Grinding ratio measurement method based on grinding wheel multi-information fusion model

Abstract

This invention relates to the field of grinding wheel monitoring technology and discloses a method for measuring grinding wheel grinding ratio based on a multi-information fusion model of the grinding wheel. The method includes: collecting vibration, acoustic emission, and current signals during the grinding process to form an initial signal set; performing time-frequency transformation to extract spectral features and generate an initial feature vector; establishing a grinding wheel service state feature library to store reference feature vector clusters of different wear degrees; inputting the initial feature vector into a multi-channel feature matching network, simultaneously calling the feature library to complete feature matching and state classification, and outputting the grinding wheel wear level; retrieving benchmark grinding ratio data based on the wear level, and coupling it with the initial feature vector under grinding process constraints to generate a real-time grinding ratio. This invention achieves accurate identification of grinding wheel wear state and real-time dynamic measurement of the grinding ratio, improving the accuracy and adaptability of grinding process monitoring.

Description

Technical Field

[0001] This invention relates to the field of grinding wheel monitoring technology, specifically to a grinding wheel grinding ratio measurement method based on a grinding wheel multi-information fusion model. Background Technology

[0002] Accurate real-time measurement of the grinding ratio is a crucial aspect of precision grinding. Existing technologies often employ single sensor signals, such as vibration or acoustic emission signals, to determine the grinding wheel's wear state by setting fixed thresholds or simple empirical models. These methods rely on calibration data under specific operating conditions, and their monitoring accuracy decreases when grinding process parameters or workpiece materials change. Traditional methods struggle to establish a dynamic relationship between wear state and grinding ratio, often relying on offline measurements or theoretical formulas to calculate the grinding ratio, which cannot adapt to fluctuations in the wear state during real-time machining.

[0003] Existing technical solutions have significant drawbacks. A single signal source provides limited information and cannot fully reflect the complex process of wheel passivation and workpiece material removal. Fixed thresholds or empirical models lack adaptability and are prone to misjudgment in variable production environments. Furthermore, the measurement of the grinding ratio often lags behind the actual machining process, failing to provide immediate feedback for process adjustments. This results in low wheel utilization, unstable machining quality, and increased production costs.

[0004] This invention addresses the challenges of accurately identifying the wear state of grinding wheels online and dynamically generating real-time grinding ratios. By fusing multi-source signals and constructing an intelligent matching mechanism, it overcomes the shortcomings of incomplete information and rigid models in existing technologies, thereby improving the accuracy and real-time performance of grinding process monitoring. Summary of the Invention

[0005] The purpose of this invention is to provide a grinding wheel grinding ratio measurement method based on a grinding wheel multi-information fusion model, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a method for measuring the grinding ratio of a grinding wheel based on a multi-information fusion model of the grinding wheel, the method comprising: The vibration, acoustic emission, and current signals of the target grinding wheel during the grinding process are collected to form an initial signal set; Perform time-frequency transformation on the initial signal set to extract the spectral features related to grinding wheel passivation and workpiece material removal, and generate an initial feature vector; A grinding wheel service status feature library is established, which stores a cluster of reference feature vectors generated from grinding wheel sample signals with different wear levels; The initial feature vector is input into a preset multi-channel feature matching network. The multi-channel feature matching network synchronously calls the reference feature vector cluster in the grinding wheel service status feature library to complete feature matching and status classification, and outputs the preliminary grinding wheel wear level. Based on the preliminary grinding wheel wear level, the corresponding benchmark grinding ratio data is retrieved from the grinding wheel service condition feature library; Under the preset grinding process constraints, the reference grinding ratio data and the initial feature vector are coupled and calculated to finally generate the real-time grinding ratio of the target grinding wheel.

[0007] Preferably, the process of performing time-frequency transformation on the initial signal set includes: Empirical mode decomposition (EMD) is performed on the vibration signal to obtain a set of intrinsic mode function (EMF) components containing different time scales. Wavelet packet decomposition is performed on the acoustic emission signal to calculate the proportion of energy in each frequency band to the total energy after decomposition, thus constructing the acoustic emission energy spectrum. Fast Fourier transform (FFT) is performed on the current signal to identify the characteristic harmonic components and their amplitudes related to grinding force fluctuations in its spectrum. The variances of the EMF components, the distribution vector of the acoustic emission energy spectrum, and the amplitudes of the characteristic harmonic components are normalized and concatenated to form the initial feature vector.

[0008] Preferably, the specific steps for establishing the grinding wheel service condition feature library are as follows: Prepare a series of grinding wheel samples at different wear stages. Under standard grinding parameters, collect the full life cycle grinding signal of each grinding wheel sample. Segment the full life cycle grinding signal to obtain signal segments corresponding to multiple equally spaced material removal volumes. Each signal segment is processed to generate a corresponding feature vector segment; based on the final weighed wear amount of the grinding wheel sample, the theoretical grinding ratio value corresponding to each feature vector segment is manually labeled to form a data pair with a one-to-one correspondence between the feature vector segment and the theoretical grinding ratio value; Data pairs from all grinding wheel samples are clustered according to the similarity of feature vector segments to form multiple reference feature vector clusters. Each reference feature vector cluster and its corresponding theoretical grinding ratio range are stored in the grinding wheel service condition feature library.

[0009] The preferred workflow of the multi-channel feature matching network is as follows: The multi-channel feature matching network includes three independent feature extraction channels corresponding to vibration features, acoustic emission features, and current features. The initial feature vector is decomposed and input into the three independent feature extraction channels respectively. In each channel, the Euclidean distance with the corresponding type of reference feature vector cluster in the grinding wheel service status feature library is calculated. The Euclidean distances calculated by the three channels are weighted and fused to obtain a comprehensive matching degree vector. The minimum value in the comprehensive matching degree vector is identified, and the wear level associated with the reference feature vector cluster corresponding to the minimum value is determined as the preliminary grinding wheel wear level.

