TWI876721B - A cascade control winding system - Google Patents

Abstract

A cascade control winding system is disclosed. The cascade control winding system comprises a winding motor, a winding tension meter, a winding-end learning parameter model, a winding controller, a wire feeding motor, a wire feeding tension meter, a wire feeding-end learning parameter model, and a wire feeding controller. The winding motor drives the iron core to rotate with torque, so that a flat wire is wound around the iron core. The winding tension meter measures a winding tension. The winding-end learning parameter model predicts torque gain parameters according to a difference between a predetermined value and a measured value of the winding tension. The winding controller outputs the torque and the target value of the wire feeding tension according to the torque gain parameters. The wire feeding motor drives the flat wire spool to rotate at a rotational speed so that the flat wire with the wire feeding tension is fed out. The wire feeding tension meter measures the wire feeding tension. The wire feeding-end learning parameter model predicts rotational speed gain parameters according to a difference between the target value of the wire feeding tension and the measured value of the winding tension. The wire feeding controller outputs the rotational speed according to the rotational speed gain parameters.

Description

Cascade Control Winding System

The present invention relates to a winding system, in particular to a cascade controlled winding system.

The cross-section of coil wire is generally circular, and later developed into a relatively wide and long flat wire. The current methods of winding flat wire can be divided into flat winding and vertical winding. The flat winding method is generally used, which is to wind the long side of the flat wire against the iron core. The method can be directly wound around the iron core, or it can be formed and then installed on the iron core.

However, the problem with flat winding is that the heat dissipation is poor. If vertical winding is used, the heat dissipation problem can be solved and the output power can be increased. However, the existing method of vertical winding of flat wires is to form them first and then install them on the iron core. The flat wire cannot be directly wound against the iron core, so there is a gap between the flat wire and the iron core. Moreover, the winding method of vertical winding and flat winding differs by 90 degrees, that is, the vertical winding of the flat wire is to wind the short side of the flat wire tightly against the stator iron core, and the difference between the inner and outer diameters is large. The outer side of the bend is greatly stretched, while the inner side is greatly contracted, which is prone to plastic deformation, such as wrinkles, so the difficulty is greatly increased.

Furthermore, in the process of winding the short side of the flat wire tightly against the iron core, the existing flat wire vertical winding device uses a winding motor to drive the iron core to rotate, and at the same time uses a wire feeding motor at the front end to drive the flat wire shaft to transport the flat wire and provide tension to the flat wire. However, since the existing flat wire vertical winding device controls the winding motor and the wire feeding motor separately and independently, it is easy to cause the tension of the flat wire at the wire feeding end and the winding end to be inconsistent, so that the accuracy of the flat wire winding on the iron core cannot be accurately controlled, and a better slot occupancy rate cannot be achieved.

Therefore, an object of the present invention is to provide a cascade controlled winding system.

Therefore, the cascade controlled winding system of the present invention is suitable for winding a flat wire from a flat wire spool onto an iron core in a vertical winding manner, and comprises: a winding motor, which drives the iron core to rotate synchronously with a torque, so that the flat wire is wound onto the iron core; a winding tension meter, which is arranged on the flat wire near the iron core and measures the winding tension of the flat wire. to generate a winding tension measurement value; a winding end learning parameter model, predicting a torque gain parameter according to a winding tension difference between a winding tension setting value and the winding tension measurement value from the winding tension meter; a winding controller, generating the torque output to the winding according to the torque gain parameter from the winding end learning parameter model motor, wherein the wire winding controller also outputs a wire feeding tension target value; a wire feeding motor, which drives the flat wire shaft to rotate with a rotational speed to feed the flat wire with wire feeding tension; a wire feeding tension meter, which is disposed on the flat wire near the flat wire shaft and measures the wire feeding tension of the flat wire to generate a wire feeding tension measurement value; a wire feeding end learning parameter model, which predicts a rotational speed gain parameter based on a wire feeding tension difference between the wire feeding tension target value from the wire winding controller and the wire feeding tension measurement value from the wire feeding tension meter; and a wire feeding controller, which generates the rotational speed output to the wire feeding motor based on the rotational speed gain parameter from the wire feeding end learning parameter model.

The efficacy of the present invention is that: since the winding end learning parameter model and the wire feeding end learning parameter model are both policy models generated based on reinforcement learning, and the winding controller outputs the wire feeding tension target value in cascade as the input of the wire feeding end, the tension of the flat wire at the wire feeding end and the winding end can be controlled to be consistent, so as to accurately control the accuracy of the flat wire being wound on the iron core, thereby achieving a better slot occupancy rate.

Referring to FIGS. 1 to 3 , the embodiment of the cascade controlled winding system of the present invention is applicable to winding the conductive flat wire 90 of the flat wire spool 9 of the winding device on the iron core 20 in a vertical winding manner. As shown in FIG. 1 , in this embodiment, the cascade controlled winding system includes a winding motor 50, a winding tension meter 51, a winding end learning parameter model 52, a winding controller 53, a wire feeding motor 60, a wire feeding tension meter 61, a wire feeding end learning parameter model 62, and a wire feeding controller 63. In this embodiment, the winding controller 53 and the wire feeding controller 63 are both proportional integral derivative (PID) controllers. The flat wire 90 from the flat wire spool 9 is first guided forward by the wire wheel 7 to pass through the wire feeding tension meter 61, and then guided forward by the wire wheel 8 to pass through the wire winding tension meter 51.

