Machine tool vibration suppression optimization method and system

Abstract

The invention discloses a machine tool vibration suppression optimization method and system. The optimization method comprises the steps that a control strategy is generated according to machining parameters; collecting vibration data of key parts of the numerical control machine tool; inputting the vibration data and the current working parameters into a vibration prediction model; and judging whether a vibration result predicted by the vibration prediction model meets machining requirements or not, and if not, adjusting machine tool working parameters in the control strategy by adopting a reinforcement learning algorithm. According to the method, the vibration is suppressed by adopting a method of combining the vibration prediction model and the reinforcement learning algorithm, so that real-time vibration prediction and self-adaptive control are realized, the vibration can be effectively suppressed, the machining precision can be remarkably improved, equipment wear and faults can be reduced, and the service life of a machine tool can be prolonged.

Description

Technical Field

[0001] The invention belongs to the field of numerically controlled machine tools, and in particular relates to a method and system for optimizing machine tool vibration suppression. Background Art

[0002] It is inevitable that CNC machine tools will vibrate during the workpiece cutting process. The vibration of machine tools is mainly caused by factors such as cutting force, structural rigidity of machine tool components and external environment. Machine tool vibration not only affects the surface quality of the workpiece, but may also cause tool damage and increase processing errors. Machine tool vibration suppression is an important means to improve processing accuracy, extend machine tool service life and ensure processing quality. It is of great significance in modern CNC machine tools and high-precision processing; however, traditional vibration suppression methods mainly rely on mechanical design and manual experience, and it is difficult to cope with complex and changeable vibration environments. Therefore, it is necessary to propose a new machine tool vibration suppression method. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is to provide a method and system for optimizing machine tool vibration suppression.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0005] A machine tool vibration suppression optimization method comprises the following steps:

[0006] S100, generating a control strategy according to the processing parameters, and generating a real-time control signal according to the control strategy to control the working parameters of the machine tool in each operation cycle;

[0007] S200, collecting vibration data of key parts of the CNC machine tool during its operation;

[0008] S300, inputting the vibration data and the current working parameters of the machine tool into a vibration prediction model to predict subsequent vibration results;

[0009] S400, judging whether the vibration result predicted by the vibration prediction model meets the processing requirements, if not, executing step S500, otherwise, returning to executing step S200;

[0010] S500: According to the vibration result predicted by the vibration prediction model, a reinforcement learning algorithm is used to adjust the machine tool working parameters in the control strategy.

[0011] Furthermore, the key parts of the machine tool include the spindle, tool, and worktable of the CNC machine tool; the working parameters include cutting speed, cutting acceleration, feed rate, cutting depth, spindle speed, tool type, and cutting route.

[0012] Furthermore, the method for collecting vibration data of key parts of a machine tool comprises the following steps:

[0013] S210, setting a vibration collection device at a key position of the machine tool to collect vibration signals at the key position of the machine tool;

[0014] S220, performing denoising processing on the vibration signal to eliminate high-frequency and low-frequency noise;

[0015] S230 . Extract frequency domain features and time domain features of the vibration signal to obtain vibration data.

[0016] Furthermore, the method for obtaining the vibration prediction model includes the following steps:

[0017] S310, determining a machine learning model or a deep learning model as a basic model of the vibration prediction model;

[0018] S320, acquiring historical vibration data of the machine tool, determining normal state data and abnormal state data in the historical vibration data according to the vibration range of the machine tool during normal operation, and marking the vibration data of the reverse point, the start point, and the stop point as well as the abnormal state data;

[0019] S330: Use the labeled vibration data to train the basic model to obtain a vibration prediction model.

[0020] Furthermore, the long short-term memory network LSTM is adopted as the basic model.

[0021] Furthermore, the reinforcement learning algorithm adopts a Q-learning algorithm, with the current working parameter values ​​and vibration data of the machine tool as the state S of the Q-learning algorithm, the adjustment of the working parameters of the machine tool as the action a of the Q-learning algorithm, and the reduction of machine tool vibration and / or the optimization of machine tool processing performance as the reward R of the Q-learning algorithm.

