperature region. In other words, tensile plastic strain in the solidification brittle temperature region is one factor that causes hot cracking in the weld.

However, it is difficult to measure the tensile strain in the solidification brittle temperature range of the weld during the process of welding the first member 110 and the second member 120.

Figure 5 is a schematic graph showing the relationship between tensile plastic strain in the solidification brittle temperature range of the weld and the displacement of width W. In Figure 5, the horizontal axis represents the tensile plastic strain in the solidification brittle temperature range of the weld, and the vertical axis represents the amount of displacement of width W. According to the findings of the inventors, there is a roughly linear relationship between the tensile plastic strain in the solidification brittle temperature range of the weld and the displacement of width W. In other words, if the displacement of width W is reduced, the tensile plastic strain in the solidification brittle temperature range of the weld is reduced, making the weld less susceptible to hot cracking.

In the welding device 100, when the displacement of the width W is equal to or greater than a predetermined threshold, the welding conditions are switched from the first welding conditions to the second welding conditions. By switching the welding conditions from the first welding conditions to the second welding conditions, the displacement of the width W becomes smaller.

In this way, in the welding device 100, the tensile plastic strain in the solidification brittle temperature region of the weld can be indirectly monitored by monitoring the displacement of the width W, and the welding conditions are adjusted based on the monitoring results to reduce the displacement of the width W, thereby suppressing the occurrence of high-temperature cracks in the weld.

Second Embodiment
The welding device 200 will be described below. Here, differences from the welding device 100 will be mainly described, and overlapping descriptions will not be repeated.

<Configuration of welding device 200>
Similar to welding device 100, welding device 200 has welding torch 10, drive mechanism 20, power supply device 30, sensor 40, and controller 50. Welding device 200 has a trained model. The trained model is stored in storage device 51 of controller 50. In welding device 200, controller 50 (CPU 52) determines first welding conditions based on the trained model. In these respects, the configuration of welding device 200 is different from the configuration of welding device 100.

The trained model may be, for example, subjected to reinforcement learning. The hot cracking index is an index indicating the susceptibility of hot cracking in a weld. The hot cracking index may be, for example, tensile plastic strain in the solidification brittle temperature region of the weld. The hot cracking index may be the displacement of the temperature gradient vector or width W in the solidification brittle temperature region of the weld. The hot cracking index may be a combination of the above parameters.

The temperature gradient vector in the solidification brittle temperature range of a weld corresponds to the crystal growth direction of the weld metal in the solidification brittle temperature range. Since hot cracking in a weld is also affected by the crystal growth direction of the weld metal in the solidification brittle temperature range, the temperature gradient vector in the solidification brittle temperature range of a weld can be an indicator of hot cracking.

The hot cracking index may be calculated based on finite element method (FEM) analysis, or may be obtained by performing measurements during actual welding.

The welding conditions (learning welding conditions) used when reinforcement learning is performed include, for example, the movement speed of the welding torch 10. The learning welding conditions may also include the output of the welding torch 10. The learning welding conditions may also include the presence or absence of a tack between the first member 110 and the second member 120, the dimensions of the tack between the first member 110 and the second member 120, the number of tacks between the first member 110 and the second member 120, the dimensions of the first member 110, the dimensions of the second member 120, the shape of the groove, and the like. There are multiple learning welding conditions. Each of the multiple learning welding conditions is different from the others with respect to at least one of the above parameters.

In reinforcement learning, a reward is given in association with the selection of learning welding conditions. The reward differs depending on which of multiple learning welding conditions is selected. In reinforcement learning, the value function is updated based on the reward given in association with the selection of learning welding conditions.

For example, a first reward is given in response to the selection of learning welding conditions. The smaller the hot cracking index corresponding to the selected learning welding conditions, the larger the first reward becomes. Reinforcement learning progresses as the value function is updated based on the first reward.

A second reward may be further awarded in conjunction with the selection of the learning welding conditions. The second reward is different from the first reward. The second reward increases as the average movement speed of the welding torch 10 increases (i.e., the welding efficiency increases). In this case, the value function is updated based on the first reward and the second reward.

The reinforcement learning may be performed in a plurality of states according to the progress stage of welding. Each of the plurality of states corresponds to, for example, which of the regions _R1 to _Rn the welding torch 10 is located in. Each of the plurality of states may correspond to which of the regions _R1 to _Rn the welding torch 10 is located in and the history of the movement speed of the welding torch 10. When the reinforcement learning is performed in a plurality of states, the selection of the learning welding conditions, the assignment of the reward, and the update of the value function are performed in each of the plurality of states.

