An on-line detection and control device and control method for grinding force

Abstract

The present application discloses an online detection and control device and control method for grinding force, which relates to the field of precision grinding force measurement technology. A spring chuck is arranged in a receiving cavity, and the outer wall of the detection shaft section is a tapered conical surface; a threaded cover is connected to the connecting shaft section to limit the axial position of the spring chuck, a mounting handle is provided with a first conical cavity, a threaded cover is provided with an avoidance hole, and a plurality of piezoelectric sensors are installed on the detection shaft section. The grinding wheel spindle acts on the plurality of piezoelectric sensors through the detection shaft section. The piezoelectric sensors are used to detect tangential force, normal force and axial force. The signal amplification and processing module can convert the charge signal output by the piezoelectric sensor into a digital signal and output it to the terminal. The terminal can control the logic control grinding head to maintain constant force grinding with a preset grinding force target value. The present application realizes all-round real-time detection and control of grinding force, improves accuracy and real-time performance, and also provides a new control method for low-damage processing of hard and brittle materials.

Description

Technical Field

[0001] The present application relates to the field of precision grinding force measurement technology, and in particular to an online grinding force detection and control device. The present application also relates to a grinding force control method. Background Art

[0002] In the field of precision manufacturing, hard and brittle materials (such as silicon carbide, silicon nitride, and optical glass) are widely used in high-end fields such as aerospace, semiconductor devices, and optical components due to their high hardness, high-temperature resistance, and excellent chemical stability. During the grinding process, these hard and brittle materials often experience surface damage in the form of cracks, pits, and plastic scratches, as well as subsurface damage in the form of subsurface cracks, which in turn affects the surface quality and product performance.

[0003] When grinding hard and brittle materials, the magnitude of the grinding force has an important influence on the processing quality. Excessive grinding force can cause serious damage such as cracks and pits on the material surface. At the same time, subsurface cracks and micro damage will deepen, weakening the mechanical properties of the material. In addition, although a small grinding force means a smaller removal amount and it is easier to obtain a smooth surface, too small a grinding force may lead to low processing efficiency, and the grinding wheel is prone to passivation, increasing the dressing cost. It can be seen that controlling the grinding force is the key to ensuring processing quality. The grinding force of the grinding wheel can be divided into the normal force F n With tangential force F t , a larger tangential force can increase the material removal rate, but may cause surface crack damage, reduce surface finish and precision. A larger normal force, on the other hand, intensifies the stress concentration on the subsurface, causing subsurface cracks and plastic deformation, weakening the mechanical properties of the material, and accelerating the wear of the grinding wheel. Therefore, controlling the balance between the tangential force and the normal force during grinding is crucial to optimizing grinding quality and efficiency. According to the formula for the maximum undeformed chip thickness, it can be concluded that by increasing the grinding wheel speed and reducing the grinding feed rate, the maximum undeformed chip thickness can be reduced, thereby ensuring that the number of cracks and pits on the grinding surface is reduced.

[0004] According to the theory of median / lateral crack systems, grinding normal force plays a major role in influencing subsurface cracks in materials. Subsurface cracks penetrate deep into the material, reducing its performance. Tangential force, on the other hand, increases material fragmentation, which improves grinding efficiency to a certain extent. However, excessive tangential force accelerates dislocations and slip, leading to dislocation accumulation at grain boundaries and second-phase particles, which in turn creates conditions for crack nucleation and accelerates damage. If the tangential force is too low, dislocation activity is insufficient, and energy may be released directly through the cracks, forming a large number of microcracks.

[0005] Currently, existing technologies lack structural and theoretical support for detecting and controlling grinding forces during grinding of hard and brittle materials. This technology has at least the following significant limitations:

[0006] (1) Traditional grinding force detection mostly uses single-axis force sensors (such as strain gauge sensors) or indirect estimation methods (such as spindle current monitoring). Such solutions can only obtain the force component in a single direction and cannot fully reflect the dynamic distribution of the three-dimensional grinding force. For example, some existing radial force detection devices can measure normal force, but are sensitive to the coupling effect of tangential force, resulting in high decoupling errors. In addition, the response frequency of strain gauge sensors is usually lower than 1kHz, making it difficult to capture microsecond force fluctuations during the grinding process, resulting in control signal lag and failure to meet the requirements of ultra-precision machining.

[0007] (2) Existing grinding force control is mostly based on fixed-parameter PID algorithms, whose proportional, integral, and differential coefficients need to be repeatedly debugged based on manual experience. However, the processing window for hard and brittle materials is extremely narrow. Once the PID parameters are fixed, they are difficult to adapt to dynamic changes such as grinding wheel wear and material batch differences. Relevant literature indicates that after one hour of continuous processing, the grinding force drift caused by grinding wheel passivation can reach 30% with traditional PID control, forcing frequent shutdowns for grinding wheel dressing, which significantly reduces processing efficiency.

[0008] (3) Although the axial force is relatively small in surface grinding, its influence is often ignored by existing technologies. In actual working conditions, the runout of the grinding wheel spindle or the tilt of the workpiece clamping can cause abnormal axial force. For example, in actual applications, during the grinding process of sapphire substrates, a sudden increase in axial force may cause the workpiece to "warp" or the grinding wheel axis to resonate, exacerbating the surface waviness. However, due to the limitations of sensor layout, traditional devices cannot effectively monitor axial force and can only detect defects through post-workpiece inspection, resulting in waste of raw materials. Summary of the Invention

[0009] The present application provides an online detection and control device for grinding force to improve or solve the technical problems of inaccurate grinding force measurement, poor real-time performance, and difficulty in precise control in the detection and control of grinding force during the grinding process of hard and brittle materials.

