Tool grinding device and method based on ring array and dynamic multi-scale convolution

Abstract

The application discloses a tool grinding wheel dressing device and method based on a ring array and dynamic multi-scale convolution, a ring array sensor monitoring device is arranged at the tool grinding wheel, the sensor is packaged into a flange, a pressure-sound dual-signal ring piezoelectric array sensor is used to collect signals, and the problems of poor anti-interference ability, single detection signal and single monitoring area of a traditional sensor are effectively solved; a multi-scale dynamic convolution network is used for signal processing, sound and dressing force dual signals are fused, precise classification of a wear state and grit collapse positioning are realized, a radial displacement self-adaptive compensation instruction is generated, and the problems of low shutdown detection efficiency and large human judgment error in a traditional dressing process can be effectively solved.

Description

Technical Field

[0001] This invention belongs to the field of tool grinding wheel dressing technology, and specifically relates to a tool grinding wheel dressing device and method based on a ring array and dynamic multi-scale convolution. Background Technology

[0002] In the field of ultra-precision grinding, ceramic-bonded tool wheels are widely used for the dressing of diamond rollers. Their dressing accuracy directly affects the surface quality and fatigue life of critical components such as gears and bearings. However, in existing technologies, ceramic grinding wheels, due to their relatively low hardness, are prone to wear during high-speed dressing, leading to dressing path deviation and accumulated tooth profile errors. Traditional monitoring methods have significant drawbacks. Firstly, it relies on a single sensor (such as force or sound signals). For example, although acoustic emission sensors can capture microscopic high-frequency signals of abrasive wear, they are easily interfered with by spindle rotation noise, causing the effective signal to be submerged and making it difficult to distinguish between grinding wheel wear and mechanical vibration. While pressure signals can monitor the trend of dressing force changes, the abrasive grains are irregularly distributed, and the average surface stress is inconsistent, making it impossible to make an accurate judgment.

[0003] Sensors with a single location cannot effectively capture wear signals from different areas around the grinding wheel. Substrate end-face mounted sensors have a lot of signal noise due to their play and vibration. Spindle stator and rotor type sensors have a complex structure and are difficult to manufacture.

[0004] Secondly, traditional judgment methods rely on human ears and eyes, which are offline, offline inspections. They rely on the volume of the sound of the adjustment contact and the visual changes in the shape of the tool grinding wheel to make judgments. They cannot capture micro-defects such as local chipping and progressive wear of the grinding wheel in real time. Moreover, the compensation amount is judged manually, which leads to compensation lag and accumulation of adjustment path deviation.

[0005] Third, existing algorithms (such as wavelet packet decomposition combined with SVM) have insufficient generalization ability under complex working conditions and are difficult to adapt to the time-varying characteristics of grinding wheel wear. This results in the loss of fragmented features in the high-frequency band (>500kHz), redundant noise in the low-frequency band (<50kHz), and diamond roller noise in the similar frequency band (≈300KHz). Feature extraction is redundant and computationally complex, and the ability to fuse multimodal data is poor.

[0006] Therefore, in order to address the problems of weak anti-interference capability, single signal acquisition, single signal monitoring area, and poor generalization ability of signal processing algorithms in existing signal acquisition devices, which lead to low monitoring accuracy, it is necessary to develop a new tool grinding wheel adaptive dressing control system. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of the prior art by providing a tool grinding wheel dressing device and method based on a ring array and dynamic multi-scale convolution, thereby improving the dressing accuracy and efficiency of tool grinding wheels.

