Abrasion prediction method for non-contact grinding high-temperature alloy taper hole of CBN grinding wheel

Abstract

The invention discloses a CBN grinding wheel non-contact type grinding high-temperature alloy taper hole abrasion prediction method, and relates to the technical field of state monitoring and intelligent prediction in the precise grinding machining process. The method mainly comprises the following steps: calibrating a dynamic contact arc acoustic propagation model by using a system identification method to obtain dynamic geometric parameters, and constructing acoustic-vibration geometric coupling characteristics in combination with multi-mode signals in a grinding process; correcting the acoustic-vibration geometric coupling features by using a maximum mean difference algorithm and an adaptive window algorithm, training a wear prediction model by using the corrected acoustic-vibration geometric coupling features, and performing regression analysis on the corrected acoustic-vibration geometric coupling features by using the trained wear prediction model to obtain a prediction result. According to the grinding wheel abrasion prediction method for high-temperature alloy taper hole grinding, the adaptability, robustness and sensitivity of abrasion state characterization under complex variable working conditions can be improved.

Description

Technical Field

[0001] This invention relates to the field of precision grinding process condition monitoring and intelligent prediction technology, and more specifically, to a method for predicting wear of high-temperature alloy tapered holes using a non-contact CBN grinding wheel. Background Technology

[0002] Ceramic-bonded CBN grinding wheels are key tools for precision grinding of internal holes in difficult-to-machine materials such as high-temperature alloys, but their wear condition directly affects machining accuracy, surface quality, and cost. Existing grinding wheel wear monitoring technologies mainly include: (1) Monitoring methods based on contact sensors: such as acoustic emission sensors and strain gauges installed on the spindle or workpiece fixture. This method is inconvenient to install, and the signal is easily affected by the overall vibration of the machine tool and the structural transmission path, resulting in a low signal-to-noise ratio.

[0003] (2) Indirect evaluation method based on a single signal: such as judging by spindle motor current or power. This method is not sensitive to micron-level wear and cannot be applied to tapered hole grinding where cutting parameters change dynamically due to geometric changes.

[0004] (3) Offline measurement method based on machine vision: This method requires interruption of processing, is inefficient, and is not suitable for closed, coolant-filled internal grinding environment.

[0005] For grinding internal tapered holes, the grinding area is closed, and the wheel-workpiece contact arc length dynamically changes with the axial feed. This causes the propagation path and attenuation characteristics of sound waves and vibration signals to change nonlinearly. Traditional features based on fixed thresholds or single signals cannot effectively characterize the wear state. Therefore, there is an urgent need for an online monitoring method for grinding wheel wear that can adapt to complex sound field changes, requires no contact installation, and can achieve high-precision prediction. Summary of the Invention

[0006] The purpose of this invention is to provide a non-contact grinding method for predicting wear of high-temperature alloy tapered holes using CBN grinding wheels, which can improve the adaptability, robustness, and sensitivity of wear condition characterization under complex and variable working conditions.

[0007] This invention provides a method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels, comprising the following steps: S1: Acquire multi-modal signals of the grinding process, including high-frequency stress wave signals, low-frequency vibration signals, and airborne sound signals; S2: The dynamic contact arc acoustic propagation model is calibrated using the system identification method to obtain the calibrated dynamic contact arc acoustic propagation model; S3: Utilize the calibrated dynamic contact arc acoustic propagation model to obtain dynamic geometric parameters, and construct acoustic-vibration geometric coupling characteristics based on the multimodal signals of the grinding process and the dynamic geometric parameters; S4: The acoustic-vibration geometric coupling features are corrected using the maximum mean difference algorithm and the adaptive window algorithm to obtain the corrected acoustic-vibration geometric coupling features; the wear prediction model is trained using the corrected acoustic-vibration geometric coupling features to obtain the trained wear prediction model. S5: Using the trained wear prediction model, regression analysis is performed on the corrected acoustic-vibration geometric coupling characteristics to obtain the grinding wheel wear amount, wear trend, and remaining life estimate.

