Intelligent tool holder for heavy-duty gear teeth, and gear tooth tool wear prediction method and system

Abstract

The application discloses a kind of intelligent tool shank for strong gear tooth, gear tooth tool wear prediction method and system, and relates to gear intelligent precision machining technical field.The intelligent tool shank includes tool shank main body and vibration sensor and force sensor integrated in the near tool position area of tool shank main body.The application also provides a kind of prediction method, the method includes: acquisition vibration, force signal obtained by intelligent tool shank and spindle current signal obtained from machine tool numerical control system as multi-source sensing signal;Multi-source sensing signal is carried out feature extraction, and target feature sequence is obtained;Target feature sequence is input into a neural network based on state space model and is handled to extract the time sequence feature representing gear tooth tool wear evolution law;And based on the time sequence feature, the wear prediction value of gear tooth tool is calculated.The application improves the accuracy and reliability of tool wear prediction by near tool position sensing and multi-source information fusion, combined with advanced time sequence model.

Description

Technical Field

[0001] This application relates to the field of intelligent precision gear machining, and in particular to a powerful intelligent tool holder for gear turning, a method and system for predicting gear cutter wear. Background Technology

[0002] Power gear turning is a highly efficient and precise gear machining method suitable for internal gears, external gears, splined gears, and complex tooth profile parts. Compared with traditional gear hobbing, gear shaping, and ordinary milling, power gear turning involves a preset axial angle between the cutting tool and the gear workpiece. Tooth profile generation and material removal are achieved through the synchronous rotation of the tool spindle and workpiece spindle, along with the feed motion. This machining process is characterized by complex motion, intermittent meshing cutting, cyclic load variations, and non-stationary signals.

[0003] As a key component in heavy-duty gear turning, the wear condition of the gear cutting cutter directly affects the gear machining accuracy, tooth profile error, tooth direction error, tooth surface roughness, and machining stability. During heavy-duty gear turning, the cutting cutter inserts are subjected to cyclic impact loads, frictional heat loads, and alternating cutting forces for extended periods, which can easily lead to failure phenomena such as flank wear, cutting edge dulling, localized chipping, and uneven wear. When the wear of the cutting cutter reaches a certain level, it can result in increased cutting forces, intensified vibration, increased spindle load, and even lead to a decline in the machining quality of the gear workpiece, tool breakage, and abnormal machine tool shutdown.

[0004] Existing tool wear monitoring methods mainly include direct detection methods and indirect monitoring methods. Direct detection methods typically use microscopes, machine vision, or contour measurement equipment to perform offline measurements of the tool wear area. These methods offer high accuracy but usually require machine downtime for inspection, making them unsuitable for the online monitoring needs of heavy-duty gear turning. Indirect monitoring methods, on the other hand, collect signals such as cutting force, vibration, acoustic emission, and current during the machining process to establish a mapping relationship between the sensor signals and the tool wear state, showing greater potential for online applications.

[0005] However, existing intelligent tool holder and tool wear monitoring methods are mostly designed for ordinary milling, turning, or drilling conditions, and are difficult to directly apply to heavy-duty gear turning scenarios. On the one hand, heavy-duty gear turning requires maintaining a strict meshing relationship, and most of the tool holders used are non-standard products, resulting in significant multi-source coupling and non-stationary characteristics during heavy-duty gear turning. On the other hand, traditional heavy-duty gear turning machining condition monitoring uses external sensors, which are far from the cutting zone and cannot accurately reflect the dynamic load and vibration response of the gear cutting tool near the tool position. In addition, existing data-driven models often suffer from insufficient long-term modeling capabilities, large redundant feature interference, and significant fluctuations in prediction results when dealing with the wear evolution of the gear cutting tool throughout its entire life cycle. Summary of the Invention

[0006] The purpose of this application is to provide a powerful intelligent tool holder for gear turning, a method and system for predicting the wear of gear turning tools, and to solve the problems in the prior art where signal distortion is caused by the sensor being far from the cutting zone, and the data processing method has low accuracy in predicting long-term wear data.

[0007] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a powerful intelligent tool holder for gear turning, comprising: Handle body; Both the vibration sensor and the force sensor are integrated in the near-tool position area of ​​the tool holder body, and are used to sense the vibration signal and the cutting force signal during the heavy-duty gear turning process, respectively. A signal processing circuit, electrically connected to the vibration sensor and the force sensor, is used to process the vibration signal and the cutting force signal of the gear. A wireless transmission module, connected to the signal processing circuit, is used to wirelessly transmit the processed vibration signal and the cutting force signal of the gear to an external device. A power supply module is used to supply power to the signal processing circuit and the wireless transmission module.

[0008] Secondly, this application provides a method for predicting the wear of turning tools, including: Collect multi-source sensor signals, including vibration signals and cutting force signals obtained through the intelligent tool holder for high-power gear turning as described in the first aspect, and spindle current signals obtained from the CNC system of the machine tool. Feature extraction is performed on the multi-source sensing signals to obtain the target feature sequence; The target feature sequence is input into a state-space model-based neural network for processing to extract time-series features characterizing the wear evolution of the cutting tool. Based on the aforementioned timing characteristics, the wear prediction value of the cutting tool is calculated and output.

[0009] Thirdly, this application provides a tooth cutting tool wear prediction system, including: As described in the first aspect, a powerful intelligent tool holder for gear turning; CNC systems for machine tools; and A host computer is communicatively connected to the intelligent tool holder and the machine tool CNC system. The host computer is configured to execute the gear cutting tool wear prediction method as described in the second aspect to calculate and output the wear prediction value of the gear cutting tool.

[0010] Fourthly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the tooth cutting tool wear prediction method described above.

[0011] Fifthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the cutting tool wear prediction method described above.

[0012] Sixthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the cutting tool wear prediction method described above.