[0022] Furthermore, the method of adjusting the machine tool working parameters in the control strategy using the reinforcement learning algorithm includes the following steps:

[0023] S510, obtaining the current processing parameters and the vibration results predicted by the vibration prediction model, and expressing them as the current state s t ;

[0024] S520, select an action a as the current action a using the ε-greedy strategy according to the current state t ;

[0025] S530, adjusting the working parameters of the machine tool in the control strategy to execute the current action a t , and execute the current action a tGet real-time vibration data and processing effect as feedback after a predetermined time;

[0026] S540, calculating the reward value r according to the feedback result t+1 ;

[0027] S550: Update the Q value of the current state and action.

[0028] S560: Determine whether the preset termination condition is met. If the termination condition is met, return to step S200; otherwise, return to step S510. t After the update, adjust the control strategy again until the requirements are met.

[0029] Furthermore, in the step S540, action a is performed. t The reward value after t+1 The calculation formula is as follows:

[0030]

[0031] Among them, a v is the weighting coefficient of the vibration data; V(t) represents the vibration of the machine tool when performing action a t The value of the vibration data fed back; i is the index of the number of parameters involved in calculating the reward value, N is the total number of parameters involved in calculating the reward value, b i is the weighting coefficient of the i-th parameter, f i (t) is the i-th parameter in executing action a t The feedback value after the parameter is calculated. When the parameter is as large as possible, its weight coefficient is positive, and when the parameter is as small as possible, its weight coefficient is negative.

[0032] Furthermore, in the step S550, the formula for updating the Q value is as follows:

[0033]

[0034] Among them, Q′(s t ,a t ) is in state s t Next, perform action a t The Q value of the current action a t The expected cumulative reward value for this strategy; Q(s t ,a t ) represents the Q value before the update; α is the learning rate, which represents the weight of the influence of new information on the existing Q value, and its value range is: 0<α≤1; r t+1 To perform action a at the current time step t tThe immediate reward obtained after the decision is made; γ is the discount factor, which indicates the influence of future rewards on the current decision, and its value range is 0≤γ≤1; For the next state S t+1 All the optional actions a′ t+1 The action with the largest Q value.

[0035] A machine tool vibration suppression optimization system, comprising:

[0036] A control strategy module is used to generate a control strategy according to the processing parameters, and to generate a real-time control signal to the machine tool actuator according to the control strategy, so as to control the working parameters of the machine tool actuator in each operation cycle;

[0037] Data acquisition module, used to collect vibration data of key parts of the machine tool during its operation;

[0038] A vibration prediction module, wherein the vibration prediction module is trained using a machine learning model or a deep learning model and is used to predict vibration results of subsequent operation cycles based on vibration data and working parameters of the machine tool in the current operation cycle; and

[0039] The parameter optimization module is used to adjust the machine tool working parameters in the control strategy by using a reinforcement learning algorithm when the vibration result predicted by the vibration prediction module does not meet the processing requirements.

[0040] In the present invention, a method combining the advance prediction of the vibration prediction model and the optimization control strategy of the reinforcement learning algorithm is used to suppress vibration, and real-time vibration prediction and adaptive control are realized, which can effectively suppress vibration, significantly improve processing accuracy, reduce equipment wear and failure, and extend the service life of the machine tool. In addition, by continuously executing the reinforcement learning algorithm to optimize the control strategy during the processing, the control strategy can also be continuously optimized, and finally the machine tool can achieve the best processing performance under the condition of minimizing vibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0042] Figure 1 It is a flow chart of an embodiment of a method for optimizing machine tool vibration suppression according to the present invention.

[0043] Figure 2 Flowchart for collecting vibration data of key parts of machine tools.

[0044] Figure 3 Flowchart for obtaining vibration prediction model.

[0045] Figure 4 Flowchart for adjusting the control strategy using reinforcement learning algorithm.

[0046] Figure 5 It is a structural block diagram of an embodiment of a machine tool vibration suppression optimization system of the present invention.

