A method for monitoring tool wear on a high-speed drilling machine

Abstract

This invention discloses a tool wear monitoring method for high-speed drilling machines, specifically relating to the field of turning and boring tool wear monitoring technology. The method includes deploying multiple types of sensors to collect tool wear data; preprocessing the collected data to generate multi-dimensional wear parameter feature vectors; fusing the feature vectors to generate a tool wear feature vector set; inputting the feature vector set into a pre-trained tool wear prediction model to output a real-time quantized value of tool wear; and comparing the quantized wear value with a preset threshold to generate tool wear status information and control commands. This invention solves the problems of single-dimensional tool wear monitoring, insufficient data quantification capability, low level of intelligence, and lack of closed-loop control by collecting tool wear data, preprocessing the data, generating a tool wear feature vector set, outputting a real-time quantized value of tool wear, and generating tool wear status information and control commands.

Description

Technical Field

[0001] This invention relates to the field of turning and boring tool wear monitoring technology, and more specifically, to a tool wear monitoring method for high-speed drilling machines. Background Technology

[0002] In the field of machining, especially in the turning and boring processes of high-speed drilling machines, the wear condition of the cutting tool, as the directly involved cutting component, directly determines the quality, precision, and production cost of the machined workpiece. Traditional methods for monitoring tool wear mainly rely on the operator's experience, judging whether the tool needs to be replaced by manual observation, touch, or even listening after the machine stops. This method is highly subjective, inefficient, and cannot achieve real-time monitoring. It is easy to cause batch workpiece scrap due to failure to replace the tool in time, or to increase tool costs due to premature tool replacement.

[0003] Existing methods for monitoring tool wear on high-speed drilling machines can meet basic monitoring needs, but they still have some shortcomings in practical applications: First, current technologies rely on a single dimension for tool wear monitoring, resulting in incomplete data acquisition from a single type of sensor and difficulty in determining the tool wear state. Second, they lack quantification capabilities; current technologies can only achieve binary judgments of normal or abnormal, failing to output continuous quantified values ​​of tool wear and thus unable to provide data support for predictive tool changes. Third, their level of intelligence is low; current technologies use simple threshold comparison methods, failing to utilize multi-source data fusion and machine learning algorithms to mine wear patterns from the data, resulting in limited model generalization ability. Fourth, they lack closed-loop control; monitoring results are typically only used for local display and alarms, failing to automatically generate execution control commands. Therefore, a method is urgently needed to address the problems of single-dimensional tool wear monitoring, insufficient data quantification capabilities, low level of intelligence, and lack of closed-loop control. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for monitoring tool wear of a high-speed drilling machine, which solves the problems mentioned in the background art through the following scheme.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring tool wear on a high-speed drilling machine, comprising:

[0006] S1: The cutting process of the high-speed drilling machine is preset as the monitoring target. Multiple types of sensors are deployed to collect tool wear data in real time during the cutting process. The wear data includes physical parameters and equipment parameters.

[0007] S2: Preprocess the collected tool wear data to generate a multi-dimensional wear parameter feature vector. The preprocessing includes outlier removal, normalization, and wear feature extraction.

[0008] The wear feature extraction includes constructing a physical parameter feature vector matrix. With device parameter feature vector matrix The physical parameter features include the wear rate S calculated based on the tool length L, and the ratio feature constructed based on the crater depth KT and the flank wear bandwidth VB. The equipment parameter characteristics include the energy percentage of the equipment power P in a specific frequency band;

[0009] S3: The multi-dimensional wear parameter feature vectors of S2 are fused to generate a tool wear feature vector set. The fusion is performed using a machine learning algorithm.

[0010] S4: Input the tool wear feature vector set in S3 into the pre-trained tool wear prediction model. The tool wear prediction model is a regression model trained based on machine learning algorithm, which outputs the real-time wear quantization value of the tool flank wear bandwidth VB.

