machine tool

Abstract

Machine tool (1), comprising: a spindle (2) that causes a tool (X) to rotate and move; a workpiece rotation mechanism (3) that causes a workpiece (W) to rotate; a control unit (4) that controls the spindle (2) and the workpiece rotation mechanism (3) according to commands from a program; and a cutting load detection unit that detects a cutting load transferred from the tool (X) to the workpiece (W), wherein the control unit (4) controls a cutting route such that the cutting depth of the workpiece (W) cut by the tool (X) in a region with a low cutting load is greater than the cutting depth in a region with a high cutting load within such a range that the cutting load detected by the cutting load detection unit does not exceed a predetermined load.

Description

[Area of ​​Invention]

[0001] The present invention relates to a machine tool. [GENERAL STATE OF THE ART]

[0002] When machining is performed using a machine tool, a workpiece is machined using a tool according to a path, speed, and spindle speed determined by a program. A control system is known for such machining operations, designed to adjust machining conditions, such as spindle speed and feed rate, in such a way that neither chatter nor tool wear / breakage occurs, thus preventing the effects of chatter and tool wear / breakage on the machined surface during machining (see, for example, patent literature 1). [List of Counterarguments][Patent Literature]

[0003] [PATENT LITERATURE 1] Publication of Japanese Patent JP 6 063 013 B1. [BRIEF DESCRIPTION OF THE INVENTION][TECHNICAL TASK]

[0004] However, there is a disadvantage that if an actual cutting load exceeds an expected cutting load, the increase in cutting load leads to an increase in synchronization error and self-excited vibration, whereby the increase in synchronization error leads to a further increase in cutting load due to the effects of abrasion and the like of the tool during machining, which requires synchronization between rotational operations of the tool and the workpiece and synchronization between rotational operations and feeding operations, such as roll milling, where the machining is carried out while the workpiece is rotated synchronously with the rotation and movement of the tool.

[0005] Although it is necessary to secure a space to prevent the cutting load from becoming too large in order to limit the occurrence of self-excited vibration, there is a disadvantage that it is not possible to fully utilize machining capability by simply securing a space for the cutting load, and machining efficiency is thereby reduced.

[0006] One object of the invention is to provide a machine tool that is able to prevent the occurrence of self-excited vibration and also to prevent the deterioration of machining efficiency when synchronization between rotation and movement of the tool and rotation of a workpiece is required. [SOLUTION TO THE TASK]

[0007] The problem is solved by a machine tool with the features of claim 1. One aspect of the invention relates to a machine tool comprising: a spindle that causes a tool to rotate and move; a workpiece rotation mechanism that causes a workpiece to rotate; a control unit that controls the spindle and the workpiece rotation mechanism according to commands from a program; and a cutting load detection unit that detects a cutting load transferred from the tool to the workpiece, wherein the control unit controls a cutting route such that the cutting depth of the workpiece cut by the tool in a region with a low cutting load is greater than the cutting depth in a region with a high cutting load within such a range that the cutting load detected by the cutting load detection unit does not exceed the predetermined load.

[0008] According to this aspect, in a case where the control unit controls the spindle and the workpiece rotation mechanism in a synchronized manner, and the workpiece attached to the workpiece rotation mechanism is cut with the tool attached to the spindle, it is possible to machine the workpiece by the cutting route that increases the cutting load in the region with the lower cutting load by adjusting the cutting depth so that it is greater in the region where the cutting load transmitted by the tool to the workpiece, as detected by the cutting load detection unit, is low than in the region with the high cutting load.Since a cutting route with a cutting load that does not exceed a predetermined load is typically selected in the region with the high cutting load, it is possible to perform the machining efficiently and thereby reduce machining time by preventing the cutting load from becoming too large, preventing the occurrence of self-excited vibration, and setting a deep cutting depth in the region with the low cutting load due to the small contact between the workpiece and the tool.

[0009] In the preceding aspect, the control unit can control the spindle in a direction where the distance between an axis of rotation of the workpiece rotated by the workpiece turning mechanism and an axis of rotation of the tool rotated by the spindle is longer in the region with the low cutting load than in the region with the high cutting load.

