A new energy decelerator gear cutter checking system

Abstract

The present application belongs to the field of mechanical engineering, and specifically relates to a new energy reducer gear cutter checking system, aiming to solve the problem of cumulative tooth profile error caused by tool wear and dynamic disturbance. The system integrates online monitoring, three-dimensional tooth profile inversion, error tracing and compensation generation modules. Through real-time acquisition of cutting force, vibration and other signals by multiple sensors, the equivalent blade morphology is inverted combined with finite element simulation, the error components are decomposed and the numerical control parameters are dynamically adjusted to realize closed-loop compensation of machining error. At the same time, a life prediction model based on improved Weibull distribution is introduced, which combines cutting force characteristics and blade degradation data to provide residual life rolling prediction and preventive maintenance warning. The system supports digital twin data interaction and edge computing deployment to improve the consistency and intelligent level of gear batch processing.

Description

Technical Field

[0001] This invention belongs to the field of mechanical engineering, specifically relating to a new energy reducer gear cutting tool verification system. Background Technology

[0002] In the field of mechanical manufacturing, gears, as the core component of power transmission, directly determine the efficiency, noise level, and service life of the transmission system through their machining accuracy. With the increasing demands on reducer performance from new energy vehicles, the precision standards for gear manufacturing are becoming increasingly stringent. High-precision, high-consistency gear machining has become a key technological bottleneck in the industry. Gear cutting tools, as the direct tools for cutting gear teeth, rely on the accuracy and stability of their geometric parameters to ensure gear machining quality. Even minute errors in the cutting tool will be amplified step by step and ultimately reflected in the finished gear. Therefore, precise calibration of the cutting tool is an indispensable and crucial step in precision gear manufacturing.

[0003] The machining of gears for new energy vehicle reducers places higher demands on the dynamic and systematic nature of tool verification. Because reducers for new energy vehicles generally employ high-speed, high-torque designs, their gears often feature special profiles and high overlap ratios, resulting in complex and highly customized tool structures. Verification of these tools involves not only measuring static geometric parameters but also comprehensively evaluating their dynamic performance under simulated cutting conditions. Traditional offline, single-point testing methods are no longer sufficient to meet the requirements of efficient, continuous, and closed-loop quality control.

[0004] Current technologies primarily rely on manually operated coordinate measuring machines (CMMs) or optical projectors for tool sampling inspection, which suffers from long inspection cycles, low sampling rates, and a lack of real-time feedback. Furthermore, the inspection data is disconnected from actual machining parameters and lacks effective integration with CNC systems and machining simulation modules, making it difficult to accurately predict tool performance in real cutting environments. In addition, the scattered storage of data from multiple batches of tools, coupled with the lack of a unified data analysis model, makes it impossible to predict tool wear trends and manage tool life, and also hinders process optimization and quality traceability. These problems are particularly pronounced in the mass production of gears for new energy vehicle reducers, severely impacting product consistency and production efficiency. Therefore, there is an urgent need for a tool verification system for new energy vehicle reducers that integrates automated inspection, data fusion analysis, and closed-loop process feedback. Summary of the Invention

[0005] The purpose of this invention is to provide a tool calibration system for gears in new energy vehicle reducers, addressing the problem of accumulated tooth profile errors caused by inaccurate tool parameters, geometric wear accumulation, and dynamic cutting force disturbances during existing gear machining processes. With the continuous improvement in the efficiency, quietness, and lifespan requirements of new energy vehicles' transmission systems, reducer gears need to meet higher precision levels and more complex profile designs. Traditional tool calibration methods rely on offline measurement and experience-based compensation, failing to respond in real-time to multi-physics coupling changes during machining, resulting in poor consistency, high scrap rates, and long debugging cycles in mass production. Furthermore, existing technologies lack systematic modeling of tool performance degradation throughout its entire lifecycle, hindering the transition from single-time calibration to predictive maintenance and restricting the deep application of intelligent manufacturing in the precision gear field.

[0006] The technical solution of this invention is to construct a closed-loop verification system integrating tool status perception, dynamic model correction, cutting force feedback compensation, and life prediction. This system includes a tool parameter initialization module, an online monitoring module, a three-dimensional tooth profile inversion module, an error source analysis module, a compensation strategy generation module, and a tool health assessment module. The tool parameter initialization module is used to input the tool's design geometric parameters, material properties, and initial cutting edge contour data, and to establish a standard cutting dynamics model. The online monitoring module uses a multi-channel sensor array deployed at the machine tool spindle and workpiece clamping end to collect in real time the triaxial cutting force, torque fluctuation signals, vibration acceleration spectrum, and local temperature field distribution during the cutting process. The three-dimensional tooth profile inversion module, based on measured cutting force time-series data and combined with a finite element cutting simulation kernel, reconstructs the equivalent cutting edge shape of the tool actually participating in cutting and calculates its normal deviation surface relative to the ideal tooth profile. The error source analysis module decomposes the normal deviation surface into systematic offset components and random fluctuation components. The former is attributed to tool installation errors, machine tool thermal deformation, and static geometric wear, while the latter is related to microscopic chipping, built-up edge formation, and transient load impacts. The compensation strategy generation module dynamically adjusts the CNC system's interpolation commands based on the type and amplitude of the error components. Systematic offsets are compensated for by modifying the differential gear ratio and axial feed coefficient of the hobbing machine, while random fluctuations trigger an adaptive damping control algorithm to adjust the matching relationship between the spindle speed and feed rate to suppress resonant mode excitation. The tool health assessment module, based on a long-term accumulated sequence of cutting force characteristic parameters, uses an improved Weibull distribution function to fit the tool reliability curve and introduces surface fatigue damage factors and the rate of increase in the cutting edge radius as degradation indicators to achieve rolling prediction of remaining service life.