[0010] The technical solutions adopted in this application are:

[0011] A grinding force online detection and control device is used to detect and control the grinding force of the grinding head during the grinding process of the hard and brittle material surface, the grinding force includes tangential force, normal force and axial force, and includes a threaded cover, a mounting handle, multiple piezoelectric sensors, a spring chuck with a hollow interior and a signal amplification and processing module, the threaded cover and the mounting handle are connected and cooperated to form a accommodating cavity for accommodating the spring chuck, the spring chuck includes a connecting shaft section and a detection shaft section in sequence along the axial direction, the outer wall of the detection shaft section is a conical surface; along the direction from the connecting shaft section to the detection shaft section, the detection shaft section gradually shrinks inward; the threaded cover limits the axial position of the spring chuck, and the mounting handle A first conical cavity is provided that is adapted to the outer wall of the detection shaft section, and the threaded cover is provided with an avoidance hole corresponding to the center of the spring chuck. The grinding wheel spindle can be inserted into the spring chuck through the avoidance hole and clamped. The multiple piezoelectric sensors are installed on the detection shaft section, so that the grinding wheel spindle acts on the multiple piezoelectric sensors through the detection shaft section. The multiple piezoelectric sensors detect the tangential force, the normal force and the axial force. The signal amplification and processing module can convert the charge signal output by the piezoelectric sensor into a digital signal and output it to the terminal. The terminal can control the grinding head to maintain constant force grinding with a preset grinding force target value according to the preset control logic.

[0012] In this technical solution, the spring chuck is confined to the accommodating cavity surrounded by the threaded cover and the mounting handle. The grinding wheel spindle can be inserted into the spring chuck through the avoidance hole of the threaded cover and clamped by the spring chuck. Then, the conical detection shaft section of the spring chuck cooperates with the first conical cavity of the mounting handle. The spring chuck reduces vibration interference during the grinding process, improves the signal acquisition accuracy of the piezoelectric sensor, and realizes accurate force transmission of the grinding wheel spindle. Moreover, the rigid force transmission path between the grinding wheel spindle and the piezoelectric sensor effectively overcomes the defect of large decoupling error of traditional single-axis sensors; the piezoelectric sensors are distributed on the detection shaft section, directly sensing Knowing the tangential, normal, and axial forces of the grinding force, the high sensitivity and fast response characteristics of the piezoelectric sensor can capture tiny changes in the grinding force in real time, providing accurate input signals for constant-force grinding control, and solving the control delay problem caused by the lag of traditional strain gauge sensors. Combined with the signal amplification and processing module, the charge signal is converted into a digital signal, realizing real-time detection of multiple dimensions such as the tangential, normal, and axial directions of the grinding force, building a full-dimensional data foundation, and providing high-precision input for subsequent control. Through the detection of the grinding axial force, the workpiece warping or grinding wheel resonance caused by the lack of axial force monitoring is avoided. In this solution, the integrated structure of the spring chuck, threaded cover, and mounting handle, as well as the arrangement of the piezoelectric sensor, achieves high rigidity, vibration resistance, and convenient installation, which not only improves the reliability and adaptability of the device, but also reduces manufacturing and maintenance costs. By setting the control logic at the terminal, it is possible to adjust at least one of the feed speed, grinding feed amount, and grinding wheel speed to achieve constant force grinding. Then, by precisely controlling the grinding force in real time, the grinding normal force, tangential force, and axial force are stabilized within the material damage threshold, directly inhibiting the expansion of subsurface cracks and reducing surface roughness. At the same time, the material removal rate is maximized while reducing abnormal wear of the grinding wheel.

[0013] The grinding force online detection and control device includes four piezoelectric sensors, which are evenly distributed along the circumference of the detection shaft segment. Among them, the force measuring ends of three piezoelectric sensors are arranged along the radial direction of the detection shaft segment for detecting the tangential force and the normal force, and the force measuring end of another piezoelectric sensor is arranged along the axial direction of the detection shaft segment for detecting the axial force.

[0014] In this technical solution, four piezoelectric sensors are arranged in three radial directions and one axial direction, of which three radial piezoelectric sensors accurately decouple the tangential force and the normal force through vector synthesis, taking into account accuracy, redundancy and cost. Although in theory some functions can be achieved with fewer sensors, in a dynamic grinding environment, the three-radial piezoelectric sensor solution can significantly improve reliability and avoid data distortion caused by single-point failure or directional coupling. An axial piezoelectric sensor is dedicated to detecting small but dangerous axial forces, which can provide early warning of workpiece clamping tilt during surface grinding and reduce scrap rate. Through the above design, excessive redundancy of the number of sensors is avoided, and the comprehensiveness of three-dimensional force detection is ensured, thereby improving measurement efficiency and accuracy.

[0015] The detection shaft section is provided with sensor mounting countersunk holes that are adapted one by one to the four piezoelectric sensors. The openings of the sensor mounting countersunk holes are covered by a cover plate, which presses and limits the piezoelectric sensors. The outer surface of the cover plate smoothly transitions to the conical surface of the detection shaft section to form a continuous curved surface. The outer surface of the cover plate has a consistent curvature with the generatrix of the conical surface to maintain the integrity of the conical surface.

[0016] In this technical solution, the piezoelectric sensor is rigidly fixed in the detection shaft section through the coordinated design of the sensor mounting countersunk hole and the cover plate. The outer surface of the cover plate and the conical surface have a smooth transition, which maintains the structural integrity and mechanical properties of the conical surface of the detection shaft section, reduces stress concentration and aerodynamic noise during high-speed rotation, extends the fatigue life of the spring chuck, and prevents cutting fluid from invading the area where the piezoelectric sensor is located.

[0017] A wire hole is provided in the sensor mounting countersunk hole, and the wire hole passes through the end of the detection shaft segment away from the connecting shaft segment. The interior of the mounting handle is hollow, and the piezoelectric sensor is connected to the data transmission line, and the data transmission line passes through the wire hole and the interior of the mounting handle.

[0018] In this technical solution, the wire hole and the internal cavity of the mounting handle form a closed wiring channel. The data transmission line is led out through the end of the detection shaft segment and the internal cavity of the mounting handle, which optimizes the wiring path, replaces the traditional exposed cables (which are susceptible to mechanical damage), avoids external cable entanglement or wear, improves the reliability of the device, and simplifies the assembly and maintenance process.

[0019] The detection shaft section and the connecting shaft section are separated by a circumferentially extending limiting groove. The inner wall of the threaded cover is provided with a limiting protrusion adapted to the limiting groove. The limiting groove and the limiting protrusion cooperate to limit the axial position of the spring chuck; the outer wall of the connecting shaft section is a conical surface, and the conical surface of the connecting shaft section and the conical surface of the detection shaft section have opposite taper directions. The threaded cover is provided with a second conical cavity adapted to the outer wall of the connecting shaft section.