[0010] This design makes it possible to easily adjust the cutting route, prevent the occurrence of self-excited vibration during machining, and also prevent the deterioration of work efficiency.

[0011] In the preceding aspect, the control unit may include a machine learning device that performs machine learning of adaptation parameters for machining conditions and an adaptation amount of the adaptation parameters during cutting, which is carried out with the tool on the workpiece, wherein the machine learning device may include a state observation unit that acquires state data, which is information that indicates a machining state during cutting, a reward calculation unit that calculates a reward based on the state data and reward conditions, a machining condition adaptation learning unit that performs machine learning of the adaptation of the machining conditions, and a machining condition adaptation unit.which, as an adaptation action, determines target parameters for adjusting the machining conditions and the amount of the adaptation based on a result of machine learning of the machining conditions adaptation, performed by the machining condition adaptation learning unit, and the state data, and adapts the machining conditions during cutting based on the adaptation action, wherein the machining condition adaptation learning unit can perform machine learning of the machining conditions adaptation as the adaptation action based on the state data acquired by the state observation unit after a machining operation, based on the output machining conditions after the adaptation and the reward calculated by the reward calculation unit, and as the reward conditions, a negative reward is ordered if vibration occurs, a positive reward is ordered.if the cutting action performed on the workpiece with the tool increases the cutting load without exceeding a predetermined load range, and a negative reward is imposed if the cutting load decreases or the cutting load exceeds the predetermined load.

[0012] With this design, it is possible to easily decide on a cutting route using machine learning, which is able to prevent the occurrence of self-excited vibration during machining and also prevent the deterioration of machining efficiency. [ADVANTAGEOUS EFFECTS OF THE INVENTION]

[0013] One effect according to the invention is that it is possible to prevent the occurrence of self-excited vibration and also to prevent the deterioration of machining efficiency when synchronization between rotation and movement of the tool and rotation of a workpiece is required. [BRIEF DESCRIPTION OF THE DRAWINGS] [ Fig. 1] Fig. Figure 1 is a block diagram illustrating a machine tool according to one embodiment of the invention. [ Fig. 2] Fig. 2 is a block diagram illustrating an example of a control unit used in the machine tool of Fig. 1 is arranged. [ Fig. 3] Fig. Figure 3 is a representation explaining a basic idea of ​​an algorithm for learning through reinforcement in the context of Fig. 1 illustrated machine tool. [ Fig. 4] Fig. 4 is a flowchart that illustrates the process of machine learning in the Fig. 1 illustrated machine tool illustrated. [ Fig. 5] Fig. Figure 5 is a schematic view to explain a machining path of the machine tool in the prior art. [ Fig. 6] Fig. Figure 6 is a schematic view to explain the contact amount between the tool and the workpiece in the diagram. Fig. 5 illustrated editing paths. [ Fig. 7] Fig. Figure 7 is a diagram illustrating the change in engine load over time during roughing, as seen three times through the machining path in Fig. 5 is carried out. [ Fig. 8] Fig. Figure 8 is a schematic view to explain the contact amount between the tool and the workpiece in a Fig. Figure 1 illustrated the optimized machining path of the machine tool. [ Fig. 9] Fig. 9 is a schematic view to explain the in Fig. Figure 1 illustrates the optimized machining path of the machine tool. [ Fig. 10] Fig. Figure 10 is a diagram showing the change in engine load over time during roughing, which is repeated three times by the in Fig. The editing path illustrated in section 9 is performed, compared to the one in Fig. 7 illustrated changes over time. [ Fig. 11] Fig. Figure 11 is a diagram illustrating an example of a structure that differs from a structure of the in Fig. 7 illustrated engine loads differ. [ Fig. 12] Fig. Figure 12 is a diagram illustrating another example of a structure that differs from the structure of the one in Fig. 7 illustrated engine loads differ. [DESCRIPTION OF EXECUTION FORMS]

[0014] A machine tool 1 according to an embodiment of the invention is described below with reference to the drawings.