[0013] According to the specific embodiments provided in this application, the following technical effects are disclosed: The intelligent tool holder for heavy-duty gear turning integrates vibration and force sensors in the near-tool position area of ​​the tool holder body. This enables in-situ and direct sensing of the dynamic response during heavy-duty gear turning, reducing signal attenuation and distortion along the transmission path. This improves the signal-to-noise ratio and sensitivity of the collected vibration and gear cutting force signals, laying a data foundation for accurate prediction.

[0014] The gear cutting tool wear prediction method constructs a multi-source sensor signal system by fusing near-tool position vibration and force signals from the smart tool holder and spindle current signals from the machine tool's CNC system. This multi-source information fusion approach can characterize the wear state of the gear cutting tool from both the local dynamic response of the tool and the overall load of the machine tool, exhibiting stronger robustness and adaptability to complex working conditions compared to a single signal source.

[0015] Meanwhile, the tooth cutting tool wear prediction method employs a state-space model-based neural network (such as the GLMamba network) to process the extracted target feature sequence, effectively capturing the long-term temporal dependencies in the slowly changing process of tooth cutting tool wear. This network structure can selectively memorize and update historical information, thereby more accurately extracting temporal features characterizing the wear evolution pattern and improving the accuracy and stability of wear prediction. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1This is a schematic diagram of the overall structure of the intelligent tool holder for high-power gear turning in this application embodiment; Figure 2 This is a cross-sectional view of the intelligent tool holder for high-power gear turning in the embodiments of this application; Figure 3 This is a schematic diagram of the main structure of the intelligent tool holder for high-power gear turning in the embodiments of this application; Figure 4 This is a schematic diagram of the auxiliary mounting component structure of the intelligent tool holder for high-power gear turning in this application embodiment; Figure 5 This is a schematic diagram of the dedicated housing structure of the intelligent tool holder for high-power gear turning in the embodiments of this application; Figure 6 This is a schematic diagram of the gear cutting tool structure of the intelligent tool holder for high-power gear turning in the embodiments of this application; Figure 7 This is a schematic diagram of the intelligent tool holder signal acquisition and transmission framework in the embodiments of this application; Figure 8 This is a flowchart illustrating the tooth cutter wear prediction method in the embodiments of this application; Figure 9 This is a schematic diagram of the GLMamba network structure in an embodiment of this application; Figure 10 This is a schematic diagram of the tooth cutting tool wear prediction system in the embodiments of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] This application aims to provide a smart tool holder for heavy-duty gear turning and a corresponding method and system for predicting tool wear, to solve the problems of distant acquisition location of heavy-duty gear turning machining status monitoring signals, poor robustness of single signals, and low accuracy of long-term wear prediction in the prior art. The technical solution proposed in this application integrates sensors near the tool position of the tool holder, fuses multi-source information, and utilizes a deep learning model to achieve accurate online prediction of the wear state of the gear turning tool.

[0021] In one embodiment of this application, a smart tool holder for heavy-duty gear turning is provided, comprising: a tool holder body; a vibration sensor and a force sensor, both integrated in the near-tool position area of ​​the tool holder body, for sensing vibration signals and gear cutting force signals during heavy-duty gear turning, respectively; a signal processing circuit electrically connected to the vibration sensor and the force sensor, for processing the vibration signals and the gear cutting force signals; a wireless transmission module connected to the signal processing circuit, for wirelessly transmitting the processed vibration signals and gear cutting force signals to an external device; and a power supply module for supplying power to the signal processing circuit and the wireless transmission module.

[0022] Reference Figure 1 and Figure 2 Specifically, the core idea of ​​the powerful intelligent toolholder for gear turning lies in deeply integrating sensing functions into the toolholder itself. The toolholder includes a main body 1 as its basic structure, on which sensors for sensing the dynamic response of the machining process are integrated. Specifically, this includes a vibration sensor 8 and a force sensor 9, both positioned in the near-tool position area of ​​the main body. This design aims to fundamentally solve the problem of signal attenuation and distortion caused by excessive distance between the signal source and the sensor, capturing the most realistic and direct cutting dynamic information through in-situ sensing.

[0023] To process these raw sensor signals, a signal processing circuit 5 is integrated inside the tool holder. This circuit is electrically connected to the vibration sensor 8 and the force sensor 9, and its function is to amplify, filter, and digitize the weak, noisy analog signals output by the sensors. This step is a crucial step in converting analog signals from the physical world into digital information that a computer can understand.

[0024] Considering that the tool holder is a high-speed rotating component in heavy-duty gear turning, wired data transmission is clearly unsuitable. Therefore, the smart tool holder in this embodiment integrates a wireless transmission module 7. This module is connected to the signal processing circuit 5 and is responsible for transmitting the processed digital signal wirelessly (e.g., via Bluetooth, Wi-Fi, etc.) to an external receiving device, such as a host computer. This design enables real-time, online data transmission, providing possibilities for subsequent analysis and prediction.

[0025] To ensure the normal operation of the aforementioned electronic components, the tool holder also has a built-in power module, for example... Figure 2 The toroidal lithium battery 2 is shown. This power module provides stable and reliable power to the signal processing circuit 5 and the wireless transmission module 7, making the entire smart tool holder an independent sensing and data acquisition unit.

[0026] In summary, this intelligent tool holder for heavy-duty gear turning integrates vibration and force sensors in the near-tool position area of ​​the tool holder body. This allows for in-situ and direct sensing of the dynamic response during heavy-duty gear turning, reducing signal attenuation and distortion along the transmission path. This improves the signal-to-noise ratio and sensitivity of the acquired vibration and cutting force signals, laying a data foundation for accurate prediction. This solves the problem of low monitoring accuracy caused by the distant acquisition location of signals in existing heavy-duty gear turning machining status monitoring technologies.