[0010] Preferably, the process of retrieving the reference grinding ratio data includes: Based on the preliminary grinding wheel wear level, the reference feature vector cluster to which it belongs is located in the grinding wheel service condition feature library; from the located reference feature vector cluster, a preset number of data pairs are randomly selected, and the theoretical grinding ratio values ​​marked on these data pairs are read; the average value of the read theoretical grinding ratio values ​​is calculated, and this average value is used as the benchmark grinding ratio data corresponding to the current grinding wheel condition.

[0011] Preferably, the steps for setting grinding process constraints are as follows: The grinding wheel linear speed, workpiece feed speed, and grinding depth used in the current grinding process are obtained to form a process parameter group; multiple typical process parameter combinations and their corresponding process influence coefficients are preset; the process parameter group is matched with the typical process parameter combinations, the matching degree is calculated, and the process influence coefficient corresponding to the typical process parameter combination with the highest matching degree is selected as the constraint coefficient of the current grinding process.

[0012] Preferably, the specific operation for generating the real-time grinding ratio through coupled computation is as follows: The acoustic emission dominant frequency amplitude variation rate, which characterizes the self-sharpening tendency of the grinding wheel, and the root mean square value of the vibration signal envelope, which characterizes the grinding stability, are separated from the initial eigenvector. The acoustic emission dominant frequency amplitude variation rate and the root mean square value of the vibration signal envelope are multiplied by the process influence coefficient to obtain the dynamic correction factor. The reference grinding ratio data is added to the dynamic correction factor, and the sum is multiplied by a normalization factor determined by the current harmonic distortion rate. The final product is defined as the real-time grinding ratio.

[0013] Preferably, when collecting vibration, acoustic emission, and current signals of the target grinding wheel during the grinding process, further synchronous calibration is performed, specifically including: A vibration acceleration sensor is installed on the grinding wheel spindle housing, an acoustic emission sensor is installed near the grinding fluid nozzle, and a current transformer is installed in the main motor power supply circuit. Before starting grinding, a synchronization trigger command is sent to the system to reset the data acquisition clocks of the three sensors and start synchronous timing. This ensures that the timestamps of the vibration signal, acoustic emission signal, and current signal are strictly aligned throughout the entire signal acquisition process.

[0014] Preferred methods for manually labeling theoretical grinding ratio values ​​for feature vector segments include: Before collecting the full life cycle grinding signals of the grinding wheel sample, the initial mass of the grinding wheel sample is accurately weighed. After completing the entire life cycle grinding experiment, the grinding wheel sample is cleaned and dried, and its final mass is accurately weighed again. The difference between the two is the total wear mass. The total mass difference of the workpiece before and after the experiment is measured to obtain the total material removal mass. The total material removal mass is divided by the total wear mass to obtain the overall average grinding ratio of the grinding wheel sample under standard parameters. According to the proportion of the material removal volume corresponding to the signal segment to the total volume, the overall average grinding ratio is proportionally allocated as the theoretical grinding ratio value corresponding to the signal segment.

[0015] Preferably, the smoothing process for the real-time grinding ratio before output includes: The real-time grinding ratios at multiple time points are continuously collected and calculated to form a real-time grinding ratio sequence. A sliding time window is used to truncate the real-time grinding ratio sequence, and the median of all real-time grinding ratio values ​​within the window is calculated. The real-time grinding ratio value at the center of the window is replaced with this median, and the window is slid until the smoothing of the entire sequence is completed, and the final smoothed grinding ratio measurement result is output.

[0016] Compared with the prior art, the beneficial effects of the present invention are: A feature library for grinding wheel service conditions was established, storing reference feature vector clusters generated from grinding wheel sample signals of different wear levels. A multi-channel feature matching network was designed. Initial feature vectors acquired in real-time were synchronously matched and classified with the reference vector clusters in the feature library through this network, directly outputting the grinding wheel wear level. Based on the multi-source signal feature library and dedicated matching network, high-precision, adaptive pattern recognition of wear states was achieved. The accuracy of wear state classification was improved, avoiding misjudgments caused by relying solely on empirical models or fixed thresholds, and enhancing the robustness of the system under different grinding conditions.

[0017] Based on the wear level output by the matching network, the corresponding benchmark grinding ratio data is retrieved from the feature library. Combined with the real-time extracted initial feature vector, a coupling operation is performed under preset grinding process constraints to generate the real-time grinding ratio of the target grinding wheel. By dynamically coupling the real-time feature vector with the benchmark data and incorporating online correction based on process constraints, the grinding ratio is adaptively updated in real-time according to the grinding wheel state and process conditions. The grinding ratio measurement results can instantly reflect changes in the processing state, improving measurement accuracy and adaptability to operating conditions, making the grinding process control more precise and efficient. Attached Figure Description

[0018] Figure 1This is a schematic diagram illustrating the working principle of the grinding ratio measurement method based on a multi-information fusion model of grinding wheels as described in this invention. Figure 2 This is a flowchart of the time-frequency transformation and feature extraction of the initial signal set; Figure 3 Flowchart for establishing a service status characteristic library for grinding wheels; Figure 4 A bar chart showing the Euclidean distance distribution of multi-channel features of a grinding wheel; Figure 5 This is a time-series comparison chart of the original and smoothed values ​​of the grinding wheel ratio. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 This invention provides a method for measuring the grinding ratio of a grinding wheel based on a multi-information fusion model. The method includes: acquiring vibration signals, acoustic emission signals, and current signals of the target grinding wheel during the grinding process, which together constitute an initial signal set; performing time-frequency transformation on the initial signal set to extract spectral features related to grinding wheel passivation and workpiece material removal, generating an initial feature vector; establishing a grinding wheel service state feature library, which stores reference feature vector clusters generated from grinding wheel sample signals with different wear levels; inputting the initial feature vector into a preset multi-channel feature matching network, which synchronously calls the reference feature vector clusters in the grinding wheel service state feature library to complete feature matching and state classification, outputting a preliminary grinding wheel wear level; based on the preliminary grinding wheel wear level, retrieving corresponding benchmark grinding ratio data from the grinding wheel service state feature library; and under preset grinding process constraints, coupling the benchmark grinding ratio data with the initial feature vector to finally generate the real-time grinding ratio of the target grinding wheel.

