Real-time dynamic compensation system for flat-end drill machining error

Abstract

The present application relates to the field of mechanical processing and intelligent control technology, in particular to a real-time dynamic compensation system for flat-end drill processing error, comprising: a signal acquisition module for acquiring dynamic feedback signals of a servo actuator; a control entropy quantization module for extracting high-frequency jitter characteristics and calculating system control entropy; a risk prediction module for predicting the failure risk probability of triggering the bottom layer derating protection based on the system control entropy; an adaptive decision module for generating control strategy instructions; an instruction execution module for adjusting control weight parameters and executing dynamic compensation operations or damping defense operations; the present application integrates processing error control and servo actuator health status into regulation and control, avoiding resonance cascade risk.

Description

Technical Field

[0001] This invention relates to the fields of machining and intelligent control technology, specifically a real-time dynamic compensation system for machining errors of flat-top drill bits. Background Technology

[0002] With the development of precision machining and CNC manufacturing technology for high-hardness materials, flat-top drills are increasingly widely used in the machining of superhard alloy blanks. In order to ensure the machining accuracy, feed stability and continuous operation capability of the production line, error control and dynamic adjustment in the servo feed process have become particularly important. During the machining of flat-top drill bits, servo actuators typically operate under conditions of high load, high feed, and long-term continuous operation. Factors such as hard agglomerates inside the blank, transient impacts at the cutting interface, electromagnetic noise interference, and feedback link transmission lag can easily lead to distortion of dynamic feedback signals, further causing problems such as increased high-frequency compensation actions, accumulated driver thermal load, and exacerbated mechanical response lag. Existing control methods mostly focus on the rapid tracking of machining errors themselves, often lacking a coordinated consideration of the health status of the servo actuator, thermal fatigue trends, and resonance instability risks. Therefore, when complex disturbances occur continuously, overcompensation, protection lag, or even triggering of underlying derating protection can easily occur, making it difficult to balance machining accuracy and system stability. Therefore, effectively acquiring and processing the dynamic feedback information of the servo actuator, and making a comprehensive judgment based on changes in machining error, the activity level of control behavior, and the operational risks of the actuator, to generate compensation or defense control strategies adapted to the current working conditions, in order to reduce the probability of instability in high-risk cutting scenarios and maintain machining continuity, is crucial to ensuring the machining quality of flat-top drill bits and the safe and reliable operation of the servo system. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time dynamic compensation system for machining errors of flat-top drill bits, and to solve the following technical problems: To avoid the servo drive accumulating heat load due to frequent forward and reverse adjustments, triggering derating protection, or even inducing spindle resonance cascade, machining error control is combined with the health status management of the servo actuator or equipment reliability maintenance. This enables the system to implement hierarchical adaptive control based on threshold constraints between error tracking and stability maintenance, according to its own health status and instability trend.

[0004] The objective of this invention can be achieved through the following technical solutions: A real-time dynamic compensation system for machining errors of flat-top drill bits, applied to the servo actuator that drives the feed of the flat-top drill bit, includes: The signal acquisition module is used to acquire the dynamic feedback signal of the servo actuator. The dynamic feedback signal includes the micro-cutting force change signal caused by the cutting of hard agglomerate blank by the flat-top drill bit. The control entropy quantization module is used to input the dynamic feedback signal into a preset real-time control theory model, extract the high-frequency jitter features in the dynamic feedback signal, and calculate the system control entropy of the servo actuator based on the high-frequency jitter features. The risk prediction module is used to predict the failure probability of the servo actuator triggering the underlying derating protection based on the system control entropy. The underlying derating protection is the mechanical hysteresis response state of the servo actuator triggered by the overheating of the servo driver. The adaptive decision-making module is used to compare the system control entropy with the preset control entropy danger threshold, and generate control strategy instructions based on the comparison results and failure risk probability. The instruction execution module is used to adjust the control weight parameters of the servo actuator in response to control strategy instructions, and to perform dynamic compensation operations or damping defense operations.

[0005] Furthermore, the signal acquisition module acquires the dynamic feedback signals from the servo actuator, specifically including: Acquire the initial operating signals of the servo actuator; Obtain the transmission delay parameters and electromagnetic noise parameters of the initial running signal; The initial running signal is phase-compensated based on the transmission delay parameter to obtain a phase-aligned signal. The phase-aligned signal is frequency-domain filtered based on electromagnetic noise parameters to generate a dynamic feedback signal.