[0008] This invention also provides a non-contact grinding wear prediction system for high-temperature alloy tapered holes using CBN grinding wheels, the system comprising the following modules: The grinding process multimodal signal module is configured to acquire grinding process multimodal signals, which include high-frequency stress wave signals, low-frequency vibration signals, and airborne sound signals. The dynamic contact arc acoustic propagation model calibration module is configured to: calibrate the dynamic contact arc acoustic propagation model using a system identification method to obtain the calibrated dynamic contact arc acoustic propagation model; The acoustic-vibration geometric coupling feature construction module is configured to: obtain dynamic geometric parameters using the calibrated dynamic contact arc acoustic propagation model, and construct acoustic-vibration geometric coupling features based on the multi-modal signals of the grinding process and the dynamic geometric parameters; The incremental training module for the wear prediction model is configured to: correct the acoustic-vibration geometric coupling features using the maximum mean difference algorithm and the adaptive window algorithm to obtain the corrected acoustic-vibration geometric coupling features; and train the wear prediction model using the corrected acoustic-vibration geometric coupling features to obtain the trained wear prediction model. The grinding wheel wear prediction module is configured to: use the trained wear prediction model to perform regression analysis on the corrected acoustic-vibration geometric coupling characteristics to obtain the grinding wheel wear amount, wear trend and remaining life estimate.

[0009] The wear prediction method for non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels provided by this invention has the following beneficial effects: This invention constructs a non-contact multimodal sensor array, specifically a non-contact three-source (acoustic emission + vibration + airborne sound) sensor array. Acoustic emission (AE) sensors, accelerometers, and microphones are deployed at non-contact positions between the worktable and the protective cover to simultaneously acquire multimodal acoustic and vibration signals during the grinding process, including high-frequency acoustic emission, low-frequency vibration, and airborne sound signals. Its wear prediction architecture, integrated with a dynamic geometric model, fundamentally solves the difficulties of contact installation and signal contamination problems. Based on the geometric characteristics of the conical hole and the grinding depth, a dynamic contact arc acoustic propagation model is constructed to accurately describe the signal propagation attenuation mechanism in the conical space and quantify the attenuation law of sound waves in the conical space. Furthermore, by fusing multimodal signal characteristics and geometric parameters, the physical model and signal characteristics are deeply integrated to form acoustic-vibration geometric coupling characteristics, driving data... By closely integrating with physical mechanisms to characterize the wear state of grinding wheels, the ability to characterize wear state under complex and variable working conditions, as well as the model's generalization, robustness, and sensitivity, are improved. To address the drift of working conditions during processing, the maximum mean difference and adaptive window algorithms are introduced for online feature correction to perform online feature drift detection. Furthermore, a lightweight gradient lifter model is adopted to realize an adaptive prediction mechanism for grinding wheel wear, enabling adaptive updates of the prediction model. This overcomes the performance degradation and inaccuracy problems caused by changes in processing conditions (workpiece geometry changes, temperature fluctuations, etc.), achieving lifelong learning and long-term reliable operation of the prediction model. Ultimately, this results in high-precision, adaptive online prediction of grinding wheel wear.

[0010] Experimental verification shows that this method can achieve highly sensitive, non-contact monitoring of wear trends in the grinding of tapered holes in high-temperature alloys, with prediction accuracy superior to [previous method name]. This provides a basis for grinding wheel life management and process optimization. The system has a simple structure and low cost, requires no modification to existing machine tools and grinding wheels, and is easy to implement and promote in industrial settings. It is particularly suitable for precision grinding of closed and complex geometric parts such as high-temperature alloy internal tapered holes. Attached Figure Description

[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of the non-contact grinding wear prediction method for high-temperature alloy tapered holes using CBN grinding wheels provided by the present invention; Figure 2 This is a schematic diagram of the overall implementation process and system architecture provided by the present invention; Figure 3 This is a geometric relationship diagram of the dynamic contact arc length Lc(z) as a function of grinding depth z, provided by the present invention. Figure 4 This is a flowchart of the acoustic-vibration geometric coupling feature construction provided by the present invention; Figure 5 This is a schematic diagram of the online feature drift detection and model adaptive update algorithm provided by the present invention; Figure 6 This is a comparison and verification chart of the grinding wheel wear prediction results and offline measurement values ​​provided by the present invention. Detailed Implementation