[0011] S5: Compare the real-time wear quantization value of S4 with the preset threshold, and generate tool wear status information and control commands based on the comparison result.

[0012] Preferably, step S4 inputs the tool wear feature vector set into a pre-trained tool wear prediction model, which is a regression model trained based on a machine learning algorithm, and outputs a real-time quantitative value of tool wear.

[0013] S401: Before inputting the feature vector set, the model loading and validity verification must be completed to ensure that the model is suitable for the tool type and machining scenario of the current high-speed drilling machine: Model loading and parameter confirmation: Load the pre-trained regression model file from the model storage path. The core of the model is the Support Vector Regression (SVR) algorithm; Read the key parameter records during model training: including the tool type, machining material, feature dimension (which must be consistent with the feature vector dimension output by S2), and wear quantization target (such as the back face wear bandwidth VB value) corresponding to the training data. Confirm whether it matches the machining scenario of the current high-speed drilling machine. If it does not match, the pre-trained model for the corresponding scenario must be replaced.

[0014] For rapid model validity verification, select 3 to 5 sets of standard feature vectors with known VB values, input them into the loaded model, compare the predicted VB values ​​output by the model with the known true VB values, and calculate the error rate M. The specific calculation formula is as follows:

[0015]

[0016] in The model prediction value representing the wear bandwidth on the flank face. This represents the actual value of the wear bandwidth on the back face. The error rate of the sample is used to determine whether the model is effective. If the error rate is too large, the model file needs to be checked again to see if it is corrupted.

[0017] S402: Adaptation processing of tool wear feature vector set, adjusting the tool wear feature vector set to a format that meets the model input requirements:

[0018] Align the feature dimensions and order, extract the vector set to be input, and compare it with the input feature dimension table of the pre-trained model; if the dimensions of the input vector set are inconsistent with the dimensions required by the model, redundant features need to be removed; if the dimensions are missing, check whether the feature extraction step has been omitted, and supplement and complete the vector set.

[0019] Read the feature mean value during model training, convert the vector set format, and convert the adapted feature vector set into the input data format supported by the model. If it is deployed in an industrial PLC or embedded system, convert it into JSON format or binary data stream to ensure that the data transmission rate meets the high-speed monitoring requirements.

[0020] S403: Inference calculation and real-time wear quantification output of the pre-trained model. The model inference maps feature vectors to real-time tool wear quantification values ​​(such as VB values) to ensure the real-time performance of the inference process. Real-time inference mode selection: For batch processing scenarios, batch inference is used, and the set of input vectors collected within the processing cycle is input into the model. The model outputs VB prediction values ​​in batches, and the inference interval must be synchronized with the processing cycle. For single-piece high-precision processing scenarios, real-time inference is used. The inference trigger frequency is set, and each set of feature vectors is input into the model. The model outputs one VB prediction value within 100ms to ensure real-time tracking of wear changes.

[0021] During the model inference calculation process, the model's inference interface (Scikit-learn's model.predict()) is called, and the adapted feature vector set is input. The model performs calculations internally. Taking the random forest regression model as an example, multiple decision trees vote on the input features and output the average VB prediction value. The inference process is monitored for timeout. If a single inference takes more than 500ms, a backup inference thread is triggered to ensure that tool wear monitoring is not interrupted.

[0022] The system outputs and stores real-time wear quantification values, outputs the VB prediction values ​​calculated by the model, and adds an inference timestamp corresponding to the S1 data acquisition timestamp to form a real-time VB quantification value key-value pair. The output results are synchronously stored in the HMI interface of the high-speed drilling machine to provide tool wear parameter change values, which is convenient for operators to monitor intuitively.

[0023] The technical effects and advantages of this invention are as follows:

[0024] This invention collects tool wear data by deploying multiple types of sensors, overcoming the limitations of existing technologies that rely on a single data source, and achieves multi-dimensional and high-precision tool status perception, significantly improving the comprehensiveness and anti-interference capability of monitoring.