[0007] Furthermore, the multi-channel sensor array in the online monitoring module includes embedded force-sensitive elements, whose placement is determined through topology optimization to ensure maximum sensitivity to pressure gradients in critical cutting areas. The sensor sampling frequency is no less than 20 kHz, supporting effective capture of high-frequency chatter components. The finite element cutting simulation kernel within the 3D gear profile inversion module pre-loads material removal models for various typical gear meshing conditions, covering dry cutting, wet cutting, and micro-lubrication modes. It can automatically call upon the corresponding friction coefficient library and thermal conductivity parameter table based on the material grade of the gear being processed. The inversion process employs an incremental iterative method, updating the equivalent cutting edge model after each tooth groove is machined to ensure the timeliness of the state estimation. The error source analysis module incorporates a wavelet packet decomposition engine, dividing the original cutting force signal into eight subspaces based on frequency bands. These subspaces correspond to different types of failure mechanisms: the 0.1 Hz to 1 Hz band is associated with slow tool wear; the 1 Hz to 50 Hz band reflects clamping looseness or machine tool clearance; the 50 Hz to 500 Hz band indicates initial microcrack propagation; and components above 500 Hz are marked as sudden chipping events. The changing trends of energy proportions in each frequency band constitute a tool condition fingerprint, used to distinguish between normal degradation and abnormal deterioration.

[0008] Furthermore, the adaptive damping control algorithm in the compensation strategy generation module designs a feedback gain matrix based on the Lyapunov stability criterion. Its real-time input variables include the current spindle angular velocity, feed acceleration, dominant vibration frequency, and their phase difference. The algorithm output is a set of constrained process parameter adjustments, the range of which is jointly limited by the machine tool dynamics boundary conditions and gear accuracy standards. When the feed compensation exceeds the set threshold three times consecutively, the system automatically locks the subsequent machining process and initiates the tool change procedure. The degradation index calculation cycle of the tool health assessment module is executed once every 100 machining cycles. Each calculation integrates the latest peak cutting force, average power consumption growth rate, and the inverted cutting edge curvature change rate. The prediction results are presented in the form of a probability density function. When the lower limit of the 95% confidence interval is less than the expected machining time of the next planned batch, the system issues a preventive replacement warning 48 hours in advance.

[0009] In one embodiment of the present invention, a data buffer is established between the three-dimensional tooth profile inversion module and the error source analysis module to temporarily store the complete inversion results of the most recent N machining cycles. The value of N ranges from 50 to 200, and the specific value is dynamically set according to the gear module. This buffer supports sliding window statistical analysis and can identify local wear patterns with periodic recurrence characteristics. When more than three out-of-tolerance records occur consecutively at the same location, the system determines that it is due to tool structure defects or improper process planning, rather than random wear. At this time, the compensation strategy generation module switches to root cause tracing mode to re-verify the consistency between the tool design parameters and the fixture positioning reference.

[0010] In one embodiment of the present invention, the tool health assessment module integrates a digital twin interface, enabling encrypted uploading of locally accumulated status data to a cloud data center to participate in the construction of an industry-level tool life prediction model. This interface adheres to a unified data format protocol, ensuring information exchange between devices from different manufacturers. The cloud model periodically distributes updated prior knowledge of degradation patterns to correct the distribution parameters in the local assessment module, thereby improving prediction accuracy. The local system only accepts remote parameter injection upon receiving a valid digital signature, ensuring information security.

[0011] In one embodiment of the present invention, the operational logic of the entire verification system is embedded in a distributed edge computing architecture, with each functional module deployed as a containerized microservice on edge nodes close to the machine tool. Inter-module communication employs a publish-subscribe message bus mechanism to ensure low-latency transmission of high-real-time data streams. Critical decision commands are verified through dual redundant channels before being issued to the CNC system to prevent safety accidents caused by misoperation. The system supports integration with the MES platform, automatically acquiring gear specification information from production orders and loading corresponding verification process templates accordingly, achieving personalized quality control for each product.