[0012] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0013] Figure 1 A schematic diagram of the wear prediction method for non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels according to this embodiment is shown. In this embodiment, the wear prediction method for non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels includes the following steps: S1: Acquire multi-modal signals of the grinding process, including high-frequency stress wave signals, low-frequency vibration signals, and airborne sound signals; As an exemplary embodiment, in step S1, the acoustic emission, acceleration and air acoustic sensors are fixed at non-contact positions on the machine tool worktable and the protective cover, respectively, to construct a sensor array, ensuring that signal acquisition is not limited by the space of the grinding area and avoiding interference with the machining process.

[0014] As an exemplary embodiment, in step S1, non-contact multimodal signals are synchronously acquired. Specifically, grinding process signals are synchronously acquired through a non-contact multimodal sensor array, wherein the acoustic emission sensor is fixed to the metal mounting surface of the machine tool worktable to acquire signals. High-frequency stress waves; accelerometers are fixed to the machine tool table or beam structure to collect data. Low-frequency vibration; an airborne sound sensor is fixed inside the machine tool protective cover to collect data. Airborne acoustic signals.

[0015] In this embodiment, the airborne sound sensor is a microphone.

[0016] In one exemplary embodiment, the acoustic emission sensor employs a resonant frequency not lower than [a certain value]. This model requires no coupling agent during installation, and the straight-line distance between the grinding wheel and the workpiece is... This is to ensure the consistency of the signal propagation path and to prevent direct flushing by the coolant.

[0017] S2: The dynamic contact arc acoustic propagation model is calibrated using the system identification method to obtain the calibrated dynamic contact arc acoustic propagation model.

[0018] In one exemplary embodiment, the calculation formula for the dynamic contact arc acoustic propagation model is as follows: , , ; in, The contact arc length is dynamically changing; Square root function; Where is the radius of the grinding wheel; The cone angle of the conical hole; For real-time grinding depth; The attenuation of high-frequency acoustic emission waves due to multiple reflections at the hole wall; This refers to the attenuation of airborne sound as it diffuses within the conical space. and These are the attenuation coefficients of high-frequency acoustic emission waves after multiple reflections at the hole wall and the diffusion attenuation coefficients of airborne sound in the conical space, respectively.

[0019] As an exemplary embodiment, the system identification method includes: obtaining the frequency response function from the machine tool table to each sensor mounting point through a hammer impact test. And the acoustic emission attenuation coefficient With airborne sound attenuation coefficient System identification is performed to ensure that the root mean square error of the simulated signal energy attenuation is consistent with that of the measured value. .

[0020] S3: The dynamic geometric parameters are obtained by using the calibrated dynamic contact arc acoustic propagation model. Based on the multimodal signals of the grinding process and the dynamic geometric parameters, the acoustic-vibration geometric coupling characteristics are constructed.

[0021] In one exemplary embodiment, the dynamic geometry parameters include dynamic contact arc length, taper angle, and grinding depth.

[0022] In one exemplary embodiment, the process of constructing the acoustic-vibration geometric coupling features is as follows: The wear-sensitive features are extracted from the multimodal signal of the grinding process, including AE band energy, vibration envelope peak value, and air acoustic spectrum centroid features. The wear-sensitive features are weighted and fused with the dynamic geometric parameters to obtain the acoustic-vibration geometric coupling features.

[0023] In one exemplary embodiment, the acoustic-vibration geometric coupling characteristics include the energy and kurtosis of the acoustic emission signal, the envelope peak value and root mean square value of the vibration signal, the spectral centroid and modulation index of the air acoustic signal, and the grinding depth, dynamic contact arc, and cone angle.