[0015] The machine tool 1 according to the embodiment comprises a spindle 2 that causes a tool X to rotate and move, a workpiece rotation mechanism 3 that causes a workpiece W to rotate, and a control unit 4 that controls the spindle 2 and the workpiece rotation mechanism 3 according to commands from a program P, as described in Fig. Figure 1 illustrates the control unit 4, which is implemented using a memory and a processor.

[0016] The control unit 4 analyzes the program P read from the memory, controls the spindle 2 and the workpiece rotation mechanism 3 in the machine tool 1 based on the control data obtained as a result of the analysis, and machines the workpiece W. The machine tool 1 includes a sensor that detects the position, speed, and rotational speed of the tool X and the rotational speed of the workpiece W, which is not illustrated in the drawing, and a sensor that detects a load (cutting load) acting on the motor of the spindle 2 during machining and vibration of the motor.For example, the position, speed and rotational speed of the tool X and the rotational speed of the workpiece W are detected by an encoder, a load acting on the motor is detected by a current sensor (cutting load detection unit: omitted in the drawing) and vibration of the motor is detected by a vibration sensor (omitted in the drawing).

[0017] The control unit 4 can acquire condition data and determination data via these sensors. The control unit 4 comprises a condition monitoring unit 5, which acquires the condition data of each component of the machine tool 1, a determination data acquisition unit 6, which acquires the determination data, and a machine learning device 7, which serves as an artificial intelligence that performs machine learning.

[0018] The state monitoring unit 5 observes the position of the tool X, the spindle speed and feed rate of the spindle 2, and the workpiece speed W, as recorded by the control data obtained as a result of the analysis performed by the program P of the control unit 4. These observations constitute state data relating to a machining state, and the unit records the observed state data in the machine learning device 7. The recorded state data is stored in the machine learning device 7 and input into a reward calculation unit 8 and a value function update unit 9, which are described below.

[0019] The entered status data can be either data recorded in a recent processing operation or data recorded in a past processing operation. It is also possible to enter, save, and output status data stored in another tax system or a centralized management system.

[0020] The determination data acquisition unit 6 acquires data indicating a motor load of the spindle 2 and vibration data acquired by the sensor as determination data and outputs the determination data to the reward calculation unit 8 of the machine learning facility 7.

[0021] It is assumed that the machine tool 1 and the control unit 4 are arranged with typical configurations of a machine tool and a numerical value control system, and a detailed description thereof is omitted in the description except for configurations that are particularly necessary to explain the processes for machine learning according to the invention. According to the Fig. In the illustrated embodiment 2, the machine learning device 7 corresponds to an agent, and each embodiment arranged in the machine tool 1, except for the machine learning device 7, corresponds to an environment.

[0022] The condition data, such as the position, speed, and rotational speed of tool X and the rotational speed and similar parameters of the workpiece W during machining using the machining tool 1, which were acquired by the condition monitoring unit 5, and the determination data, such as the motor load and vibration data, which were acquired by the determination data acquisition unit 6, are input into the machine learning device 7 as information to specify an environment. This includes data acquired from each component of the machine tool 1 and data acquired by the control unit 4.

[0023] In this embodiment, a command to adjust a machining path of tool X and to adjust a cutting depth is used as output data in an action output that the machine learning facility 7 outputs to the environment.

[0024] In this embodiment, rewards such as motor load (one positive reward and one negative reward), the occurrence of wear / breakage of tool X (negative reward), the occurrence of vibration (negative reward), and the like are used to be assigned in response to machine learning. An operator can appropriately configure which data are to be used as the basis for deciding on a reward.

[0025] Facility 7, dedicated to machine learning, performs machine learning based on the state data, determination data, and reward previously described. In machine learning, (1) the observation of a state s t an environment using the agent, (2) the selection of an action at which the agent itself is based on the observed state s tand past learning and execution of the action at can be used, (3) a change from the state s t to the next state s t+1 the environment due to the execution of the action a t , (4) the receipt of a reward r t+1 using the agent based on a state change as a result of action a t and (5) continuing the learning using the agent based on the state s t , the action at, the reward r t+1 and the result of past learning between the agent and the environment at a specific clock time t, as in Fig. 3 illustrates.