The first welding conditions are determined based on the value function of the trained model. More specifically, the first welding conditions are determined so as to maximize the value function of the trained model in each of a plurality of states. The controller 50 operates the welding torch 10 according to the movement speed and output of the welding torch 10 at each point in time included in the determined first welding conditions.

The above reinforcement learning is, for example, Q-learning. However, the trained model may be one that has undergone reinforcement learning other than Q-learning. Q-learning is performed by repeatedly updating the value function according to formula (1).

Q(s(t), a(t)) is the value function at time t. s(t) represents the state at time t. s(t) can take multiple values depending on the state at time t. a(t) represents the action at time t. a(t) can take multiple values depending on the action selected at time t. α is the learning rate and is a number greater than 0 and less than or equal to 1. γ is the discount rate and is a number greater than 0 and less than or equal to 1.

max[Q(s(i),a(i)] is the maximum value of the value function at time i. Time i can be any time in the past or future relative to time t. r is the reward given corresponding to the selected action (welding condition).

In Q-learning, first, an arbitrary initial value is given to Q(s(t), a(t)). Second, Q(s(t), a(t)) is repeatedly updated according to equation (1). When performing this update, for example, the ε-greedy method is used. That is, when performing this update, in principle, the welding conditions that maximize the value of the value function are selected, but there is a certain probability that random welding conditions are selected regardless of the magnitude of the value function. This certain probability decreases as the learning progresses (as the number of repetitions increases).

s(t) corresponds to, for example, the position of welding torch 10 at time t and the history of welding conditions before time t. The position of welding torch 10 at time t corresponds to which of regions _R1 to _Rn welding torch 10 is located in. The history of welding conditions before time t is, for example, the history of the movement speed of welding torch 10 before time t or the history of the movement speed of welding torch 10 before time t. How far back in time the history of welding conditions is to be considered as s(t) is appropriately selected.

a(t) corresponds to, for example, the welding conditions at time t. The welding conditions at time t considered as a(t) are, for example, the output of the welding torch 10 at time t and the movement speed of the welding torch 10 at time t.

However, the welding conditions at time t considered as a(t) are not limited to the moving speed of the welding torch 10 and the output of the welding torch 10. For example, the welding conditions at time t considered as a(t) are the dimensions of the tack between the first member 110 and the second member 120, the number of tacks between the first member 110 and the second member 120, the dimensions of the first member 110, the dimensions of the second member 120, the shape of the groove, etc.

r includes, for example, a first reward that is awarded based on a hot crack index under the welding conditions at time t. r may further include a second reward that is awarded based on an index indicating the welding efficiency under the welding conditions at time t (for example, the average movement speed of the welding torch 10).

The first welding conditions are determined based on the learned value of Q(s(t), a(t)). More specifically, the first welding conditions at each time point are determined to correspond to a(t) at which the value of Q(s(t), a(t)) is maximized.

<Effects of the welding device 200>
Many parameters are involved in the occurrence of hot cracking, so it is difficult to optimize the welding conditions by repeating actual welding procedures to suppress the occurrence of hot cracking while maintaining welding efficiency.

On the other hand, for example, by using reinforcement learning, it is possible to easily optimize welding conditions that can suppress the occurrence of hot cracking while maintaining welding efficiency. In addition, by calculating the hot cracking index using finite element analysis, it is possible to easily prepare a large amount of data set required for reinforcement learning, making this optimization even more efficient.

<Example of determining first welding conditions based on reinforcement learning>
The width of the first member 110 and the second member 120 in the first direction DR1 was set to 70 mm. The length of the first member 110 and the second member 120 in the second direction DR2 was set to 300 mm. The first member 110 and the second member 120 were each divided into four regions, i.e., region _R1 , region _R2 , region _R3 , and region _R4 .

The time when the welding torch 10 was in region _R1 was designated time 1, the time when the welding torch 10 was in region _R2 was designated time 2, the time when the welding torch 10 was in region _R3 was designated time 3, and the time when the welding torch 10 was in region _R4 was designated time 4.

The welding condition considered as a(t) was the movement speed of the welding torch 10. Four types of movement speed of the welding torch 10, 5 mm/sec, 10 mm/sec, 15 mm/sec, and 20 mm/sec, were selectable as a(t). Note that the welding conditions other than the movement speed of the welding torch 10 were fixed to constant values. More specifically, the heat input per unit length from the welding torch 10 was set to 70 J/mm, and no tack was provided between the first member 110 and the second member 120.

s(t) was determined based on the position of the welding torch 10 at time t (in which of regions _R1 to _R4 the welding torch 10 is located) and the movement speed of the welding torch 10 at time (t-1). However, for s(1), since there is no movement speed of the welding torch 10 at an earlier time point, the movement speed of the welding torch 10 at time (t-1) was not taken into consideration.

r was given a first reward based on the hot cracking index and a second reward based on the average speed of the welding torch 10. The higher the hot cracking index was for the welding conditions, the higher the first reward was given, and the higher the average speed of the welding torch 10 was for the welding conditions, the higher the second reward was given. The hot cracking index was determined by the tensile plastic strain in the solidification brittle temperature range of the weld.