[0008] To solve the above technical problems, the technical solution adopted by the present invention is as follows: A tool wheel dressing device based on ring array and dynamic multi-scale convolution includes a grinding machine and a tool wheel and diamond roller mounted on the grinding machine, and a ring array sensor monitoring device is also provided at the tool wheel. The ring array sensor monitoring device includes a rear cover of the encapsulated flange, a front cover of the encapsulated flange, and a conductive slip ring. The front cover of the encapsulated flange is also equipped with a pressure film sensor and an acoustic emission sensor. The signal transmission lines of the pressure film sensor and the acoustic emission sensor are connected to the rotor side of the conductive slip ring via cables, and the stator side of the conductive slip ring is fixed to the other side of the grinding machine. The flange is assembled and clamped onto the tool grinding wheel and installed onto the grinding machine spindle, and the internal radial pressure and internal acoustic emission signal inside the hole are monitored in real time. The outer layer of the pressure film sensor and acoustic emission sensor is also provided with straight-hole microchannels for gas exchange.

[0009] The acoustic emission sensor is also provided with an elastic silicone damping layer inside, and the elastic silicone damping layer is arranged in a ring-shaped contact within the acoustic emission sensor.

[0010] The conductive slip ring is also equipped with a preamplifier, and the data is transmitted to the computer control terminal via a data acquisition card.

[0011] The pressure diaphragm sensors are evenly distributed at 120° intervals inside the front cover of the encapsulation flange.

[0012] The acoustic emission sensors are evenly distributed at 120° intervals on the inside of the pressure film sensor.

[0013] A method for tool wheel dressing devices based on ring arrays and dynamic multi-scale convolution includes the following steps: (1) Use digital filters to filter out high-frequency noise >500KHz and low-frequency noise <50KHz, and use differential vibration to suppress diamond roller noise ≈300KHz; The pressure signal is processed by sliding smoothing, dynamic pulse entropy extraction is performed, and sound and pressure signals are collected. (2) Perform a short-time Fourier transform on the sound signal and output the feature vector; Pressure pulse entropy is extracted from the pressure signal, a feature vector is output, and then fused with the sound signal. (3) Perform multi-scale convolution branches and divide them into three branches: micro-wear, medium-wear and macro-wear, and then merge the branches; (4) Use attention mechanisms for frequency band weighting and deweighting; (5) The fully connected classification layer outputs the levels as micro wear level, medium wear level and macro wear level respectively; (6) Determine the radial compensation amount based on the wear level; (7) Input the radial compensation amount into the computer for G-code update generation; (8) The updated G code is input into the grinding machine control system to adjust the grinding wheel trajectory in real time and realize adaptive closed-loop control.

[0014] In step (3), Micro-wear branch (1*1 convolution kernel): Moderately worn branch (3x3 convolution kernel): Macroscopic wear branch (5*5 convolution kernel): In the formula, W n b is the convolution kernel weight matrix; n For bias terms; The convolutional branches are then merged and concatenated using the following formula: .

[0015] In step (4), the attention mechanism enhances the key frequency band features: In the formula, Q and K are the query matrix and key matrix, respectively, derived from... Obtained by transformation; For feature dimensions; Frequency band attention weights; The weighted feature is .

[0016] In step (5), the fully connected classification layer is used to output the rank according to the following formula. In the formula, c represents the output level.

[0017] The beneficial effects of this invention are: (1) This invention discloses a tool grinding wheel dressing device and method based on a ring array and dynamic multi-scale convolution. A ring array sensor monitoring device is set at the tool grinding wheel. The sensor is encapsulated in the flange. The pressure-sound dual signal ring piezoelectric array sensor is used to collect the signal, which effectively solves the problems of poor anti-interference ability, single detection signal and single monitoring area of ​​traditional sensors. The signal processing is performed by using a multi-scale dynamic convolution network to fuse the dual signals of sound and dressing force, realize the accurate classification of wear state and the location of abrasive chipping, and generate radial displacement adaptive compensation command. This can effectively solve the problems of low downtime detection efficiency and large human judgment error in the traditional dressing process.

[0018] (2) By reducing the downtime monitoring time of tool grinding wheels through both structure and algorithm, the dressing accuracy and efficiency of diamond rollers are improved, thereby improving the processing quality of high value-added parts such as aerospace gears and precision reducer gears.

[0019] (3) The integrated elastic silicone damping layer isolates the vibration noise of the main shaft, and the straight hole microchannel achieves physical cooling through gas exchange, solving the problems of poor anti-interference ability and thermal noise impact of traditional monitoring systems.