[0024] S4: The acoustic-vibration geometric coupling features are corrected using the maximum mean difference algorithm and the adaptive window algorithm to obtain the corrected acoustic-vibration geometric coupling features; the wear prediction model is trained using the corrected acoustic-vibration geometric coupling features to obtain the trained wear prediction model.

[0025] In one exemplary embodiment, the process of correcting the acoustic-vibration geometric coupling characteristics is as follows: The early acoustic-vibration geometric coupling characteristics of multiple groups under normal wear conditions are used as the benchmark distribution; The maximum mean difference between the acoustic-vibration geometric coupling characteristics and the baseline distribution within the current time window is calculated using the maximum mean difference algorithm. If the maximum mean difference value exceeds the preset threshold three times consecutively, and a sudden change is detected using the adaptive window algorithm, it is determined that a significant drift has occurred. Data of acoustic-vibration geometric coupling features that have experienced significant drift are removed, and the latest multiple sets of acoustic-vibration geometric coupling features are added to obtain the corrected acoustic-vibration geometric coupling features.

[0026] In one exemplary embodiment, the threshold is 0.05.

[0027] In one exemplary embodiment, the kernel function of the maximum mean difference algorithm is a Gaussian radial basis function, and the bandwidth parameter of the maximum mean difference algorithm is selected using the median heuristic rule; the significance level of the adaptive window algorithm is 0.05, and the minimum window length is 100 data points.

[0028] As an example, before the model is deployed, offline training is required through stepped grinding experiments with known wear. The training data must cover the entire life cycle from new grinding wheel to complete wear, and cover the typical machining range of tapered holes from small taper angle to large taper angle.

[0029] In one exemplary embodiment, the wear prediction model is: Model or online ridge regression model; In one exemplary embodiment, the wear prediction model is: The model, the The model's hyperparameters are determined through Bayesian optimization, the maximum tree depth does not exceed 8 layers, and the learning rate is no greater than 0.1.

[0030] S5: Using the trained wear prediction model, regression analysis is performed on the corrected acoustic-vibration geometric coupling characteristics to obtain the grinding wheel wear amount, wear trend, and remaining life estimate.

[0031] In one exemplary embodiment, the method for predicting wear of high-temperature alloy tapered holes using a CBN grinding wheel in non-contact grinding further includes comparing and verifying the wear amount of the grinding wheel with the results of offline microscopy measurements. The method specifies that the absolute value of the prediction error between the wear amount of the grinding wheel and the offline microscopy measurements should not exceed a certain threshold over the entire lifespan of the grinding wheel. The prediction error of the remaining lifetime estimate does not exceed the total lifetime. If the prediction result is satisfactory, then the prediction result is deemed acceptable.

[0032] It's worth noting that Light GBM (Light Gradient Boosting Machine) is a high-efficiency machine learning framework developed by Microsoft based on Gradient Boosting Decision Trees (GBDT). Through several optimization techniques, it significantly improves training speed and memory efficiency while maintaining high accuracy. Its key technological innovations include: Histogram Algorithm: Discretizing continuous feature values ​​into k bins reduces the time complexity of calculating split gain. Leaf-wise growth strategy: Selecting the leaf node with the largest loss decrease for splitting each time is more efficient than traditional level-wise growth, but may overfit small datasets. One-sided gradient sampling (GOSS): Retaining large gradient samples and randomly sampling small gradient samples reduces computation without significantly affecting accuracy. Mutually exclusive feature bundling (EFB): Merging sparse and mutually exclusive features reduces feature dimensionality and accelerates training. These designs enable Light GBM to perform exceptionally well in large-scale data competitions and industrial applications.

[0033] Ridge regression, also known as Tikhonov regularization, is an improvement on ordinary least squares (OLS) specifically designed to handle multicollinearity among independent variables. When multiple features are highly correlated, the coefficient estimates in ordinary linear regression become unstable or even unsolvable. Ridge regression improves model stability and generalization ability by incorporating an L2 regularization term (the sum of squares of the regression coefficients) into the loss function to constrain the magnitude of the coefficients.

