Processing method based on deep learning and multi-sensor fusion

Abstract

The invention relates to the technical field of machine tool intelligent machining, and discloses a machining method based on deep learning and multi-sensor fusion, which comprises the following steps: collecting data in the machining process of a numerical control machine tool, and preprocessing the data in the machining process of the numerical control machine tool to obtain a data set; inputting the obtained data set corresponding to the data acquired by the plurality of sensors into a Kalman filtering algorithm for Kalman recursive filtering processing to obtain fused actual observation data acquired by the sensors; inputting a data set corresponding to the fused actual observation data acquired by the sensor and the data acquired by the numerical control system into a depth Q network algorithm, and predicting to obtain an optimal tool path; a magnetic field is applied to the machined workpiece through the excitation module; and machining a machined workpiece model along the optimal tool path by adopting an ultrasonic tool high-frequency machining method. According to the invention, the surface quality of the to-be-processed workpiece can be obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent machining of machine tools, and in particular to a machining method based on deep learning and multi-sensor fusion. Background Art

[0002] At present, the product quality of the machining industry depends largely on the technical level and experience of the machine tool operator. Although the development of automation equipment and CNC technology has improved the machining accuracy and efficiency to a certain extent, many key links still require manual intervention. Operators need to adjust machining parameters such as cutting speed, feed speed and tool path based on experience to ensure the machining quality of the product. However, this method of relying on human experience is easily affected by many factors, including the operator's technical level, working status and external environment. This uncertainty leads to the instability of machining quality, increases the scrap rate and rework rate, and limits production efficiency and product consistency.

[0003] The application of deep learning technology in industrial manufacturing provides a new approach to solving the above problems. By introducing deep learning algorithms, a large amount of data in the machining process (such as cutting force, vibration signals, temperature changes, etc.) can be analyzed and processed in real time. Based on this data, the deep learning model can automatically adjust the machining parameters, optimize the tool path, and reduce human errors, thereby achieving stable and high-quality machining results. The deep learning algorithm can predict and adjust possible machining deviations in advance by learning a large amount of historical machining data to ensure that each product can achieve the best quality. This can not only significantly improve machining efficiency and product consistency, but also reduce the workload of operators and improve the intelligence level of the entire production line.

[0004] At the same time, during the machining process, the problem of excessive surface hardness and toughness of the processed material is often encountered, which will significantly reduce the processing quality and greatly shorten the service life of the tool. Excessive surface hardness and toughness of the material will increase the cutting force, causing increased tool wear, which will affect the processing accuracy and surface finish. In addition, the micro stress and micro defects that may exist inside the processed material will further affect the processing quality, causing internal stress concentration and potential failure risks of the product. Magnetic treatment of the workpiece by magnetic field can effectively relieve residual stress and repair micro defects by adjusting the magnetic domain structure inside the workpiece, thereby improving the processing performance of the material and the quality of the finished product.

[0005] When it comes to the processing of brittle and hard materials, traditional machining methods often cannot avoid surface defects. These materials are susceptible to problems such as overheating, uneven material removal, and microcracks during processing. Overheating can cause thermal stress inside the material, which in turn causes cracks and failures; uneven material removal can increase surface roughness and affect the performance of the product; and microcracks can reduce the fatigue resistance and life of the material. Ultrasonic tool processing technology can effectively reduce cutting force and processing heat, improve the uniformity of material removal, and reduce the formation of surface microcracks by applying high-frequency vibration during the processing. Ultrasonic vibration makes the cutting process smoother, reduces the fluctuation of cutting force and the accumulation of heat, thereby improving processing accuracy and surface quality. This technology has shown significant advantages in the field of high-precision and high-demand manufacturing, especially in applications involving brittle and hard materials, significantly improving processing quality and efficiency, and reducing manufacturing costs. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a processing method based on deep learning and multi-sensor fusion, which can improve the surface quality of the workpiece to be processed.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A processing method based on deep learning and multi-sensor fusion, comprising the following steps:

[0009] Collect data during the processing of a CNC machine tool, and pre-process the data during the processing of the CNC machine tool to obtain a data set, wherein the data during the processing of the CNC machine tool includes cutting parameters, machine tool status information, and processing results; the cutting parameters include cutting speed, feed speed, cutting depth, ultrasonic frequency, and tool path; the machine tool status parameters include vibration, temperature, power on / off status, and running time; wherein the cutting speed, feed speed, cutting depth, ultrasonic frequency, vibration, and temperature data are collected by sensors, and the tool path, power on / off status, running time, and processing results are recorded by a CNC system of the CNC machine tool;

[0010] The obtained data set corresponding to the data collected by several sensors is input into the Kalman filter algorithm for Kalman recursive filtering to obtain the actual observation data collected by the fused sensors;

[0011] The data set corresponding to the actual observation data collected by the fused sensor and the data collected by the CNC system is input into the deep Q network algorithm to predict the optimal tool path;

[0012] According to the optimal tool path, the control parameters of the CNC machine tool are adjusted to achieve the best cutting effect.

[0013] Furthermore, the method for preprocessing the data in the CNC machine tool processing process is:

[0014] Perform timestamp alignment on the data in the CNC machine tool processing process to obtain aligned data;

[0015] Perform outlier identification on the aligned data to obtain outliers, and process the outliers to obtain data after outlier processing;

[0016] The data after exception processing is labeled to obtain a data set.