[0025] This invention uses a regression model trained based on machine learning algorithms to directly map the fused feature vectors into continuous quantitative values ​​of tool wear, thereby achieving real-time quantitative prediction of tool wear and providing accurate data support for predictive maintenance of tool wear.

[0026] This invention uses principal component analysis algorithm to perform feature fusion, automatically extracts tool wear-related data features, reduces the subjectivity of manual selection and data dimensionality, improves the intelligence level of the monitoring method, and enhances the generalization ability of the model.

[0027] This invention compares the predicted wear quantification value with multiple preset thresholds and automatically generates tool wear status information and control commands based on the results, forming an automated monitoring closed loop of perception, analysis, decision-making, and control. This achieves a complete closed loop from status perception to decision-making and control, improving the automation level of the machining process. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the tool wear monitoring method of the present invention.

[0029] Figure 2 This is a schematic diagram of the tool wear monitoring feedback structure of the present invention.

[0030] Figure 3 This is a schematic diagram of the monitoring process of the present invention. Detailed Implementation

[0031] 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.

[0032] Please see Figures 1-3As shown, this invention provides a tool wear monitoring method for a high-speed drilling machine. The method involves deploying multiple types of sensors to collect tool wear data in real time during the cutting process; preprocessing the tool wear data to generate multi-dimensional wear parameter feature vectors; fusing the feature vectors to generate a tool wear feature vector set; inputting the wear feature vector set into a pre-trained tool wear prediction model to output real-time tool wear quantization values; and generating wear status information and control commands based on the quantization values. This invention discloses a tool wear monitoring method for a high-speed drilling machine, the method comprising:

[0033] S1: The cutting process of the high-speed drilling machine is preset as the monitoring target. Multiple types of sensors are deployed to collect tool wear data in real time during the cutting process. The wear data includes physical parameters and equipment parameters.

[0034] S2: Preprocess the collected tool wear data to generate a multi-dimensional wear parameter feature vector. The preprocessing includes outlier removal, normalization, and wear feature extraction.

[0035] The wear feature extraction includes constructing a physical parameter feature vector matrix. With device parameter feature vector matrix The physical parameter features include the wear rate S calculated based on the tool length L, and the ratio feature constructed based on the crater depth KT and the flank wear bandwidth VB. The equipment parameter characteristics include the energy percentage of the equipment power P in a specific frequency band;

[0036] S3: The multi-dimensional wear parameter feature vectors of S2 are fused to generate a tool wear feature vector set. The fusion is performed using a machine learning algorithm.

[0037] S4: Input the tool wear feature vector set in S3 into the pre-trained tool wear prediction model. The tool wear prediction model is a regression model trained based on machine learning algorithm, which outputs the real-time wear quantization value of the tool flank wear bandwidth VB.

[0038] S5: Compare the real-time wear quantization value of S4 with the preset threshold, and generate tool wear status information and control commands based on the comparison result.

[0039] In S1, the cutting process of the preset high-speed drilling machine is the monitoring target. Multiple types of sensors are deployed to collect tool wear data in real time during the cutting process. This wear data includes physical parameters and equipment parameters.

[0040] S101: Preset cutting process of high-speed drilling machine. The main cutting process is divided into turning and boring. The main difference between the two processes is the different processing methods and application scenarios.

[0041] Turning is a cutting process that uses a single-edged cutting tool on a lathe to cut a rotating workpiece. The tool moves along the workpiece axially or radially to remove material. It is mainly used to process cylindrical parts and can complete the machining of outer circles, inner holes, end faces and threads. It is suitable for mass production of high-precision rotating parts.

[0042] Boring is the process of enlarging the inner diameter of a pre-drilled hole in a workpiece using a single-edged boring tool. It is usually performed on a boring machine or machining center. It is mainly used for machining cylindrical holes, threaded holes, and end faces of parts such as housings and brackets. Unlike turning, boring is often used for finishing pre-drilled holes, such as holes that have been cast or forged.