[0027] In one specific implementation, such as Figure 2 and Figure 3 As shown, the tool holder body 1 can be structurally divided into an integrally formed spindle connection part 102 and a gear cutting tool mounting part. The spindle connection part 102 is provided with a spindle connection hole 101 and a spindle connection flange 103 for mating with the machine tool spindle to ensure a rigid connection and precise positioning between the tool holder and the machine tool. The gear cutting tool mounting part is further subdivided into a tool holder connection part connected to the spindle connection part 102 and a tool holder direct connection part 108 for directly connecting the gear cutting tool 11 tool holder. The tool holder direct connection part 108 is provided with a tool holder connection hole 109 for mounting the gear cutting tool 11 tool holder to transmit torque and prevent the tool holder from rotating. This segmented design ensures a reliable connection with the machine tool spindle and also provides a specific area for sensor installation.

[0028] In this specific embodiment, the core electronic components, such as the signal processing circuit 5, the wireless transmission module 7, and the power module (ring-shaped lithium battery 2), are not randomly arranged, but are compactly mounted on the periphery of the tool holder connection part using a specially designed auxiliary mounting part 3. Simultaneously, to ensure precise and stable mounting of the vibration sensor 8, a mounting groove 107 is specially provided on the outer wall of the tool holder connection part near the tool holder direct connection part 108, and the vibration sensor 8 is installed in this groove. This embedded mounting method not only protects the sensor but also ensures its rigid connection with the tool holder body 1, which is beneficial for the effective transmission of vibration signals.

[0029] For example, such as Figure 4 As shown, the auxiliary mounting component 3 is a cylindrical structure, with an internal battery positioning flange 301 for fixing the toroidal lithium battery 2 and a signal processing circuit mounting flange 303 for mounting the signal processing circuit 5. Its outer wall may have a housing mounting threaded hole 304 for connecting to the dedicated housing 6. This auxiliary mounting component 3 can be tightly fitted to the tool holder body 1 via a heat-fitting sleeve 302, thereby reliably fixing the electronic components to the tool holder body 1.

[0030] This embodiment proposes an innovative solution for the installation of the force sensor. The force sensor 9 is in the form of a flexible conformal force sensor, which can be wrapped around and attached to the periphery of the tool holder direct connection portion 108 like a thin film. The ingenuity of this design lies in the fact that when the cutting tool 11 undergoes a slight deformation under force during cutting, this deformation is transmitted to the tool holder direct connection portion 108, causing the flexible conformal force sensor 9 surrounding it to generate a corresponding strain, thereby outputting an electrical signal proportional to the cutting force. This solves the problem of accurately measuring the cutting force on a rotating, irregular tool holder surface.

[0031] like Figure 6 As shown, the cutting tool 11 itself includes a cutting tool shank 1101 for connection with the tool holder body 1 and a cutting insert 1102 for performing cutting functions. The smart tool holder of this application is designed to monitor the wear condition of the cutting insert 1102.

[0032] In another preferred embodiment, to further optimize the structural layout and mechanical properties, such as Figure 3 As shown, the tool holder connection is further refined into an integrally connected tool holder connection extension 104, a tool holder connection conical surface portion 105, and a tool holder connection contraction portion 106. The tool holder connection extension 104 is connected to the spindle connection portion 102, while the tool holder connection contraction portion 106 is connected to the tool holder direct connection portion 108. This coarse-to-fine transitional structure design helps to ensure smooth stress transmission and improves the overall rigidity and fatigue resistance of the tool holder.

[0033] Based on this structure, the mounting positions of each functional component have been further optimized. The vibration sensor mounting groove 107 is located on the outer wall of the tool holder connection retraction portion 106, which is closer to the cutting area and has a more sensitive vibration response. The relatively large electronic modules, namely the signal processing circuit 5, the wireless transmission module 7, and the power supply module, are mounted on the periphery of the relatively spacious tool holder connection extension portion 104 via auxiliary mounting parts 3. This layout places the sensitive sensor as close as possible to the signal source, while placing the electronic processing unit in a space with ample room and relatively low vibration, achieving an effective combination of structure and function.

[0034] Furthermore, to protect the precision sensors and electronic circuits integrated inside the tool holder from corrosion and physical impact from coolant, metal chips, oil, etc., in the machining environment, this specific embodiment also includes a tool holder housing, such as... Figure 1 and Figure 5The dedicated housing 6 is shown. This housing 6 is located around the body 1 of the tool holder, providing comprehensive encapsulation and sealing protection for the body 1 and all core components mounted thereon, including the signal processing circuit 5, wireless transmission module 7, power module, vibration sensor 8, and force sensor 9. This helps improve the durability and reliability of the smart tool holder in harsh industrial environments, ensuring its long-term stable operation.

[0035] For example, such as Figure 5 As shown, the dedicated housing 6 can be designed in segments for easy installation and maintenance. For example, it may include an electrical section mounting housing 601 corresponding to the tool holder connection extension 104, a frustoconical transition housing 602 corresponding to the tool holder connection conical section 105, a vibration sensor section mounting housing 603 corresponding to the tool holder connection retraction section 106, and a flexible conformal force sensor section mounting housing 604 corresponding to the tool holder direct connection section 108. Each housing segment is connected via mounting holes 607 and bolts. The housing may also be equipped with a signal circuit switch 605 for controlling the on / off state of the circuit and a magnetic charging port 606 for charging the internal power supply, thus providing comprehensive protection while ensuring ease of operation. To enhance the sealing effect, rubber sealing rings 10 can be installed at the connection between the dedicated housing 6 and the tool holder body 1 or at the connection between its segmented housings. Furthermore, a support flange 4 can be installed at specific locations on the tool holder body 1, such as near the spindle connection flange 103, to assist in supporting or fixing the dedicated housing 6 and enhance the overall structural stability.

[0036] It should be noted that the various specific implementation methods in the above embodiments can be combined to form different embodiments.

[0037] The following is a detailed embodiment of a powerful intelligent tool holder for gear turning.