[0021] In one embodiment of the present invention, see [reference] Figure 2Empirical mode decomposition (EMD) is performed on the vibration signal to obtain a set of intrinsic mode function (EMF) components containing different time scales. Wavelet packet decomposition is performed on the acoustic emission signal to calculate the proportion of energy in each frequency band to the total energy, thus constructing the acoustic emission energy spectrum. Fast Fourier Transform (FFT) is performed on the current signal to identify the characteristic harmonic components and their amplitudes related to grinding force fluctuations in its spectrum. The variances of the EMF components, the distribution vector of the acoustic emission energy spectrum, and the amplitudes of the characteristic harmonic components are normalized and concatenated to form an initial eigenvector.

[0022] In practical implementation, the process of performing time-frequency transformation on the initial signal set to extract spectral features related to wheel passivation and workpiece material removal and generate an initial feature vector can be clearly illustrated by combining specific grinding scenarios, such as using a ceramic-bonded CBN grinding wheel to perform surface grinding on a 45# steel workpiece. After acquiring the vibration signal, acoustic emission signal, and current signal of the target grinding wheel during the grinding process, a time-synchronized initial signal set is formed. For the vibration signal, empirical mode decomposition is performed. This decomposition process adaptively decomposes the non-stationary original vibration waveform into a series of intrinsic mode function components arranged from high frequency to low frequency. For example, eight intrinsic mode function components are obtained, forming a set of intrinsic mode function components containing different time scales.

[0023] In some embodiments, wavelet packet decomposition is used to process the acoustic emission signal. The 'db4' wavelet basis function is selected for a 4-level decomposition, dividing the full frequency band of the acoustic emission signal into 16 independent sub-bands. The signal energy of each sub-band is calculated, and then the proportion of each sub-band's energy to the total energy is calculated, forming a 16-dimensional distribution vector. This distribution vector is the acoustic emission energy spectrum, characterizing the distribution of the acoustic emission signal energy in the frequency domain. For the current signal, a fast Fourier transform is performed to convert the time-domain current waveform into a spectrum. Characteristic harmonic components corresponding to the periodic fluctuations of the grinding force are identified in the spectrum, such as the second and third harmonics of the 50Hz power frequency, and the amplitudes of these characteristic harmonic components are recorded.

[0024] Understandably, after extracting the independent features of each signal, data integration is required to construct an initial feature vector. The variances of each intrinsic mode function component obtained from empirical mode decomposition are calculated, for example, resulting in eight variance values. The set of variances of the intrinsic mode function components, the 16-dimensional distribution vector of the acoustic emission energy spectrum, and the set of amplitudes of the identified characteristic harmonic components are then normalized to eliminate dimensional influences. All normalized values ​​are then concatenated in a predetermined order, such as vibration characteristics, acoustic emission characteristics, and current characteristics, ultimately forming a multi-dimensional comprehensive initial feature vector.

[0025] It is understandable that the construction of the acoustic emission energy spectrum involves specific mathematical calculations. For an acoustic emission signal after wavelet packet decomposition, the energy proportion of its i-th sub-band is calculated using the following formula: ; Where: in the formula This represents the proportion of energy in the i-th sub-band to the total energy, as shown in the formula. This represents the k-th discrete data point of the reconstructed signal in the i-th sub-band, in the formula. The formula represents the total number of data points for each sub-band signal. The formula represents the total number of subbands after wavelet packet decomposition. For sub-band indexing, This represents an approximate representation of the energy of the k-th data point of the i-th sub-band signal.

[0026] In practical implementation, the essence of this feature extraction method can be clarified through specific data comparison. The original vibration time-domain waveform, acoustic emission time-domain waveform, and current time-domain waveform are all one-dimensional sequences that change with time. After the described time-frequency transformation processing, the vibration signal is transformed into a variance sequence that represents the distribution of its energy on different intrinsic mode function components, the acoustic emission signal is transformed into a vector that represents the proportion of its energy distribution in different frequency intervals, and the current signal is transformed into a set of amplitudes that represent the intensity of specific frequency components. In practical implementation, the essence of this feature extraction method can be clarified through specific data comparison. For example, in the case of surface grinding of a 45# steel workpiece with a ceramic-bonded CBN grinding wheel, the original vibration time-domain waveform is a discrete sequence containing 1024 sampling points. After empirical mode decomposition, eight intrinsic mode function components are generated. The variance of each component is calculated to form an 8-dimensional variance sequence, which is used to characterize the distribution of vibration energy at different time scales. The original acoustic emission time-domain waveform is also a sequence of 1024 sampling points. It is divided into 16 sub-bands through wavelet packet decomposition. The energy proportion of each sub-band is calculated to form a 16-dimensional distribution vector, which intuitively reflects the distribution ratio of acoustic emission energy in the frequency range. After the original current time-domain waveform is processed by fast Fourier transform, the second harmonic amplitude of 50Hz power frequency with an amplitude of 0.5 volts and the third harmonic amplitude with an amplitude of 0.3 volts are identified in the spectrum, forming an amplitude set. This conversion from the time domain to the time-frequency domain or frequency domain allows the periodicity, impact, and stability information hidden in the original signal that is closely related to the wear state of the grinding wheel to be highlighted.