[0034] This embodiment provides a non-contact grinding wear prediction system for high-temperature alloy tapered holes using CBN grinding wheels. The system includes the following modules: The grinding process multimodal signal module is configured to acquire grinding process multimodal signals, which include high-frequency stress wave signals, low-frequency vibration signals, and airborne sound signals. The dynamic contact arc acoustic propagation model calibration module is configured to: calibrate the dynamic contact arc acoustic propagation model using a system identification method to obtain the calibrated dynamic contact arc acoustic propagation model; The acoustic-vibration geometric coupling feature construction module is configured to: obtain dynamic geometric parameters using the calibrated dynamic contact arc acoustic propagation model, and construct acoustic-vibration geometric coupling features based on the multi-modal signals of the grinding process and the dynamic geometric parameters; The incremental training module for the wear prediction model is configured to: correct the acoustic-vibration geometric coupling features using the maximum mean difference algorithm and the adaptive window algorithm to obtain the corrected acoustic-vibration geometric coupling features; and train the wear prediction model using the corrected acoustic-vibration geometric coupling features to obtain the trained wear prediction model. The grinding wheel wear prediction module is configured to: use the trained wear prediction model to perform regression analysis on the corrected acoustic-vibration geometric coupling characteristics to obtain the grinding wheel wear amount, wear trend and remaining life estimate.

[0035] In some embodiments, the above-described method for predicting wear of high-temperature alloy tapered holes using non-contact grinding with CBN grinding wheels can also be implemented in the following ways.

[0036] In this embodiment, the grinding wheel wear prediction method for grinding high-temperature alloy tapered holes includes: Step 1: Non-contact multimodal signal synchronous acquisition: Grinding process signals are synchronously acquired using a non-contact multimodal sensor array, wherein the acoustic emission sensor is fixed to the metal mounting surface of the machine tool worktable to acquire signals. High-frequency stress waves; accelerometers are fixed to the machine tool table or beam structure to collect data. Low-frequency vibration; an airborne sound sensor (microphone) is fixed inside the machine tool protective cover to collect data. Airborne acoustic signals; In one exemplary embodiment, the acoustic emission sensor employs a resonant frequency not lower than [a certain value]. This model requires no coupling agent during installation, and the straight-line distance between the grinding wheel and the workpiece is... This is to ensure the consistency of the signal propagation path and to prevent direct flushing by coolant.

[0037] As an exemplary embodiment, in step 1, acoustic emission, acceleration, and airborne acoustic sensors are fixed at non-contact positions on the machine tool worktable and the protective cover, respectively, to construct a sensor array, ensuring that signal acquisition is not limited by the space of the grinding area and avoiding interference with the machining process.

[0038] Step 2: Construct a dynamic contact arc acoustic propagation model: Based on the cone angle of the tapered hole With real-time grinding depth Calculate the dynamically changing contact arc length: , And establish an attenuation model for high-frequency acoustic emission waves after multiple reflections at the hole wall: , The diffusion and attenuation model of airborne sound in a conical space: ; It provides a physical basis for feature construction.

[0039] In one exemplary embodiment, the dynamic contact arc acoustic propagation model needs to be calibrated in the frequency domain, and the frequency response function from the machine tool table to each sensor mounting point is obtained through a hammer impact test. And the acoustic emission attenuation coefficient With airborne sound attenuation coefficient System identification is performed to ensure that the root mean square error of the simulated signal energy attenuation is consistent with that of the measured value. .

[0040] Step 3: Constructing the Acoustic-Vibration Geometric Coupling Feature (SGCF): Extract wear-sensitive features from the multimodal signal: AE band energy, vibration envelope peak value, and airborne acoustic spectrum centroid features, and correlate them with dynamic geometric parameters. (i.e., dynamic contact arc length) and cone angle The weighted fusion is performed to form the SGCF feature vector. This high-dimensional feature vector can comprehensively reflect the changes in contact stiffness and sound propagation path caused by wear, and characterize the wear state of the grinding wheel. In one exemplary embodiment, the acoustic-vibration geometric coupling feature SGCF is a 12-dimensional feature vector, specifically including: the AE signal in The energy and kurtosis of the frequency band, the vibration signal in The peak envelope and root mean square (RMS) values ​​of the frequency band, the spectral centroid and modulation index of the airborne acoustic signal, and the grinding depth z and contact arc length. and cone angle Three geometric parameters.