[0020] In this technical solution, the connecting shaft section cooperates with the threaded cover through the limiting groove and the limiting protrusion to achieve reliable axial limitation of the spring collet. Moreover, the contraction and expansion characteristics of the spring collet are utilized to facilitate the cooperation of the limiting groove and the limiting protrusion with the threaded cover. The taper directions of the connecting shaft section and the detection shaft section are opposite, forming a bidirectional self-centering locking structure, which helps to reduce the coaxiality error of the spring collet assembly, avoid axial force abnormalities caused by the runout of the grinding wheel spindle, enhance the connection stability between the spring collet and the grinding wheel spindle, reduce vibration offset during the grinding process, improve the symmetry of force transmission, and help suppress the plastic deformation of the spring collet during heavy-load grinding, thereby extending its service life.

[0021] The spring chuck is provided with a plurality of first elastic grooves extending from the connecting shaft segment toward the detecting shaft segment and a plurality of second elastic grooves extending from the detecting shaft segment toward the connecting shaft segment. The first elastic grooves and the second elastic grooves are alternately arranged along the circumference of the spring chuck.

[0022] In this technical solution, the spring chuck is provided with alternately distributed first elastic grooves and second elastic grooves, which significantly improves the radial elastic deformation capacity of the spring chuck, ensures that the grinding wheel spindle is subjected to uniform force, and avoids plastic deformation or fatigue fracture caused by excessive local stress.

[0023] Preferably, the threaded gland is provided with an internal thread, and the mounting handle is provided with an external thread, and the threaded gland and the mounting handle are detachably connected by screwing the internal thread and the external thread together.

[0024] In this technical solution, the threaded cover and the mounting handle are connected by threads to achieve detachable assembly, which is convenient for quick maintenance or replacement of spring chucks or sensor modules of different specifications, adapting to diverse workpiece processing needs, and reducing maintenance costs.

[0025] Another object of the present application is to propose a grinding force control method, which is applied to the grinding force online detection and control device as described above, including: setting the initial grinding wheel speed, initial feed speed and initial grinding feed amount, and setting the target value of the grinding force based on the critical crack pressure of the hard and brittle material being ground; the piezoelectric sensor collects the grinding force signal in real time, and the grinding force signal is processed by the signal amplification and processing module and output to the PID feedback control system, and the PID feedback control system adopts a quasi-ductile critical adjustment model according to the deviation between the actual grinding force detection value and the target value, and dynamically coordinates and controls the grinding wheel speed and / or feed speed in real time to maintain constant force grinding at the target value; during the control process of the PID feedback control system, sensor signals, processing parameters and control output data are collected to construct a training sample set for the feedforward deep learning network; supervised learning is used to train the deep learning model to learn the mapping relationship from sensor input to control signal output; the proportion of feedforward control and feedback control is dynamically adjusted based on the control strategy, and the control weight is gradually transitioned from feedback control to feedforward control.

[0026] The quasi-ductility critical adjustment model specifically includes:

[0027] a) Construct the composite control quantity equation: , where p is the dynamic adjustment factor, specifically 0.1-0.5, and τ is the system response time constant. is the composite control quantity, is the actual detection value of grinding force, is the target value of the grinding force. The real-time grinding force error is mapped to the adjustment amount of the grinding wheel speed and feed speed through the dynamic adjustment factor p and the system response time constant 𝜏;

[0028] b) Implement a secondary error response mechanism:

[0029] i) When the absolute value of the grinding force error is greater than 15% of the target value, adjust the feed speed;

[0030] ii) When the absolute value of the grinding force error is less than or equal to 15% of the target value, adjust the grinding wheel speed.

[0031] In a preferred embodiment, in the quasi-ductility critical regulation model, a dynamic limit mechanism can also be used to constrain the parameter adjustment range, such as limiting the feed speed to 1.2 times the initial value, the speed fluctuation does not exceed ±20%, and limiting the single adjustment range (≤5% of the current value). hour, ;when hour, ; Intelligent adjustment factor design is used for adjustment factors, , in order to achieve rapid adjustment when the error is large and fine control when the error is small, among which, is the feed speed, is the initial feed speed, is the maximum adjustment of feed speed, is the grinding wheel speed, is the initial grinding wheel speed, is the maximum adjustment of the grinding wheel speed, is the initial adjustment factor.

[0032] The control method further includes: the control strategy adopts a dynamic weighted control strategy, and the dynamic weighted control strategy specifically defines a feedforward control weight and feedback control weights , initially =0, =1; when the mean square error E between the feedforward control signal output by the deep learning model and the PID feedback signal is less than the set threshold, gradually increase and reduce , until =1, =0.

[0033] The control method further includes optimizing control actions using a deep deterministic policy gradient (DDPG) algorithm and generating continuous control signals via an actor-critic network; designing a reward function that combines force error, motion smoothness, and machining efficiency to optimize the control strategy; and achieving a smooth transition from PID control to feedforward control through experience replay and a target network soft update mechanism. The control system required for grinding force control consists of multiple modules, including an environment modeling module, an actor-critic network module, an experience replay module, a reward design module, and a dynamic weighting module. The environment modeling module transforms the grinding force control problem into a Markov decision process based on reinforcement learning, defining states (sensor signals, machining parameters, force error, etc.), actions (servo motor control signals), and a reward function (combining force error, oscillation amplitude, and motion smoothness).

[0034] The technical effects achieved by adopting the above-mentioned grinding force control method are as follows: a grinding force threshold setting method based on the critical crack pressure is combined with PID feedback control to adjust the grinding wheel parameters in real time. A feedforward control model is constructed through deep learning, achieving a transition from experience-based control to intelligent prediction, solving the problems of response lag and parameter rigidity in traditional control. Based on the composite control quantity equation: through the dynamic coupling of the dynamic adjustment factor p and the time constant τ, the grinding force error is mapped into a joint adjustment of speed and feed, solving the response lag problem caused by traditional PID parameter rigidity and shortening the system response time. For large errors, the feed speed is adjusted first, while for small errors, the speed and feed rate are adjusted jointly, balancing rapid correction and stability. A dynamic weighting strategy is used to gradually transition control weights. Initially, PID feedback is relied upon to ensure stability. As data accumulates, the proportion of feedforward control is increased, ultimately achieving low-latency, high-precision constant-force grinding, balancing system robustness and response speed. The DDPG algorithm is introduced to optimize control actions, and the reward function is designed based on force error, action smoothness, and machining efficiency. Through the Actor-Critic network and experience replay mechanism, the system can autonomously adapt to complex working conditions such as grinding wheel wear and material changes, thereby improving the generalization ability of control. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] 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:

[0036] Figure 1 Assembly of the grinding force online detection and control device provided in the embodiment of the present application Figure 1 ;

[0037] Figure 2 Assembly of the grinding force online detection and control device provided in the embodiment of the present application Figure 2 ;

[0038] Figure 3 A cross-sectional view of a grinding force online detection and control device provided in an embodiment of the present application;

[0039] Figure 4 A schematic diagram of the cooperation between the spring chuck and the cover plate provided in an embodiment of the present application;

[0040] Figure 5 A cross-sectional view of a threaded gland provided in an embodiment of the present application;

[0041] Figure 6 A cross-sectional view of the mounting handle provided in an embodiment of the present application;

[0042] Figure 7 Schematic diagram of the structure of the spring chuck provided in the embodiment of the present application Figure 1 ;

[0043] Figure 8 Schematic diagram of the structure of the spring chuck provided in the embodiment of the present application Figure 2 ;

[0044] Figure 9 A cross-sectional view of the cooperation between the spring chuck and the piezoelectric sensor provided in an embodiment of the present application;

[0045] Figure 10 A control flow chart of the grinding force control method provided in an embodiment of the present application;

[0046] Figure 11 This is a DDPG reinforcement learning control flow chart provided in an embodiment of the present application.

[0047] List of parts and reference numerals:

[0048] 1 threaded gland, 11 avoidance hole, 12 limiting protrusion, 13 second tapered cavity, 14 internal thread;

[0049] 2 mounting handle, 21 first tapered cavity, 22 external thread;

[0050] 3 piezoelectric sensor;

[0051] 4 spring chuck, 41 connecting shaft section, 42 detection shaft section, 43 sensor mounting countersunk hole, 44 wire hole, 45 limiting groove, 46 first elastic groove, 47 second elastic groove;

[0052] 5 Cover. DETAILED DESCRIPTION

[0053] In order to more clearly illustrate the overall concept of the present application, a detailed description is given below in an illustrative manner in conjunction with the accompanying drawings.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present application is not limited to the specific embodiments disclosed below.

[0055] In addition, in the description of the present application, it should be understood that the terms "upper", "lower", "top", "bottom", "inside", "outside", "axial", "radial", "circumferential", "lateral", "longitudinal", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application 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 should not be understood as a limitation on the present application.

[0056] In this application, unless otherwise expressly specified or limited, terms such as "installed," "connected," "connect," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection, electrical connection, or communication; direct connection or indirect connection through an intermediate medium; and internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on specific circumstances.

[0057] In this application, unless otherwise expressly specified and limited, a first feature "above" or "below" a second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments or examples.

[0058] This application provides an online grinding force detection and control device. For ease of explanation and understanding, the following content provided in this application is based on the illustrated product structure. Of course, those skilled in the art will understand that the above structure is merely a specific example and schematic illustration and does not constitute a specific limitation of the technical solution provided in this application.

[0059] The grinding force online detection and control device of the present application is used to detect and control the grinding force of the grinding head during the grinding process of hard and brittle materials (such as silicon carbide, silicon nitride, optical glass, etc.). The grinding force includes tangential force, normal force and axial force. In terms of overall composition, Figures 1 to 9As shown, it includes a threaded cover 1, a mounting handle 2, a plurality of piezoelectric sensors 3, a spring chuck 4 with a hollow interior, and a signal amplification and processing module. The mounting handle 2 can be connected to the main shaft of a machine tool. The threaded cover 1 and the mounting handle 2 are connected and cooperated to form a receiving cavity for accommodating the spring chuck 4. The spring chuck 4 includes a connecting shaft section 41 and a detection shaft section 42 in the axial direction. The outer wall of the detection shaft section 42 is a conical surface. The detection shaft section 42 gradually shrinks inward along the direction from the connecting shaft section 41 to the detection shaft section 42. The threaded cover 1 is connected to the connecting shaft section 41 to limit the axial position of the spring chuck 4. The mounting handle 2 is provided with a connection to the detection shaft section 42. The outer wall of the first conical cavity 21 is adapted to the threaded cover 1, and the threaded cover 1 is provided with an avoidance hole 11 corresponding to the center of the spring collet 4. The grinding wheel spindle can be inserted into the spring collet 4 through the avoidance hole 11 and clamped. Multiple piezoelectric sensors 3 are installed on the detection shaft segment 42. The grinding wheel spindle acts on the multiple piezoelectric sensors 3 through the detection shaft segment 42. The multiple piezoelectric sensors 3 detect tangential force, normal force and axial force. The signal amplification and processing module can convert the charge signal output by the piezoelectric sensor 3 into a digital signal and output it to the terminal. The terminal can control the grinding head to maintain constant force grinding with a preset grinding force target value according to the preset control logic.

[0060] It should be noted that the present application does not impose any specific restrictions on the connection and cooperation between the threaded gland 1 and the mounting handle 2. The two can be connected by non-detachable means such as ultrasonic welding. As a preferred solution, the two can also be detachably connected by means of snap connection, threaded connection, etc. In a preferred embodiment, as Figure 2 、 Figure 3 、 Figure 5 and Figure 6 As shown, the threaded cover 1 is provided with an internal thread 14, and the mounting handle 2 is provided with an external thread 22. The threaded cover 1 and the mounting handle 2 are detachably connected by screwing the internal thread 14 and the external thread 22 together, thereby facilitating the rapid replacement of spring chucks 4 or sensor modules of different specifications, adapting to diverse workpiece processing requirements, and reducing maintenance costs.