[0043] S102: Deploy multiple types of sensors to collect tool wear data in real time during the cutting process. The tool wear data includes physical parameters and equipment parameters.

[0044] Physical parameters include: tool length L, flank wear width VB, and crater depth KT; equipment parameters include: cutting temperature W, equipment power P, and tool speed V.

[0045] Further explanation is needed regarding the deployment of multiple types of sensors to collect tool wear data in real time. A non-contact laser displacement sensor is installed on a fixed bracket on the side of the spindle box of the high-speed drilling machine (150-200mm from the tool shank and parallel to the tool axis) to collect data on the change in tool length L in real time.

[0046] An adjustable column is set on the side of the high-speed drilling machine's worktable, and a high-speed industrial camera and a macro waterproof lens are installed on the column to collect real-time data on the changes in the back face wear bandwidth VB and the crescent depth KT. The installation position requires the height to be flush with the cutting area of ​​the tool, and the lens to be at an angle of 30 to 45 degrees to the cutting edge of the tool.

[0047] An infrared thermal imager is installed on the cantilever bracket above the spindle of the high-speed drilling machine to collect the temperature parameter W of the cutting zone in real time. The installation distance is required to be 300-500mm from the cutting zone.

[0048] A current transformer and a power transmitter are installed in the spindle motor control cabinet of a high-speed drilling machine to collect the power parameters P of the equipment in real time. The installation position needs to be connected in series in the spindle motor power supply circuit.

[0049] A spindle encoder with the machine tool's own components is installed at the rear end of the high-speed drilling machine spindle to collect the tool speed parameter V in real time at a sampling frequency of 10Hz.

[0050] In S2, the preprocessing of the collected tool wear data to generate a multi-dimensional wear parameter feature vector specifically includes:

[0051] Outlier removal involves receiving tool wear data from sensors and using... The principle is to remove outliers from tool wear data. Specific steps include: taking tool length L as an example, firstly, establishing a tool length dataset. Where n represents the total number of tool length data points collected. The first part represents the specific value of the collected tool length data; the second part represents the average value of the tool length data. and standard deviation The calculation, specifically the formula is as follows: Specific standard deviation The calculation formula is in Indicates the first Each tool length data value; when a certain tool length data value satisfy When this happens, the tool length data is determined to be abnormal and is discarded.

[0052] Normalization is necessary because the collected tool wear data have different dimensions. Normalization is required to unify the scale of the tool wear data, and the normalization result is then mapped to... Range; to prevent excessive differences in data units from overshadowing smaller tool wear values ​​during tool wear diagnosis, the specific normalization steps include: normalizing the collected tool wear data... Normalization processing is performed, and the maximum and minimum values ​​of the data collected by each sensor are calculated and denoted as follows: , The normalization calculation formula is as follows: in This represents the normalized value of tool wear data, and its value range is within... Within the interval, This represents the collected wear data of a single tool, which includes: tool length data, flank wear bandwidth VB, crater depth KT, cutting temperature parameter W, equipment power parameter P, and tool speed parameter V;

[0053] Wear feature extraction: Raw tool wear data can only reflect a single state. It is necessary to extract time-domain, frequency-domain, and statistical domain features of the data to construct a multi-dimensional wear feature vector. The specific extraction logic is carried out according to physical parameters and equipment parameters.

[0054] Physical parameter feature extraction directly reflects the tool wear morphology: For the tool length L, the length wear rate is calculated. As a feature, it is used to reflect changes in wear, where This represents the difference between the current tool length value and the original tool length value. The rate of change of the difference in tool length over time, relative to the wear rate. The slope of the linear overfit is calculated to reflect the tool wear rate;

[0055] For the flank wear bandwidth VB, the sliding mean of VB is calculated as a feature. The average VB value of five consecutive time stamps is taken to reflect the smooth instantaneous tool wear fluctuation. For the state feature, wear level labels are marked according to the VB value (0 indicates normal VB < 0.2mm; 1 indicates moderate wear 0.2mm ≤ VB < 0.4mm; 2 indicates severe wear VB ≥ 0.4mm).