[0012] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0013] This solution establishes a two-way closed-loop feedback mechanism from the physical cutting process to the digital model, achieving full-process observability and controllability of tool status. Unlike traditional open-loop verification methods, this system can complete a full "sensing-inversion-analysis-compensation" cycle in each machining cycle, enabling tooth profile errors to be identified and corrected at the nascent stage, significantly improving the dimensional consistency and surface integrity of batch products. This solution is the first to use cutting force frequency domain fingerprinting for the classification and identification of tool failure modes, breaking through the limitations of relying solely on a single threshold alarm, and shifting maintenance decisions from passive response to proactive prediction. Based on a multi-source data fusion-based remaining life prediction model, which comprehensively considers the interactive effects of mechanical load, thermal effects, and material fatigue, the prediction error rate is reduced by at least 40% compared to traditional methods, significantly reducing unplanned downtime and resource waste caused by excessive replacement. This solution adopts a modular and service-oriented software architecture, possessing good scalability and compatibility, and can be adapted to various brands and models of gear machining centers, providing generalized technical support for the intelligent manufacturing of core components of new energy reducers. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall technical architecture of the new energy reducer gear tool verification system proposed in this invention. Detailed Implementation

[0015] Please refer to Figure 1 This application provides a new energy vehicle reducer gear tool calibration system, aiming to solve the problem of continuously increasing tooth profile errors caused by inaccurate tool geometry parameters, accumulated edge wear, and dynamic cutting force disturbances during the machining of new energy vehicle reducer gears under high precision and high efficiency requirements. As the requirements for NVH performance, load-bearing capacity, and fatigue life of transmission systems become increasingly stringent, gear tooth profile modification designs are becoming increasingly complex. Traditional tool calibration methods relying on manual experience and offline testing can no longer meet the technical requirements for machining consistency, process stability, and quality traceability in the intelligent manufacturing environment. Existing technologies typically perform a static tool calibration before mass production begins and evaluate machining quality through sampling after several batches. This model has inherent defects such as response lag, coarse compensation, and lack of real-time feedback, leading to the continuous accumulation of small errors during continuous machining, ultimately causing the entire batch of workpieces to exceed tolerances and be scrapped. Furthermore, current mainstream solutions have not established a quantitative model of the multi-physics field coupling degradation behavior throughout the tool's entire life cycle, making it difficult to achieve a paradigm shift from "post-failure maintenance" to "predictive maintenance."

[0016] This invention constructs a closed-loop, adaptive tool calibration system. Its core architecture consists of six functional modules: tool parameter initialization module, online monitoring module, 3D tooth profile inversion module, error source analysis module, compensation strategy generation module, and tool health assessment module. These modules work collaboratively according to strict temporal logic and data flow relationships, forming a complete technical closed loop of "perception—modeling—diagnosis—decision—execution—prediction." The system is deployed on an industrial control platform with edge computing capabilities, directly connected to the real-time data bus of a CNC gear hobbing machine or gear shaping machine, supporting millisecond-level response latency and terabyte-level historical data storage. The entire system's operating cycle is based on the machining of a single gear tooth groove as the basic time unit. After each tooth groove cutting action is completed, the system initiates a complete state assessment and compensation update process, ensuring the timeliness and continuity of error correction.

[0017] The tool parameter initialization module serves as the starting point for system operation, undertaking the task of building the basic model. The module receives the original tool CAD geometric model from the product design end, extracts all design parameters including module, pressure angle, helix angle, addendum coefficient, displacement coefficient, and tool tip radius, and converts them into a structured dataset in an internally unified format. Simultaneously, it inputs information such as tool material type, heat treatment hardness grade, coating type, and its thickness distribution for assigning physical property parameters in subsequent cutting thermodynamic simulations. For newly installed tools, operators use a high-precision optical profilometer to scan the initial cutting edge microstructure, obtaining a two-dimensional cross-sectional coordinate sequence of sampling points every 0.1 mm along the entire cutting edge length, collecting no fewer than 2000 spatial points to form an initial digital twin of the cutting edge. Based on the above parameters, the system automatically constructs a standard cutting dynamics model. This model uses a lumped parameter method to describe the dynamic coupling relationship between the tool, workpiece, and machine tool, and includes three core sub-matrices: mass matrix, damping matrix, and stiffness matrix. The stiffness matrix is ​​nonlinearly corrected based on the tool overhang length, clamping method, and spindle front bearing preload to reflect the structural flexibility characteristics under actual installation conditions. All initialization data is hash-verified and written to the local security database, generating a unique device fingerprint for subsequent version comparison and change tracking.

[0018] The online monitoring module is responsible for acquiring multi-dimensional physical signals during the machining process, and is a key component for achieving full-process observability of the system. The module integrates a multi-channel sensor array, which consists of four types of sensing units: a triaxial piezoelectric force gauge, a torque sensor, a miniature accelerometer, and an infrared thermal imaging probe. The triaxial piezoelectric force gauge is mounted at the end flange of the tool spindle, employing a quartz crystal sensing element, possessing wide frequency response characteristics, a measurement range covering 0 to 50 kN, and a resolution of 10 N, used to capture transient cutting force components in the X, Y, and Z orthogonal directions. The torque sensor is embedded in the spindle drive link, recording dynamic torque fluctuations in the rotational direction in real time, with a sampling accuracy better than ±0.5% FS and a response bandwidth of 3 kHz. The miniature accelerometer is attached to the workpiece fixture base surface in a triangular layout, with a sampling frequency set to 20480 Hz, effectively capturing high-frequency vibration components within 5000 Hz, and exhibiting high sensitivity, particularly for resonant modes induced by tool chatter. The infrared thermal imaging probe is fixed inside the protective housing, covering the entire cutting area with a field of view. The frame rate is set to 60 fps, and the spatial resolution is 320×240 pixels, generating 60 local temperature field distribution maps per second. The minimum detectable temperature difference is 0.1℃. All sensors are equipped with independent signal conditioning circuits to perform filtering, amplification, and analog-to-digital conversion. Raw data is transmitted to the edge computing node via Gigabit Ethernet in floating-point array format. The data packet header includes a timestamp, device ID, sampling sequence number, and CRC checksum to ensure transmission integrity. The system is set to trigger synchronous sampling every 50 microseconds to ensure that the phase consistency error between channels is less than 1°, meeting the stringent coherence requirements of subsequent frequency domain analysis.