[0047] The accompanying drawings in the specification are as follows:

[0048] Control strategy module-100; machine tool actuator-200; data acquisition module-300; vibration prediction module-400; parameter optimization module-500. DETAILED DESCRIPTION

[0049] The following describes the implementation methods of the present invention through specific examples. The illustrations provided in the following embodiments are only used to schematically illustrate the basic concept of the present invention. The following embodiments and features in the embodiments may be combined with each other without conflict.

[0050] At present, vibration suppression mainly adopts spindle vibration suppression technology, feed axis vibration suppression technology and tool vibration suppression technology.

[0051] There are three main methods for spindle vibration suppression technology, namely variable speed cutting, spindle bearing preload control and spindle system self-balancing control, but all of them have obvious disadvantages. Among them, variable speed cutting technology uses periodic and continuous changes in cutting speed to avoid unstable cutting areas, thereby suppressing cutting vibration. Its disadvantages are: a large number of cutting processing tests are required for the machining system to establish the system stability limit diagram. If any part of the spindle, tool, fixture, or workpiece in the machining system changes, the stability limit diagram will also change, so it is necessary to re-plan the vibration prediction data. Spindle bearing preload control uses the method of increasing preload at low speed and high torque and reducing preload at high speed and low torque. Its disadvantage is: it can only fix the bearing preload for a single application scenario. Spindle system self-balancing control uses a related balancing mechanism to self-balance the high-speed rotation of the spindle. Its disadvantages are: the balancing mechanism has a complex structure and is difficult to promote and apply. Generally, the balance of the spindle is only calibrated before machining.

[0052] The vibration suppression of the feed axis adopts the method of improving the structure of mechanical transmission components to suppress vibration, mainly using new linear motors, electromagnetic shielding of electrical cabinets and other devices and servo tuning. Its disadvantages are: it takes a lot of time and energy, and the economic benefits are not obvious.

[0053] Tool vibration suppression mainly adopts methods such as improving tool bar materials, optimizing fixtures, and optimizing process parameters. Its disadvantages are: it takes a lot of time and energy, and the economic benefits are not obvious.

[0054] See also Figure 1 , Figure 1This is a flow chart of an embodiment of a machine tool vibration suppression optimization method of the present invention. The machine tool vibration suppression optimization method of this embodiment includes the following steps:

[0055] S100, the CNC machine tool generates a control strategy according to the processing parameters, and generates a real-time control signal according to the control strategy to control the working parameters of the machine tool in each operation cycle. The processing parameters are generally the cutting speed, feed rate, cutting depth, processing route and other parameters pre-entered during processing.

[0056] S200, collects vibration data of key parts of the machine tool during its operation. Figure 2 , the method for collecting vibration data of key parts of a machine tool may include the following steps:

[0057] S210, arranging vibration collection devices such as acceleration sensors, displacement sensors and strain gauges at key parts of the machine tool to collect vibration signals of the key parts of the machine tool. The key parts of the machine tool generally include the main shaft, tool, workbench (including feed shaft) and other positions of the machine tool.

[0058] S220: Use a filter (such as Kalman filtering, bandpass filtering, etc.) to denoise the collected vibration signal to eliminate high-frequency and low-frequency noise.

[0059] S230, using Fourier transform (FFT), wavelet transform and other methods to extract frequency domain features and time domain features of the vibration signal to obtain vibration data.

[0060] S300, input the vibration data and the current working parameters of the machine tool into the vibration prediction model to predict the subsequent vibration results. The working parameters of the machine tool generally include cutting speed, cutting acceleration, feed rate, cutting depth, spindle speed, tool type, cutting route, etc. The operation cycle is just a cycle artificially divided for the convenience of control. The duration of the operation cycle can be selected according to actual needs. The control strategy can send multiple real-time control signals within an operation cycle. For example, 1ms can be used as an operation cycle. Of course, 0.1ms or other durations can also be selected as an operation cycle, as long as the final control effect can be achieved.

[0061] See also Figure 3 , the method of pre-training the vibration prediction model can include the following steps:

[0062] S310, determine a machine learning model (such as support vector machine SVM, random forest) or a deep learning model (such as convolutional neural network CNN, long short-term memory network LSTM) as the basic model of the vibration prediction model. In this embodiment, the long short-term memory network LSTM is used as the basic model.