[0061] In addition, the present application does not specifically limit the manner in which the threaded gland 1 axially limits the spring chuck 4. In a preferred embodiment, Figure 3 、 Figure 5 and Figure 7As shown, the detection shaft section 42 and the connecting shaft section 41 are separated by a circumferentially extending limiting groove 45. The inner wall of the threaded gland 1 is provided with a limiting protrusion 12 adapted to the limiting groove 45. The limiting groove 45 cooperates with the limiting protrusion 12 to reliably limit the axial position of the spring collet 4. Moreover, the contraction and expansion characteristics of the spring collet 4 are utilized to facilitate the matching and disengagement of the limiting groove 45 and the limiting protrusion 12 with the threaded gland 1. In other embodiments, the threaded gland 1 can also use other methods to limit the axial position of the spring collet 4, for example, a screw-lock groove is provided on the spring collet 4, and a screw-lock protrusion is provided on the threaded gland 1, so that the spring collet 4 and the threaded gland 1 are rotationally matched through the screw-lock groove and the screw-lock protrusion.

[0062] In the present technical solution, the spring chuck 4 is confined in the accommodating cavity surrounded by the threaded cover 1 and the mounting handle 2. The grinding wheel spindle can be inserted into the spring chuck 4 through the avoidance hole 11 of the threaded cover 1 and is clamped by the spring chuck 4. Then, the conical detection shaft section 42 of the spring chuck 4 cooperates with the first conical cavity 21 of the mounting handle 2. The spring chuck 4 reduces the vibration interference during the grinding process, improves the signal acquisition accuracy of the piezoelectric sensor 3, and realizes the precise force transmission of the grinding wheel spindle. Moreover, the rigid force transmission path between the grinding wheel spindle and the piezoelectric sensor 3 effectively overcomes the defect of large decoupling error of the traditional single-axis sensor; the piezoelectric sensor 3 is distributed in the detection The measuring shaft section 42 directly senses the tangential, normal, and axial forces of the grinding force. Through the high sensitivity and fast response characteristics of the piezoelectric sensor 3, it can capture small changes in the grinding force in real time, providing accurate input signals for constant force grinding control, solving the control delay problem caused by the hysteresis of traditional strain gauge sensors. Combined with the signal amplification processing module, the charge signal is converted into a digital signal, realizing real-time detection of multiple dimensions such as the tangential, normal, and axial directions of the grinding force, building a full-dimensional data foundation, and providing high-precision input for subsequent control. By detecting the axial force of the grinding, workpiece warping or grinding wheel resonance caused by the lack of axial force monitoring is avoided. In this solution, the integrated structure of the spring chuck 4, the threaded cover 1, and the mounting handle 2, as well as the arrangement of the piezoelectric sensor 3, achieves high rigidity, vibration resistance, and convenient installation, which not only improves the reliability and adaptability of the device, but also reduces manufacturing and maintenance costs. Specifically, the signal amplification processing module includes a signal conditioning module and a data transmission / acquisition / processing and analysis module. The signal amplification processing module converts the charge signal output by the piezoelectric sensor 3 into a digital signal and transmits it to the terminal. By setting the control logic at the terminal, constant force grinding is achieved by adjusting the feed speed, grinding feed, grinding wheel speed, and other methods. Furthermore, through real-time and precise control of the grinding force, the grinding normal force, tangential force, and axial force are stabilized within the material damage threshold, directly inhibiting subsurface crack propagation and reducing surface roughness. This also maximizes the material removal rate while minimizing abnormal grinding wheel wear. There are no restrictions on the control logic set at the terminal. For example, PID control logic, PID-feedforward composite control logic, or a three-level control logic of PID feedback + feedforward compensation + dynamic limiting can be used. The control logic of specific examples is detailed in the control method.

[0063] Regarding the structure of the connecting shaft section 41, in a preferred embodiment, as shown in FIG. Figure 3 、 Figure 5 and Figure 7As shown, the outer wall of the connecting shaft segment 41 is a tapered surface, and the taper direction of the tapered surface of the connecting shaft segment 41 is opposite to that of the tapered surface of the detection shaft segment 42. The threaded gland 1 is provided with a second tapered cavity 13 adapted to the outer wall of the connecting shaft segment 41. It will be understood by those skilled in the art that the taper directions of the connecting shaft segment 41 and the detection shaft segment 42 are opposite, forming a bidirectional self-centering locking structure, which helps to reduce the coaxiality error of the spring chuck 4 during assembly, avoids abnormal axial force caused by the runout of the grinding wheel spindle, enhances the connection stability of the spring chuck 4 and the grinding wheel spindle, reduces vibration offset during the grinding process, improves the symmetry of force transmission, and helps to suppress the plastic deformation of the spring chuck 4 during heavy-load grinding, thereby extending its service life.

[0064] In order to realize that the spring clamp 4 has a reliable and stable elastic deformation capability, in a preferred embodiment, as shown in FIG. Figure 7 As shown, the spring chuck 4 is provided with a plurality of first elastic grooves 46 extending from the connecting shaft section 41 toward the detection shaft section 42 and a plurality of second elastic grooves 47 extending from the detection shaft section 42 toward the connecting shaft section 41, and the first elastic grooves 46 and the second elastic grooves 47 are alternately arranged along the circumference of the spring chuck 4. In this technical solution, the spring chuck 4 is provided with the first elastic grooves 46 and the second elastic grooves 47 that are alternately distributed, which significantly improves the radial elastic deformation capacity of the spring chuck 4, ensures that the grinding wheel spindle is evenly stressed, and avoids plastic deformation or fatigue fracture caused by excessive local stress. Moreover, through the above design, the proportion of effective expansion or contraction of the spring chuck 4 is large, which can effectively reduce the volume and make it easy to operate, and the expansion or contraction amplitude of the detection shaft section 42 is larger, which helps to clamp the grinding wheel spindle in the hole.

[0065] In a preferred embodiment, the piezoelectric sensor 3 can be a piezoelectric ceramic sensor, which can produce a more significant charge change when subjected to external force. It can keenly sense and convert tiny changes in physical quantities such as pressure and vibration into electrical signals, so that it can accurately detect weak external stimuli and respond to external forces in a very short time, quickly converting mechanical signals into electrical signals. This fast response characteristic makes it suitable for dynamic measurement of grinding force, capture of rapidly changing signals, and other situations.