[0056] For the crater depth KT, the maximum value of the tool wear data characteristic KT is statistically analyzed, and the ratio of KT to VB is calculated. As a characteristic, when the ratio is greater than 0.5, it indicates that the rake face is worn and requires close attention.

[0057] Equipment parameter feature extraction, specifically targeting cutting temperature, involves calculating the peak value of W. This refers to the highest temperature within a single processing cycle, and the correlation coefficient between W and P is calculated based on the correlation characteristics to extract the corresponding time-domain features; for equipment power, the mean value of the power data P is calculated. Fourier transform (FFT) is performed on the mean P value to extract the energy proportion in the 50-100Hz frequency band as a time-domain feature, which is used to reflect the tool wear amount; for the tool rotation speed, the fluctuation coefficient of V is extracted. As a time-domain feature, i.e., the standard deviation of rotational speed divided by the mean rotational speed multiplied by 100%; integrating physical parameter features and equipment parameter features, a feature vector matrix is ​​constructed to generate a multi-dimensional wear parameter feature vector, wherein the feature vector matrix includes:

[0058]

[0059]

[0060] in Represents the eigenvector matrix of physical parameters, Represents the feature vector matrix of device parameters, The time step representing tool wear data, This indicates the total number of sensors.

[0061] In S3, the process of fusing multi-dimensional wear parameter feature vectors to generate a tool wear feature vector set employs a machine learning algorithm, specifically including:

[0062] S301: Understanding input and output. Input is a multi-dimensional wear parameter feature matrix, where each row represents a data feature sample and each column represents a data time step; output is a tool wear feature vector set, which is a new matrix with the same number of rows but significantly reduced number of columns, and each new feature is a linear combination of the original features, which can better represent the tool state.

[0063] The objectives include: reducing the number of features through dimensionality reduction, thereby lowering the computational complexity of subsequent machine learning models; eliminating correlations between original features, making the features input into the model more independent; and creating new, more representative and discriminative features to improve the accuracy of model predictions.

[0064] S302: Feature fusion is performed using an unsupervised machine learning dimensionality reduction algorithm, Principal Component Analysis (PCA). The algorithm principle is as follows: PCA projects the original high-dimensional data into a new low-dimensional coordinate system through linear transformation. Specific steps include:

[0065] Input the eigenvector matrix obtained from S2 into the PCA algorithm; the algorithm calculates the covariance matrix of the data and its eigenvalues ​​and eigenvectors; sort the eigenvalues ​​from largest to smallest, and the corresponding eigenvectors are the principal component directions; select the first k principal components, where k represents the target dimension after dimensionality reduction; project the original data onto these k principal component directions to obtain a new dataset after dimensionality reduction, which is the fused tool wear feature vector set.

[0066] In step S4, the tool wear feature vector set is input into a pre-trained tool wear prediction model, which is a regression model trained based on a machine learning algorithm, and outputs a real-time quantitative value of tool wear.

[0067] S401: Before inputting the feature vector set, the model loading and validity verification must be completed to ensure that the model is suitable for the tool type and machining scenario of the current high-speed drilling machine: Model loading and parameter confirmation: Load the pre-trained regression model file from the model storage path. The core of the model is the Support Vector Regression (SVR) algorithm; Read the key parameter records during model training: including the tool type, machining material, feature dimension (which must be consistent with the feature vector dimension output by S2), and wear quantization target (such as the back face wear bandwidth VB value) corresponding to the training data. Confirm whether it matches the machining scenario of the current high-speed drilling machine. If it does not match, the pre-trained model for the corresponding scenario must be replaced.