Under the above conditions, formula (1) was updated 50 times. The values of Q(s(t), a(t)) at the time when formula (1) was updated 50 times are shown in Table 1. Note that, since four different movement speeds of welding torch 10 can be selected at each of time points 1 to 4, there are a total of 4 ⁴ = 256 patterns.

As shown in Table 1, the selection of a(t) that maximizes the value of Q(s(t), a(t)) at each time point is to set the movement speed of the welding torch 10 at time point 1 to 5 mm/sec, the movement speed of the welding torch 10 at time point 2 to 5 mm/sec, the movement speed of the welding torch 10 at time point 3 to 15 mm/sec, and the movement speed of the welding torch 10 at time point 4 to 20 mm/sec.

In addition to the determination of the welding conditions using the trained model that had undergone the above Q-learning, the presence or absence of hot cracks was evaluated based on finite element analysis. This analysis was performed 256 times in the same manner as above. As a result, the most preferable welding conditions from the viewpoints of preventing hot cracks and welding efficiency were that the moving speeds of the welding torch 10 in the regions _R1 , _R2 , _R3 , and _R4 were 5 mm/sec, 5 mm/sec, 15 mm/sec, and 20 mm/sec, respectively. That is, the optimal welding conditions obtained by the finite element analysis and the optimal welding conditions obtained by the trained model were consistent with each other.

In this way, the above comparison reveals that the first welding conditions can be appropriately determined based on a trained model that has undergone Q-learning.

<Modification>
In the above, reinforcement learning is applied to the trained model, but the learning applied to the trained model is not limited to this. For example, the trained model may output a hot crack index when a welding condition is applied to an input welding condition, and supervised learning may be performed using a data set including the learning welding condition and the hot crack index when the learning welding condition is applied as training data.

The first welding conditions may also be determined by applying a genetic algorithm. More specifically, the storage device 51 stores an evaluation function that outputs an evaluation value of the welding conditions when the welding conditions are input. This evaluation value includes a hot cracking index. The CPU 52 may determine the first welding conditions by referencing this evaluation value and applying a genetic algorithm (for example, searching for welding conditions that minimize the hot cracking index using a genetic algorithm).

Furthermore, the method for determining the first welding conditions (optimization method) is not limited to the above. The CPU 52 may determine the first welding conditions by applying, for example, a complex method, a steepest descent method, or a random search.

Although the embodiment of the present invention has been described above, the above embodiment can be modified in various ways. Furthermore, the scope of the present invention is not limited to the above embodiment. The scope of the present invention is indicated by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims.

10 welding torch, 20 driving mechanism, 30 power supply, 40 sensor, 50 controller, 51 storage device, 52 CPU, 100, 200 welding device, 110 first member, 110a, 110b end face, 110c first main surface, 110d second main surface, 120 second member, 120a, 120b end face, 120c first main surface, 120d second main surface, 130 weld metal, DR1 first direction, DR2 second direction, W width.

Claims (12)