[0019] The core task of the 3D tooth profile inversion module is to inversely deduce the actual equivalent cutting edge shape of the current tool from the measured cutting force signal and calculate its normal deviation surface relative to the ideal tooth profile. The module has a built-in finite element cutting simulation kernel, which is developed based on an explicit dynamics algorithm and adopts adaptive mesh generation technology to control the solution time to less than 200 milliseconds while ensuring calculation accuracy. The simulation kernel is preset with three typical machining environment configurations: in dry cutting mode, the air convection heat transfer coefficient is set to 15 W / (m²·K), with no lubricating medium involved; in wet cutting mode, the coolant flow rate is set to 40 L / min, the nozzle angle is adjusted to form a 60° angle with the cutting zone, and the thermal conductivity is increased to 80 W / (m²·K); in micro-lubrication mode, the oil mist concentration is controlled at 5 ml / h, the lubricating film thickness is modeled as 0.5 μm, and the friction coefficient is reduced to 0.08. The system automatically matches the corresponding mode based on the process card information issued by the MES platform and loads the physical property library of the corresponding material. For example, the yield strength of 20CrMnTi steel is 980 MPa, the elastic modulus is 210 GPa, the Poisson's ratio is 0.3, and the coefficient of thermal expansion is 12 × 10⁻⁶ / ℃. The inversion process adopts the incremental iteration method, and the specific steps are as follows: First, the machining path of the current tooth groove is discretized into 10,000 tiny slices, and the cutting depth is assumed to be constant within each slice interval; second, the finite element kernel is called to simulate the material removal process under the slice, and the theoretical cutting force curve is output; then, the simulation results are correlated with the measured force signal, and the objective function is defined as the mean square error of both.

[0020]

[0021] in, The cost function represents the degree of deviation between the simulation and the actual measurement. This is the measured three-dimensional resultant force timing signal; Based on the current parameter set The simulated output force signal; The vector of cutting edge morphology parameters to be optimized includes the radial offset, tangential tilt angle, and local blunt circle radius at each sampling point; This refers to the time length of a single tooth groove machining cycle. The system employs an improved conjugate gradient method for... Iterative optimization is performed, with the cutting edge model updated and the simulation rerun in each iteration, until... The value drops below the preset convergence threshold of 0.05, or the number of iterations reaches the upper limit of 50. The final optimal output... This constitutes the equivalent cutting edge digital model of the current tool. Based on this model, the system further calculates the tooth profile curve generated under the ideal meshing trajectory and compares it point by point with the theoretical tooth profile to generate a normal deviation surface with a resolution of not less than 0.01 mm, which is stored in grayscale image format. A positive deviation value indicates overcutting of material, and a negative value indicates undercutting.

[0022] The error source analysis module receives the normal deviation surface and original cutting force time series data output from the 3D tooth profile inversion module, and performs in-depth fault mechanism decomposition. The module has a built-in wavelet packet decomposition engine, which selects the Daubechies db4 wavelet basis function to perform 4-level decomposition on the three-dimensional cutting force signal, generating a total of 8 frequency band sub-signals covering the full frequency range from 0.1 Hz to 10 kHz. The frequency bands are divided as follows: Band 1 (0.1–1 Hz) corresponds to extremely low frequency drift components, mainly reflecting the rise in cutting force baseline caused by slow and uniform tool wear; Band 2 (1–50 Hz) covers slow-varying interference sources such as machine tool thermal deformation, foundation settlement, and fixture loosening; Band 3 (50–500 Hz) is associated with gear blank eccentricity, blank positioning error, and initial microcrack propagation; Band 4 (500–2000 Hz) corresponds to the tool's natural modal vibration, which easily excites regenerative chatter; Band 5 (2000–5000 Hz) marks sudden damage events such as local edge chipping and coating peeling; Band 6 (5000–8000 Hz) reflects the periodic formation and shedding process of built-up edge; Band 7 (8000–10000 Hz) captures subsurface damage induced by grain boundary slip and phase transformation; Band 8 (>10000 Hz) (Hz) is reserved as a noise isolation zone and is not included in the analysis. The system calculates the energy percentage for each frequency band:

[0023]

[0024] in For the first The first frequency band Wavelet coefficients. 8-dimensional energy vector. This constructs a tool condition fingerprint map, serving as input features for classification and recognition. The module pre-trains a support vector machine classifier using a historical labeled dataset for supervised learning. Labels include five categories: "normal wear," "installation eccentricity," "thermal deformation-dominated," "chatter risk," and "chipping warning." When a new sample arrives, the classifier outputs the most probable failure mode category and its confidence score. If the confidence score is below 85%, a secondary discriminant analysis is initiated, combining the spatial distribution characteristics of the normal deviation surface (e.g., whether it exhibits circumferential symmetry, axial gradual change, or local concentration) for fusion judgment. Ultimately, the system decomposes the normal deviation surface into two parts: a low-frequency smooth component, attributed to systematic offsets, including tool installation errors, machine tool geometric errors, and overall wear trends; and a high-frequency random fluctuation component, attributed to transient disturbances such as micro-chipping, built-up edge adhesion, and impact loads. These two components are output to different processing branches of the compensation strategy generation module.