[0068] For rapid model validity verification, select 3 to 5 sets of standard feature vectors with known VB values, input them into the loaded model, compare the predicted VB values ​​output by the model with the known true VB values, and calculate the error rate M. The specific calculation formula is as follows:

[0069]

[0070] in The model prediction value representing the wear bandwidth on the flank face. This represents the actual value of the wear bandwidth on the back face. The error rate of the sample is used to determine whether the model is effective. If the error rate is too large, the model file needs to be checked again to see if it is corrupted.

[0071] S402: Adaptation processing of tool wear feature vector set, adjusting the tool wear feature vector set to a format that meets the model input requirements:

[0072] Align the feature dimensions and order, extract the vector set to be input, and compare it with the input feature dimension table of the pre-trained model; if the dimensions of the input vector set are inconsistent with the dimensions required by the model, redundant features need to be removed; if the dimensions are missing, check whether the feature extraction step has been omitted, and supplement and complete the vector set.

[0073] Read the feature mean value during model training, convert the vector set format, and convert the adapted feature vector set into the input data format supported by the model. If it is deployed in an industrial PLC or embedded system, convert it into JSON format or binary data stream to ensure that the data transmission rate meets the high-speed monitoring requirements.

[0074] S403: Inference calculation and real-time wear quantification output of the pre-trained model. The model inference maps feature vectors to real-time tool wear quantification values ​​(such as VB values) to ensure the real-time performance of the inference process. Real-time inference mode selection: For batch processing scenarios, batch inference is used, and the set of input vectors collected within the processing cycle is input into the model. The model outputs VB prediction values ​​in batches, and the inference interval must be synchronized with the processing cycle. For single-piece high-precision processing scenarios, real-time inference is used. The inference trigger frequency is set, and each set of feature vectors is input into the model. The model outputs one VB prediction value within 100ms to ensure real-time tracking of wear changes.

[0075] During the model inference calculation process, the model's inference interface (Scikit-learn's model.predict()) is called, and the adapted feature vector set is input. The model performs calculations internally. Taking the random forest regression model as an example, multiple decision trees vote on the input features and output the average VB prediction value. The inference process is monitored for timeout. If a single inference takes more than 500ms, a backup inference thread is triggered to ensure that tool wear monitoring is not interrupted.

[0076] The system outputs and stores real-time wear quantification values, outputs the VB prediction values ​​calculated by the model, and adds an inference timestamp corresponding to the S1 data acquisition timestamp to form a real-time VB quantification value key-value pair. The output results are synchronously stored in the HMI interface of the high-speed drilling machine to provide tool wear parameter change values, which is convenient for operators to monitor intuitively.

[0077] In S5, the real-time wear quantification value is compared with a preset threshold, and tool wear status information and control commands are generated based on the comparison result:

[0078] Based on tool characteristics and machining requirements, preset thresholds and classifications are determined. Before comparing real-time wear quantification values, the threshold system must be clearly defined to avoid misjudgment or omission due to unreasonable threshold settings.

[0079] It needs further explanation that the threshold setting is based on collected data. The core basic data collected includes the tool material, tool type, material hardness, and machining accuracy requirements currently used on the high-speed drilling machine. Historical data is referenced, and the ISO tool wear standard is followed. Combined with the company's historical machining data, a multi-source data fusion analysis and experimental verification calibration method is used to determine the basic range of the threshold.

[0080] The experimental verification and calibration method is as follows: Taking the verification and calibration of cutting experiments as an example, a single-variable control experiment is conducted. Under the same machining conditions (fixed speed and feed rate), the tool VB value is manually controlled (through pre-wear treatment), and the machining quality corresponding to different VB values ​​is tested. When VB = 0.15mm, it is checked whether the tool is still within the usable range; when VB = 0.25mm, the wear state of the tool is recorded. A dynamic calibration mechanism is used, and each time new tool wear data is collected, the threshold range needs to be recalculated (using the sliding window method, with the window size being the most recent 300 batches of machining records). When the batch of machining materials or the tool changes, the experimental verification needs to be re-executed.