A welding device that welds a first member and a second member in a state in which a first end surface that is an end surface of a first member in a first direction and a second end surface that is an end surface of a second member in the first direction face each other,
a welding torch that moves over a welded portion where the first member and the second member are welded along a second direction that intersects the first direction;
a sensor that measures a width of a groove formed by the first end surface and the second end surface in the first direction while moving along the second direction together with the welding torch,
The welding device is configured to perform welding according to a first welding condition when the displacement of the width is less than a predetermined threshold, and to perform welding according to a second welding condition different from the first welding condition when the displacement of the width is equal to or greater than the threshold,
the width variation under the second welding condition is smaller than the width variation under the first welding condition,
A welding apparatus , wherein an output of the welding torch under the second welding condition is smaller than an output of the welding torch under the first welding condition . The welding device of claim 1, wherein the sensor measures the width displacement in front of the welding torch. The welding device according to claim 1 or 2, wherein the movement speed of the welding torch under the second welding condition is smaller than the movement speed of the welding torch under the first welding condition. 4. The welding device according to claim 1 , wherein under the second welding condition, the welding torch is weaved along the first direction or the second direction. A welding device that welds a first member and a second member in a state in which a first end surface that is an end surface of a first member in a first direction and a second end surface that is an end surface of a second member in the first direction face each other,
a welding torch that moves over a welded portion where the first member and the second member are welded along a second direction that intersects the first direction;
a sensor that measures a width of a groove formed by the first end surface and the second end surface in the first direction while moving along the second direction together with the welding torch,
The welding device is configured to perform welding according to a first welding condition when the displacement of the width is less than a predetermined threshold, and to perform welding according to a second welding condition different from the first welding condition when the displacement of the width is equal to or greater than the threshold,
The width variation under the second welding condition is smaller than the width variation under the first welding condition,
The welding apparatus further includes a memory device and a processor.
The storage device stores a trained model,
The trained model has undergone reinforcement learning,
In the reinforcement learning, when any one of a plurality of learning welding conditions is selected, a first reward is given;
The first reward increases as a hot crack index, which is an index of the susceptibility of hot cracking in the welded portion, decreases,
In the reinforcement learning, a value function is updated based on the first reward;
The processor determines the first welding condition based on the value function . The welding device according to claim 5 , wherein the hot cracking index when each of the plurality of learning welding conditions is selected is calculated based on a finite element method analysis. The welding device according to claim 6 , wherein the hot cracking index is at least one of a tensile plastic strain in a solidification brittle temperature region of the weld, a temperature gradient vector in the solidification brittle temperature region, and a displacement of the width. In the reinforcement learning, when any one of the plurality of learning welding conditions is selected, a second reward is further given;
the second reward increases as the average moving speed of the welding torch increases,
The welding device according to claim 5 , wherein in the reinforcement learning, the value function is updated based on the first reward and the second reward. The welding device according to any one of claims 5 to 8 , wherein the trained model has undergone Q-learning. A welding device that welds a first member and a second member in a state in which a first end surface that is an end surface of a first member in a first direction and a second end surface that is an end surface of a second member in the first direction face each other,
a welding torch that moves over a welded portion where the first member and the second member are welded along a second direction that intersects the first direction;
a sensor that measures a width of a groove formed by the first end surface and the second end surface in the first direction while moving along the second direction together with the welding torch,
The welding device is configured to perform welding according to a first welding condition when the displacement of the width is less than a predetermined threshold, and to perform welding according to a second welding condition different from the first welding condition when the displacement of the width is equal to or greater than the threshold,
the width variation under the second welding condition is smaller than the width variation under the first welding condition,
The welding apparatus further includes a memory device and a processor.
The storage device stores a trained model,
When welding conditions are input, the trained model outputs a hot cracking index that is an index of the likelihood of hot cracking occurring in the weld when the welding conditions are applied;
The trained model is subjected to supervised learning using training data including a learning welding condition and the hot cracking index when the learning welding condition is applied,
The processor determines the first welding condition based on the hot cracking index output from the trained model . A welding device that welds a first member and a second member in a state in which a first end surface that is an end surface of a first member in a first direction and a second end surface that is an end surface of a second member in the first direction face each other,
a welding torch that moves over a welded portion where the first member and the second member are welded along a second direction that intersects the first direction;
a sensor that measures a width of a groove formed by the first end surface and the second end surface in the first direction while moving along the second direction together with the welding torch,
The welding device is configured to perform welding according to a first welding condition when the displacement of the width is less than a predetermined threshold, and to perform welding according to a second welding condition different from the first welding condition when the displacement of the width is equal to or greater than the threshold,
The width variation under the second welding condition is smaller than the width variation under the first welding condition,
The welding apparatus further includes a memory device and a processor.
an evaluation function is stored in the storage device, which outputs an evaluation value of a welding condition when the welding condition is input;
The evaluation value includes a hot crack index which is an index of the susceptibility of hot cracking to occur in the weld when the welding conditions are applied,
The processor determines the first welding condition by applying a genetic algorithm based on the evaluation value. A welding method for welding a first member and a second member in a state in which a first end surface that is an end surface of a first member in a first direction and a second end surface that is an end surface of a second member in the first direction face each other, the method comprising the steps of:
welding the first member and the second member together by moving a welding torch along a second direction intersecting the first direction;
and measuring a width of the groove formed by the first end surface and the second end surface in the first direction by a sensor moving along the second direction together with the welding torch,
The welding of the first member and the second member is performed according to a first welding condition when the displacement of the width is less than a predetermined threshold value, and according to a second welding condition different from the first welding condition when the displacement of the width is equal to or greater than the threshold value,
The width variation under the second welding condition is smaller than the width variation under the first welding condition,
A welding method , wherein an output of the welding torch under the second welding condition is smaller than an output of the welding torch under the first welding condition .

Images