[0038] Reference Figures 1 to 7 In one detailed embodiment, the powerful intelligent tool holder for gear turning includes: a tool holder body 1, a ring-shaped lithium battery 2, an auxiliary mounting component 3, a support flange 4, a signal processing circuit 5, a dedicated housing 6, a wireless transmission module 7, a vibration sensor 8, a flexible conformal force sensor 9, a rubber sealing ring 10, and a gear turning cutter 11.

[0039] The tool holder body 1, from top to bottom, consists of a spindle connection hole 101, a spindle connection part 102, a spindle connection flange 103, a tool holder connection extension part 104, a tool holder connection tapered part 105, a tool holder connection retractable part 106, a vibration sensor mounting groove 107, a tool holder direct connection part 108, and a tool holder connection hole 109. Keyways 110 are provided on both sides of the spindle connection flange 103.

[0040] The auxiliary mounting component 3 includes a battery positioning flange 301, a heat-fitting sleeve 302, a signal processing circuit mounting flange 303, and a housing mounting threaded hole 304. The auxiliary mounting component 3 is installed on the tool bar connection extension 104 by heat fitting.

[0041] There are four annular lithium batteries 2 in total, which are positioned by the battery positioning flange 301 and attached to the heat-fitting sleeve 302.

[0042] The support flange 4 has a two-sided split structure, which is bonded to the annular lithium battery 2, the heat-fitted sleeve 302 and itself respectively.

[0043] The signal processing circuit 5 also has a two-sided split structure and is mounted on the signal processing circuit mounting flange 303 via copper pillars; meanwhile, the wireless transmission module 7 is brazed onto the signal processing circuit 5.

[0044] The dedicated housing 6 consists of, from top to bottom, an electrical section mounting housing 601, a frustoconical transition housing 602, a vibration sensor section mounting housing 603, and a flexible conformal force sensor section mounting housing 604. The electrical section mounting housing 601 encloses the toroidal lithium battery 2, auxiliary mounting parts 3, support flange 4, signal processing circuit 5, and wireless transmission module 7, and is equipped with a signal circuit switch 605 and a magnetic charging port 606. The gap between the bottom opening of the dedicated housing 6 and the main body of the handle 1 is filled with a rubber sealing ring 10, and the top is installed in the housing mounting threaded hole 304 through the mounting hole 607 by bolts, and is closed by the battery positioning flange 301.

[0045] There are two vibration sensors 8 in total, which are symmetrically installed in the vibration sensor mounting groove 107 and enclosed by the vibration sensor segment mounting housing 603.

[0046] There are four flexible conformal force sensors 9 in total. They are made of polyvinylidene fluoride, a piezoelectric thin film material with positive piezoelectric effect, and are uniformly pasted on the direct connection part 108 of the tool holder. They are also wrapped by the flexible conformal force sensor segment mounting housing 604.

[0047] The gear cutting cutter 11 includes a gear cutting cutter bar 1101 and a gear cutting insert 1102, which are threadedly connected to the cutter bar connecting hole 109 through the external thread at the upper end of the gear cutting cutter bar 1101.

[0048] The toroidal lithium battery 2, wireless transmission module 7, vibration sensor 8, flexible conformal force sensor 9, signal circuit switch 605, and magnetic charging port 606 are all connected to the signal processing circuit 5.

[0049] During the high-power gear turning process, when the gear cutting insert 1102 cuts the gear, the gear cutting tool holder 1101 deforms under the cutting force, generating corresponding strain that is transmitted to the flexible conformal force sensor 9. The charge signal generated by the flexible conformal force sensor 9 is transmitted to the signal processing circuit 5. The vibration sensor senses the acceleration change caused by the machining vibration of the gear cutting insert and transmits the voltage signal to the signal processing circuit 5. The signal processing circuit 5 encodes the sensor signal and transmits it to the host computer. The signal processing circuit includes a signal preprocessing module, a digital-to-analog conversion module, a wireless transmission module, a power supply module, and a control module. The signal preprocessing module is connected to the vibration sensor 8 and the flexible conformal force sensor 9, and performs signal conditioning and amplification on the analog signal output by the sensor. Then, it is transmitted to the digital-to-analog conversion module to convert the analog signal into a digital signal. Next, the encoded digital signal is transmitted to the wireless transmission module, and then the encoded digital signal is transmitted to the host computer based on the wireless transmission module. In addition, the power module is connected to the toroidal lithium battery to power the signal processing circuit; the signal circuit switch 605 determines whether the signal processing circuit 5 is started; the magnetic charging port 606 is connected to the power module charging circuit to charge the toroidal battery when its power is low.

[0050] The host computer receives vibration and turning force signals from the smart tool holder via a wireless transmission module. At the same time, it communicates with the CNC system via a serial communication protocol to receive the tool spindle current signal and preprocesses the vibration, force and current signals to ensure that the start and end times of acquisition are synchronized.

[0051] The above is the powerful intelligent tool holder for gear turning provided in this application. Based on the powerful intelligent tool holder for gear turning, this application also provides a method for predicting the wear of gear turning tools.

[0052] Reference Figure 8 In one embodiment of this application, a method for predicting the wear of a cutting tool is provided, which specifically includes: Step S110: Collect multi-source sensor signals, including vibration signals and cutting force signals obtained through the intelligent tool holder for high-power gear turning provided in this application, as well as spindle current signals obtained from the machine tool CNC system.

[0053] Step S120: Extract features from the multi-source sensor signals to obtain the target feature sequence.

[0054] Step S130: Input the target feature sequence into a neural network based on a state-space model for processing to extract time-series features that characterize the wear evolution of the cutting tool.

[0055] Step S140: Based on the time series characteristics, calculate and output the wear prediction value of the turning tool.