[0041] Step 4: Perform online feature drift correction and adaptive model update: Use the maximum mean difference algorithm to detect the time-varying characteristics (drift) of the SGCF feature distribution, and combine it with an adaptive window algorithm to identify signal abrupt change points. When the drift exceeds a threshold... In this way, the parameters of the wear prediction model can be adaptively corrected online to ensure long-term prediction stability; In one exemplary embodiment, in the online feature drift correction, the kernel function of the maximum mean difference algorithm is a Gaussian radial basis function, and the bandwidth parameter is selected through the median heuristic rule; the significance level of the adaptive window algorithm is set to 0.05, and the minimum window length is 100 data points.

[0042] As an example, before the model is deployed, offline training is required through stepped grinding experiments with known wear. The training data must cover the entire life cycle from new grinding wheel to complete wear, and cover the typical machining range of tapered holes from small taper angle to large taper angle.

[0043] Step 5: Adaptive Testing and Output of Grinding Wheel Wear: A lightweight machine learning model is used to perform regression analysis on the corrected SGCF features. Specifically, the corrected SGCF features are input into a wear prediction model based on Light GBM or online ridge regression, and the grinding wheel wear amount is output in real time. Wear trends (trend curves) and remaining life estimates provide data support for production decisions.

[0044] In one exemplary embodiment, when the wear prediction model adopts Light GBM, its hyperparameters are determined by Bayesian optimization, the maximum depth of the tree does not exceed 8 layers, and the learning rate is not greater than 0.1, so as to ensure the lightweight and real-time performance of the model and meet the latency requirements of online prediction.

[0045] It should be noted that the grinding wheel wear amount ultimately output by the above method is... The results need to be compared and verified with offline microscope measurements, and the absolute value of the prediction error must not exceed a certain limit over the entire lifespan. The remaining lifetime prediction error does not exceed the total lifetime. .

[0046] In some embodiments, the above-described CBN wheel non-contact grinding wear prediction system for high-temperature alloy tapered holes can also be implemented in the following manner. In this embodiment, the CBN wheel non-contact grinding wear prediction system for high-temperature alloy tapered holes includes: a sensor module for non-contact signal acquisition; an industrial control computer and data acquisition card for signal synchronization and processing; and a computing module equipped with feature extraction, drift detection, and prediction algorithm software for executing the wear prediction method and graphically displaying the results.

[0047] In some embodiments, the above-described method for predicting wear of high-temperature alloy tapered holes using non-contact grinding with CBN grinding wheels can also be implemented in the following ways.

[0048] like Figure 2 The diagram shows the overall implementation process and system architecture. In this embodiment, the grinding wheel wear prediction method for grinding high-temperature alloy tapered holes includes: 1. System hardware configuration and signal acquisition; Sensor selection and placement: The acoustic emission sensor selected is from Physical Acoustics (PAC). Type resonant sensor, center frequency It is fixed to a flat area on the side of the machine tool worktable by a magnetic mounting base; the accelerometer sensor is an ICP type triaxial accelerometer from PCB company, with a range of... It is threaded onto the front crossbeam of the worktable; the airborne sound sensor uses a high-frequency measuring microphone from GRAS, with a frequency range of... It is fixed to the top of the machine tool protective cover by a bracket, facing the processing area; The data acquisition system uses the NI PXIe platform and is equipped with three synchronous acquisition cards; the sampling rate of the AE signal is set to 2MHz, and the sampling rate of the vibration and sound signals is set to 50kHz. All signals are strictly synchronized through hardware triggering.

[0049] 2. Calibration of the acoustic propagation model for dynamic contact arcs; like Figure 3 The figure shows the dynamic contact arc length. With grinding depth A diagram showing the changing geometric relationships; In offline mode, the lead breakage signal was used to simulate an AE source, and experiments were conducted at different depths of the conical hole, recording the responses of each sensor.