[0027] In one embodiment of the present invention, see [reference] Figure 3A series of grinding wheel samples at different wear stages were prepared. Under standard grinding parameters, the full lifecycle grinding signal of each grinding wheel sample was collected. The full lifecycle grinding signal was segmented to obtain signal segments corresponding to multiple equally spaced material removal volumes. Each signal segment was processed to generate a corresponding feature vector segment. Based on the final weighed wear amount of the grinding wheel sample, the theoretical grinding ratio value corresponding to each feature vector segment was manually labeled, forming a one-to-one correspondence between feature vector segments and theoretical grinding ratio values. The data pairs from all grinding wheel samples were clustered according to the similarity of the feature vector segments to form multiple reference feature vector clusters. Each reference feature vector cluster and its corresponding theoretical grinding ratio value range were stored in the grinding wheel service status feature library. The method for manually labeling the theoretical grinding ratio values ​​for the feature vector segments included accurately weighing the initial mass of the grinding wheel sample before collecting the full lifecycle grinding signal. After completing the entire full lifecycle grinding experiment, the grinding wheel sample was cleaned and dried, and its final mass was accurately weighed again; the difference between the two was the total wear mass. The total mass difference of the workpiece before and after the experiment is measured to obtain the total material removal mass. The total material removal mass is divided by the total wear mass to obtain the overall average grinding ratio of the grinding wheel sample under standard parameters. Based on the proportion of the material removal volume corresponding to the signal segment to the total volume, the overall average grinding ratio is proportionally allocated as the theoretical grinding ratio value corresponding to that signal segment.

[0028] In practical implementation, an example scenario is used: a cylindrical grinding experiment is conducted on a batch of titanium alloy workpieces using five identical resin-bonded white corundum grinding wheels. A series of grinding wheel samples at different wear stages are prepared, representing the complete wear state from brand new to fully passivated. Under standard grinding parameters—fixed wheel linear speed, workpiece rotation speed, and constant depth of cut feed—the full lifecycle grinding signal of each grinding wheel sample is collected from the start of use until it reaches a predetermined wear standard. Each segment of the full lifecycle grinding signal is segmented based on the volume of material removed from the workpiece, obtaining signal segments corresponding to multiple equally spaced material removal volumes. For example, a 2-second signal segment is extracted for every 10 cubic millimeters of workpiece material removed.

[0029] In some embodiments, each signal segment obtained by dividing the signal by material removal volume is processed, namely, empirical mode decomposition is performed on the vibration signal within the segment and the variance of the intrinsic mode function components is calculated; wavelet packet decomposition is performed on the acoustic emission signal and the acoustic emission energy spectrum is constructed; fast Fourier transform is performed on the current signal and the amplitude of characteristic harmonic components is identified; finally, the normalized features are spliced ​​together to generate a feature vector segment corresponding to the signal segment. Based on the final weighed wear amount of the grinding wheel sample, the theoretical grinding ratio value corresponding to each feature vector segment is manually labeled, forming a data pair with a one-to-one correspondence between the feature vector segment and the theoretical grinding ratio value.

[0030] The method for manually labeling theoretical grinding ratio values ​​for feature vector segments can be understood to include the following: Before acquiring the full lifecycle grinding signal of any grinding wheel sample, the initial mass of the grinding wheel sample is accurately weighed using a precision balance and recorded. After the grinding wheel sample completes the entire lifecycle grinding experiment, it is thoroughly cleaned and dried using a detergent and an ultrasonic cleaner. Its final mass is then accurately weighed again using the same precision balance. The difference between the initial mass and the final mass is the total wear mass of the grinding wheel sample. Simultaneously, the total mass of all experimental workpieces ground under this grinding wheel sample before and after grinding is measured, and the difference in total mass is calculated to obtain the total material removal mass from the grinding wheel sample. The total material removal mass is divided by the total wear mass to obtain the overall average grinding ratio of the grinding wheel sample under standard parameters. Based on the proportion of the material removal volume corresponding to each signal segment to the total material removal volume over the entire lifecycle, the overall average grinding ratio is allocated according to this proportion, and the calculated value is used as the theoretical grinding ratio value corresponding to that signal segment. The allocation process follows the following relationship: ; Where: in the formula This represents the theoretical grinding ratio of the current signal segment to be labeled, in the formula. This represents the calculated average grinding ratio of the entire grinding wheel sample, as shown in the formula. This represents the volume of material removed at equal intervals corresponding to the current signal segment, in the formula. This represents the total material removal volume during the entire lifecycle grinding experiment of the grinding wheel sample.

[0031] In practice, all data pairs generated by all five grinding wheel samples throughout their entire lifecycle are collected, with the total number of data pairs equal to the sum of the number of all signal segments. Clustering algorithms, such as K-Means clustering, are applied to cluster feature vector segments based on their similarity in the feature space, forming multiple reference feature vector clusters with similar characteristics. Each reference feature vector cluster contains several feature vector segments, and the theoretical grinding ratio values ​​associated with these feature vector segments form a numerical range. The central vector of each reference feature vector cluster and the corresponding theoretical grinding ratio value range are correlated and stored in a database, thus constructing a grinding wheel service status feature library. Through specific data comparison, it can be seen that the original data pairs without clustering are discrete and numerous points, while the reference feature vector clusters formed after clustering represent the typical characteristic patterns of the grinding wheel at a specific wear stage and their corresponding grinding ratio level ranges.