[0020] (4) A ring array of pressure and sound sensors is used, which are evenly distributed in staggered layers at 120° intervals to cover the entire circumferential area of ​​the grinding wheel, thus solving the problem of single sensor signal and insufficient area coverage in traditional single sensors.

[0021] (5) By extracting wear level features of different branches through different convolution kernels, and using attention mechanism to weight key frequency bands, feature partitioning is enhanced, thus solving the problem of weak generalization ability of existing wavelet packet decomposition and support vector machine algorithms. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall device structure of the present invention; Figure 2 This is a cross-sectional view of the ring array sensor monitoring device of the present invention; Figure 3 This is a schematic diagram of the layout of a ring array sensor; Figure 4 This is the flow control diagram of the present invention. Detailed Implementation

[0023] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.

[0024] Please see Figure 1It should be understood that the structures, proportions, sizes, etc., illustrated in the accompanying drawings are merely for illustrative purposes to aid those skilled in the art and to facilitate understanding and reading. They are not intended to limit the scope of the invention and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, without affecting the effectiveness and purpose of the invention, should still fall within the scope of the technical content disclosed in this invention. Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention's implementation.

[0025] This invention provides a tool grinding wheel dressing device and method based on a ring array and dynamic multi-scale convolution, such as... Figures 1 to 4 As shown.

[0026] A tool wheel dressing device based on a ring array and dynamic multi-scale convolution includes a grinding machine 1 and a tool wheel 4 and a diamond roller 5 mounted on the grinding machine 1. A ring array sensor monitoring device is also provided at the tool wheel 4. The ring array sensor monitoring device includes a rear cover 11 of the encapsulated flange, a front cover 12 of the encapsulated flange, and a conductive slip ring 17. A pressure film sensor 13 and an acoustic emission sensor 19 are also provided inside the front cover 12 of the encapsulated flange. The signal transmission lines of the pressure film sensor 13 and the acoustic emission sensor 19 are connected to the rotor side of the conductive slip ring 17 through a low-noise cable 16. The stator side of the conductive slip ring 17 is fixed to the other side of the grinding machine 1, which solves the winding problem caused by the spindle rotation.

[0027] The flange is assembled and fixed to the tool grinding wheel 4 by threaded clamping and installed to the grinding machine spindle 3 by bolts 18, and the internal radial pressure and internal acoustic emission signal inside the hole are monitored in real time.

[0028] The outer layer of the pressure film sensor 13 and acoustic emission sensor 19 is also provided with a straight-hole microchannel 15 for gas exchange. The high-speed rotation of the main shaft facilitates gas exchange, thereby achieving physical cooling and reducing the impact of high temperature on the pressure-sensitive material and thermal noise interference. The conductive slip ring 17 is also connected to a preamplifier 8, and the data is transmitted to the computer control terminal 10 via the data acquisition card 9. The pressure film sensors 13 are evenly distributed at 120° intervals inside the front cover 12 of the encapsulation flange. They use piezoelectric signals to collect the stress changes generated when the tool grinding wheel 4 contacts the diamond roller 5 to characterize the pressure changes under different dressing amounts. The acoustic emission sensors 19 are evenly distributed at 120° intervals inside the pressure film sensors 13. They are integrated inside the tool grinding wheel 4 to directly monitor the abrasive wear and chipping sound signals during the dressing process.

[0029] The acoustic emission sensor 19 is also provided with an elastic silicone damping layer 20 inside, and the elastic silicone damping layer 20 is arranged in a ring contact inside the acoustic emission sensor 19. The elastic silicone damping layer 20 is in direct contact with the acoustic emission sensor 19 to isolate the rotation noise level of the grinding machine spindle 3 and the vibration noise of the grinding machine.