[0050] Using system identification methods, the attenuation coefficient in the model is derived from the measured signal attenuation level. and This makes the correlation coefficient between the model-predicted decay curve and the measured data... .

[0051] 3. Online calculation of acoustic-vibration geometric coupling characteristics; like Figure 4 The diagram shown is a flowchart for constructing acoustic-vibration geometric coupling features. Develop a real-time feature extraction program using Python on an industrial control computer.

[0052] Perform wavelet packet transform on the AE signal to extract... Subband energy, also known as AE band energy.

[0053] Perform Hilbert transform on the vibration signal, calculate the envelope spectrum, and extract... The peak value of the frequency band, also known as the peak value of the vibration envelope.

[0054] Perform a short-time Fourier transform on the airborne sound signal to calculate its spectral centroid, i.e., the airborne sound spectral centroid feature.

[0055] Read the current Z-axis coordinate in real time from the CNC system, and combine it with the known cone angle. Online computing .

[0056] After normalizing all the above features and geometric parameters, they are combined into a 12-dimensional SGCF feature vector.

[0057] 4. Drift detection and model adaptation implementation; like Figure 5 The diagram shown illustrates the principle of the online feature drift detection and model adaptive update algorithm. During the initialization phase, the first 100 sets of SGCF data under normal wear conditions were collected as the baseline distribution.

[0058] During online operation, the maximum mean discrepancy (MMD) between the SGCF feature and the baseline distribution within the current time window is calculated every 200ms.

[0059] Set threshold When the MMD exceeds the threshold three times consecutively, and the Adaptive Window algorithm (ADWIN) also detects the mutation, a significant drift is determined to have occurred.

[0060] Trigger a model update by adding the latest 200 sets of data to the training set and removing the oldest data. Then, incrementally train the LightGBM model and update the model parameters.

[0061] 5. Experimental verification and performance evaluation; like Figure 6 The figure shown is a comparison and verification chart of the predicted grinding wheel wear and the offline measurement values.

[0062] Performing tapered hole (taper angle) of GH4169 high-temperature alloy on a coordinate grinding machine Grinding test.

[0063] After grinding 10 holes, the machine is stopped and the actual wear of the grinding wheel is measured offline using a tool microscope as the true value.

[0064] Compare online predicted values ​​with offline measured values, calculate the prediction error over the entire life cycle, and require... The above data point errors are in Within.

[0065] The system ultimately displays the current wear level, historical trend curve, and estimated number of remaining machined holes in real time on the HMI interface.

[0066] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels, characterized in that, Includes the following steps: S1: Acquire multi-modal signals of the grinding process, including high-frequency stress wave signals, low-frequency vibration signals, and airborne sound signals; S2: The dynamic contact arc acoustic propagation model is calibrated using the system identification method to obtain the calibrated dynamic contact arc acoustic propagation model; S3: Utilize the calibrated dynamic contact arc acoustic propagation model to obtain dynamic geometric parameters, and construct acoustic-vibration geometric coupling characteristics based on the multimodal signals of the grinding process and the dynamic geometric parameters; S4: The acoustic-vibration geometric coupling features are corrected using the maximum mean difference algorithm and the adaptive window algorithm to obtain the corrected acoustic-vibration geometric coupling features; the wear prediction model is trained using the corrected acoustic-vibration geometric coupling features to obtain the trained wear prediction model. S5: Using the trained wear prediction model, regression analysis is performed on the corrected acoustic-vibration geometric coupling characteristics to obtain the grinding wheel wear amount, wear trend, and remaining life estimate.

2. The method for predicting wear of high-temperature alloy tapered holes using a CBN grinding wheel in non-contact grinding according to claim 1, characterized in that, The calculation formula for the dynamic contact arc acoustic propagation model is as follows: ， ， ； in, The contact arc length is dynamically changing; Square root function; Where is the radius of the grinding wheel; The cone angle of the conical hole; For real-time grinding depth; The attenuation of high-frequency acoustic emission waves due to multiple reflections at the hole wall; This refers to the attenuation of airborne sound as it diffuses within the conical space. and These are the attenuation coefficients of high-frequency acoustic emission waves after multiple reflections at the hole wall and the diffusion attenuation coefficients of airborne sound in the conical space, respectively.