[0032] In one embodiment of the present invention, the multi-channel feature matching network includes three independent feature extraction channels corresponding to vibration features, acoustic emission features, and current features. The initial feature vector is decomposed and input into the three independent feature extraction channels respectively. Within each channel, the Euclidean distance to the corresponding type of reference feature vector cluster in the grinding wheel service state feature library is calculated. The Euclidean distances calculated by the three channels are weighted and fused to obtain a comprehensive matching degree vector. The minimum value in the comprehensive matching degree vector is identified, and the wear level associated with the reference feature vector cluster corresponding to the minimum value is determined as the preliminary grinding wheel wear level. The process of retrieving the benchmark grinding ratio data includes locating the reference feature vector cluster to which the preliminary grinding wheel wear level belongs in the grinding wheel service state feature library. From the located reference feature vector cluster, a preset number of data pairs are randomly selected, and the theoretical grinding ratio values ​​labeled on these data pairs are read. The average value of the read theoretical grinding ratio values ​​is calculated, and this average value is used as the benchmark grinding ratio data corresponding to the current grinding wheel state.

[0033] In practical implementation, the description is tailored to specific grinding scenarios, such as precision grinding of mold steel using CBN grinding wheels. The pre-defined multi-channel feature matching network includes three independent feature extraction channels corresponding to vibration features, acoustic emission features, and current features. Each independent feature extraction channel has the same internal structure but independent parameters. When the initial feature vector generated from the real-time grinding signal is input, it is decomposed into three parts—vibration feature sub-vector, acoustic emission feature sub-vector, and current feature sub-vector—according to a pre-defined dimensionality partitioning rule, and then input to their respective independent feature extraction channels.

[0034] In some embodiments, within each independent feature extraction channel, the system synchronously calls the corresponding type of reference feature vector cluster stored in the grinding wheel service status feature library. Specifically, the vibration feature independent feature extraction channel calls the vibration reference feature vector cluster composed of the vibration feature parts of all reference feature vector clusters in the grinding wheel service status feature library; the acoustic emission feature independent feature extraction channel calls the acoustic emission reference feature vector cluster composed of the acoustic emission feature parts of all reference feature vector clusters in the grinding wheel service status feature library; and the current feature independent feature extraction channel calls the current reference feature vector cluster composed of the current feature parts of all reference feature vector clusters in the grinding wheel service status feature library. Subsequently, within each independent feature extraction channel, the Euclidean distance between the input real-time feature sub-vector and each reference feature vector in the corresponding reference feature vector cluster is calculated. Taking the vibration channel as an example, the Euclidean distance between the real-time vibration feature sub-vector and the j-th vibration reference feature vector in the grinding wheel service status feature library is calculated using the following formula: ; Where: in the formula The Euclidean distance between the real-time vibration feature vector and the j-th vibration reference feature vector is expressed in the formula. The formula represents the total dimension of the vibration eigenvectors. The value of the k-th element in the real-time vibration feature vector is given in the formula. This represents the value of the kth element of the j-th vibration reference feature vector in the grinding wheel service condition feature library.

[0035] In practical implementation, the Euclidean distance set calculated from the three independent feature extraction channels is used. , , A weighted fusion is performed to obtain a comprehensive matching degree vector. Each element of the comprehensive matching degree vector corresponds to a reference feature vector cluster in the grinding wheel service condition feature library, and its value is obtained by weighted summation of the distance values ​​of that cluster across the three channels. The minimum value in the comprehensive matching degree vector is identified, and the wear level pre-associated in the grinding wheel service condition feature library for the reference feature vector cluster corresponding to the minimum value is determined as the initial grinding wheel wear level of the current target grinding wheel. For example, data comparison shows that when the comprehensive matching degree between the real-time signal features and the reference feature vector cluster representing "moderate wear" is the highest, the calculated comprehensive matching degree vector has the smallest value corresponding to the "moderate wear" cluster, thus the initial grinding wheel wear level is determined to be "moderate wear".

[0036] It is understandable that after outputting the initial grinding wheel wear level, the process of retrieving the reference grinding ratio data immediately begins. Based on the initial grinding wheel wear level, such as "moderate wear level 2," a specific reference feature vector cluster associated with it is located in the grinding wheel service condition feature library. From this located specific reference feature vector cluster, a preset number of data pairs are randomly selected; the preset number can be set to 5 to 10. The theoretical grinding ratio values ​​labeled on these randomly selected data pairs are read. The arithmetic mean of all read theoretical grinding ratio values ​​is calculated, and this arithmetic mean is used as the reference grinding ratio data corresponding to the current grinding wheel condition. Through specific data comparison, 8 data pairs are randomly selected from a "moderate wear level 2" reference feature vector cluster containing 20 data pairs. Their theoretical grinding ratio values ​​are 32.1, 33.5, 31.8, 32.9, 33.0, 32.4, 32.6, and 31.9, respectively. The calculated average value of 32.53 is retrieved as the reference grinding ratio data for the current condition.

[0037] See Figure 4This is a bar chart showing the Euclidean distance distribution of multi-channel features of a grinding wheel, primarily displaying the feature matching distances of vibration, acoustic emission, and current channels under different wear levels. The distances for all three channels corresponding to "Moderate Wear Level 2" are the smallest among all levels, indicating the highest matching degree between the real-time features and the reference features for this level. The distances for other levels are significantly larger, reflecting the discriminative power of the feature matching. This chart is used in the grinding wheel feature matching stage, visually presenting the matching effect of multi-channel features, helping technicians quickly locate the current wear level of the grinding wheel, and is a core reference tool in the grinding wheel grinding ratio measurement process.