[0056] Specifically, regarding the above method for predicting the wear of cutting tools: First, during the high-power gear turning process, a multi-source sensor signal acquisition step is performed (step S110). This includes acquiring vibration signals and gear cutting force signals directly related to the tool state using the intelligent tool holder for high-power gear turning of this application. Simultaneously, a spindle current signal reflecting the overall load of the machine tool is acquired from the machine tool CNC system. By fusing multiple pieces of information from both the tool end and the machine tool end, a more comprehensive data foundation is constructed, which is the first step in improving prediction robustness.

[0057] In a preferred embodiment, two key preprocessing steps are included before feature extraction from the acquired multi-source sensor signals. First, the vibration signal, cutting force signal, and spindle current signal are time-aligned. Since the vibration / force signal and spindle current signal come from different acquisition systems, their start and stop times may differ slightly. Time alignment calibrates these signals on the time axis, ensuring that each data point in subsequent analysis corresponds to the same machining moment—a prerequisite for multi-source information fusion. Second, the time-aligned multi-source sensor signals are preprocessed, including in-cutout rejection and signal denoising. In-cutout rejection aims to remove transient signal portions at the start and end of machining, as these portions have different signal characteristics than the stable cutting process and can interfere with the assessment of wear status. In-cutout rejection includes removing a predetermined proportion of transient signal portions at the beginning and end of each signal segment. Signal denoising includes one or more of filtering, wavelet denoising, empirical mode decomposition denoising, ensemble empirical mode decomposition denoising, or variational mode decomposition denoising. These preprocessing steps significantly improve the quality of the signal, providing a foundation for subsequent feature extraction and modeling.

[0058] Furthermore, after acquiring the original signal, it is necessary to extract its features (step S120) to obtain a target feature sequence that can quantify the changes in the signal.

[0059] In one specific implementation, feature extraction is performed on multi-source sensing signals to obtain a target feature sequence, specifically including: extracting time-domain features, frequency-domain features, and time-frequency-domain features of the target type from the vibration signal, the gear cutting force signal, and the spindle current signal, respectively; wherein, the time-domain features, frequency-domain features, and time-frequency-domain features of the target type are constructed into a target feature sequence according to the processing time order; the target types of the time-domain features include one or more of the following: mean, root mean square, variance, standard deviation, peak value, peak-to-peak value, skewness, kurtosis, waveform factor, peak factor, impulse factor, and margin factor; the target types of the frequency-domain features include one or more of the following: spectral centroid, dominant frequency amplitude, band energy, spectral entropy, mean square frequency, and frequency standard deviation; the target types of the time-frequency-domain features include one or more of the following: wavelet packet energy, wavelet packet energy entropy, short-time Fourier transform energy, empirical mode component energy, and Hilbert marginal spectrum features.

[0060] This specific implementation extracts statistical features reflecting the inherent laws of the preprocessed vibration signal, gear cutting force signal, and spindle current signal from multiple dimensions, including the time domain, frequency domain, and time-frequency domain. Time-domain features, such as mean, root mean square, and kurtosis, describe the overall amplitude and distribution of the signal; frequency-domain features, such as spectral centroid and dominant frequency amplitude, reveal the energy distribution of the signal across frequencies; and time-frequency domain features, such as wavelet packet energy, capture the dynamic process of signal frequency components changing over time. By constructing such a multi-domain feature set, signal changes caused by tool wear can be captured from different angles and comprehensively.

[0061] It should be noted that there are many types of time-domain features, frequency-domain features, and time-frequency-domain features, and not all time-domain features, frequency-domain features, and time-frequency-domain features are suitable for extraction and wear prediction. In a preferred embodiment, the target type of time-domain features, frequency-domain features, and time-frequency-domain features is determined during the training phase of the state-space model-based neural network.

[0062] During the training phase, multiple or all types of time-domain features, frequency-domain features, and time-frequency-domain features are extracted from vibration signals, cutting force signals, and spindle current signals to construct an original feature set. The comprehensive sensitivity index between each feature in the original feature set and the cutting tool wear value is calculated. The comprehensive sensitivity index is composed of the absolute Pearson correlation coefficient, mutual information, and monotonicity index. The features are sorted in descending order of comprehensive sensitivity index, and the top-k feature types are selected as the target types.

[0063] The comprehensive sensitivity index is not a single evaluation criterion, but rather a weighted average of the absolute Pearson correlation coefficient (representing linear correlation), mutual information (representing nonlinear correlation), and monotonicity index (representing trend consistency). All features are ranked using this comprehensive index, and the top-k feature types are selected as the final target types for modeling. This data-driven feature selection method helps improve the quality and relevance of input information, thereby enhancing the model's training efficiency and prediction accuracy.

[0064] Then, the target feature sequence constructed according to the processing time order is input into a neural network based on a state-space model for processing (step S130) to extract temporal features that can characterize the evolution law of tool wear. Tool wear is a typical temporal process, and its current state is closely related to its historical state. The state-space model is a powerful tool for dealing with such problems, as it can capture the inherent dynamic law of the system state evolving over time.

[0065] In one specific implementation, the state-space model-based neural network is a GLMamba network, the structure of which is shown in the diagram below. Figure 9 As shown, this network is a deep learning model designed to handle long sequence problems. Internally, it contains several carefully designed components, including an input projection module, a local gated enhancement module, a selective state space module, a long sequence memory fusion module, a residual normalization module, and an output mapping module. Through its unique selective state space mechanism and gating structure, the GLMamba network can effectively learn and memorize wear evolution information throughout the entire tool life cycle, overcoming the "forgetting" problem that traditional models easily encounter when dealing with long sequences. Its technical advantage lies in its ability to accurately extract deep temporal features strongly correlated with the wear accumulation process from complex feature sequences, laying the foundation for accurate prediction.

[0066] Finally, based on the deep temporal features extracted by the neural network, the wear prediction value of the cutting tool is calculated and output through an output module (step S140). This prediction value can be the specific value of the tool flank wear width (VB value), providing users with intuitive and quantifiable tool health status information.