[0017] Furthermore, a specific method for performing Kalman recursive filtering to obtain actual observation data collected by the fused sensor is as follows: the Kalman recursive filtering includes a prediction step and an update step; in the prediction step, a state equation is used to predict the next state of the dynamic system to obtain predicted state data, and the covariance of the prediction error is estimated to reflect the uncertainty of the state of the dynamic system; in the update step, the actual observation data provided by each sensor is compared with the predicted state data using the observation equation, and the difference between the predicted state data and the actual observation data is measured by calculating the observation residual.

[0018] Furthermore, the data set corresponding to the actual observation data collected by the fused sensor and the data collected by the CNC system is input into the deep Q network algorithm, and the specific method for predicting the optimal tool path is as follows:

[0019] Establish an environment model, the environment model includes a state space S, an action space A and a reward function R; the state space S includes the front tool position , target position g, cutting speed V, feed rate F and cutting depth D; the action space A includes the moving direction and step length of the tool; the reward function R is used to design the reward function according to the processing results.

[0020] Positive Reward: If the size, shape, surface roughness, etc. of the processing results meet the expected processing accuracy requirements; if the processing is completed in a relatively short time and the quality meets the requirements; if the surface finish of the workpiece after processing is high and there are no obvious defects (such as scratches and cracks), a high positive reward will be given.

[0021] Negative rewards: If the machining result exceeds the expected tolerance range, resulting in a disqualified workpiece, a negative reward is given; if the machining time is too long, even if the quality meets the requirements, a lower positive reward or negative reward is given to prevent inefficient path selection; if the workpiece surface has defects or is too rough, a negative reward is given to guide the algorithm to avoid selecting machining paths or parameters that will cause surface quality problems.

[0022] The Q value update formula is:

[0023]

[0024] in, : In state Next action The Q value represents the expected return after taking a certain action in the current state; : Learning rate, which determines the weighted ratio of the new and old Q values; : The reward obtained at time step t; : Discount factor, which represents the present value of future rewards, and its value ranges from 0 to 1; : Next state The maximum Q value under , which indicates the optimal expected return in the new state; : Indicates that the state The action taken when , that is, the current action; : represents the action at the next time step t+1, that is, the future action;

[0025] The loss function is:

[0026]

[0027] in, is the loss function, which measures the difference between the current Q network and the target Q network; is the parameter of the current Q network; are the parameters of the target Q network; is the expected value; is the target Q value, which represents the sum of current reward and future reward; is the output of the current Q value network.

[0028] Furthermore, the magnetic field-assisted ultrasonic machining method based on deep learning multi-sensor fusion of the present invention also includes the following steps:

[0029] Applying a magnetic field to the workpiece being processed through an excitation module;

[0030] The ultrasonic tool high frequency processing method is used to process the workpiece model along the optimal tool path, and then cooling treatment is performed;

[0031] Perform stress relief heat treatment on the cooled workpiece;

[0032] The surface of the workpiece is polished using an automated grinding and polishing process.

[0033] Furthermore, the excitation module includes a magnet, a magnet clamp, a fixed plate and an adjustment frame. The fixed plate is provided with an adjustment hole. The magnet clamp is connected to the fixed plate through the adjustment hole and can move on the fixed plate along the adjustment hole; one end of the adjustment frame is fixedly connected to the fixed plate, and the other end is connected to the adjustment mechanism. The adjustment mechanism is a servo motor and a ball screw mechanism, which can adjust the distance between the fixed plate and the machine tool spindle; the excitation module is installed on the machine tool body, and the magnet is fixed to both sides of the workpiece to be processed through the fixed plate. At the same time, the applied magnetic field orientation of the magnet is adjusted through the adjustment frame so that the workpiece to be processed is always located in the center of the magnetic field; the magnet clamp clamps the magnet, and the magnet is located on both sides of the workpiece to be processed.

[0034] Furthermore, the specific method of using ultrasonic tool high frequency processing method to process the workpiece model is as follows:

[0035] Determine the positioning and fixing method of the workpiece to prevent displacement during processing;

[0036] Setting the vibration frequency of the tool on the processing equipment, wherein the vibration frequency is 20 to 50 KHz;

[0037] Adjust the cutting speed to less than one-third of the critical cutting speed;

[0038] The tool moves along the optimal tool path, and the cutting direction is consistent with the tool vibration direction.

[0039] Furthermore, the specific method of performing stress relief heat treatment on the cooled workpiece is as follows:

[0040] The workpiece is heated to a stress relief heat treatment temperature by a heat treatment furnace or flame, and is kept warm for a preset heat preservation time, wherein the stress relief heat treatment temperature is 500-700° C. and the heat preservation time is 2-6 hours.