[0038] In one embodiment of the present invention, the step of setting grinding process constraints is as follows: The grinding wheel linear velocity, workpiece feed rate, and grinding depth used in the current grinding process are obtained to form a process parameter set. Multiple typical process parameter combinations and their corresponding process influence coefficients are pre-set. The process parameter set is matched with the typical process parameter combinations, the matching degree is calculated, and the process influence coefficient corresponding to the typical process parameter combination with the highest matching degree is selected as the constraint coefficient of the current grinding process. When collecting vibration, acoustic emission, and current signals of the target grinding wheel during the grinding process, further synchronization calibration is performed. Specifically, this includes installing a vibration acceleration sensor on the grinding wheel spindle housing, an acoustic emission sensor near the grinding fluid nozzle, and a current transformer in the main motor power supply circuit. Before starting grinding, a synchronization trigger command is sent to the system to reset the data acquisition clocks of the three sensors and start synchronous timing. It is ensured that the timestamps of the vibration signal, acoustic emission signal, and current signal are strictly aligned throughout the entire signal acquisition process.

[0039] In practical implementation, a surface grinding scenario is used as an example. For instance, when grinding a stainless steel workpiece with a white corundum wheel on a CNC surface grinder, the wheel linear velocity, workpiece feed rate, and grinding depth are acquired. These three parameters constitute a set of process parameters describing the current processing conditions. Multiple typical process parameter combinations and their corresponding process influence coefficients are pre-defined in the system as a data table. These typical process parameter combinations cover common rough grinding, semi-finish grinding, and finish grinding processes. The real-time acquired process parameter set is then matched with the pre-defined typical process parameter combinations. The process requires dimensionless processing of each parameter in the current process parameter set and its corresponding parameter in a typical process parameter combination. This involves dividing each parameter by its own typical reference value to eliminate dimensions, converting the grinding wheel speed, workpiece feed rate, and grinding depth into dimensionless ratios. Then, the Euclidean distance between the processed current parameter set and each typical process parameter combination in multidimensional space is calculated. The process influence coefficient corresponding to the typical process parameter combination with the smallest distance is selected and used as the constraint coefficient for the current grinding process, used for subsequent correction calculations. The matching degree can be calculated based on the following relationship: ; Where: in the formula This represents the dimensionless distance between the current set of process parameters and the i-th typical combination of process parameters, as shown in the formula. This represents the current linear velocity of the grinding wheel, in the formula. The linear velocity of the grinding wheel for the i-th typical process parameter combination is given in the formula. The reference value representing the linear velocity of the grinding wheel is found in the formula. This represents the current workpiece feed rate, in the formula. This represents the workpiece feed rate for the i-th typical process parameter combination, in the formula. The reference value representing the workpiece feed rate is found in the formula. This indicates the current grinding depth, in the formula. The formula represents the grinding depth for the i-th typical process parameter combination. The reference value representing the depth of grinding, a dimensionless distance. The smaller the value, the higher the similarity. The process influence coefficient of the typical process parameter combination with the smallest distance is selected. See Table 1.

[0040] Table 1. Typical combinations of process parameters and their process influence coefficients: In some embodiments, when acquiring vibration, acoustic emission, and current signals of the target grinding wheel during the grinding process, strict synchronization calibration is required to ensure the consistency of multi-source signals in the time dimension. Specific synchronization calibration operations include rigidly mounting a vibration acceleration sensor on the grinding wheel spindle housing to acquire vibration signals, installing an acoustic emission sensor near the grinding fluid nozzle to acquire acoustic emission signals in the grinding zone, and installing a current transformer in the three-phase circuit of the main motor power supply circuit to acquire the main motor current signal. Before starting the grinding program, the control system sends a hardware synchronization trigger command to the data acquisition units of the three sensors. This synchronization trigger command simultaneously resets the data acquisition clocks of the vibration sensor, acoustic emission sensor, and current transformer to zero and begins synchronous timing with the same time base. Through this hardware synchronization mechanism, it is ensured that the timestamps of the three data streams—vibration signal, acoustic emission signal, and current signal—are strictly aligned throughout the entire signal acquisition process. Subsequent time-frequency transformation and feature extraction ensure that all features correspond to the grinding state at the same physical moment. It is understandable that the process parameter set can be obtained by reading it in real time through the CNC system, or by the operator manually inputting the current grinding wheel linear speed, workpiece feed speed and grinding depth values ​​on the human-machine interface. Typical process parameter combinations and their process influence coefficients are usually pre-established based on previous process experiments or domain knowledge bases.

[0041] In one embodiment of the present invention, the specific operation of generating the real-time grinding ratio through coupling operation is as follows: The rate of change of the acoustic emission dominant frequency amplitude, representing the self-sharpening tendency of the grinding wheel, and the root mean square value of the vibration signal envelope, representing grinding stability, are separated from the initial feature vector. The rate of change of the acoustic emission dominant frequency amplitude and the root mean square value of the vibration signal envelope are multiplied by the process influence coefficient to obtain a dynamic correction factor. The reference grinding ratio data is added to the dynamic correction factor, and the sum is multiplied by a normalization factor determined by the current harmonic distortion rate. The final product is defined as the real-time grinding ratio. Smoothing processing of the real-time grinding ratio before output includes: continuously acquiring and calculating the real-time grinding ratio at multiple time points to form a real-time grinding ratio sequence; using a sliding time window to truncate the real-time grinding ratio sequence and calculating the median of all real-time grinding ratio values ​​within the window; replacing the real-time grinding ratio value at the center of the window with this median, and sliding the window until the smoothing processing of the entire sequence is completed, outputting the smoothed final grinding ratio measurement result.