[0030] In this invention, a digital filter is used to filter out high-frequency noise >500KHz and low-frequency noise <50KHz, and differential vibration is used to suppress diamond roller noise ≈300KHz. The pressure signal is processed by sliding smoothing, and dynamic pulse entropy extraction is performed to collect sound and pressure signals. The sound signal is subjected to short-time Fourier transform to output a 64-dimensional feature vector. The pressure signal is subjected to pressure pulse entropy extraction to output a 16-dimensional feature vector, which is then fused into an 80-dimensional feature vector.

[0031] The method for dressing a tool grinding wheel based on a ring array and dynamic multi-scale convolution includes the following steps: (1) Use digital filters to filter out high-frequency noise >500KHz and low-frequency noise <50KHz, and use differential vibration to suppress diamond roller noise ≈300KHz; The pressure signal is processed by sliding smoothing, dynamic pulse entropy extraction is performed, and sound and pressure signals are collected. (2) Perform a short-time Fourier transform on the sound signal and output the feature vector; Pressure pulse entropy is extracted from the pressure signal, a feature vector is output, and then fused with the sound signal. (3) Perform multi-scale convolution branches and divide them into three branches: micro-wear, medium-wear and macro-wear, and then merge the branches; Micro-wear branch (1*1 convolution kernel): Moderately worn branch (3x3 convolution kernel): Macroscopic wear branch (5*5 convolution kernel): In the formula, W n b is the convolution kernel weight matrix; n For bias terms; The convolutional branches are then merged and concatenated using the following formula: .

[0032] (4) Use attention mechanisms for frequency band weighting and deweighting; In the formula, Q and K are the query matrix and key matrix, respectively, derived from... Obtained by transformation; For feature dimensions; Frequency band attention weights; The weighted feature is .

[0033] (5) The fully connected classification layer outputs the levels as micro-wear level, medium wear level, and macro-wear level, respectively; the fully connected classification layer outputs the levels according to the following formula. In the formula, c represents the output level.

[0034] (6) Determine the radial compensation amount based on the wear level.

[0035] (7) Input the radial compensation amount into the computer to generate G-code update.

[0036] (8) The updated G code is input into the grinding machine control system to adjust the grinding wheel trajectory in real time and realize adaptive closed-loop control.

[0037] This invention utilizes a pressure-acoustic emission dual-layer ring piezoelectric array sensor to collaboratively acquire pressure and sound signals during the grinding process via a conductive slip ring equipped with a low-noise cable. The structure is simple, and the circumferential signal measurement is uniform and complete, effectively solving the problems of weak anti-interference capability, single signal acquisition, and single signal monitoring area in existing detection fields. It employs a multi-scale convolutional-attention network (MSCAN) to fuse the acoustic emission signal's temporal spectrum and dressing force data, extracting a grinding wheel wear level map to achieve accurate classification of wear states and location of abrasive chipping. This generates adaptive radial displacement compensation commands, effectively solving the problems of low downtime detection efficiency and large human judgment errors in traditional dressing processes. By reducing tool grinding wheel downtime monitoring time through both structural and algorithmic improvements, it enhances the accuracy and efficiency of diamond roller dressing, thereby improving the processing quality of high-value-added parts such as aerospace gears and precision reducer gears.

[0038] If this patent uses terms such as "first" and "second" to define components, those skilled in the art should know that the use of "first" and "second" is merely for the convenience of describing the invention and simplifying the description, and the above terms have no special meaning.

[0039] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection of this invention is defined by the appended claims and their equivalents.

[0040] In the description of this invention, it should be understood that the terms "front", "rear", "left", "right", "center", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only used to facilitate the description of this invention and to simplify the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the scope of protection of this invention.

[0041] The endpoints and any values ​​of the ranges disclosed herein are not limited to the exact ranges or values, and these ranges or values ​​should be understood to include values ​​close to these ranges or values. For numerical ranges, the endpoint values ​​of the various ranges, the endpoint values ​​of the various ranges and individual point values, and individual point values ​​can be combined with each other to obtain one or more new numerical ranges, which should be considered as specifically disclosed herein.