[0041] The present invention collects data during the processing of CNC machine tools, including cutting parameters, machine tool status information and processing results, pre-processes the data, and then uses a Kalman filter algorithm to effectively process and fuse multi-sensor data, uses a deep Q network algorithm to predict the optimal tool path, and dynamically adjusts the control parameters of the CNC machine tool according to the prediction results to achieve the optimal cutting effect; then a magnetic field is applied to the surface to be processed through an excitation module to make the distribution of hard points inside the workpiece material to be processed more uniform, and significantly reduce the residual stress inside the workpiece to be processed; then an ultrasonic tool is used to ultrasonically vibrate at a frequency of 20-50KHz, and cutting is performed in a high-speed vibration manner along the cutting direction, and the workpiece after processing is subjected to post-processing stress relief heat treatment at a temperature of 500-700°C and a stress relief heat treatment insulation time of 2-6h; finally, a replaceable grinding wheel is used to change the automated grinding and polishing process for applying grinding pressure to polish the workpiece after processing, so that its surface roughness can reach above submicron level.

[0042] Compared with the prior art, the present invention provides a processing method based on deep learning and multi-sensor fusion, which has the following beneficial effects:

[0043] (1) The present invention uses a machine learning algorithm to process multi-sensor data and a powerful recursive filtering algorithm specifically for accurate estimation of dynamic system states. During the process of machining a workpiece on a machine tool, the data provided by multiple sensors may contain noise and errors. The algorithm effectively fuses these sensor signals through two steps: prediction and update. The prediction step uses the state equation to predict the next state of the system and estimates the covariance of the prediction error to reflect the uncertainty of the system state; the update step combines the observed data and corrects the predicted state and error covariance through the gain matrix. Through this dynamic adjustment and correction, the algorithm can improve the accuracy and stability of state estimation, thereby providing more reliable system state information. This method not only improves the accuracy of the machining process, but also reduces the impact of sensor noise on system control, ensuring machining quality and efficiency.

[0044] (2) The present invention uses a deep Q network algorithm to optimize the machine tool tool path planning through reinforcement learning, which significantly improves the efficiency and accuracy of the surface processing of the workpiece being processed. The DQN algorithm continuously interacts with the environment by defining the state space, action space and reward function to learn the optimal action strategy. Specifically, DQN plans the optimal tool movement path based on the current tool position, processing target and machine tool operating parameters. The Q value update mechanism makes the decision of each step tend to the global optimum, thereby reducing tool wear and improving processing efficiency. In addition, DQN uses experience replay and target Q network to improve the stability of the learning process and the reliability of the strategy. In practical applications, after deploying the trained DQN model, the machine tool can adjust the tool path in real time, dynamically respond to changes in the processing process, and further optimize the processing parameters. This intelligent path planning method not only shortens the processing time and reduces production costs, but also improves the consistency and quality of the product.

[0045] (3) The present invention uses a magnetic field to perform magnetic treatment on the workpiece to be processed. The magnetic field reduces the plastic deformation inside the cold-rolled steel, and the hardness of the workpiece to be processed is reduced. At the same time, the greater the deformation of the workpiece to be processed, the higher the residual stress, and the more obvious the hardness reduction after magnetic treatment. Magnetic treatment can also make the distribution of hard points inside the workpiece material to be processed more uniform, significantly reducing the residual stress inside the workpiece to be processed. At the same time, the magnetic field can magnetize the processing tool to significantly increase the service life of the tool. Magnetic treatment processing technology can not only improve the physical and mechanical properties of the material during the processing process, but also further improve its comprehensive performance by pre-processing or post-processing the workpiece to be processed. In the pre-processing stage, the magnetic field can enhance the processing adaptability and stability of the workpiece to be processed by changing the grain structure and stress distribution of the material; in the post-processing stage, the magnetic field can improve the service life and reliability of the workpiece to be processed by eliminating residual stress and micro defects.

[0046] (4) The present invention also uses ultrasonic machining, which uses ultrasonic vibration energy to efficiently remove materials. It has shown significant advantages in many industrial applications. This technology can achieve high-precision and high-quality processing effects. Since ultrasonic machining can be performed without generating a heat-affected zone, thermal damage to the material can be minimized, thereby retaining the original physical and chemical properties of the material. In addition, this processing method has extremely high control over the material removal rate, so that the processing process can be accurate to the micron or even nanometer level, which is very suitable for the field of precision machining and micromachining. Ultrasonic machining also has good flexibility and adaptability. It can process a variety of hard materials. Compared with traditional processing methods, ultrasonic machining can achieve dry processing without the use of coolants and lubricants. It not only improves production efficiency and processing quality, but also reduces the cost of waste disposal and environmental impact.

[0047] (5) The present invention conducts process analysis and simulation tests on the workpiece to be processed, replaces the traditional grinding method with modern composite grinding and polishing, and replaces manual and semi-automatic operations with automation, thereby improving the surface quality of the workpiece to be processed. The obtained workpiece has no end face collapse and deformation defects, and optimizes the tool path through deep learning, magnetic field assistance and ultrasonic tool processing, and uses automated grinding and polishing technology to make the surface roughness reach above sub-micron level.