[0042] In practical implementation, the specific operation of generating the real-time grinding ratio through coupling computation can be described in the scenario of using a ceramic CBN grinding wheel to perform creep-feed grinding on a high-temperature alloy workpiece. From the initial feature vector generated based on the real-time grinding signal, the rate of change of the acoustic emission dominant frequency amplitude, representing the self-sharpening tendency of the grinding wheel, and the root mean square value of the vibration signal envelope, representing grinding stability, are separated. The rate of change of the acoustic emission dominant frequency amplitude, the root mean square value of the vibration signal envelope, and the process influence coefficient are multiplied to obtain a dynamic correction factor. The retrieved benchmark grinding ratio data is added to the dynamic correction factor, and the sum is then multiplied by a normalization factor determined by the current harmonic distortion rate. The final product is defined as the real-time grinding ratio. Before substituting into the formula, the rate of change of the acoustic emission dominant frequency amplitude and the root mean square value of the vibration signal envelope need to be divided by a preset standardization coefficient to eliminate dimensions, making them dimensionless values. The calculation relationship can be expressed as: ; Where: in the formula This represents the final real-time grinding ratio, as shown in the formula. This represents the reference grinding ratio data retrieved from the grinding wheel service condition feature library. In the formula... This represents the original value of the rate of change of the acoustic emission dominant frequency amplitude separated from the initial eigenvector. In the formula... The standardized coefficient representing the rate of change of the acoustic emission dominant frequency amplitude is given in the formula. This represents the original value of the root mean square value of the vibration signal envelope separated from the initial eigenvector. In the formula... The standardized coefficients representing the root mean square value of the vibration signal envelope are given in the formula. This represents the process influence coefficient obtained by matching the current process parameters. In the formula... This represents the calculated current harmonic distortion rate, in the formula. This is the normalization factor determined by the harmonic distortion rate of the current.

[0043] In some embodiments, the process of separating the acoustic emission dominant frequency amplitude change rate from the initial eigenvector involves determining the dominant frequency band in the acoustic emission signal energy spectrum and calculating the difference ratio of the dominant frequency band amplitude between adjacent signal analysis segments. The process of separating the vibration signal envelope root mean square value from the initial eigenvector involves performing a Hilbert transform on the original vibration signal to extract the envelope, and then calculating the root mean square value of the envelope. The current harmonic distortion rate is obtained by performing spectral analysis on the current signal and calculating the ratio of the sum of squares of each harmonic amplitude to the fundamental amplitude. The normalization factor is inversely proportional to the current harmonic distortion rate; the normalization factor decreases as the current harmonic distortion rate increases. Specific data comparisons show that, assuming the reference grinding ratio data... The original value of the rate of change of the acoustic emission dominant frequency amplitude is 32.5. -150s -1 Its standardized coefficient For 1000s -1 The original value of the root mean square value of the vibration signal envelope. 4.0 m / s 2 Its standardized coefficient The flow rate is 5.0 m / s², and the process influence coefficient is... The harmonic distortion rate is 1.1. If it is 0.02, then it is dimensionless. dimensionless The normalization factor is Substituting into the formula, the real-time grinding ratio is calculated. .

[0044] It is understandable that smoothing is performed on the real-time grinding ratio before output. The real-time grinding ratio is continuously collected and calculated at multiple time points. These real-time grinding ratio values ​​arranged in chronological order constitute a real-time grinding ratio sequence. A sliding time window of fixed width is used to truncate the real-time grinding ratio sequence, for example, a window containing 5 data points. The median of all real-time grinding ratio values ​​within the window is calculated, and the calculated median replaces the original real-time grinding ratio value corresponding to the center time of the window. Then, the sliding time window is moved forward by one time point, and the operations of truncation, median calculation, and replacement of the center point value are repeated until the sliding time window traverses the entire real-time grinding ratio sequence, completing the smoothing of the entire sequence. Finally, the smoothed final grinding ratio measurement result is output.

[0045] See Figure 5This is a time-series comparison chart of the original and smoothed values ​​of the grinding wheel's grinding ratio, primarily showcasing the dynamic changes in the grinding ratio and the effect of the smoothing process. The original grinding ratio fluctuates wildly, reflecting noise interference from real-time measurements; the smoothed grinding ratio curve is flatter, filtering out instantaneous fluctuations and better reflecting the trend of grinding wheel wear. This chart is used in the output stage of the grinding wheel's grinding ratio, visually presenting the noise in the original measurement data and the optimization effect of the smoothing process. It helps technicians obtain a more stable grinding ratio trend and is a core reference tool for grinding wheel condition assessment and process adjustment.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for measuring the grinding ratio of a grinding wheel based on a multi-information fusion model of the grinding wheel, characterized in that, The method includes: The vibration, acoustic emission, and current signals of the target grinding wheel during the grinding process are collected to form an initial signal set; Perform time-frequency transformation on the initial signal set to extract the spectral features related to grinding wheel passivation and workpiece material removal, and generate an initial feature vector; A grinding wheel service status feature library is established, which stores a cluster of reference feature vectors generated from grinding wheel sample signals with different wear levels; The initial feature vector is input into a preset multi-channel feature matching network. The multi-channel feature matching network synchronously calls the reference feature vector cluster in the grinding wheel service status feature library to complete feature matching and status classification, and outputs the preliminary grinding wheel wear level. Based on the preliminary grinding wheel wear level, the corresponding benchmark grinding ratio data is retrieved from the grinding wheel service condition feature library; Under the preset grinding process constraints, the reference grinding ratio data and the initial feature vector are coupled and calculated to finally generate the real-time grinding ratio of the target grinding wheel.

2. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 1, characterized in that, The process of performing time-frequency transformation on the initial signal set includes: Empirical mode decomposition (EMD) is performed on the vibration signal to obtain a set of intrinsic mode function (EMF) components containing different time scales. Wavelet packet decomposition is performed on the acoustic emission signal to calculate the proportion of energy in each frequency band to the total energy after decomposition, thus constructing the acoustic emission energy spectrum. Fast Fourier transform (FFT) is performed on the current signal to identify the characteristic harmonic components and their amplitudes related to grinding force fluctuations in its spectrum. The variances of the EMF components, the distribution vector of the acoustic emission energy spectrum, and the amplitudes of the characteristic harmonic components are normalized and concatenated to form the initial feature vector.

3. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 2, characterized in that, The specific steps for establishing a grinding wheel service condition characteristic database are as follows: Prepare a series of grinding wheel samples at different wear stages. Under standard grinding parameters, collect the full life cycle grinding signal of each grinding wheel sample. Segment the full life cycle grinding signal to obtain signal segments corresponding to multiple equally spaced material removal volumes. Each signal segment is processed to generate a corresponding feature vector segment; based on the final weighed wear amount of the grinding wheel sample, the theoretical grinding ratio value corresponding to each feature vector segment is manually labeled to form a data pair with a one-to-one correspondence between the feature vector segment and the theoretical grinding ratio value; Data pairs from all grinding wheel samples are clustered according to the similarity of feature vector segments to form multiple reference feature vector clusters. Each reference feature vector cluster and its corresponding theoretical grinding ratio range are stored in the grinding wheel service condition feature library.

4. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 3, characterized in that, The workflow of a multi-channel feature matching network is as follows: The multi-channel feature matching network includes three independent feature extraction channels corresponding to vibration features, acoustic emission features, and current features; The initial feature vector is decomposed and input into three independent feature extraction channels. In each channel, the Euclidean distance with the corresponding type of reference feature vector cluster in the grinding wheel service status feature library is calculated. The Euclidean distances calculated by the three channels are weighted and fused to obtain a comprehensive matching degree vector. The minimum value in the comprehensive matching degree vector is identified, and the wear level associated with the reference feature vector cluster corresponding to the minimum value is determined as the preliminary grinding wheel wear level.

5. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 4, characterized in that, The process of retrieving the reference grinding ratio data includes: Based on the preliminary grinding wheel wear level, the reference feature vector cluster to which it belongs is located in the grinding wheel service condition feature library; from the located reference feature vector cluster, a preset number of data pairs are randomly selected, and the theoretical grinding ratio values ​​marked on these data pairs are read; the average value of the read theoretical grinding ratio values ​​is calculated, and this average value is used as the benchmark grinding ratio data corresponding to the current grinding wheel condition.

6. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 5, characterized in that, The steps for setting grinding process constraints are as follows: The linear velocity of the grinding wheel, the workpiece feed rate, and the depth of grinding used in the current grinding process are obtained to form a set of process parameters; Pre-set multiple typical process parameter combinations and their corresponding process influence coefficients; The process parameter set is matched with typical process parameter combinations, the matching degree is calculated, and the process influence coefficient corresponding to the typical process parameter combination with the highest matching degree is selected as the constraint coefficient of the current grinding process.

7. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 6, characterized in that, The specific operation for generating the real-time grinding ratio through coupling calculation is as follows: The acoustic emission dominant frequency amplitude variation rate, which characterizes the self-sharpening tendency of the grinding wheel, and the root mean square value of the vibration signal envelope, which characterizes the grinding stability, are separated from the initial feature vector. The dynamic correction factor is obtained by multiplying the acoustic emission dominant frequency amplitude variation rate and the root mean square value of the vibration signal envelope by the process influence coefficient. The reference grinding ratio data is added to the dynamic correction factor, and the sum is then multiplied by a normalization factor determined by the current harmonic distortion rate. The final product is defined as the real-time grinding ratio.

8. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 1, characterized in that, When collecting vibration, acoustic emission, and current signals of the target grinding wheel during the grinding process, further synchronous calibration is performed, specifically including: A vibration acceleration sensor is installed on the grinding wheel spindle housing, an acoustic emission sensor is installed near the grinding fluid nozzle, and a current transformer is installed in the main motor power supply circuit. Before starting grinding, a synchronization trigger command is sent to the system to reset the data acquisition clocks of the three sensors and start synchronous timing. This ensures that the timestamps of the vibration signal, acoustic emission signal, and current signal are strictly aligned throughout the entire signal acquisition process.

9. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 3, characterized in that, Methods for manually labeling theoretical grinding ratio values ​​for feature vector segments include: Before collecting the full life cycle grinding signals of the grinding wheel sample, the initial mass of the grinding wheel sample is accurately weighed. After completing the entire life cycle grinding experiment, the grinding wheel sample is cleaned and dried, and its final mass is accurately weighed again. The difference between the two is the total wear mass. The total mass difference of the workpiece before and after the experiment is measured to obtain the total material removal mass. The total material removal mass is divided by the total wear mass to obtain the overall average grinding ratio of the grinding wheel sample under standard parameters. According to the proportion of the material removal volume corresponding to the signal segment to the total volume, the overall average grinding ratio is proportionally allocated as the theoretical grinding ratio value corresponding to the signal segment.

10. The grinding ratio measurement method based on a multi-information fusion model of grinding wheels according to claim 7, characterized in that, Smoothing of the real-time grinding ratio before output includes: The real-time grinding ratios at multiple time points are continuously collected and calculated to form a real-time grinding ratio sequence. A sliding time window is used to truncate the real-time grinding ratio sequence, and the median of all real-time grinding ratio values ​​within the window is calculated. The real-time grinding ratio value at the center of the window is replaced with this median, and the window is slid until the smoothing of the entire sequence is completed, and the final smoothed grinding ratio measurement result is output.

Images