[0063] S320. Obtain the historical vibration data of the machine tool, determine the normal state data and abnormal state data in the historical vibration data according to the vibration range of the machine tool during normal operation, and mark the vibration data and abnormal state data of the reverse point, start point and stop point. When processing a new workpiece, the vibration data can be obtained as the historical vibration data of the machine tool by trial cutting the workpiece. During the processing, large vibrations are generally generated at the reverse point, start point and stop point, so they are marked together with the abnormal state data during training, and the rest of the data is normal state data. Normal state data and abnormal state data are generally distinguished according to the historical vibration range. For example, if the historical data shows that the vibration value is between 0 and 1, the data value within this range is normal state data, and the data value outside this range is abnormal state.

[0064] S330, training the basic model using the marked vibration data to obtain a vibration prediction model. Of course, when the workpiece is formally processed, the vibration data and related working parameters in the processing process can also be used as historical vibration data to continue training the model after each cycle is completed, thereby continuously improving the prediction accuracy of the vibration prediction model.

[0065] S400, judging whether the vibration result predicted by the vibration prediction model meets the processing requirements, if not, executing step S500, otherwise, returning to executing step S200.

[0066] Due to the need to consider the processing cost, the number of trial cutting of the workpiece is relatively limited. Training the model only through the data of the trial cutting of the workpiece can only preliminarily meet the processing requirements, and cannot make the control strategy reach the optimal state. Therefore, in the initial processing stage of the workpiece, the S500 step is generally executed to continuously optimize the control strategy. As the number of workpiece processing times increases, the model is continuously trained by using the vibration data of each processing, so that the control strategy of the workpiece processing gradually becomes stable after continuous optimization. At this time, under normal conditions, optimization is generally no longer required, and it will directly return to execute the S200 step, and only monitor to intervene in time when unexpected situations occur during processing. Of course, if the performance of the machine tool changes due to a collision or other circumstances, or a new workpiece is replaced, it will need to be optimized again.

[0067] S500, according to the vibration result predicted by the vibration prediction model, a reinforcement learning algorithm (such as Q-learning, deep Q network DQN and other algorithms) is used to adjust the machine tool working parameters in the control strategy.

[0068] In this embodiment, the reinforcement learning algorithm adopts the Q-learning algorithm, and uses the current machine tool working parameter value and vibration data (including vibration data predicted by the vibration prediction model and vibration data fed back after the action is executed) as the state S of the Q-learning algorithm, and the adjustment of the machine tool working parameters as the action a of the Q-learning algorithm. For example, the parameter value of a certain working parameter can be stepped up or down as an action a, or the adjustment of the cutting route can be used as an action a. The reduction of machine tool vibration and the optimization of machine tool processing performance are used as the reward R of the Q-learning algorithm. Of course, only the reduction of machine tool vibration can be used as the reward R of the Q-learning algorithm.

[0069] See also Figure 4 , when the reinforcement learning algorithm adopts the Q-learning algorithm, this step may include the following sub-steps:

[0070] S510, obtaining the current processing parameters and the vibration results predicted by the vibration prediction model, and expressing them as the current state s t ; Therefore, the possible vibration situation can be processed in advance according to the vibration results predicted by the vibration prediction model, and the subsequent vibration can be reduced by adjusting the control strategy to achieve the effect of suppressing vibration.

[0071] S520, according to the current state, adopt the ∈-greedy strategy (i.e., the decaying greedy strategy) to select an action a as the current action a t , the formula is as follows:

[0072]

[0073] in, is the action with the largest Q value in the current state. ∈ is the probability of exploration, that is, random actions are selected with a probability of ∈ to explore new possibilities to prevent falling into local optimality, 0<∈<1. 1-∈ is the probability of utilization, that is, the action with the largest current Q value is selected with a probability of 1-∈ to further optimize the control strategy ∈-greedy strategy, and the value of ∈ can gradually decrease with the increase of the number of times.