[0081] Multi-level threshold classification is set, and the wear degree is divided into three levels based on historical processing data. Taking the real-time wear quantification value VB as an example, the three levels are normal threshold, warning threshold and failure threshold.

[0082] What needs further explanation is the normal threshold. At this point, the machining dimensions and surface roughness meet the requirements, the tool is in the normal wear stage, its performance is good, and no intervention is required;

[0083] Warning threshold At this point, the tool is still suitable for short-term machining, but the wear is close to the critical value, and it is necessary to prepare to replace the tool.

[0084] Failure threshold When the tool wear has reached the end of its lifespan, continued use may lead to a deterioration in machining quality, and it is necessary to stop the machine immediately and replace the tool.

[0085] Based on the comparison results, tool wear status information and control commands are generated, specifically including:

[0086] The core parameter comparison first compares the VB value with the VB threshold, then compares... and Threshold, records the threshold level to which the two parameters belong (e.g., VB belongs to the warning threshold T2, (belongs to the normal threshold T1);

[0087] For auxiliary parameter comparison, if the parameter is at the warning threshold T2 or the failure threshold T3, it is necessary to further compare the auxiliary parameters (such as whether KT exceeds the warning threshold by 0.15mm, and whether the wear correlation value corresponding to W exceeds the wear threshold of 400℃). If the auxiliary parameter also exceeds the corresponding threshold, confirm the comparison result of the parameter; if the auxiliary parameter does not exceed the threshold, it is necessary to check whether there is an error in the measurement of the parameter.

[0088] Different control commands are matched according to the judgment level, and the triggering conditions and execution actions of the commands are clearly defined. When the normal threshold is reached, a continue processing command is generated; when the warning threshold is reached, a tool change preparation prompt command is generated; when the failure threshold is reached, an emergency stop and tool change command is generated. Control commands are transmitted through the communication interface (such as EtherCAT) between the high-speed drilling machine's PLC and the monitoring system. Based on the system's automatic detection of the command execution results, if the execution fails, a second command is immediately triggered and a status message indicating command execution failure is displayed, requesting manual shutdown.

[0089] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0090] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A tool wear monitoring method for a high speed drilling machine, characterized by, include: S1: The cutting process of the high-speed drilling machine is preset as the monitoring target. Multiple types of sensors are deployed to collect tool wear data in real time during the cutting process. The wear data includes physical parameters and equipment parameters. S2: Preprocess the collected tool wear data to generate a multi-dimensional wear parameter feature vector. The preprocessing includes outlier removal, normalization, and wear feature extraction. The wear feature extraction includes constructing a physical parameter feature vector matrix and a device parameter feature vector matrix , wherein the physical parameter features include a wear rate S calculated based on a tool length L, a ratio feature constructed based on a crater depth KT and a flank wear band width VB , and the device parameter features include an energy proportion in a specific frequency band based on a device power P; S3: The multi-dimensional wear parameter feature vectors of S2 are fused to generate a tool wear feature vector set. The fusion is performed using a machine learning algorithm. S4: Input the tool wear feature vector set in S3 into the pre-trained tool wear prediction model. The tool wear prediction model is a regression model trained based on machine learning algorithm, which outputs the real-time wear quantization value of the tool flank wear bandwidth VB. S5: Compare the real-time wear quantization value of S4 with the preset threshold, and generate tool wear status information and control commands based on the comparison result.

2. The method for monitoring tool wear on a high-speed drilling machine according to claim 1, characterized in that, The cutting process of the high-speed drilling machine preset in S1 is the monitoring target: The preset high-speed drilling machine cutting process includes turning and boring. Turning is a single-edged tool used to perform axial or radial cutting on a rotating workpiece, which is used to process cylindrical parts. Boring is a single-edged boring tool used to enlarge the inner diameter of the pre-drilled hole in the workpiece, which is used to process holes in box-type and bracket-type parts.