[0066] Regarding the number and installation method of the piezoelectric sensors 3, in a preferred embodiment, as shown in FIG. Figure 3 、 Figure 4 and Figure 9As shown, the online grinding force detection and control device includes four piezoelectric sensors 3, evenly distributed along the circumference of a detection shaft segment 42. The force-measuring ends of three piezoelectric sensors 3 are arranged radially along the detection shaft segment 42 to detect tangential and normal forces, while the force-measuring end of one piezoelectric sensor 3 is arranged axially along the detection shaft segment 42 to detect axial force. The four piezoelectric sensors 3 are arranged in a three-radial and one-axial configuration. The three radial piezoelectric sensors accurately decouple the tangential and normal forces through vector synthesis, balancing accuracy, redundancy, and cost. While theoretically, some functionality could be achieved with fewer sensors, the three-radial piezoelectric sensor solution significantly improves reliability in dynamic grinding environments, avoiding data distortion caused by single-point failures or directional coupling. One axial piezoelectric sensor is specifically designed to detect small but dangerous axial forces, providing early warning of workpiece clamping tilt during surface grinding and reducing scrap. This design avoids excessive sensor redundancy while ensuring comprehensive three-dimensional force detection, improving measurement efficiency and accuracy.

[0067] Furthermore, if Figure 4 、 Figure 7 、 Figure 8 and Figure 9 As shown, the detection shaft section 42 is provided with sensor mounting countersunk holes 43 that are adapted to the four piezoelectric sensors 3 one by one. The openings of the sensor mounting countersunk holes 43 are covered by a cover plate 5. The cover plate 5 presses and limits the piezoelectric sensors 3. The outer surface of the cover plate 5 smoothly transitions with the conical surface of the detection shaft section 42, thereby forming a continuous curved surface. The outer surface of the cover plate 5 has the same generatrix curvature as that of the conical surface to maintain the integrity of the conical surface. Specifically, the force measuring ends of the three radially arranged piezoelectric sensors 3 can be close to the bottom of the groove of the sensor mounting countersunk hole 43, and the force measuring end of the axially arranged piezoelectric sensor 3 can be close to the groove side of the sensor mounting countersunk hole 43 and face the connecting shaft section 41. Preferably, a fixing glue can be used to fix the piezoelectric sensor 3 to the spring chuck 4 in the sensor mounting countersunk hole 43. The sensor mounting countersunk hole 43 and the cover plate 5 are designed to cooperate and rigidly secure the piezoelectric sensor 3 within the detection shaft segment 42. The outer surface of the cover plate 5 smoothly transitions with the tapered surface, maintaining the structural integrity and mechanical properties of the tapered surface of the detection shaft segment 42. This reduces stress concentration and aerodynamic noise during high-speed rotation, extends the fatigue life of the spring chuck 4, and prevents cutting fluid from invading the area where the piezoelectric sensor 3 is located. Preferably, the cover plate 5 can also be secured to the spring chuck 4 with adhesive to limit the piezoelectric sensor 3.

[0068] Furthermore, if Figure 7 and Figure 8As shown, a wire hole 44 is provided in the sensor mounting counterbore 43. The wire hole 44 passes through the end of the detection shaft section 42 away from the connecting shaft section 41. The interior of the mounting handle 2 is hollow. The piezoelectric sensor 3 is connected to a data transmission line (not shown in the figure), which passes through the wire hole 44 and the internal cavity of the mounting handle 2 to exit. Since the detection shaft section 42 is provided with sensor mounting counterbore 43 that matches the four piezoelectric sensors 3 one by one, for a total of four sensor mounting counterbore 43, a wire hole 44 can be provided in each sensor mounting counterbore 43. The data transmission line connected to the piezoelectric sensor 3 passes through the wire hole 44 and exits from the spring chuck 4. The wire hole 44 and the internal cavity of the mounting handle 2 form a closed wiring channel. The data transmission line is led out through the end of the detection shaft section 42 and the internal cavity of the mounting handle 2, optimizing the wiring path, replacing traditional exposed cables (which are susceptible to mechanical damage) to avoid external cable entanglement or wear, improving the reliability of the device, and simplifying the assembly and maintenance process.

[0069] The assembly process of the grinding force online detection and control device: First, the four piezoelectric sensors 3 are respectively installed in the sensor mounting countersunk holes 43 of the detection shaft segment 42, and the data transmission line passes through the wire hole 44. The four piezoelectric sensors 3 are fixed by fixing glue, and the force measuring ends of three piezoelectric sensors 3 are arranged along the radial direction of the detection shaft segment 42 and close to the bottom of the groove, and the other is arranged along the axial direction and close to the side of the groove toward the direction of the connecting shaft segment 41. Then, the cover plate 5 is used to cover the opening of the sensor mounting countersunk hole 43. The cover plate 5 is fixed by fixing glue to press and limit the piezoelectric sensor 3 to ensure a smooth transition between the outer surface of the cover plate 5 and the conical surface of the detection shaft section 42; then, the connecting shaft section 41 of the spring chuck 4 is matched with the second conical cavity 13 of the threaded cover 1, and the limiting protrusion 12 is matched with the limiting groove 45 to realize the assembly of the spring chuck 4 and the threaded cover 1; thereafter, the threaded cover 1 is threadedly connected to the mounting handle 2, so that the detection shaft section 42 is matched with the first conical cavity 21 of the mounting handle 2, and the data transmission line passes through the internal cavity of the mounting handle 2 to the outside of the detection device, and then is connected to the signal processing module.

[0070] The present application provides a grinding force control method, which is applied to the above grinding force online detection and control device, with reference to Figure 10 and Figure 11 As shown, including:

[0071] Set the initial grinding wheel speed, initial feed speed and initial grinding feed, and set the target value of the grinding force based on the critical crack pressure of the hard and brittle material being ground;

[0072] The piezoelectric sensor 3 collects the grinding force signal in real time. The grinding force signal is processed by the signal amplification and processing module and output to the PID feedback control system. The PID feedback control system adopts a quasi-ductility critical adjustment model based on the deviation between the actual grinding force detection value and the target value. Through dynamic coordination and real-time regulation of the grinding wheel speed and / or feed speed, the grinding force is maintained at the target value.

[0073] During the PID feedback control system, sensor signals, processing parameters, and control output data are collected to build a training sample set for the feedforward deep learning network. Supervised learning is used to train the deep learning model to learn the mapping relationship from sensor input to control signal output.

[0074] Based on the control strategy, the proportion of feedforward control and feedback control is dynamically adjusted, and the control weight is gradually transitioned from feedback control to feedforward control.