[0067] The following is a detailed embodiment of a method for predicting the wear of cutting tools.

[0068] In one detailed embodiment, the tooth cutting tool wear prediction method includes: 1. Multi-source sensor signal acquisition.

[0069] During high-intensity gear turning, vibration signals and gear cutting force signals are acquired using the intelligent tool holder provided in this application. The vibration signals are acquired by a vibration sensor installed in the vibration sensor mounting groove, and the gear cutting force signals are acquired by a flexible conformal force sensor attached to the outer surface of the direct connection part of the tool holder.

[0070] The signal processing circuit inside the intelligent tool holder conditions, amplifies, converts analog to digital and encodes vibration signals and cutting force signals, and sends them to the host computer via a wireless transmission module.

[0071] Meanwhile, the host computer communicates with the machine tool's CNC system via a serial communication protocol to acquire the spindle current signal. The spindle current signal is used to reflect the changes in machine tool load during heavy-duty gear turning, and together with the near-tool position vibration signal and the gear cutting force signal, it constitutes a multi-source sensing signal.

[0072] 2. Time-align the vibration signal, the cutting force signal of the gear, and the spindle current signal so that the start and end times of the acquisition of the three types of signals are consistent.

[0073] 3. Preprocess the multi-source sensor signals that have completed time alignment.

[0074] (1) Cut-in and cut-out rejection process: Because there are obvious transient impacts during the entry and exit stages of high-power gear turning, the signal characteristics of these impacts differ from those of the stable meshing cutting stage. Therefore, the transient portion of the signal with a preset ratio at the beginning and end is removed, and the signal of the stable meshing cutting interval is retained.

[0075] For example, the transient portion of each signal segment is removed at the beginning and end, and the middle 90% of the stable meshing and cutting interval signal is retained.

[0076] (2) Signal noise reduction processing: Noise reduction is performed on the signal in the stable meshing cutting range. Noise reduction methods include one or more of the following: filtering, wavelet denoising, empirical mode decomposition denoising, ensemble empirical mode decomposition denoising, or variational mode decomposition denoising.

[0077] 4. Perform feature extraction on multi-source sensor signals.

[0078] Time-domain features, frequency-domain features, and time-frequency-domain features were extracted from the noise-reduced vibration signal, gear cutting force signal, and spindle current signal, respectively.

[0079] Time-domain features include one or more of the following: mean, root mean square, variance, standard deviation, peak value, peak-to-peak value, skewness, kurtosis, waveform factor, peak factor, impulse factor, and margin factor.

[0080] Frequency domain characteristics include one or more of the following: spectral centroid, dominant frequency amplitude, band energy, spectral entropy, mean square frequency, and frequency standard deviation.

[0081] The time-frequency domain features include one or more of the following: wavelet packet energy, wavelet packet energy entropy, short-time Fourier transform energy, empirical mode component energy, and Hilbert marginal spectrum features.

[0082] For example, 32 features are extracted from each sensor signal.

[0083] 5. Feature normalization processing.

[0084] The extracted features are normalized using the min-max normalization method to obtain a standardized original feature set: ; In the formula: Indicates the first j Based on sensor signals in a single processing sample s Extracted features i , and Let represent the minimum and maximum values ​​of the corresponding features in the training set, respectively. This represents the normalized eigenvalues.

[0085] 6. Sensitive feature screening (determining the target types of time-domain features, frequency-domain features, and time-frequency-domain features).

[0086] To reduce the impact of redundant and noisy features on model training, a comprehensive sensitivity index between each feature in the standardized original feature set and the wear value of the cutting tool is calculated.

[0087] The comprehensive sensitivity index is composed of a weighted average of the absolute Pearson correlation coefficient, mutual information, and monotonicity index: ; In the formula: Indicates based on sensor signals s Extracted features i The corresponding comprehensive sensitivity indicators, Representation of features x s,i Wear value of cutting tool y The absolute Pearson correlation coefficient between them Representation of features x s,i Wear value of cutting tool y Inter-information Representation of features x s,i Monotonicity index that varies with the processing procedure; , , These are the corresponding weights that sum to 1. In this example, they are 0.5, 0.4, and 0.1 respectively.

[0088] at the same time, Calculate using the following formula: ; In the formula: N Indicates the number of training samples; Indicates the first j The training sample of the th training sample i One characteristic, This represents the average value of the corresponding feature. Indicates the first j The actual wear values ​​of the cutting tools corresponding to each training sample. This represents the average wear value of the actual cutting tool.

[0089] Calculate using the following formula: In the formula: p ( x , y ) represents the joint probability distribution between the feature and the wear value of the cutting tool. p ( x )and p ( y ) represent the marginal probability distributions of the feature and the wear value of the cutting tool, respectively.

[0090] Calculate using the following formula: ; In the formula: N pos This indicates the number of times the feature increment is greater than zero in adjacent processed samples. N neg This indicates the number of times the feature increment is less than zero in adjacent processed samples.

[0091] After obtaining the comprehensive sensitivity index between each feature and the wear value of the cutting tool, the features are sorted in descending order of the comprehensive sensitivity index, and the top-k features are selected as the sensitive feature set (the type of the sensitive feature is the target type, which includes the time domain feature, frequency domain feature and time-frequency domain feature of the target type).

[0092] It should be noted that sensitive feature screening is only performed on the training set, while the validation set, test set, and online monitoring data use the same feature types as the training set.

[0093] 7. Input the target feature sequence into the GLMamba network for processing to extract the temporal features characterizing the wear evolution of the cutting tool.

[0094] The sensitive feature set is constructed into a sensitive feature sequence according to the processing time order or the processing sample order, and then input into the GLMamba network.

[0095] The GLMamba network includes an input projection module, a local control enhancement module, a selective state space module, a long sequence memory fusion module, a residual normalization module, and an output mapping module.