[0025] The compensation strategy generation module formulates differentiated compensation schemes based on the two types of error components output by the error source analysis module. For systematic offset components, the module adopts a feedforward compensation mechanism, actively correcting the error by modifying the G-code interpolation parameters of the CNC system. Specifically, when a continuously large radial dimension is detected, indicating uniform tool wear, the system automatically calculates the required compensation amount. ,in This is the integral mean of the average values ​​of the most recent 5 tooth groove normal deviation surfaces, in μm. This compensation amount is converted into an adjustment command for the differential gear ratio of the gear hobbing machine, using the following formula:

[0026]

[0027] in, This is the corrected differential ratio; The original setting value; The gear module; This represents the number of teeth on the gear being machined. Simultaneously, the axial feed coefficient... Scaling proportionally To coordinate the spiral unfolding speed, all parameter adjustments are injected into the CNC system PLC registers after double redundancy verification, with the update cycle synchronized with the gear slot machining cycle. For random fluctuation components, the module initiates an adaptive damping control algorithm, which designs a state feedback controller based on Lyapunov stability theory. The system state vector is defined. ,in Main axis angular velocity, Its rate of change, For feed acceleration, The phase difference is the dominant vibrational frequency. Construct the Lyapunov function. ,in It is a positive definite symmetric matrix, obtained by solving the Riccati equation. Feedback gain matrix. Depend on The calculation shows that, among which To control the input matrix, This is the weighting adjustment factor. The controller output is the spindle speed adjustment amount. With feed rate correction value Both are constrained by machine tool dynamics boundaries: the maximum spindle acceleration does not exceed the square of 300 rad / s², the feed system response delay is less than 10 ms, and the adjustment range must not exceed ±15% of the nominal value. The controller executes cyclically at a frequency of 100 Hz, monitoring the proportion of vibration energy in the fourth frequency band in real time. When this proportion exceeds the threshold of 70% within three consecutive sampling windows, the chatter risk level is determined to be high, forcibly switching to a speed reduction mode, reducing the spindle speed to 85% of its original value, and activating the active damping hydraulic support system. When the feedforward compensation exceeds the ±50 μm limit three times consecutively, the system determines that the tool has entered a stage of rapid wear, immediately locks the subsequent machining process, prohibits the start of new workpieces, and displays a red alarm on the human-machine interface, requiring the operator to execute the tool replacement procedure.

[0028] The tool health assessment module is designed for full tool lifecycle management, enabling rolling prediction of remaining service life. The module uses 100 complete gear machining cycles as one assessment period. Upon startup, it extracts a sequence of key characteristic parameters for the tool from the historical database over the most recent N cycles. N is a set of 128, dynamically adjusted based on the module of the gear being machined; for every 1 increase in module, N increases by 8, with a maximum of 200. Extracted features include: peak maximum cutting force per cycle. Average power consumption Total vibrational energy The inversion obtained cutting edge blunt circle radius and its rate of change of curvature The system calculates the growth rate sequence of the above parameters, for example... The module models the tool reliability function using an improved two-parameter Weibull distribution based on long-term observation data and fits its temporal evolution trend.

[0029]

[0030] in, For a moment The reliability of the tool is the probability that it has not failed. , where is the scale parameter, representing the feature lifetime; The shape parameter reflects the trend of failure rate change. <1 indicates early failure. =1 indicates random failure. >1 indicates wear and tear failure. Traditional Weibull models rely solely on time or processing quantity; this invention introduces two degradation covariates to enhance predictive ability: surface fatigue damage factor. With the growth rate of the blunt radius of the cutting edge . Calculated from the cumulative plastic strain energy density, the formula is as follows: ,in For the first stress of individual elements For its plastic strain, It is a volume element. The most recent 10 linear regressions were used to... Measurements were obtained. The system constructs an accelerated degradation model, which will determine the effective working time. Defined as:

[0031]

[0032] in, =0.35, =0.62, which is an empirical coefficient calibrated through extensive accelerated life testing. The final reliability function is updated to... .parameter and The system employs maximum likelihood estimation to update data online from the field, refitting the distribution curve after each evaluation cycle. Predictive results are output as a probability density function, providing the probability distribution of failure time over the next 1000 machining cycles. A safety threshold is set: when the lower limit of the 95% confidence interval's lifespan is less than the estimated machining time for the next planned batch, it is considered a high-risk state. A preventative replacement warning is sent via the MES system 48 hours in advance, and a spare parts requisition form is automatically generated. If three consecutive out-of-tolerance records occur at the same location within the sliding window, the root cause tracing mode is activated. The system retrieves the most recent 150 inversion results stored in the data buffer, performs spatial correlation analysis, and identifies the existence of a fixed-phase periodic wear pattern. If found, a "structural defect" alarm is reported, and it is recommended to check the tool manufacturing tolerances or the repeatability of the fixture positioning datum.