3. The method for monitoring tool wear on a high-speed drilling machine according to claim 1, characterized in that, The S1 collects tool wear data in real time during the cutting process by deploying multiple types of sensors, specifically including: A non-contact laser displacement sensor is installed on the side of the high-speed drilling machine spindle box to collect the tool length L; a high-speed industrial camera and a macro waterproof lens are installed on the side of the high-speed drilling machine worktable to collect the flank wear bandwidth VB and the crescent depth KT; an infrared thermal imager is installed diagonally above the high-speed drilling machine spindle to collect the cutting zone temperature parameter W; a current transformer and a power transmitter are installed in the high-speed drilling machine spindle motor control cabinet to collect the equipment power parameter P; and the tool speed parameter V is collected using the machine tool's built-in spindle encoder.

4. The method for monitoring tool wear on a high-speed drilling machine according to claim 1, characterized in that, S2 preprocesses the tool wear data to generate a multi-dimensional wear parameter feature vector: based on Outlier removal is performed according to the principle, a dataset for each parameter is created, and the mean of the dataset is calculated. with standard deviation Eliminate those that meet the requirements Abnormal data; Min-Max normalization is used to calculate the maximum value (max) and minimum value (min) of each parameter, mapping the data to... Interval.

5. The method for monitoring tool wear on a high-speed drilling machine according to claim 1, characterized in that, S3 fuses the feature vectors of multi-dimensional wear parameters: Principal Component Analysis (PCA) algorithm from machine learning is used to reduce the dimensionality of multi-dimensional wear parameter feature vectors and fuse them. The feature vector matrix is ​​constructed from the multi-dimensional wear parameter feature vectors and then input into the PCA algorithm. Calculate the covariance matrix of the eigenvector matrix, and solve for the eigenvalues ​​and eigenvectors of the covariance matrix; sort the eigenvalues ​​in descending order, and their corresponding eigenvectors are the principal component directions; select the principal component directions corresponding to the first k eigenvalues; project the original eigenvector matrix onto the first k principal component directions to obtain the dimensionality-reduced tool wear eigenvector set.

6. The method for monitoring tool wear on a high-speed drilling machine according to claim 1, characterized in that, The S4 tool wear prediction model is a regression model trained based on machine learning algorithms, which outputs real-time quantitative values ​​of tool wear. Model loading and validity verification: Load the pre-trained regression model file from the model storage path. The core of the model is either the Support Vector Regression (SVR) algorithm or the Random Forest Regression algorithm. Read the model training parameters and confirm that they match the current high-speed drilling machine processing scenario. Select 3-5 sets of standard feature vectors with known true values ​​of VB and input them into the model to calculate the error rate M. Feature vector set adaptation involves comparing the input feature vector set with the model input feature dimension table, removing redundant features or supplementing missing features to ensure that the dimensions are consistent with the model requirements; adjusting the vector set arrangement according to the feature order during model training; and converting the adapted vector set into JSON format. Model inference and quantization output: Batch inference is used for batch processing scenarios; real-time inference is used for single-piece high-precision processing scenarios; the model.predict() interface of Scikit-learn is called to perform inference, and timeout monitoring is performed on the inference process; VB prediction values ​​are output and inference timestamps are attached.

7. The method for monitoring tool wear on a high-speed drilling machine according to claim 1, characterized in that, S5 compares the real-time wear quantification value with a preset threshold, and generates tool wear status information and control commands based on the comparison result: Based on the tool material, model, hardness of the machined material, and ISO tool wear standards, the threshold is divided into three levels, specifically including: normal threshold, warning threshold, and failure threshold. Control commands are generated as follows: when the core parameters are at the normal threshold, a continue machining command is generated; when they are at the warning threshold, a tool change preparation prompt command is generated; and when they are at the failure threshold, an emergency stop and tool change command is generated. The control commands are transmitted to the high-speed drilling machine PLC terminal through the EtherCAT communication interface, and the system automatically detects the command execution results.

Images