[0026] In particular, the adjustment of the machining path (cutting route) and the adjustment of the cutting depth, which are carried out on the spindle 2 and the workpiece turning mechanism 3, correspond to the action at, a value that is based on the state s t+1The reward r is evaluated and calculated as a result of performing the action at, and corresponds to a reward r. t+1 and the learning is performed by applying them to a value function update expression according to the machine learning algorithm.

[0027] The machine learning device 7 comprises a reward calculation unit (reward calculation unit) 8, a value function update unit (processing condition adaptation learning unit) 9, and a decision-making unit (processing condition adaptation unit) 10. The machine learning device 7 can be located in the control unit 4 or it can be located in a personal computer or the like outside the control unit 4, as shown in Fig. 2 illustrated.

[0028] The reward calculation unit 8 calculates a reward using the state data and the determination data acquired by the state observation unit 5 or the determination data acquisition unit 6 based on conditions under which a reward is ordered in machine learning set by an operator or the like, and outputs the reward to the value function update unit 9.

[0029] The reward includes a positive reward and a negative reward and can be adjusted appropriately.

[0030] The reward calculation unit 8 analyzes the data input by the state observation unit 5 and the determination data acquisition unit 6 based on the set reward conditions and outputs the calculated reward to the value function update unit 9.

[0031] The following are examples of reward conditions set by the reward calculation unit 8. "Reward 1: Occurrence of abrasion / breakage of tool X (negative reward)"

[0032] If tool X wears down to an amount not less than a predetermined reference value during machining using machine tool 1, a negative reward is imposed according to the degree of wear. A negative reward is also imposed if tool X breaks. "Reward 2: Occurrence of oscillation (negative reward)"

[0033] If vibration of a magnitude not less than a predetermined reference value occurs during machining using machine tool 1, a negative reward will be imposed according to the degree of the vibration. "Reward 3: Engine load (positive reward, negative reward)"

[0034] If the measured motor load increases within a range during machining, which is carried out after adjusting the previous machining conditions for machine tool 1, such that the motor load does not exceed a predetermined reference value (predetermined load), a positive reward is awarded according to its degree. If the predetermined reference value is exceeded or the motor load decreases, a negative reward is awarded.

[0035] The value function update unit 9 performs machine learning based on the state data acquired by the state observation unit 5 and the reward calculated by the reward calculation unit 8. The decision regarding the value function used for learning is made according to the learning algorithm being applied. For example, if Q-learning, which is one of the representative algorithms for learning by reinforcement, is used, learning is performed by updating an action value function Q(s). t , a t ) carried out according to Math. 1. Q(St,at)←Q(st,at)+a(rt+1+γmaxaQ(st+1,a)−Q(st,at))

[0036] Here, α represents a learning coefficient, γ represents a discount rate, and these are defined within the ranges 0 < α ≤ 1 and 0 < γ ≤ 1.

[0037] Decision-making unit 10 selects an action a t(an action with the highest value), with which the reward in the future (r t+1 + r t+2 + ...) in the present state s t The maximum is determined using a value function generated through past learning. For the purpose of learning progress, a random action is also selected during the learning process with a specific probability when choosing an action using the decision-making unit 10 (ε-greedy procedure).

[0038] Referring to the flowchart in Fig. Section 4 describes a machine learning process performed by the value function update unit 9.

[0039] First, when machine learning is started, the state observation unit 5 captures data relating to a machining state of the machine tool 1, and the determination data acquisition unit 6 captures determination data (step S1).

[0040] The value function update unit 9 gives a current state s t based on the processing status data recorded by the condition monitoring unit 5 and the determination data recorded by the determination data acquisition unit 6 (step S2).

[0041] Decision-making unit 10 selects an action a t based on a past learning outcome and the state s specified in step S2 t from (step S3).

[0042] Next, machine tool 1 executes the action selected in step S3 (step S4).

[0043] Then, the acquisition of state data using the state monitoring unit 5 and the acquisition of the determination data using the determination data acquisition unit 6 are carried out (step S5). At this stage, the state of the machine tool 1 has changed due to the action performed in step S4 with the time transition from a cycle time t to a cycle time t+1.