[0033] To enhance system robustness and intelligence, a data buffer is established between the 3D gear profile inversion module and the error source analysis module. This buffer temporarily stores the complete inversion results of the most recent N machining cycles, with N ranging from 50 to 200. The specific value is dynamically set based on the gear module size; the larger the module, the more historical data is required to ensure statistical significance. This buffer uses a circular queue structure to manage memory and supports sliding window-based statistical analysis. The system periodically performs pixel-level clustering analysis on all normal deviation surfaces within the buffer, identifying local out-of-tolerance points with fixed spatial locations and similar shapes. When a point appears more than three times consecutively in an area with an absolute deviation greater than 25 μm, the system determines that this phenomenon has strong periodicity, ruling out random disturbances and attributing it to weak points in the tool's structure or uncompensated system errors in the process planning. At this point, the compensation strategy generation module switches to root cause tracing mode, suspends the issuance of regular compensation commands, and instead calls the tool design parameter verification engine to compare the consistency between the current measured wear mode and the theoretical stress concentration area. If the matching degree exceeds 80%, a process optimization suggestion report will be generated, recommending adjustments to the cutting sequence, changes to the tool path, or replacement with a tool material of a higher toughness grade.

[0034] To further enhance the generalization ability of the prediction model, the tool health assessment module integrates a digital twin interface, supporting encrypted uploads of local status data to the cloud data center. The interface protocol follows the data structure defined by the ISO 23219 standard, transmitting anonymized feature parameter sequences, machining condition labels, and failure timestamps, without any enterprise identification information. The upload frequency is once daily by default, but manual triggering for instant synchronization is possible in emergencies. The cloud data center aggregates similar tool operation data from different regions and various machine models, using a federated learning framework to train a global life prediction model, avoiding cross-domain flow of raw data. After training, model parameters are distributed to each local terminal in the form of encrypted firmware packages to update the prior distribution parameters in the tool health assessment module. The local system only accepts remote injection after verifying the validity of the digital signature and the trustworthiness of the certificate chain, preventing malicious tampering. The updated model significantly improves the ability to identify rare failure modes, especially in predicting accuracy under extreme conditions such as high temperature and humidity, heavy load, and variable speed.

[0035] The entire verification system's operational logic is embedded in a distributed edge computing architecture. Each functional module is deployed as a containerized microservice on edge computing nodes close to the machine tool. Each module is encapsulated as an independent Docker container, sharing the host operating system kernel but possessing isolated file systems, network stacks, and process spaces. Inter-container communication is achieved through a lightweight publish-subscribe message bus using the MQTT protocol. The topic naming rule is " / machine / {id} / module / {name} / data", with a QoS level of 2 to ensure no message loss or duplication. High-real-time data streams (such as raw cutting force signals) are broadcast using UDP multicast, with latency controlled within 2 milliseconds; low-frequency control commands (such as compensation parameter updates) are reliably transmitted via TCP. Critical decision commands (such as machine shutdown / locking) must be verified through dual redundant channels: one path delivers the signal to the CNC system PLC via industrial Ethernet, and the other path is connected via a dedicated safety relay hardwired connection. The two signals must reach consensus within 100 milliseconds before execution. The system supports bidirectional integration with the MES platform, automatically acquiring gear specification information from production orders, including module, number of teeth, helix angle, precision grade, and profile modification curve type. Based on this, it loads the corresponding verification process template, achieving personalized quality control for each product. All operation logs, alarm records, and compensation history are written to the blockchain evidence storage system in real time, with hash values ​​recorded on the chain, ensuring that the entire process is auditable and non-repudiable.

[0036] This invention establishes a two-way closed-loop feedback mechanism from the physical cutting process to the digital model, achieving full-process observability and controllability of tool status. Unlike traditional open-loop verification methods, this system can complete a full "sensing-inversion-analysis-compensation" cycle within each machining cycle, enabling tooth profile errors to be identified and corrected at the nascent stage, significantly improving the dimensional consistency and surface integrity of batch products. This solution is the first to use cutting force frequency domain fingerprinting for the classification and identification of tool failure modes, breaking through the limitations of relying solely on single threshold alarms and shifting maintenance decisions from passive response to proactive prediction. Based on a multi-source data fusion-based remaining life prediction model, which comprehensively considers the interactive effects of mechanical load, thermal effects, and material fatigue, the prediction error rate is reduced by at least 40% compared to traditional methods, significantly reducing unplanned downtime and resource waste caused by excessive replacements. This solution adopts a modular, service-oriented software architecture with good scalability and compatibility, adaptable to various brands and models of gear machining centers, providing generalized technical support for the intelligent manufacturing of core components of new energy reducers.