[0075] Preferably, the quasi-ductility critical adjustment model specifically includes:

[0076] a) Construct the composite control quantity equation: , where p is the dynamic adjustment factor (0.1-0.5), τ is the system response time constant, is the composite control quantity, is the actual detection value of grinding force, is the target value of the grinding force. The real-time grinding force error is mapped to the adjustment amount of the grinding wheel speed and feed speed through the dynamic adjustment factor p and the system response time constant 𝜏;

[0077] b) Implement a secondary error response mechanism:

[0078] i) When the absolute value of the grinding force error is greater than 15% of the target value, adjust the feed speed;

[0079] ii) When the absolute value of the grinding force error is less than or equal to 15% of the target value, adjust the grinding wheel speed.

[0080] In a preferred embodiment, in the quasi-ductility critical regulation model, a dynamic limit mechanism can also be used to constrain the parameter adjustment range, such as limiting the feed speed to 1.2 times the initial value, the speed fluctuation does not exceed ±20%, and limiting the single adjustment range (≤5% of the current value). hour, ;when hour, ; Intelligent adjustment factor design is used for adjustment factors, , in order to achieve rapid adjustment when the error is large and fine control when the error is small, among which, is the feed speed, is the initial feed speed, is the maximum adjustment of feed speed, is the grinding wheel speed, is the initial grinding wheel speed, is the maximum adjustment of the grinding wheel speed, is the initial adjustment factor.

[0081] Preferably, the control strategy adopts a dynamic weighted control strategy: .

[0082] in, is the output of the feedforward network, is the output of the PID feedback network, is the feedforward control weight, is the feedback control weight.

[0083] Initial settings =0, =1, gradually increases with the increase of sample data and reduce , when the error between the output of the feedforward control and the output of the feedback control is less than the set threshold, =1, = 0. The mean square error can be used to measure the difference between the feedforward control output and the feedback control output:

[0084]

[0085] When the mean square error E between the feedforward control signal output by the deep learning model and the PID feedback signal is less than the set threshold, a complete transition from PID control to feedforward control is achieved.

[0086] Preferably, the control method further includes: using a deep deterministic policy gradient (DDPG) algorithm to optimize control actions and generate continuous control signals through an actor-critic network; designing a reward function to optimize the control strategy by combining force error, action smoothness and processing efficiency; and achieving a smooth transition from PID control to feedforward control through experience replay and target network soft update mechanism.

[0087] The technical effects achieved by adopting the above-mentioned grinding force control method are as follows: a grinding force threshold setting method based on the critical crack pressure, combined with PID feedback control to adjust the grinding wheel parameters in real time, and a feedforward control model constructed through deep learning to achieve a transition from experience dependence to intelligent prediction, solving the problems of response lag and parameter solidification in traditional control. Through a dynamic weighted strategy, the control weight is gradually transitioned. In the initial stage, PID feedback is relied upon to ensure stability. As data accumulates, the proportion of feedforward control is increased, ultimately achieving low-latency, high-precision constant force grinding and balancing system robustness and response speed. The DDPG algorithm is introduced to optimize control actions, and a reward function is designed based on force error, action smoothness, and processing efficiency. Through the Actor-Critic network and experience replay mechanism, the system can autonomously adapt to complex working conditions such as grinding wheel wear and material changes, thereby improving the generalization ability of control.

[0088] The specific implementation route of this method is to calculate the critical pressure for crack initiation in the hard and brittle materials being ground based on theoretical low-damage calculations (for example, the critical normal load for surface crack initiation during grinding can be calculated using a median / lateral crack theory model based on the fracture toughness and elastic modulus of the hard and brittle materials, and the safety threshold range of the grinding force can be set accordingly to achieve low-damage grinding). The initial grinding force target value is set, including the safety target values of the grinding tangential force, the grinding normal force, and the grinding axial force. The initial grinding wheel speed, initial feed speed, and initial grinding feed of the servo motor are also set. During the grinding process, the force signal is collected by the piezoelectric ceramic sensor, and the signal amplification and processing module implements data processing and transmits it to the initial PID feedback control system. The initial PID feedback control system is fully responsible for the grinding process. By analyzing the deviation between the actual grinding force detection value and the target value, the concept of "quasi-ductile" control is proposed. The nonlinear decoupling of the grinding force and motion parameters is achieved through a composite control quantity equation. An intelligent response mechanism based on error classification is designed to maintain the grinding wheel grinding at the target constant force. Data is collected during the feedback control phase, providing samples for the feedforward deep learning network. The deep learning model learns the behavior of the PID feedback control system through supervised learning, mapping sensor inputs to control signal outputs. To dynamically adjust the balance between feedforward and feedback control, a weighted control strategy is introduced, achieving a gradual transition from feedback control to feedforward control. A feedforward control strategy based on deep deterministic policy gradient (DDPG) is employed, achieving a direct mapping from sensor signals to control signals through reinforcement learning. First, during the initial grinding phase, the PID feedback control system is responsible for constant-force grinding. Force signals, machining parameters, and control outputs are dynamically collected as training samples for the reinforcement learning network. Subsequently, the DDPG algorithm learns the control policy using an actor-critic architecture, in which an actor network generates continuous control signals and a critic network evaluates the value of actions. Stable training is achieved through experience replay and a target network. The entire control system consists of multiple modules, including an environment modeling module, an actor-critic network module, an experience replay module, a reward design module, and a dynamic weighting module. The environmental modeling module transforms the grinding force control problem into a Markov decision process in reinforcement learning, defining the state (sensor signal, processing parameters, force error, etc.), action (servo motor control signal) and reward function (combining force error, oscillation amplitude and action smoothness). The Actor network is responsible for generating control signals, and the Critic network evaluates the value of the action. Both are stably trained through experience replay and the soft update mechanism of the target network. The working path is as follows: First, in the initial stage of grinding, the PID controller is responsible for constant force grinding, and the state St is collected in real time through the sensor, and combined with the feedforward control signal generated by the Actor network and , using the dynamic weighting module according to the weight and Calculate the final control signal Subsequently, the state, action, reward, and next state are interactively recorded and stored in the experience replay module. The actor-critic network is then optimized during each training cycle. In the reward design module, the reward function is optimized by combining target force error, oscillation amplitude, and action smoothness, guiding the system to learn more efficient control strategies. As data accumulates and training progresses, the feedforward control weight is gradually increased, ultimately transitioning to full reliance on feedforward control to achieve precise and stable constant-force grinding control.