[0044] The reward calculation unit 8 calculates a reward r t+1 based on data from an assessment result recorded in step S5 (step S6).

[0045] Then the value function update unit 9 runs with machine learning based on the state s specified in step S2. t , the action selected in step S3 and the reward calculated in step S6 t+1 The process continues (step S7) and the procedures from step S2 are repeated.

[0046] In this way, the machine learning device 7 continues learning by repeating steps (1) to (5) as described above. It is possible to perform the learning to be adapted to a new environment through additional learning, even if the machine learning device 7 is placed in the new environment after learning has been completed in a previous environment. Thus, by means of an application for adapting the machining path and cutting depth, performed by the control unit 4, which controls the machine tool 1 for machining the workpiece W, it is possible to learn the adaptation of the machining path and the like in a short time by performing additional learning in a new environment, in addition to previous learning of the adaptation of the machining path and the like, even in an application for controlling a new machine tool 1 according to the invention.

[0047] If machine tool 1 causes tool X to rotate and move, causing workpiece W to rotate synchronously with the rotation and movement of tool X, and also performs machining operations such as milling, then in the case of a typical machining path, roughing is performed a multitude of times, for example three times, and finishing is performed once while maintaining a constant cutting depth with respect to workpiece W, as shown in Fig. 5 illustrates.

[0048] In this case, the engine load changes, as in Fig. Figure 7 illustrates that the amounts of contact between the tool X and the workpiece W at the start of cutting the workpiece W and at the end of cutting the workpiece W with the tool X are small during roughing, as shown in Fig. 6 illustrates.

[0049] In this case, however, performing the cutting at the time of starting and ending the cutting of workpiece W requires more time, and the cutting time as a whole is extended if a cutting depth is set such that the motor load during cutting does not exceed the maximum permissible load (predetermined load).

[0050] According to the embodiment, the machining path is adjusted in a direction in which the cutting load increases at the start of cutting and at the end of cutting when the cutting load is low, and the cutting depth thus increases at the start of cutting and at the end of cutting when the cutting load is low, as shown in Fig. Figure 8 illustrates this. That is, to obtain a constant cutting depth in the machining path in the prior art, the amounts of contact between the tool X and the workpiece W at the time of the start of cutting and at the time of the end of cutting are small, and the space in the motor load is more than sufficient.

[0051] Meanwhile, the embodiment employs the machining path, which increases the cutting depth at the start and end of the cutting process when the cutting load is low, as shown in Fig. Figure 9 illustrates this. That is, the control unit 4 controls the spindle 2 in the direction where the distance between the axis of rotation of the workpiece W, rotated by the workpiece turning mechanism 3, and the axis of rotation of the tool X, rotated by the spindle 2, is greater in the region of low motor load than in the region of high motor load. This has the advantage of stabilizing the fluctuation in the amount of cutting from the beginning to the end of the cutting process each time roughing is performed, increasing the amount of cutting in each roughing operation, reducing the overall cutting time, and thus improving machining efficiency, as shown in Figure 9. Fig. 10 illustrated.

[0052] There is an advantage in that it is possible to adjust the machining path in a direction where the motor load decreases, thereby limiting the occurrence of self-excited vibration due to a synchronization error between the rotation of the tool X and the rotation of the workpiece W by arranging a negative reward for the occurrence of vibration during machining.

[0053] Although the action-value function used in Q-Learning was presented as an example of a value function in this embodiment, a state-value function, such as that used in TD-Learning, which is a reinforcement learning algorithm, can be used instead. The state-value function presented in Math. 2 can be presented as an example. V(st)←V(st)+a[rt+1+γV(st+1)−V(st)]

[0054] In reinforcement learning, a large number of agents are connected to a network or the like to construct a system. Information concerning states s, actions a, rewards r, and the like is shared among the agents and used for learning within the agents. The corresponding agents can thus perform efficient learning through distributed learning by reinforcement, in which the corresponding agents carry out the learning taking into account the environments of other agents as well.According to the embodiment, it is still possible to ensure that the multitude of agents controlling a multitude of environments efficiently learn the adaptation of machining conditions, which is carried out by the control unit 4 that controls the machine tool 1 to machine the workpiece W, through distributed machine learning in a state in which the multitude of agents are connected via a network or the like.