The winding end learning parameter model 52 and the transmission end learning parameter model 62 are both policy models generated based on reinforcement learning. In this embodiment, since the winding end learning parameter model 52 and the delivery end learning parameter model 62 are both strategy models generated based on the Advantage Actor Critic technology of enhanced learning, the winding end strategy model 54 and the delivery end strategy model 64 must first be trained using the training process of Figure 2. After the training process is completed, the winding end strategy model 54 and the delivery end strategy model 64 are respectively deployed as the winding end learning parameter model 52 and the delivery end learning parameter model 62 of Figures 1 and 3.

As shown in FIG. 2 , since the embodiment of the cascade control winding system of the present invention must undergo a pre-training process, it also includes the winding end strategy model 54, the winding end reward model 55, the winding end memory parameter database 56, the feeding end strategy model 64, the feeding end reward model 65, and the feeding end memory parameter database 66.

In the training process, first, the winding motor 50 drives the iron core 20 to rotate synchronously with the torque, so that the flat wire 90 is wound on the iron core 20. Then, the winding end strategy model 54 can generate a torque gain parameter according to the winding tension measurement value and output it to the winding controller 53, and store the torque gain parameter in the winding end memory parameter database 56. Then, the winding controller 53 generates the torque output to the winding motor 50 and the winding end reward model 55 according to the torque gain parameter. Then, the winding end reward model 55 and the winding end memory parameters in the winding end memory parameter database 56 are used to optimize the winding end strategy model 54.

In addition, the wire winding controller 53 also outputs the wire feeding tension target value to the wire feeding end strategy model 64. Then, the wire feeding end strategy model 64 generates a speed gain parameter according to the wire feeding tension difference between the wire feeding tension target value and the wire feeding tension measurement value, and outputs it to the wire feeding controller 63, and stores the speed gain parameter in the wire feeding end memory parameter database 66. Then, the wire feeding controller 63 generates a speed output to the wire feeding motor 60 and the wire feeding end reward model 65 according to the speed gain parameter. Finally, the wire feeding end reward model 65 and the wire feeding end memory parameters in the wire feeding end memory parameter database 66 are used to optimize the wire feeding end strategy model 64.

Therefore, after the training process is completed as shown in FIG. 2 , the winding end strategy model 54 can be deployed as the winding end learning parameter model 52, and the delivery end strategy model 64 can be deployed as the delivery end learning parameter model 62, as shown in FIGS. 1 and 3 .

Therefore, as shown in Figs. 1 and 3, in this embodiment, the cascade control process of the cascade control winding system of the present invention is that the winding motor 50 applies the torque to drive the iron core 20 to rotate synchronously, so that the flat wire 90 is wound on the iron core 20. Therefore, the winding tension meter 51 disposed on the flat wire 90 near the iron core 20 can measure the winding tension of the flat wire 90 to generate a winding tension measurement value.

Next, the winding end learning parameter model 52 predicts a torque gain parameter based on a winding tension difference between a winding tension setting value and a winding tension measurement value from the winding tension meter 51, wherein the torque gain parameter includes a torque gain parameter proportional portion, a torque gain parameter integral portion, and a torque gain parameter differential portion.

Then, the winding controller 53 generates the torque output to the winding motor 50 according to the torque gain parameter from the winding end learning parameter model 52. The winding controller 53 also outputs the wire feeding tension target value.

At the wire feeding end, the wire feeding motor 60 applies a rotational speed to drive the flat wire shaft 9 to rotate so as to feed the flat wire 90 with wire feeding tension.

Therefore, the wire feeding tension meter 61 disposed on the flat wire 90 near the flat wire shaft 9 can measure the wire feeding tension of the flat wire 90 to generate a wire feeding tension measurement value.

Next, the wire feeding end learning parameter model 62 predicts a speed gain parameter according to the wire feeding tension difference between the wire feeding tension target value from the wire winding controller 53 and the wire feeding tension measurement value from the wire feeding tension meter 61. The speed gain parameter includes a speed gain parameter proportional part, a speed gain parameter integral part, and a speed gain parameter differential part.

Then, the wire feeding controller 63 can generate the speed output to the wire feeding motor 60 according to the speed gain parameter from the wire feeding end learning parameter model 62.

Therefore, in this embodiment, since the winding end learning parameter model 52 and the wire feeding end learning parameter model 62 are both policy models generated based on reinforcement learning, and the winding controller 53 outputs the wire feeding tension target value in series as the input of the wire feeding end, the tension of the flat wire 90 at the wire feeding end and the winding end can be controlled to be consistent, thereby being able to accurately control the accuracy of the flat wire 90 wound around the iron core 20 to achieve a better slot occupancy rate.