[0037] Existing technologies in gear cutting tool calibration generally employ periodic shutdowns for inspection and manual intervention, essentially representing an open-loop control structure. Once machining begins, the actual cutting state of the tool is no longer dynamically tracked; operation is maintained solely by preset process parameters until a significant anomaly occurs or deviations are detected during sampling, at which point remedial measures are taken. This approach cannot address the gradual performance degradation of tools under continuous loads, nor can it identify sudden damage events, leading to a large number of potentially defective products flowing into the next process. Furthermore, existing methods lack the ability to finely analyze the sources of error, often attributing all deviations to "tool wear," resulting in a one-size-fits-all compensation strategy that is neither accurate nor economical. This invention fundamentally changes this situation. By deploying a high-density sensor network and a real-time inversion algorithm, a digital mirror of the actual cutting behavior of the tool is constructed, achieving a leap from "black box operation" to "transparent manufacturing." The system not only knows "whether deviations are out of tolerance," but also clearly answers four core questions: "why deviations are out of tolerance," "where is the damage," "how to correct it," and "when to replace," forming a complete intelligent decision-making chain.

[0038] In the error source analysis stage, the wavelet packet frequency band division mechanism and state fingerprint spectrum concept proposed in this invention constitute the essential features that distinguish it from existing technologies. Traditional methods typically only monitor whether the mean or peak value of the cutting force exceeds the limit, at most supplemented by FFT transformation to observe the dominant frequency component, making it difficult to distinguish different failure mechanisms. This scheme divides the entire frequency band into 8 subspaces with clear physical meaning and establishes a mapping relationship between the energy proportion of each frequency band and specific fault types, enabling the system to read an electrocardiogram like a doctor. Figure 1 In this way, specific pathological features can be identified from complex signal backgrounds. For example, if the energy in the first frequency band continues to rise while other frequency bands remain stable, it can be judged as uniform wear; if a sudden pulse energy burst occurs in the fifth frequency band, it strongly suggests a chipping event. This classification capability based on frequency domain fingerprints gives the system a rudimentary cognitive reasoning function, far exceeding simple threshold comparison logic.

[0039] In terms of lifespan prediction, the improved Weibull model proposed in this invention, which integrates the surface fatigue damage factor and the growth rate of the cutting edge blunt radius, significantly outperforms traditional methods that rely solely on time or cutting length. Traditional models assume a uniform degradation process, neglecting the influence of actual operating condition fluctuations. This solution incorporates effective working time. The concept of this invention converts every severe load, every sudden temperature change, and every microscopic damage into an equivalent aging increment, ensuring that lifespan calculations truly reflect actual usage intensity. This model has been validated through comparisons at three different gear manufacturers. Under the same batch of cutting tools, the deviation between the average remaining life predicted by this invention and the actual failure time is 12.3%, while the deviation of the traditional method is as high as 21.7%, representing an improvement of nearly 44%. More importantly, this solution can issue an effective early warning an average of 7.2 hours before failure, reserving sufficient buffer time for production scheduling and avoiding the paralysis of the entire production line due to sudden shutdowns.

[0040] At the system architecture level, this invention adopts a technical approach combining containerized microservices and edge computing, solving the problems of slow response, difficult expansion, and poor compatibility in traditional centralized control systems. The decoupled design of each module allows for independent upgrades, replacements, or horizontal expansion. For example, when a new type of sensor needs to be connected, only the corresponding data acquisition microservice container needs to be developed, without modifying the entire system architecture. The publish-subscribe communication mechanism ensures data throughput in high-concurrency scenarios; in actual testing, even with a single device generating 1.2 GB of raw data per second, it still maintains a 99.99% message delivery success rate. The dual-redundant instruction verification mechanism fundamentally eliminates the risk of misoperation due to software bugs or network jitter, meeting the requirements of ISO 13849-1 safety level PL e.

[0041] In summary, this invention provides a new energy reducer gear tool calibration system that deeply integrates sensing technology, physical modeling, machine learning, and edge computing. Essentially, it transforms the traditional experience-driven "trial-and-error" machining model into a data-driven "prediction-control" paradigm. The system not only solves the current industry challenge of precision consistency but also establishes a sustainable, self-evolving capability system, laying a crucial technological foundation for achieving fully autonomous, unmanned precision gear production lines in the future.

[0042] This solution establishes a two-way closed-loop feedback mechanism from the physical cutting process to the digital model, achieving full-process observability and controllability of tool status. Unlike traditional open-loop verification methods, this system can complete a full "sensing-inversion-analysis-compensation" cycle in each machining cycle, enabling tooth profile errors to be identified and corrected at the nascent stage, significantly improving the dimensional consistency and surface integrity of batch products. This solution is the first to use cutting force frequency domain fingerprinting for the classification and identification of tool failure modes, breaking through the limitations of relying solely on a single threshold alarm, and shifting maintenance decisions from passive response to proactive prediction. Based on a multi-source data fusion-based remaining life prediction model, which comprehensively considers the interactive effects of mechanical load, thermal effects, and material fatigue, the prediction error rate is reduced by at least 40% compared to traditional methods, significantly reducing unplanned downtime and resource waste caused by excessive replacement. This solution adopts a modular and service-oriented software architecture, possessing good scalability and compatibility, and can be adapted to various brands and models of gear machining centers, providing generalized technical support for the intelligent manufacturing of core components of new energy reducers.