[0074] S530, adjusting the working parameters of the machine tool in the control strategy to execute the current action a t , and execute the current action a t After a predetermined time, the real-time vibration data and processing effect are obtained as feedback results. For example, for locations with large vibrations due to the assembly and structure of the machine tool itself, the vibration can be reduced by reducing the cutting speed. If adjusting the cutting speed and other parameter values ​​still cannot solve the problem, the cutting route can be adjusted to avoid excessive vibration by bypassing the location.

[0075] S540, calculating the reward value r according to the feedback result t+1 Since the purpose of this embodiment is to suppress vibration, the feedback result will include vibration data. For example, the execution action a can be set. t The reward value after t+1 The calculation formula is as follows:

[0076] r t+1 =-V(t)

[0077] Among them, V(t) represents the time when the machine tool performs action a. t The value of the vibration data fed back after the machining process is completed; the value of the vibration data can be the vibration amplitude value, or the comprehensive value of vibration performance related parameters such as vibration acceleration and displacement. Of course, in the actual machining process, there may not only be requirements for vibration, but also certain requirements for machining efficiency and other machining effect performance. At this time, the machining efficiency and other machining effect performance can also be added to the reward value r t+1 At this time, execute action a t The reward value after t+1 The calculation formula is as follows:

[0078]

[0079] Among them, a v is the weighting coefficient of the vibration data value; i is the index of the number of parameters involved in calculating the bonus value, and i=0 means that the performance of processing efficiency and other processing effects is not considered. N is the total number of parameters involved in calculating the bonus value, b i is the weighting coefficient of the i-th parameter, f i (t) is the i-th parameter in executing action a t The feedback value after.

[0080] In this embodiment, the parameters involved in calculating the reward value include machine tool vibration performance parameters, processing efficiency and workpiece surface roughness; therefore, executing action a t The reward value after t+1 The calculation formula is as follows:

[0081] r t+1 =-a v V(t)+b1f1(t)+b2f2(t)

[0082] Among them, b1 is the weighted coefficient of machine tool processing efficiency, f1(t) is the weighted coefficient of machine tool processing efficiency when executing action a. t B2 is the weighted coefficient of the workpiece surface roughness, f2(t) is the processing efficiency after performing action a. t The surface roughness of the workpiece is fed back.

[0083] S550, update the Q value of the current state and action. When the Q-learning algorithm is executed for the first time, the Q value Q(s,a) of the state and action will be initialized to the initial value; usually the initial value can be set to 0 or a small random value. After that, the Q-learning algorithm will continue to update the Q value of the state and action until the Q value of the state and action tends to be stable. The updated Q value will also be used as the current action a in step S520. t The formula for updating the Q value is as follows:

[0084]

[0085] Among them, Q′(s t ,a t ) is in state s t Next, perform action a t The Q value of the current action a t The expected cumulative reward value for this strategy; Q(s t ,a t ) represents the Q value before the update; α is the learning rate, which represents the weight of the influence of new information on the existing Q value, and its value range is: 0<α≤1; r t+1 To perform action a at the current time step t t The immediate reward obtained after the decision is made; γ is the discount factor, which indicates the influence of future rewards on the current decision, and its value range is 0≤γ≤1; For the next state S t+1 All the optional actions a′ t+1 The action with the largest Q value.

[0086] S560, determine whether the preset termination conditions are met, and the termination conditions are that the system can converge to the optimal processing parameters (for example: when the change in Q value is less than a certain threshold, or the reward value is stable at a higher level within a number of steps, it can be determined that the system converges to the optimal processing parameters), minimize vibration, or meet other predetermined conditions (such as processing efficiency, surface roughness, etc.). If the termination conditions are met, it is considered that the control strategy has basically reached the optimal state and no further optimization is required, and the process returns to step S200 to continue monitoring the processing process. If the termination conditions are not met, it is considered that there is still room for further optimization of the control strategy, and the current state s t After the update, return to step S510 to adjust the control strategy again until the requirements are met.

[0087] In this embodiment, a method combining the advance prediction of the vibration prediction model and the optimization control strategy of the reinforcement learning algorithm is used to suppress vibration. The vibration prediction model can improve the prediction of the vibration of the machine tool, so that the control strategy can be optimized in advance by the reinforcement learning algorithm before a large vibration may occur, so as to avoid large vibration during the processing and affect the processing accuracy. In addition, by continuously executing the reinforcement learning algorithm to optimize the control strategy during the processing, the control strategy can also be continuously optimized, and finally the machine tool can achieve the best processing performance with the minimum vibration.