[0096] The specific processing steps for features by each module in the GLMamba network are as follows: (1) Input projection module.

[0097] Let the first l Layer t Input features at any time Mapped to latent space representation The calculation process is expressed as follows: ; In the formula: and They represent the first l The weight matrix and bias vector of the layer input projection module.

[0098] (2) Local control enhancement module.

[0099] The local control enhancement module is used to fuse the current input representation and the historical hidden state to generate update gates, selection gates, and memory gates. The calculation process is represented as follows: ; In the formula: Indicates the first l The layer was hidden a moment ago. This represents a vector concatenation operation. LayerNorm ( ) indicates layer standardization. , , These represent the update gate, selection gate, and memory gate, respectively. σ ( ) represents the Sigmoid activation function, and tanh() represents the hyperbolic tangent activation function.

[0100] (3) Selective state space module.

[0101] The selective state-space module is used to dynamically update the state-space vector according to the update gate. Its calculation process is expressed as follows: ; In the formula: and Indicates the first l The state space vectors of the previous and current time steps of the layer. Al and B l These are the state transition matrix and the input mapping matrix, respectively, and ⊙ represents element-wise multiplication.

[0102] (4) Long sequence memory fusion module.

[0103] The long sequence memory fusion module is used to fuse historical memory states and candidate memory states according to the selection gate. Its calculation process is expressed as follows: ; In the formula: Indicates candidate memory states, and They represent the first l The long sequence memory state of the previous time step and the current time step of the layer, C l This is the candidate memory mapping matrix.

[0104] (5) Residual normalization module and output mapping module.

[0105] The residual normalization module and the output mapping module are used to fuse the state space vector and the long sequence memory state to obtain the hidden output at the current time step. Their calculation process is represented as follows: ; In the formula: D l and E l These represent the memory state mapping matrix and the state space mapping matrix, respectively. Indicates the first l The layer currently hides its output. R l This represents the residual mapping matrix when the input dimension and the hidden output dimension are the same. R l Unit mapping.

[0106] For a network composed of multiple layers of GLMamba units, the first l Layer t The output at time t is used as the first l +1 floor t The input at time t is represented as: ; In the formula: l =1,2, …, L , L This indicates the total number of layers in the GLMamba network.

[0107] After multiple recursive steps, the final time series features are obtained. H .

[0108] The final time-series characteristics output by the GLMamba network H Input the wear prediction module and output the wear prediction value of the cutting tool: ; In the formula: This represents the predicted wear value of the cutting tool. This indicates the wear prediction module.

[0109] The wear prediction module employs one of the following: a multilayer perceptron, a linear regression layer, a one-dimensional convolutional prediction layer, or a gated regression layer.

[0110] In summary, the cutting tool wear prediction method in this application constructs a multi-source sensing signal system by fusing near-tool position vibration and force signals from the smart tool holder and spindle current signals from the machine tool CNC system. This multi-source information fusion approach can characterize the wear state of the cutting tool from both the local dynamic response of the tool and the overall load of the machine tool, exhibiting stronger robustness and adaptability to complex working conditions compared to a single signal source.

[0111] Meanwhile, the tooth cutting tool wear prediction method employs a state-space model-based neural network (such as the GLMamba network) to process the extracted target feature sequence, effectively capturing the long-term temporal dependencies in the slowly changing process of tooth cutting tool wear. This network structure can selectively memorize and update historical information, thereby more accurately extracting temporal features characterizing the wear evolution pattern and improving the accuracy and stability of wear prediction.

[0112] Reference Figure 10 In one embodiment of this application, a gear cutting tool wear prediction system is also provided. This system is a concrete implementation of the aforementioned intelligent tool holder and gear cutting tool wear prediction method. The system includes the powerful intelligent tool holder for gear cutting provided in this application, used to provide vibration and force signals near the tool position; a machine tool CNC system, used to provide machine tool status signals such as spindle current; and a host computer as the core control unit of the system. The host computer communicates with the intelligent tool holder and the machine tool CNC system via wireless and wired connections, collecting all data. The host computer is configured to execute the gear cutting tool wear prediction method provided in this application to calculate and output the predicted wear value of the gear cutting tool, and present it to the user through a display alarm unit.

[0113] This application provides an end-to-end technical solution. In a specific application scenario, such as a large-scale production line for automotive gearbox gears, the gear cutting tool wear prediction system provided in this application can play a significant role. When a machine tool equipped with an intelligent tool holder performs heavy-duty gear turning, the display and alarm unit of the host computer will display the current gear cutting tool wear prediction curve in real time. Operators can intuitively see the health status of the tool. When the predicted wear value approaches the preset alarm threshold (e.g., the wear width reaches 0.2mm), the system will automatically issue an audible and visual alarm and prompt "It is recommended to replace the tool." This allows operators to plan tool replacement before the tool completely fails, leading to workpiece scrap or machine tool damage, thus achieving predictive maintenance.

[0114] Furthermore, the tool wear prediction system can record multi-source sensor data and wear data throughout the entire tool lifecycle, which constitutes industrial big data assets. Process engineers can utilize this data for retrospective analysis. For example, they can analyze under which cutting parameters (speed, feed) the tool wears fastest, thereby optimizing machining processes, extending tool life, and reducing production costs. Simultaneously, this data can also be used to train and iterate wear prediction models, making them increasingly adaptable to different working conditions and materials, forming a continuously self-optimizing intelligent manufacturing closed-loop system.

[0115] In summary, this application achieves in-situ, high-fidelity acquisition of key information from the cutting process by highly integrating sensors, signal processing, wireless transmission, and power modules into the near-tool position area of ​​a high-power gear-turning tool holder. Furthermore, it proposes a multi-source information fusion method that integrates tool holder sensor signals and machine tool status signals, utilizing a GLMamba network to model and predict the complex, long-term process of tool wear. The intelligent tool holder, prediction method, and system provided in this application offer technical support for achieving intelligent, precise, and efficient gear machining.