[0048] (6) In the present invention, tool path optimization, magnetic field assistance and ultrasonic machining technology each bring significant process improvements, and produce additional benefits through their synergistic effect. Tool path optimization can accurately control the movement of the tool and reduce error accumulation; magnetic field assistance reduces cutting force, making the tool movement smoother and reducing wear; ultrasonic machining reduces cutting resistance through vibration, making the entire machining process smoother. The combination of the three significantly reduces vibration and friction during machining, improving machining accuracy and surface quality. In addition, this synergistic effect also extends the service life of the tool. The optimized path and magnetic field assistance reduce thermal damage to the tool, while ultrasonic machining further reduces mechanical wear, thereby reducing the frequency of tool changes and improving production efficiency. By optimizing the cutting path and reducing friction and temperature fluctuations, the three together extend the life of the tool and reduce machining costs. The integration of these technologies also improves material removal efficiency, especially in hard material machining, where the synergistic effect greatly increases the removal rate. At the same time, the optimized machining path and reduced vibration and friction also optimize the energy consumption of the entire machining process, which not only reduces production costs but also reduces environmental impact. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic flow chart of the magnetic field assisted ultrasonic machining method of the present invention;

[0050] Figure 2 It is a three-dimensional structural schematic diagram of the excitation module in the present invention;

[0051] Figure 3 A schematic diagram of an ultrasonic tool machining a workpiece in the present invention (the excitation module and the grinding wheel are not shown in the figure);

[0052] Figure 4 This is a schematic diagram of the automated grinding and polishing process of the present invention (the excitation module is not shown in the figure).

[0053] The meanings of the reference numerals in the figure are as follows: 1-magnet; 2-magnet fixture; 3-fixed plate; 4-adjusting frame; 5-ultrasonic tool; 6-machined workpiece; 7-workpiece fixture; 8-grinding wheel; 9-machine tool body. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Unless otherwise specifically stated, the relative arrangement, numerical expressions and numerical values ​​of the parts and steps set forth in these embodiments do not limit the scope of the present invention. Meanwhile, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to actual proportional relationships. The technology, methods and equipment known to those of ordinary skill in the relevant art may not be discussed in detail, but in appropriate cases, the technology, methods and equipment should be considered as a part of the specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may also include different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, and therefore, once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0056] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the protection content of the present invention.

[0057] Example 1

[0058] like Figure 1 As shown, the processing method based on deep learning and multi-sensor fusion proposed in this embodiment includes the following steps:

[0059] Step S1: Collect data during the CNC machine tool processing process, which includes cutting parameters, machine tool status information and processing results, and pre-process the data to obtain a data set. Cutting parameters include cutting speed, feed speed, cutting depth, ultrasonic frequency and tool path; machine tool status parameters include vibration, temperature, power on / off status and running time. Among them, cutting speed, feed speed, cutting depth, ultrasonic frequency, vibration and temperature data are collected by sensors; tool path, power on / off status, running time and processing results are recorded by the CNC system of the CNC machine tool. Processing results include measuring the size, shape, accuracy, surface quality and processing efficiency of the workpiece.

[0060] In a specific implementation of this embodiment: In step S1, a number of sensors are installed on the CNC machine tool, and each sensor is used to capture the actual observation data during the processing in real time. The actual observation data includes cutting speed, feed speed, cutting depth, ultrasonic frequency, vibration and temperature. For example, the vibration sensor is used to collect actual vibration data, and the temperature sensor is used to collect actual temperature data. The sensor is connected to the CNC system of the CNC machine tool for communication, and is used to transmit the collected cutting speed, feed speed, cutting depth, ultrasonic frequency, vibration and temperature data to the CNC system of the CNC machine tool. At the same time, the CNC system of the CNC machine tool records the tool path, cutting parameters, machine status information and processing results. The CNC system preprocesses the data collected from each sensor and the CNC system to obtain a unified data set.

[0061] In a specific implementation of this embodiment, the data preprocessing method is as follows: (1) aligning the timestamps of all cutting parameters, machine tool status parameters and processing result data to ensure the time consistency from different data sources.

[0062] (2) The CNC system identifies outliers in the data after timestamp alignment and processes them, thus ensuring the integrity and accuracy of the data set.

[0063] Outliers refer to values ​​that deviate greatly from normal data. The way to deal with identified outliers is to directly delete them.

[0064] (3) Based on the actual processing results, the data after outlier processing is labeled and marked as target variables, which ensures that each label can accurately correspond to its relevant features.

[0065] It should be noted that the numerical control system in the numerical control machine tool of the present invention is a mature numerical control system in the prior art, and the numerical control system can be used for data collection, recording, data abnormal value identification, deletion of identified abnormal values, etc. The method of labeling the data after abnormal value processing can be calibrated manually, or the integrated module of the numerical control system can be used for labeling.

[0066] Step S2: Input the data set corresponding to the data collected by several sensors obtained in step S1 into the external host for Kalman recursive filtering to obtain the actual observation data collected by the fused sensors to improve the accuracy and stability of the machine tool state estimation.

[0067] In a specific implementation of the present embodiment, the Kalman recursive filtering algorithm is used to accurately estimate the state of the overall cutting dynamic process. The basic principle is to continuously correct the state of the dynamic system through two steps of prediction and update. First, in the prediction step, the state equation is used to predict the next state of the dynamic system to obtain the predicted state data, and the covariance of the prediction error is estimated at the same time to reflect the uncertainty of the dynamic system state. Next, in the update step, the actual observation data (i.e., observation value) provided by each sensor is compared with the predicted state data (i.e., predicted value) using the observation equation, and the difference between the predicted value and the observed value is measured by calculating the observation residual. The Kalman gain matrix plays a key role in this process. It determines how to weight the predicted state data and the actual observation data, thereby correcting the covariance of the predicted state data and the error. This method can effectively fuse the data collected by multiple sensors, and improve the accuracy and stability of the estimation by dynamically adjusting the overall cutting dynamic system state estimation, thereby providing more reliable state information for step S3.