[0088] See also Figure 5 , Figure 5 The structure block diagram of an embodiment of the machine tool vibration suppression optimization system of the present invention. The machine tool vibration suppression optimization system of this embodiment includes a control strategy module 100, a data acquisition module 300, a vibration prediction module 400 and a parameter optimization module 500. The control strategy module 100 is used to generate a control strategy according to processing parameters, and generate a real-time control signal to the machine tool actuator 200 according to the control strategy, so as to control the working parameters of the machine tool actuator 200 in each operation cycle.

[0089] The data acquisition module 300 is used to collect vibration data of key parts of the machine tool in the machine tool actuator 200 during the operation of the machine tool. The data acquisition module 300 may include a vibration acquisition unit, a denoising processing unit and a feature extraction unit. The vibration acquisition unit is used to collect vibration signals of key parts of the machine tool such as the spindle, tool, workbench, etc. in the machine tool actuator 200, and may include vibration acquisition devices such as multiple acceleration sensors, displacement sensors and strain gauges set at key parts of the machine tool. The data collected by the above vibration acquisition devices can be transmitted to the denoising processing unit through the industrial Internet of Things. The denoising processing unit is used to use a filter (such as Kalman filtering, bandpass filtering, etc.) to denoise the collected vibration signal to eliminate high-frequency and low-frequency noise. The feature extraction unit is used to extract the frequency domain characteristics and time domain characteristics of the vibration signal by Fourier transform, wavelet transform and other methods to obtain vibration data.

[0090] The vibration prediction module 400 is used to predict the vibration results of subsequent operation cycles based on the vibration data and the working parameters of the machine tool in the current operation cycle. The vibration prediction module 400 can be obtained by training with a machine learning model or a deep learning model, for example, by executing steps S310 to S330 in the embodiment of the above-mentioned machine tool vibration suppression optimization method.

[0091] The parameter optimization module 500 is used to adjust the machine tool operating parameters in the control strategy using a reinforcement learning algorithm when the vibration result predicted by the vibration prediction module does not meet the processing requirements. For example, the machine tool operating parameters in the control strategy can be adjusted by executing steps S510 to S560 in the embodiment of the above-mentioned machine tool vibration suppression optimization method, thereby realizing adaptive control of the CNC machine tool.

[0092] The machine tool vibration suppression optimization system of this embodiment can effectively suppress vibration, significantly improve machining accuracy, reduce equipment wear and failure, and extend the service life of machine tools through real-time vibration prediction and adaptive control. In addition, through predictive maintenance, sudden failures and downtime can be reduced, and maintenance costs can be reduced.

[0093] The above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A machine tool vibration suppression optimization method, characterized in that: The following steps are involved: S100, generating a control strategy according to the processing parameters, and generating a real-time control signal according to the control strategy to control the working parameters of the machine tool in each operation cycle; S200, collecting vibration data of key parts of the CNC machine tool during its operation; S300, inputting the vibration data and the current working parameters of the machine tool into a vibration prediction model to predict subsequent vibration results; S400, judging whether the vibration result predicted by the vibration prediction model meets the processing requirements, if not, executing step S500, otherwise, returning to executing step S200; S500: According to the vibration result predicted by the vibration prediction model, a reinforcement learning algorithm is used to adjust the machine tool working parameters in the control strategy.

2. The machine tool vibration suppression optimization method according to claim 1, characterized in that: The key parts of the machine tool include the spindle, tool and worktable of the CNC machine tool; the working parameters include cutting speed, cutting acceleration, feed rate, cutting depth, spindle speed, tool type and cutting route.

3. The machine tool vibration suppression optimization method according to claim 1, characterized in that: The method for collecting vibration data of key parts of a machine tool includes the following steps: S210, setting a vibration collection device at a key position of the machine tool to collect vibration signals at the key position of the machine tool; S220, performing denoising processing on the vibration signal to eliminate high-frequency and low-frequency noise; S230 . Extract frequency domain features and time domain features of the vibration signal to obtain vibration data.