3. The method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels according to claim 1, characterized in that, The dynamic geometric parameters include dynamic contact arc length, taper angle, and grinding depth.

4. The method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels according to claim 1, characterized in that, The process of constructing the acoustic-vibration geometric coupling characteristics is as follows: The wear-sensitive features are extracted from the multimodal signal of the grinding process, including AE band energy, vibration envelope peak value, and air acoustic spectrum centroid features. The wear-sensitive features are weighted and fused with the dynamic geometric parameters to obtain the acoustic-vibration geometric coupling features.

5. The method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels according to claim 1, characterized in that, The acoustic-vibration geometric coupling characteristics include the energy and kurtosis of the acoustic emission signal, the envelope peak value and root mean square value of the vibration signal, the spectral centroid and modulation index of the air acoustic signal, and the grinding depth, dynamic contact arc and cone angle.

6. The method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels according to claim 1, characterized in that, The process of correcting the aforementioned acoustic-vibration geometric coupling characteristics is as follows: The early acoustic-vibration geometric coupling characteristics of multiple groups under normal wear conditions are used as the benchmark distribution; The maximum mean difference between the acoustic-vibration geometric coupling characteristics and the baseline distribution within the current time window is calculated using the maximum mean difference algorithm. If the maximum mean difference value exceeds the preset threshold three times consecutively, and a sudden change is detected using the adaptive window algorithm, it is determined that a significant drift has occurred. Data of acoustic-vibration geometric coupling features that have experienced significant drift are removed, and the latest multiple sets of acoustic-vibration geometric coupling features are added to obtain the corrected acoustic-vibration geometric coupling features.

7. The method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels according to claim 1, characterized in that, The wear prediction model is either the Light GBM model or the online ridge regression model.

8. The method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels according to claim 1, characterized in that, The wear prediction model is the Light GBM model. The hyperparameters of the Light GBM model are determined by Bayesian optimization, the maximum depth of the tree does not exceed 8 layers, and the learning rate is not greater than 0.

1.

9. The method for predicting wear in non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels according to claim 1, characterized in that, It also includes comparing and verifying the grinding wheel wear amount with the offline microscope measurement results. The absolute value of the prediction error between the grinding wheel wear amount and the offline microscope measurement results should not exceed [a certain value] within the entire lifespan. The prediction error of the remaining lifetime estimate does not exceed the total lifetime. If the prediction result is satisfactory, then the prediction result is deemed acceptable.

10. A wear prediction system for non-contact grinding of high-temperature alloy tapered holes using CBN grinding wheels, characterized in that, The system includes the following modules: The grinding process multimodal signal module is configured to acquire grinding process multimodal signals, which include high-frequency stress wave signals, low-frequency vibration signals, and airborne sound signals. The dynamic contact arc acoustic propagation model calibration module is configured to: calibrate the dynamic contact arc acoustic propagation model using a system identification method to obtain the calibrated dynamic contact arc acoustic propagation model; The acoustic-vibration geometric coupling feature construction module is configured to: obtain dynamic geometric parameters using the calibrated dynamic contact arc acoustic propagation model, and construct acoustic-vibration geometric coupling features based on the multi-modal signals of the grinding process and the dynamic geometric parameters; The incremental training module for the wear prediction model is configured to: correct the acoustic-vibration geometric coupling features using the maximum mean difference algorithm and the adaptive window algorithm to obtain the corrected acoustic-vibration geometric coupling features; and train the wear prediction model using the corrected acoustic-vibration geometric coupling features to obtain the trained wear prediction model. The grinding wheel wear prediction module is configured to: use the trained wear prediction model to perform regression analysis on the corrected acoustic-vibration geometric coupling characteristics to obtain the grinding wheel wear amount, wear trend and remaining life estimate.

Images