[0068] In a specific implementation of this embodiment, the state equation describes the dynamic changes of the state of the overall cutting dynamic system, providing a basis for predicting the future state of the dynamic system. The state equation is as follows:

[0069]

[0070] in, is the state vector at time k, representing the current state of the system. is the state transfer matrix, which describes the change of state from the previous moment to the current moment. is the control input matrix, which describes the impact of the control input on the state. is the control input vector, which represents the external control input to the system. is the process noise, which represents the random disturbance of the system during the transfer process.

[0071] In a specific implementation of this embodiment, the observation equation links the system state with the actual observation data, so that the estimation of the system state can be corrected based on the actual observation data. The observation equation is as follows:

[0072]

[0073] in, is the observation vector at time k, which represents the observation data measured by the sensor. is the observation matrix, which describes the relationship between the state vector and the observation vector. is the observation noise, which represents the random error in the measurement process.

[0074] In a specific implementation of this embodiment, the prediction step is divided into state prediction and error covariance prediction, where the state prediction is used to predict the state at the current moment, which is based on the state and control input at the previous moment. The state prediction equation is as follows:

[0075]

[0076] in, is the predicted state vector, which represents the prediction of the state before time k. is the state vector at the previous moment.

[0077] In a specific implementation of this embodiment, the error covariance prediction formula calculates the covariance matrix of the predicted state to evaluate the uncertainty of the prediction. The error covariance prediction formula is as follows:

[0078]

[0079] in, is the predicted error covariance matrix, which measures the uncertainty of the predicted state. : is the error covariance matrix of the previous moment. is the process noise covariance matrix. is the state transfer matrix The transpose of .

[0080] In a specific implementation of this embodiment, the update step includes Kalman gain, state update and error covariance update. The Kalman gain matrix determines how to weight the predicted state and the actual observed data to obtain the best state estimate:

[0081]

[0082] in, is the Kalman gain matrix, which represents the influence weight of observation data on state estimation. is the observation noise covariance matrix. is the observation matrix The transpose of .

[0083] The state update corrects the predicted state by weighting the observation residual to obtain a more accurate state estimate:

[0084]

[0085] in, : Updated state estimate. : Observation residual, which represents the difference between the actual observed value and the predicted observed value.

[0086] In a specific implementation of this embodiment, the error covariance update updates the uncertainty of the state estimate, reflecting the reduction of the uncertainty of the state estimate due to the new actual observation data:

[0087]

[0088] in, : Updated error covariance matrix.

[0089] The data from vibration sensors, temperature sensors and force sensors are processed by Kalman filtering, and multi-sensor data is fused to obtain a more accurate state estimate. The fused data is used to subsequently adjust the processing parameters and tool paths in real time to optimize the processing process.

[0090] Step S3: Input the actual observation data collected by the fused sensor obtained in step S2 and the data set recorded by the numerical control system obtained in step S1 into the deep Q network algorithm to predict the optimal tool path.

[0091] In a specific implementation of this embodiment, a deep Q network (DQN) algorithm is used to optimize the decision-making strategy through reinforcement learning. The core is to use the Q value update mechanism to learn and estimate the expected benefits of taking a specific action in a given state. Specifically, the DQN algorithm gradually approaches the true action value by continuously updating the Q value table. It interacts with the environment to obtain the immediate reward of the current state and action, and updates the Q value in combination with the maximum possible reward in the future. This process uses a technique called experience replay to randomly sample past experience to break data correlation, thereby improving the stability of learning. In addition, DQN introduces a target Q network to generate a more stable target value, thereby further reducing instability during training. The loss function is used to measure the difference between the predicted Q value and the target Q value. By optimizing the loss function and adjusting the network parameters, the strategy continues to approach the optimal.

[0092] First, the environment model is established, which is divided into state space S, action space A and reward function R. The state space S includes: the current tool position , target position g, cutting speed V, feed rate F and cutting depth D. The action space A includes the moving direction and step length of the tool, such as the step length of up and down, left and right, and forward and backward movement. Reward function R: The reward function is designed according to the processing efficiency and processing quality. For example, a positive reward is obtained when the target position is approached quickly, and a negative reward is obtained when it deviates from the target or the processing quality decreases.

[0093] Specifically, the Q-value update formula adjusts the current Q-value to reflect new information. Each update takes into account the weighted sum of the reward of the current action and the possible future rewards, so that the Q-value gradually approaches the actual reward:

[0094]

[0095] in, : In state Next action The Q value represents the expected return after taking a certain action in the current state; : Learning rate, which determines the weighted ratio of the new and old Q values; : The reward obtained at time step t; : Discount factor, which represents the present value of future rewards, and its value ranges from 0 to 1; : Next state The maximum Q value under , which indicates the optimal expected return in the new state; : Indicates that the state The action taken when , that is, the current action; : represents the action at the next time step t+1, that is, the future action.

[0096] Furthermore, the loss function is used to guide the adjustment of network parameters. By minimizing the loss function, the output of the current Q network can be made closer to the actual return, improving the accuracy of the strategy:

[0097]

[0098] in, is the loss function, which measures the difference between the current Q network and the target Q network; is the parameter of the current Q network; are the parameters of the target Q network; is the expected value; is the target Q value, which represents the sum of current reward and future reward; is the output of the current Q value network.