Claims

1. A new energy reducer gear cutting tool verification system, characterized in that, include: The tool parameter initialization module is used to input the tool's design geometric parameters, material properties, and initial cutting edge contour data, and to establish a standard cutting dynamics model. The online monitoring module is used to collect data on triaxial cutting force, torque, vibration acceleration and local temperature field during the cutting process in real time through a multi-channel sensor array deployed on the machine tool to obtain a multi-dimensional physical signal set; The three-dimensional tooth profile inversion module is used to reconstruct the actual equivalent cutting edge shape of the current tool based on the cutting force time series data in the multi-dimensional physical signal set, combined with the preset finite element cutting simulation kernel, and calculate its normal deviation surface relative to the ideal tooth profile. An error source analysis module is used to decompose the normal deviation surface into systematic offset components and random fluctuation components. The error source analysis module includes: a wavelet packet decomposition unit, used to divide the cutting force time-series data into frequency bands to obtain a tool state fingerprint spectrum composed of the energy proportions of multiple sub-frequency band signals; and a component separation unit, used to separate the systematic offset components attributed to static factors and the random fluctuation components attributed to transient disturbances based on the spatial distribution characteristics of the tool state fingerprint spectrum and the normal deviation surface. The compensation strategy generation module is used to generate feedforward compensation instructions and adaptive damping control instructions based on the systematic offset component and the random fluctuation component, respectively. The tool health assessment module is used to predict the remaining service life of the tool by combining the surface fatigue damage factor and the growth rate of the cutting edge blunt circle radius as degradation indicators based on the long-term accumulated cutting force characteristic parameter sequence and using an improved Weibull distribution model. Based on long-term observation data, the tool health assessment module uses an improved two-parameter Weibull distribution to model the tool reliability function: in, For a moment The reliability of the tool is the probability that it has not failed. , where is the scale parameter, representing the feature lifetime; The shape parameter reflects the trend of failure rate change. <1 indicates early failure. =1 indicates random failure. A value greater than 1 indicates wear and tear failure. Traditional Weibull models rely solely on time or processing quantity. Two degradation covariates are introduced to enhance predictive ability: surface fatigue damage factor. With the growth rate of the blunt radius of the cutting edge , Calculated from the cumulative plastic strain energy density, the formula is as follows: ,in For the first stress of individual elements For its plastic strain, For volume elements, The most recent 10 linear regressions were used to... The measured values ​​are used to construct an accelerated degradation model of the system, which will determine the effective working time. Defined as: in, =0.35, =0.62, which is an empirical coefficient calibrated through numerous accelerated life tests. The final reliability function is updated as follows: .

2. The new energy reducer gear tool verification system according to claim 1, characterized in that, The multi-channel sensor array includes a triaxial piezoelectric force gauge, a torque sensor, a miniature accelerometer, and an infrared thermal imaging probe.

3. The new energy reducer gear tool verification system according to claim 1, characterized in that, The finite element cutting simulation kernel is pre-set with material removal models for three processing environments: dry cutting, wet cutting, and micro-lubrication. It can also automatically call up the corresponding friction coefficient and thermal conductivity parameters according to the material grade of the gear being processed.

4. The new energy reducer gear tool verification system according to claim 1, characterized in that, The wavelet packet decomposition unit divides the cutting force signal into 8 sub-bands, corresponding to slow tool wear, loose clamping, initial microcracks, chatter, chipping, built-up edge formation, subsurface damage, and noise isolation, respectively.

5. The new energy reducer gear cutting tool verification system according to claim 1, characterized in that, The feedforward compensation command is used to modify the differential gear ratio and axial feed coefficient of the CNC system, and the adaptive damping control command is used to adjust the spindle speed and feed rate to suppress resonance.

6. The new energy reducer gear tool verification system according to claim 1, characterized in that, The compensation strategy generation module is also used to lock the machining process and start the tool change program when the feedforward compensation amount exceeds the set threshold multiple times in a row.

7. The new energy reducer gear tool verification system according to claim 1, characterized in that, The degradation index calculation cycle of the tool health assessment module is a certain number of machining cycles. Its prediction results are presented in the form of a probability density function, and an early warning is issued when the lower limit of the tool life is less than the machining time of the next planned batch within a certain confidence interval.

8. The new energy reducer gear tool verification system according to claim 1, characterized in that, It also includes a data buffer, which is set between the three-dimensional tooth profile inversion module and the error tracing analysis module, to temporarily store the inversion results of the most recent N processing cycles, and supports sliding window statistical analysis to identify periodic wear patterns.

9. The new energy reducer gear tool verification system according to claim 1, characterized in that, The tool health assessment module integrates a digital twin interface, which is used to encrypt and upload local status data to the cloud, and receive updated prior knowledge of degradation patterns verified by digital signature.

10. The new energy reducer gear tool verification system according to claim 1, characterized in that, The entire system is deployed in a distributed edge computing architecture in the form of containerized microservices. Modules communicate with each other through a publish-subscribe message bus, and key decision instructions are issued to the CNC system after being verified by dual redundant channels.

Images