Claims

1. A method for dressing a tool grinding wheel based on a ring array and dynamic multi-scale convolution, characterized in that: The dressing device includes a grinding machine and a tool grinding wheel and a diamond roller mounted on the grinding machine, and a ring array sensor monitoring device is also provided at the tool grinding wheel; The ring array sensor monitoring device includes a rear cover of the encapsulated flange, a front cover of the encapsulated flange, and a conductive slip ring. The front cover of the encapsulated flange is also equipped with a pressure film sensor and an acoustic emission sensor. The signal transmission lines of the pressure film sensor and the acoustic emission sensor are connected to the rotor side of the conductive slip ring via cables, and the stator side of the conductive slip ring is fixed to the other side of the grinding machine. The flange is assembled and clamped onto the tool grinding wheel and installed onto the grinding machine spindle, and the internal radial pressure and internal acoustic emission signal inside the hole are monitored in real time. The outer layer of the pressure film sensor and acoustic emission sensor is also provided with straight-hole microchannels for gas exchange. The method for adjusting the device includes the following steps: (1) Use digital filters to filter out high-frequency noise >500KHz and low-frequency noise <50KHz, and use differential vibration to suppress diamond roller noise ≈300KHz; The pressure signal is processed by sliding smoothing, dynamic pulse entropy extraction is performed, and sound and pressure signals are collected. (2) Perform a short-time Fourier transform on the sound signal and output the feature vector; Pressure pulse entropy is extracted from the pressure signal, a feature vector is output, and then fused with the sound signal. (3) Perform multi-scale convolution branches and divide them into three branches: micro-wear, medium-wear and macro-wear, and then merge the branches; (4) Use attention mechanisms for frequency band weighting and deweighting; (5) The fully connected classification layer outputs the levels as micro wear level, medium wear level and macro wear level respectively; (6) Determine the radial compensation amount based on the wear level; (7) Input the radial compensation amount into the computer for G-code update generation; (8) The updated G code is input into the grinding machine control system to adjust the grinding wheel trajectory in real time and realize adaptive closed-loop control.

2. The method of the tool grinding wheel dressing device based on ring array and dynamic multi-scale convolution according to claim 1, characterized in that, In step (3), Micro-wear branch - 1*1 convolution kernel: ； Moderate wear branch - 3*3 convolution kernel: ； Macroscopic wear branch - 5*5 convolution kernel: ； In the formula, W n b is the convolution kernel weight matrix; n For bias terms; The convolutional branches are then merged and concatenated using the following formula: 。 3. The method of the tool grinding wheel dressing device based on ring array and dynamic multi-scale convolution according to claim 1, characterized in that, In step (4), the attention mechanism enhances the key frequency band features: ; In the formula, Q and K are the query matrix and key matrix, respectively, derived from... Obtained by transformation; For feature dimensions; Frequency band attention weights; The weighted feature is: 。 4. The method of the tool grinding wheel dressing device based on ring array and dynamic multi-scale convolution according to claim 1, characterized in that, In step (5), the fully connected classification layer is used to output the rank according to the following formula. ； In the formula, c represents the output level.

5. The method for dressing a tool grinding wheel based on a ring array and dynamic multi-scale convolution according to claim 1, characterized in that: The acoustic emission sensor is also provided with an elastic silicone damping layer inside, and the elastic silicone damping layer is arranged in a ring-shaped contact within the acoustic emission sensor.

6. The method of the tool grinding wheel dressing device based on ring array and dynamic multi-scale convolution according to claim 1, characterized in that: The conductive slip ring is also equipped with a preamplifier, and the data is transmitted to the computer control terminal via a data acquisition card.

7. The method for dressing a tool grinding wheel based on a ring array and dynamic multi-scale convolution according to claim 1, characterized in that: The pressure diaphragm sensors are evenly distributed at 120° intervals inside the front cover of the encapsulation flange.

8. The method for dressing a tool grinding wheel based on a ring array and dynamic multi-scale convolution according to claim 7, characterized in that: The acoustic emission sensors are evenly distributed at 120° intervals on the inside of the pressure film sensor.

Images