[0099] Step S4: applying a magnetic field to the workpiece 6 through the excitation module.

[0100] Specifically, Figure 2 As shown, Figure 2The red arrow in the middle indicates the direction of the magnetic field (the direction of the arrow is only for reference, and it changes with the change of the direction of the magnetic field). In step S4, the excitation module includes a magnet 1, a magnet fixture 2, a fixed plate 3 and an adjustment frame 4. The fixed plate 3 is provided with an adjustment hole. The magnet fixture 2 is connected to the fixed plate 3 through the adjustment hole and can move on the fixed plate 3 along the adjustment hole. One end of the adjustment frame 4 is fixedly connected to the fixed plate 3, and the other end is connected to the adjustment mechanism. The adjustment mechanism is a servo motor and a ball screw mechanism, which can adjust the distance between the fixed plate 3 and the machine tool spindle. The excitation module is installed on the machine tool body 9. The magnet 1 is fixed to both sides of the workpiece 6 to be processed by using the fixed plate 3. At the same time, the adjustment frame 4 can be used to adjust the specific magnetic field orientation of the magnet 1 so that the workpiece 6 to be processed is always located in the center of the magnetic field. The magnet fixture 2 clamps the magnet 1 on both sides of the workpiece 6 to be processed. The magnet fixture 2 is fixed to the lower part of the fixed plate 3 by screws and nuts. The distance between the magnet 1 and the workpiece 6 to be processed can be adjusted through the adjustment hole on the fixed plate 3. The fixed plate 3 and the adjustment frame 4 are fixedly connected, and the adjustment frame 4 can be adjusted forward and backward to ensure the freedom of the magnet 1 in the forward and backward direction. The magnet 1 can be a permanent magnet or an electromagnet, and a magnetic field of 0.3 Tesla must be formed between the workpiece 6 to be processed. The distance between the magnet 1 and the workpiece 6 to be processed is controlled by a high-precision servo motor and a ball screw pair on the adjustment frame 4 to adjust the size of the magnetic field.

[0101] Step S5: using the ultrasonic tool 5 to process a model of the workpiece 6 in a high frequency processing method, and then performing a cooling treatment.

[0102] like Figure 3 As shown, the workpiece 6 to be processed is first positioned and fixed by the workpiece fixture 7 to prevent displacement during the processing, and then the vibration frequency of the ultrasonic tool 5 is set on the processing equipment to ensure that the vibration frequency is within the range of 20-50KHz. This vibration frequency range can effectively improve the processing efficiency and surface quality while reducing material damage. According to the material of the workpiece 6 to be processed and the expected processing results, the cutting speed is adjusted to be less than one-third of the critical cutting speed to obtain a good ultrasonic vibration cutting effect. The ultrasonic tool 6 moves along the optimal tool path, and the cutting direction should be consistent with the vibration direction of the ultrasonic tool 6 to ensure the processing effect and accuracy.

[0103] Step S6: performing stress relief heat treatment on the cooled workpiece 6.

[0104] Specifically, the stress relief heat treatment temperature is 500-700°C, and the stress relief heat treatment insulation time is 2-6h; the material is heated to a certain temperature using a heat treatment furnace or flame and kept warm for a period of time to relax the residual stress inside the material.

[0105] Step S7: polishing the workpiece 6 using an automated grinding and polishing process.

[0106] Specifically, the automatic grinding and polishing CNC lathe is equipped with a grinding wheel 8, a polishing wheel and a tool magazine, which can be automatically replaced with the corresponding grinding wheel 8 and polishing wheel according to the process requirements. Before the grinding wheel 8 and the polishing wheel are replaced, the grinding liquid or polishing liquid is automatically added according to the requirements. When the grinding wheel 8 and the polishing wheel need to be replaced, the cylinder clamp removes the chuck and moves to the installation position to complete the replacement. The polishing wheel completes the grinding and polishing of the parts according to the program. At the same time, the synchronous pulley speed increase design is adopted, and the maximum speed of the spindle will reach 5000r / min, which can better adapt to the changes in the aperture size and have a wider speed regulation range. At the same time, in order to improve the grinding efficiency, when the polishing wheel contacts the workpiece 6 to be processed, the pressure of the contact surface can be accurately controlled by the pressure of the cylinder to control the pressure, and then the pressure value with higher efficiency and better polishing effect can be found through experimental verification. In order to improve efficiency, it is planned to use two different polishing wheels for fine grinding and polishing, and process them in sequence to achieve the final requirements. At the same time, in order to avoid the particles after grinding scratching the surface of the workpiece 6 to be processed, after fine grinding, the surface of the workpiece 6 to be processed is cleaned with a cleaning liquid, and then polished. In the process of polishing, additional polishing liquid can also be added to achieve better results.

[0107] It should be noted that the automated grinding and polishing CNC lathe and the grinding wheel 8, polishing wheel, and tool magazine configured thereon are all existing equipment or structures, and the present invention does not improve them.