4. The machine tool vibration suppression optimization method according to claim 1, characterized in that: The method for obtaining the vibration prediction model includes the following steps: S310, determining a machine learning model or a deep learning model as a basic model of the vibration prediction model; S320, acquiring historical vibration data of the machine tool, determining normal state data and abnormal state data in the historical vibration data according to the vibration range of the machine tool during normal operation, and marking the vibration data of the reverse point, the start point, and the stop point as well as the abnormal state data; S330: Use the labeled vibration data to train the basic model to obtain a vibration prediction model.

5. The machine tool vibration suppression optimization method according to claim 1, characterized in that: The long short-term memory network LSTM is used as the basic model.

6. The machine tool vibration suppression optimization method according to any one of claims 1 to 5, characterized in that: The reinforcement learning algorithm adopts the Q-learning algorithm, with the current working parameter values ​​and vibration data of the machine tool as the state S of the Q-learning algorithm, the adjustment of the working parameters of the machine tool as the action a of the Q-learning algorithm, and the reduction of machine tool vibration and / or the optimization of machine tool processing performance as the reward R of the Q-learning algorithm.

7. The machine tool vibration suppression optimization method according to claim 6, characterized in that: The method of adjusting the machine tool working parameters in the control strategy using the reinforcement learning algorithm includes the following steps: S510, obtaining the current processing parameters and the vibration results predicted by the vibration prediction model, and expressing them as the current state s t ; S520, select an action a as the current action a using the ε-greedy strategy according to the current state t ; S530, adjusting the working parameters of the machine tool in the control strategy to execute the current action a t , and execute the current action a t Get real-time vibration data and processing effect as feedback after a predetermined time; S540, calculating the reward value r according to the feedback result t+1 ; S550: Update the Q value of the current state and action. S560: Determine whether the preset termination condition is met. If the termination condition is met, return to step S200; otherwise, return to step S510. t After the update, adjust the control strategy again until the requirements are met.

8. The machine tool vibration suppression optimization method according to claim 7, characterized in that: In the step S540, action a is performed. t The reward value after t+1 The calculation formula is as follows: Among them, a v is the weighting coefficient of the vibration data; V(t) represents the vibration of the machine tool when performing action a t The value of the vibration data fed back; i is the index of the number of parameters involved in calculating the reward value, N is the total number of parameters involved in calculating the reward value, b i is the weighting coefficient of the i-th parameter, f i (t) is the i-th parameter in executing action a t The feedback value after the parameter is calculated. When the parameter is as large as possible, its weight coefficient is positive, and when the parameter is as small as possible, its weight coefficient is negative.

9. The machine tool vibration suppression optimization method according to claim 8, characterized in that: In the step S550, the formula for updating the Q value is as follows: Among them, Q′(s t ,a t ) is in state s t Next, perform action a t The Q value of the current action a t The expected cumulative reward value for this strategy; Q(s t ,a t ) represents the Q value before the update; α is the learning rate, which represents the weight of the influence of new information on the existing Q value, and its value range is: 0<α≤1; r t+1 To perform action a at the current time step t t The immediate reward obtained after the decision is made; γ is the discount factor, which indicates the influence of future rewards on the current decision, and its value range is 0≤γ≤1; For the next state S t+1 All the optional actions a′ t+1 The action with the largest Q value.

10. A machine tool vibration suppression optimization system, characterized in that: include A control strategy module is used to generate a control strategy according to the processing parameters, and to generate a real-time control signal to the machine tool actuator according to the control strategy, so as to control the working parameters of the machine tool actuator in each operation cycle; Data acquisition module, used to collect vibration data of key parts of the machine tool during its operation; A vibration prediction module, wherein the vibration prediction module is trained using a machine learning model or a deep learning model and is used to predict vibration results of subsequent operation cycles based on vibration data and working parameters of the machine tool in the current operation cycle; as well as The parameter optimization module is used to adjust the machine tool working parameters in the control strategy by using a reinforcement learning algorithm when the vibration result predicted by the vibration prediction module does not meet the processing requirements.

Images