In summary, since the winding end learning parameter model 52 and the wire feeding end learning parameter model 62 of the present invention are both policy models generated based on the Advantage Actor Critic technology of reinforcement learning, and the winding controller 53 outputs the wire feeding tension target value in cascade as the input of the wire feeding end, the tension of the flat wire 90 at the wire feeding end and the winding end can be controlled to be consistent, so the accuracy of the flat wire 90 being wound around the iron core 20 can be accurately controlled to achieve a better slot occupancy rate; therefore, the purpose of the present invention can be achieved.

However, the above is only an embodiment of the present invention and should not be used to limit the scope of implementation of the present invention. All simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the content of the patent specification are still within the scope of the present patent.

20: Iron core 50: Winding motor 51: Winding tension gauge 52: Winding end learning parameter model 53: Winding controller 54: Winding end strategy model 55: Winding end reward model 56: Winding end memory parameter database 60: Wire feeding motor 61: Wire feeding tension gauge 62: Wire feeding end learning parameter model 63: Wire feeding controller 64: Wire feeding end strategy model 65: Wire feeding end reward model 66: Wire feeding end memory parameter database 7: Wire guide wheel 8: Wire guide wheel 9: Flat wire spool 90: Flat wire

Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, wherein: FIG. 1 is a schematic diagram illustrating an embodiment of the cascade control winding system of the present invention, which is suitable for winding a flat wire of a flat wire spool on an iron core using a vertical winding method and is constructed based on enhanced learning; FIG. 2 is a block diagram illustrating the training process of enhanced learning in the present embodiment; and FIG. 3 is a block diagram illustrating the cascade control architecture of the cascade control winding system of the present invention in the present embodiment.

20: Iron core

50: Winding motor

51: Winding tension gauge

52: Winding end learning parameter model

53: Winding controller

60: Wire feeding motor

61: Wire feeding tension gauge

62: Transmitter learning parameter model

63: Wire feeding controller

7: Wire wheel

8: Wire wheel

9: Flat wire spool

90: Flat wire

Claims (7)

A cascade control winding system is suitable for winding a flat wire of a flat wire spool onto an iron core using a vertical winding method, and comprises: A winding motor drives the iron core to rotate synchronously with a torque, so that the flat wire is wound onto the iron core; A winding tension meter is disposed on the flat wire near the iron core, and measures the winding tension of the flat wire to generate a winding tension measurement value; A winding end learning parameter model predicts a torque gain parameter based on the winding tension difference between a winding tension setting value and the winding tension measurement value from the winding tension meter; A winding controller generates the torque output to the winding motor according to the torque gain parameter from the winding end learning parameter model, wherein the winding controller also outputs a wire feeding tension target value; A wire feeding motor drives the flat wire shaft to rotate at a rotation speed to feed the flat wire with wire feeding tension; A wire feeding tension meter is disposed on the flat wire near the flat wire shaft and measures the wire feeding tension of the flat wire to generate a wire feeding tension measurement value; A wire feeding end learning parameter model predicts a speed gain parameter according to a wire feeding tension difference between the wire feeding tension target value from the winding controller and the wire feeding tension measurement value from the wire feeding tension meter; and The wire feeding controller generates the speed output to the wire feeding motor according to the speed gain parameter from the wire feeding end learning parameter model. A cascade controlled winding system as described in claim 1, wherein the winding controller and the wire feeding controller are both proportional integral derivative controllers. A cascade controlled winding system as described in claim 1, wherein the winding end learning parameter model and the feeding end learning parameter model are both strategy models generated based on reinforcement learning. A cascade controlled winding system as described in claim 3, wherein the winding end learning parameter model and the feeding end learning parameter model are both strategy models generated based on the advantage actuator evaluator technology of enhanced learning. The cascade control winding system as described in claim 4 further includes a winding end strategy model, a winding end reward model, and a winding end memory parameter database for a training process, wherein during the training process, the winding end strategy model generates the torque gain parameter output to the winding controller according to the winding tension measurement value, and outputs the torque gain parameter to the winding controller. The gain parameter is stored in the winding end memory parameter database, and then the winding controller generates the torque output to the winding motor and the winding end reward model according to the torque gain parameter, and then optimizes the winding end strategy model using the winding end reward model and the winding end memory parameters in the winding end memory parameter database. The cascade control winding system as described in claim 5 further includes a wire feeding end strategy model, a wire feeding end reward model, and a wire feeding end memory parameter database for the training process, wherein, during the training process, the winding controller also outputs the wire feeding tension target value to the wire feeding end strategy model, and the wire feeding end strategy model generates a wire feeding tension difference between the wire feeding tension target value and the wire feeding tension measurement value. The speed gain parameter is output to the wire feeding controller, and the speed gain parameter is stored in the wire feeding end memory parameter database. Then the wire feeding controller generates the speed output to the wire feeding motor and the wire feeding end reward model according to the speed gain parameter, and then optimizes the wire feeding end strategy model by using the wire feeding end reward model and the wire feeding end memory parameters in the wire feeding end memory parameter database. A cascade controlled winding system as described in claim 6, wherein after the training process is completed, the winding end strategy model is deployed as the winding end learning parameter model, and the delivery end strategy model is deployed as the delivery end learning parameter model.

Images