[0108] Kalman filtering is a powerful recursive filtering algorithm used for accurate estimation of dynamic system states. During the process of machining workpieces 6 by machine tools, the data provided by multiple sensors may contain noise and errors. Kalman filtering effectively fuses these sensor signals through two steps: prediction and update. The prediction step uses the state equation to predict the next state of the system, and estimates the covariance of the prediction error to reflect the uncertainty of the system state; the update step combines the observed data and corrects the predicted state and error covariance through the Kalman gain matrix. Through this dynamic adjustment and correction, Kalman filtering can improve the accuracy and stability of state estimation, thereby providing more reliable system state information. This method not only improves the accuracy of the machining process, but also reduces the impact of sensor noise on system control, ensuring machining quality and efficiency.

[0109] The Deep Q Network (DQN) algorithm optimizes the tool path planning of machine tools through reinforcement learning, significantly improving the efficiency and accuracy of the workpiece being machined. The DQN algorithm continuously interacts with the environment by defining the state space, action space, and reward function to learn the optimal action strategy. Specifically, DQN plans the optimal tool movement path based on the current tool position, processing target, and machine tool operating parameters. The Q value update mechanism makes the decision at each step tend to the global optimum, thereby reducing tool wear and improving processing efficiency. In addition, DQN uses experience replay and target Q networks to improve the stability of the learning process and the reliability of the strategy. In actual applications, after deploying the trained DQN model, the machine tool can adjust the tool path in real time, dynamically respond to changes in the processing process, and further optimize the processing parameters. This intelligent path planning method not only shortens processing time and reduces production costs, but also improves product consistency and quality.

[0110] After the workpiece is treated with a magnetic field, the magnetic field reduces the plastic deformation inside the cold-rolled steel, and the hardness of the workpiece is reduced. At the same time, the greater the deformation of the workpiece, the higher the residual stress, and the more obvious the hardness reduction after magnetic treatment. Magnetic treatment can also make the distribution of hard points inside the workpiece material more uniform, significantly reduce the residual stress inside the workpiece, and the magnetic field can magnetize the processing tool, significantly improving the service life of the tool.

[0111] Ultrasonic machining uses ultrasonic vibration energy to efficiently remove materials, and it has shown significant advantages in many industrial applications. This technology can achieve high-precision and high-quality machining results. Since ultrasonic machining can be performed without generating a heat-affected zone, thermal damage to the material can be minimized, thereby retaining the original physical and chemical properties of the material. In addition, this machining method has extremely high control over the material removal rate, allowing the machining process to be accurate to the micron or even nanometer level, which is very suitable for precision machining and micromachining.

[0112] Ultrasonic machining also has good flexibility and adaptability. It can process various hard materials. Compared with traditional machining methods, ultrasonic machining can achieve dry machining without the use of coolant and lubricant, which not only improves production efficiency and machining quality, but also reduces the cost of waste disposal and environmental impact.

[0113] Example 2

[0114] The difference between Example 2 and Example 1 is that the valve body is specifically selected as the processing target of this example, and the specific method of each step is the same.

[0115] The magnetic field-assisted ultrasonic machining method based on deep learning multi-sensor fusion of this embodiment includes the following steps:

[0116] Step S1: Collect data during the processing of a CNC machine tool, and preprocess the data during the processing of the CNC machine tool to obtain a data set, wherein the data during the processing of the CNC machine tool includes cutting parameters, machine tool status information, and processing results; the cutting parameters include cutting speed, feed speed, cutting depth, ultrasonic frequency, and tool path; the machine tool status parameters include vibration, temperature, power on / off status, and running time; wherein the cutting speed, feed speed, cutting depth, ultrasonic frequency, vibration, and temperature data are collected by sensors, and the tool path, power on / off status, running time, and processing results are recorded by the CNC system of the CNC machine tool.

[0117] Step S2: Input the obtained data set corresponding to the data collected by the multiple sensors into the Kalman filter algorithm for Kalman recursive filtering to obtain the actual observation data collected by the fused sensors.

[0118] Step S3: Input the data set corresponding to the actual observation data collected by the fused sensor and the data collected by the CNC system into the deep Q network algorithm to predict the optimal tool path.

[0119] Step S4: applying a magnetic field to the material to be processed through an excitation module.

[0120] Step S5: using an ultrasonic tool high frequency processing method to process the valve body model, and then performing a cooling process.

[0121] Step S6: performing post-processing stress relief heat treatment on the cooled valve sealing surface.

[0122] Step S7: polishing the valve body sealing surface after the stress relief heat treatment using an automated grinding and polishing process.

[0123] This invention improves the processing quality, work efficiency and equipment automation of the valve body through the research and analysis of the magnetic field-assisted ultrasonic processing method based on deep learning multi-sensor fusion. This new processing technology has brought significant improvements to the production field of machine tool processing. Its introduction will have a profound impact on traditional processing methods, achieve a perfect balance between output, quality and efficiency, and significantly enhance the ability to innovate key technologies for intelligent processing of machine tools. With the industrial application of this process, it will promote the acceleration of the intelligent transformation of the manufacturing industry and promote the development of intelligent processing technology from single to diversified, thereby realizing the transformation from rigidity to flexibility. This not only marks the optimization of the production process, but also represents the active exploration and innovation of future manufacturing models.

[0124] It should be noted that, in this application, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or equipment including the elements.