[0089] Anything not described in this application can be achieved by adopting or drawing on existing technologies.

[0090] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0091] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A grinding force control method using an online grinding force detection and control device, wherein the online grinding force detection and control device is used to detect and control the grinding force of a grinding head during grinding of a hard and brittle material surface, wherein the grinding force includes tangential force, normal force and axial force, and is characterized in that: The grinding force online detection and control device includes a threaded gland, a mounting handle, multiple piezoelectric sensors, a hollow spring chuck, and a signal amplification and processing module. The threaded gland and the mounting handle are connected and cooperated to form a housing for the spring chuck. The spring chuck is provided with a connecting shaft section and a detection shaft section along the axial direction. The outer wall of the detection shaft section is a tapered surface. The detection shaft section gradually shrinks inward in the direction from the connecting shaft section to the detection shaft section. The threaded gland limits the axial position of the spring chuck, the mounting handle is provided with a first tapered cavity adapted to the outer wall of the detection shaft section, the threaded gland is provided with a avoidance hole corresponding to the center of the spring chuck, so that the grinding wheel spindle can be inserted into the spring chuck through the avoidance hole and clamped, the multiple piezoelectric sensors are installed on the detection shaft section, the grinding wheel spindle acts on the multiple piezoelectric sensors through the detection shaft section, the multiple piezoelectric sensors detect the tangential force, the normal force and the axial force, the signal amplification and processing module can convert the charge signal output by the piezoelectric sensor into a digital signal and output it to the terminal, and the terminal can control the grinding head to maintain constant force grinding at a preset grinding force target value according to a preset control logic; The grinding force control method using the grinding force online detection and control device includes: Set the initial grinding wheel speed, initial feed speed and initial grinding feed, and set the target value of the grinding force based on the critical crack pressure of the hard and brittle material being ground; The piezoelectric sensor collects the grinding force signal in real time. The grinding force signal is processed by the signal amplification and processing module and output to the PID feedback control system. The PID feedback control system adopts a quasi-ductility critical adjustment model based on the deviation between the actual grinding force detection value and the target value. Through dynamic coordination and real-time control of the grinding wheel speed and / or feed speed, the quasi-ductility critical adjustment model is maintained at the target value constant force grinding. The quasi-ductility critical adjustment model specifically includes: a) Construct the composite control quantity equation: , where p is the dynamic adjustment factor, is the system response time constant, is the composite control quantity, is the actual detection value of grinding force, is the target value of grinding force, which is adjusted by dynamic adjustment factor p and system response time constant , the real-time grinding force error is mapped to the adjustment amount of grinding wheel speed and feed speed; b) Implement a secondary error response mechanism: i) When the absolute value of the grinding force error is greater than 15% of the target value, adjust the feed speed; ii) When the absolute value of the grinding force error is less than or equal to 15% of the target value, adjust the grinding wheel speed; During the PID feedback control system, sensor signals, processing parameters, and control output data are collected to build a training sample set for the feedforward deep learning network. Supervised learning is used to train the deep learning model to learn the mapping relationship from sensor input to control signal output. Based on the control strategy, the proportion of feedforward control and feedback control is dynamically adjusted, and the control weight is gradually transitioned from feedback control to feedforward control.

2. The grinding force control method using the grinding force online detection and control device according to claim 1, characterized in that: The grinding force online detection and control device includes four piezoelectric sensors, which are evenly distributed along the circumference of the detection shaft segment. Among them, the force measuring ends of three piezoelectric sensors are arranged along the radial direction of the detection shaft segment for detecting the tangential force and the normal force, and the force measuring end of another piezoelectric sensor is arranged along the axial direction of the detection shaft segment for detecting the axial force.

3. The grinding force control method using the grinding force online detection and control device according to claim 2, characterized in that: The detection shaft section is provided with sensor mounting countersunk holes that are adapted one by one to the four piezoelectric sensors. The openings of the sensor mounting countersunk holes are covered by a cover plate, which presses and limits the piezoelectric sensors. The outer surface of the cover plate smoothly transitions to the conical surface of the detection shaft section to form a continuous curved surface. The outer surface of the cover plate has a consistent curvature with the generatrix of the conical surface to maintain the integrity of the conical surface.

4. The grinding force control method using the grinding force online detection and control device according to claim 3, characterized in that: A wire hole is provided in the sensor mounting countersunk hole, and the wire hole passes through the end of the detection shaft segment away from the connecting shaft segment. The interior of the mounting handle is hollow, and the piezoelectric sensor is connected to the data transmission line, and the data transmission line passes through the wire hole and the interior of the mounting handle.

5. The grinding force control method using the grinding force online detection and control device according to claim 1, characterized in that: The detection shaft section and the connecting shaft section are separated by a circumferentially extending limiting groove. The inner wall of the threaded cover is provided with a limiting protrusion adapted to the limiting groove. The limiting groove and the limiting protrusion cooperate to limit the axial position of the spring chuck; the outer wall of the connecting shaft section is a conical surface, and the conical surface of the connecting shaft section and the conical surface of the detection shaft section have opposite taper directions. The threaded cover is provided with a second conical cavity adapted to the outer wall of the connecting shaft section.

6. The grinding force control method using the grinding force online detection and control device according to claim 1, characterized in that: The spring chuck is provided with a plurality of first elastic grooves extending from the connecting shaft segment toward the detecting shaft segment and a plurality of second elastic grooves extending from the detecting shaft segment toward the connecting shaft segment. The first elastic grooves and the second elastic grooves are alternately arranged along the circumference of the spring chuck.

7. The grinding force control method according to claim 1, wherein: The control strategy adopts a dynamic weighted control strategy, and the dynamic weighted control strategy is specifically: Define feedforward control weights and feedback control weights , initially =0, =1; When the mean square error E between the feedforward control signal output by the deep learning model and the PID feedback signal is less than the set threshold, gradually increase and reduce , until =1, =0.

8. The grinding force control method according to claim 1, wherein: The control method further comprises: The deep deterministic policy gradient (DDPG) algorithm is used to optimize control actions and generate continuous control signals through the actor-critic network. Design a reward function that combines force error, motion smoothness, and processing efficiency to optimize the control strategy; Through experience replay and target network soft update mechanism, a smooth transition from PID control to feedforward control is achieved.

Images