[0116] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0117] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0118] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0119] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0120] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0121] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0122] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0123] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A powerful intelligent tool holder for turning gears, characterized in that, include: Handle body; Both the vibration sensor and the force sensor are integrated in the near-tool position area of ​​the tool holder body, and are used to sense the vibration signal and the cutting force signal during the heavy-duty gear turning process, respectively. A signal processing circuit, electrically connected to the vibration sensor and the force sensor, is used to process the vibration signal and the cutting force signal of the gear. A wireless transmission module, connected to the signal processing circuit, is used to wirelessly transmit the processed vibration signal and the cutting force signal of the gear to an external device. A power supply module is used to supply power to the signal processing circuit and the wireless transmission module.

2. The powerful intelligent tool holder for gear turning according to claim 1, characterized in that, The tool holder body includes an integral spindle connection part and a gear cutting tool mounting part; The gear cutting tool mounting part includes a tool holder connecting part connected to the spindle connecting part and a tool holder direct connecting part for connecting the gear cutting tool holder; The signal processing circuit, the wireless transmission module, and the power module are mounted on the periphery of the tool holder connection part via auxiliary mounting components; Additionally, a mounting groove is provided on the outer wall of the tool holder connecting portion near the tool holder direct connection portion; The vibration sensor is installed in the mounting groove; The force sensor is a flexible conformal force sensor and is mounted around the periphery of the tool holder direct connection portion.

3. The powerful intelligent tool holder for gear turning according to claim 2, characterized in that, The tool holder connection part includes an integral tool holder connection extension part, a tool holder connection conical surface part, and a tool holder connection retractable part. The tool holder connection extension part is connected to the spindle connection part, and the tool holder connection retractable part is connected to the tool holder straight connection part. The mounting groove is provided on the outer wall of the connecting and retracting part of the tool bar; The signal processing circuit, the wireless transmission module, and the power module are mounted on the periphery of the tool holder connection extension via the auxiliary mounting component.

4. The high-strength intelligent tool holder for gear turning according to claim 3, characterized in that, Also includes: The handle housing is disposed around the handle body and is used to enclose and protect the handle body, the signal processing circuit, the wireless transmission module, the power module, the vibration sensor, and the force sensor.

5. A method for predicting the wear of a turning tool, characterized in that, include: Collect multi-source sensor signals, including vibration signals and cutting force signals obtained by the intelligent tool holder for high-power gear turning as described in any one of claims 1-4, and spindle current signals obtained from the CNC system of the machine tool; Feature extraction is performed on the multi-source sensing signals to obtain the target feature sequence; The target feature sequence is input into a state-space model-based neural network for processing to extract time-series features characterizing the wear evolution of the cutting tool. Based on the aforementioned timing characteristics, the wear prediction value of the cutting tool is calculated and output.

6. The method for predicting tooth cutter wear according to claim 5, characterized in that, The neural network based on the state-space model is a GLMamba network; The GLMamba network includes an input projection module, a local control enhancement module, a selective state space module, a long sequence memory fusion module, a residual normalization module, and an output mapping module.

7. The method for predicting tooth cutter wear according to claim 5, characterized in that, Before performing feature extraction on the multi-source sensing signals, the method further includes: Time alignment is performed on the vibration signal, the cutting force signal of the gear, and the spindle current signal; The multi-source sensor signal that has completed the time alignment is preprocessed; wherein, the preprocessing includes cut-in and cut-out rejection processing and signal denoising processing, the cut-in and cut-out rejection processing includes rejecting the transient signal portions at the beginning and end of each signal segment according to a preset proportion, and the signal denoising processing includes one or more of filtering, wavelet denoising, empirical mode decomposition denoising, ensemble empirical mode decomposition denoising, or variational mode decomposition denoising.

8. The method for predicting tooth cutter wear according to claim 5, characterized in that, Feature extraction is performed on the multi-source sensing signals to obtain the target feature sequence, specifically including: Extract time-domain features, frequency-domain features, and time-frequency-domain features of the target type from the vibration signal, the gear cutting force signal, and the spindle current signal, respectively. The target feature sequence is constructed by the time-domain features, frequency-domain features, and time-frequency-domain features of the target type in the order of processing time. The target types of the time-domain features include one or more of the following: mean, root mean square, variance, standard deviation, peak value, peak-to-peak value, skewness, kurtosis, waveform factor, peak factor, impulse factor, and margin factor. The target types of the frequency domain features include one or more of the following: spectral centroid, dominant frequency amplitude, band energy, spectral entropy, mean square frequency, and frequency standard deviation. The target types of the time-frequency domain features include one or more of wavelet packet energy, wavelet packet energy entropy, short-time Fourier transform energy, empirical mode component energy, and Hilbert marginal spectrum features.

9. The method for predicting tooth cutter wear according to claim 8, characterized in that, The target types of the time-domain features, the frequency-domain features, and the time-frequency-domain features are determined during the training phase of the state-space model-based neural network. During the training phase, multiple or all types of time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the vibration signal, the cutting force signal of the gear, and the spindle current signal, respectively, to construct the original feature set; Calculate the comprehensive sensitivity index between each feature in the original feature set and the wear value of the cutting tool. The comprehensive sensitivity index is composed of a weighted average of the absolute Pearson correlation coefficient, mutual information, and monotonicity index. The features are sorted in descending order of the comprehensive sensitivity index, and the top-k feature types are selected as the target type.

10. A tooth cutting tool wear prediction system, characterized in that, include: Intelligent tool holder for high-strength gear turning as described in any one of claims 1 to 4; CNC system for machine tools; as well as A host computer is communicatively connected to the intelligent tool holder and the machine tool CNC system. The host computer is configured to execute the tooth cutting tool wear prediction method as described in any one of claims 5 to 9, so as to calculate and output the wear prediction value of the tooth cutting tool.

Images