[0125] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A processing method based on deep learning and multi-sensor fusion, characterized in that: The following steps are involved: Collect data during the processing of a CNC machine tool, and pre-process the data during the processing of the CNC machine tool to obtain a data set, wherein the data during the processing of the CNC machine tool includes cutting parameters, machine tool status information, and processing results; the cutting parameters include cutting speed, feed speed, cutting depth, ultrasonic frequency, and tool path; the machine tool status parameters include vibration, temperature, power on / off status, and running time; wherein the cutting speed, feed speed, cutting depth, ultrasonic frequency, vibration, and temperature data are collected by sensors, and the tool path, power on / off status, running time, and processing results are recorded by a CNC system of the CNC machine tool; The obtained data set corresponding to the data collected by several sensors is input into the Kalman filter algorithm for Kalman recursive filtering to obtain the actual observation data collected by the fused sensors; The data sets corresponding to the actual observation data collected by the fused sensor and the data collected by the CNC system are input into the deep Q-network algorithm to predict the optimal tool path.

2. The processing method based on deep learning and multi-sensor fusion according to claim 1, characterized in that: The method for preprocessing data in the CNC machine tool processing process is: Perform timestamp alignment on the data in the CNC machine tool processing process to obtain aligned data; Perform outlier identification on the aligned data to obtain outliers, and process the outliers to obtain data after outlier processing; The data after exception processing is labeled to obtain a data set.

3. The processing method based on deep learning and multi-sensor fusion according to claim 1, characterized in that: The specific method of performing Kalman recursive filtering processing to obtain the actual observation data collected by the fused sensor is as follows: the Kalman recursive filtering processing includes a prediction step and an update step; in the prediction step, the state equation is used to predict the next state of the dynamic system to obtain the predicted state data, and the covariance of the prediction error is estimated to reflect the uncertainty of the state of the dynamic system; In the update step, the actual observation data provided by each sensor is compared with the predicted state data using the observation equation, and the difference between the predicted state data and the actual observation data is measured by calculating the observation residual.

4. The processing method based on deep learning and multi-sensor fusion according to claim 1, characterized in that: The data set corresponding to the actual observation data collected by the fused sensor and the data collected by the CNC system is input into the deep Q network algorithm to predict the optimal tool path as follows: Establish an environment model, the environment model includes a state space S, an action space A and a reward function R; the state space S includes the front tool position , target position g, cutting speed V, feed rate F and cutting depth D; the action space A includes the moving direction and step length of the tool; the reward function R is used to design a reward function according to the processing result; The Q value update formula is: ； in, : In state Next action The Q value represents the expected return after taking a certain action in the current state; : Learning rate, which determines the weighted ratio of the new and old Q values; : The reward obtained at time step t; : Discount factor, which represents the present value of future rewards, and its value ranges from 0 to 1; : Next state The maximum Q value under , which indicates the optimal expected return in the new state; : Indicates that the state The action taken when , that is, the current action; : represents the action at the next time step t+1, that is, the future action; The loss function is: ； in, is the loss function, which measures the difference between the current Q network and the target Q network; is the parameter of the current Q network; are the parameters of the target Q network; is the expected value; is the target Q value, which represents the sum of current reward and future reward; is the output of the current Q value network.

5. The processing method based on deep learning and multi-sensor fusion according to claim 1, characterized in that: The following steps are also included: Applying a magnetic field to the workpiece being processed through an excitation module; The ultrasonic tool high frequency processing method is used to process the workpiece model along the optimal tool path, and then cooling treatment is performed; Perform stress relief heat treatment on the cooled workpiece; The surface of the workpiece is polished using an automated grinding and polishing process.

6. The processing method based on deep learning and multi-sensor fusion according to claim 5, characterized in that: The excitation module includes a magnet, a magnet fixture, a fixed plate and an adjustment frame. The fixed plate is provided with an adjustment hole. The magnet fixture is connected to the fixed plate through the adjustment hole and can move on the fixed plate along the adjustment hole. One end of the adjustment frame is fixedly connected to the fixed plate, and the other end is connected to the adjustment mechanism. The adjustment mechanism is a servo motor and a ball screw mechanism, which can adjust the distance between the fixed plate and the main shaft of the machine tool. The excitation module is installed on the main body of the machine tool, and the magnet is fixed to both sides of the workpiece to be processed through the fixed plate. At the same time, the orientation of the applied magnetic field of the magnet is adjusted through the adjustment frame, so that the workpiece to be processed is always located in the center of the magnetic field. The magnet fixture holds the magnet, and the magnet is located on both sides of the workpiece being processed.

7. The processing method based on deep learning and multi-sensor fusion according to claim 5, characterized in that: The specific method of using ultrasonic tool high frequency processing method to process the workpiece model is as follows: Determine the positioning and fixing method of the workpiece to prevent displacement during processing; Setting the vibration frequency of the tool on the processing equipment, wherein the vibration frequency is 20 to 50 KHz; Adjust the cutting speed to less than one-third of the critical cutting speed; The tool moves along the optimal tool path, and the cutting direction is consistent with the tool vibration direction.

8. The processing method based on deep learning and multi-sensor fusion according to claim 5, characterized in that: The specific method of stress relief heat treatment for the cooled workpiece is: The workpiece is heated to a stress relief heat treatment temperature by a heat treatment furnace or flame, and is kept warm for a preset heat preservation time, wherein the stress relief heat treatment temperature is 500-700° C. and the heat preservation time is 2-6 hours.

Images