[0055] Various methods are well known as algorithms for learning through reinforcement, such as Q-learning, a SARSA method, TD-learning and an AC method, and any one of these algorithms can be used as an algorithm that is applied in the implementation.

[0056] Although the embodiment describes as an example the case in which the control unit 4 optimizes the machining path by machine learning such that the cutting depth of the workpiece W, which is cut with the tool X in the region with the low motor load, is greater than the cutting depth in the region with the high motor load within a range such that the motor load does not exceed a maximum permissible load, the embodiment is not limited to this.

[0057] This means that the control unit 4 can control the spindle 2 without using machine learning in a direction where the distance between the axis of rotation of the workpiece W, rotated by the workpiece turning mechanism 3, and the axis of rotation of the tool X, rotated by the spindle 2, is greater in the region of low motor load than in the region of high motor load. The machining path can be optimized by repeating the machining process while modifying the machining path within a range such that the motor load does not exceed the maximum permissible load.

[0058] The shape of the motor load, which optimizes the machining path according to the embodiment, is not trapezoidal, as in Fig. 7 and Fig. Figure 10 illustrates, limits, and the embodiment can be applied to a motor load in any form, as shown in Fig. 11 and Fig. 12 illustrated. [List of reference symbols] 1 machine tool 2 spindles 3 Workpiece rotation mechanism 4 Control unit 5 Condition monitoring unit 7. Machine learning setup 8 Reward Calculation Unit (Unit for calculating the reward) 9 Value Function Update Unit (Processing Condition Adaptation Learning Unit) 10 Decision-making unit (processing condition adjustment unit) P Program W workpiece X Tool

Claims

[1] Machine tool (1), comprising: a spindle (2) which causes a tool (X) to rotate and move; a workpiece rotating mechanism (3) that causes a workpiece (W) to rotate; a control unit (4) which controls the spindle (2) and the workpiece rotating mechanism (3) according to instructions from a program; and a cutting load detection unit that detects a cutting load transmitted from the tool (X) to the workpiece (W), wherein the control unit (4) controls a cutting route such that a cutting depth of the workpiece (W) cut with the tool (X) in a region with a low cutting load is greater than the cutting depth in a region with a high cutting load within a range such that the cutting load detected by the cutting load detection unit does not exceed a predetermined load. [2] The machine tool (1) according to claim 1, wherein the control unit (4) controls the spindle (2) in a direction in which a distance between a rotation axis of the workpiece (X) rotated by the workpiece rotating mechanism (3) and a rotation axis of the tool (W) rotated by the spindle (2) is longer in the low cutting load region than in the high cutting load region. [3] Machine tool (1) according to claim 1 or 2, wherein the control unit (4) comprises a machine learning device (7) which performs machine learning of adjustment parameters for machining conditions and an adjustment amount of the adjustment parameters when cutting with the tool (W) on the workpiece (X), wherein the machine learning device (7) comprises: a state observation unit (5) which acquires state data which is information indicating a processing state during cutting, a reward calculation unit (8) that calculates a reward based on the state data and reward conditions, a machining condition adaptation learning unit (9) that performs machine learning of the adaptation of the machining conditions, and a machining condition adjustment unit (10) that decides, as an adjustment action, target parameters for adjusting the machining conditions and the adjustment amount based on a result of machine learning of the machining condition adjustment performed by the machining condition adjustment learning unit (9) and the state data, and adjusts the machining conditions during cutting based on the adjustment action, the machining condition adjustment learning unit (9) performs machine learning of the adjustment of the machining conditions as the adjustment action based on the state data acquired by the state observation unit (5) after a machining operation based on the output machining conditions after the adjustment and the reward calculated by the reward calculation unit (8), and as the reward conditions, a negative reward is arranged when a vibration occurs, a positive reward is arranged when the cutting performed with the tool (X) on the workpiece (W) increases the cutting load, without exceeding a predetermined load range, and a negative reward is provided when the cutting load decreases or the cutting load exceeds the predetermined load.

Images