[0006] Furthermore, the control entropy quantization module calculates the system control entropy of the servo actuator based on high-frequency jitter characteristics, specifically including: Get the preset time sliding window; Within a time sliding window, frequency statistics are performed on high-frequency jitter features to obtain the high-frequency jitter rate of the signal; Obtain real-time temperature data corresponding to the servo actuator; If the real-time temperature data is greater than the preset temperature safety threshold, the thermal fatigue accumulation degree is quantified based on the magnitude and duration of the real-time temperature data exceeding the temperature safety threshold; if the real-time temperature data is less than or equal to the temperature safety threshold, the thermal fatigue accumulation degree is determined to be zero. The system control entropy is obtained based on the high-frequency jitter rate of the signal and the cumulative thermal fatigue.

[0007] Furthermore, the risk prediction module predicts the probability of failure that the servo actuator will trigger the underlying derating protection, specifically including: Construct the state-space equations for the servo actuator; The system control entropy and dynamic feedback signal are input into the state space equation to map the current system poles of the servo actuator. Obtain the preset resonance cascade pole boundaries; Calculate the phase space distance between the current system poles and the boundary of the resonant cascade poles; The failure risk probability is obtained based on the phase space distance and the preset limit tolerance distance.

[0008] Furthermore, the adaptive decision-making module generates control strategy instructions based on the comparison results and failure risk probability, specifically including: If the system control entropy is greater than or equal to the control entropy danger threshold, then the failure risk probability is used as the control gain adjustment parameter to generate a dimension-reduced damping control command, and the dimension-reduced damping control command is used as the control strategy command. If the system control entropy is less than the control entropy danger threshold, the failure risk probability is used as the compensation step size adjustment parameter to generate a high-frequency tracking compensation command, and the high-frequency tracking compensation command is used as the control strategy command.

[0009] Furthermore, the instruction execution module performs damping defense operations, specifically including: In response to the dimension-reduced damping control command, obtain the current command bandwidth of the servo actuator; The current command bandwidth is multiplied by the preset damping attenuation coefficient to perform frequency reduction processing, thereby obtaining the target low-frequency bandwidth; A preset virtual mechanical dead zone parameter is introduced into the control loop corresponding to the servo actuator. The virtual mechanical dead zone parameter is the absolute value threshold of the deviation between the command control position and the feedback actual position. Under the constraints of the target low-frequency bandwidth and virtual mechanical dead zone parameters, the servo actuator is subjected to drive operations that limit the maximum output torque and reduce acceleration, thus completing the damping defense operation.

[0010] Furthermore, the instruction execution module performs dynamic compensation operations, specifically including: In response to the high-frequency tracking compensation command, the dynamic feedback signal is used as the state space input and input to the preset deep reinforcement learning network model; Under the constraint of reducing processing error and system control entropy as reward functions, the corresponding compensation amount is output by the policy network in the deep reinforcement learning network model as a dynamic correction instruction. The dynamic correction command is sent to the servo actuator, which then performs the dynamic compensation operation.

[0011] The beneficial effects of this invention are: 1. This invention effectively solves the problem that when tracking only instantaneous processing errors during the processing of flat-top drill bits, the accumulation of driver heat load, mechanical hysteresis amplification, and increased risk of resonance cascade are easily caused by material hard point impact, feedback link delay, electromagnetic noise, and excessive high-frequency compensation. 2. This invention obtains transmission delay parameters and electromagnetic noise parameters from the initial operating signal, and performs phase compensation processing and frequency domain filtering processing in sequence. This ensures that the dynamic feedback signal on which subsequent control judgment is based has time consistency and clear physical meaning, and avoids the controller misjudging transmission misalignment and field interference as real cutting disturbances. 3. This invention performs frequency statistics on high-frequency jitter characteristics within a time sliding window and combines the continuous integral results of real-time temperature data after exceeding the safety threshold to form the system control entropy. This enables the control system to simultaneously characterize the activity level of compensation behavior and the thermal fatigue evolution trend of the actuator, thereby transforming the risk of overcompensation from empirical judgment into a quantifiable state indicator. 4. This invention constructs the state-space equation of the servo actuator and maps the system control entropy and dynamic feedback signal together to the current system poles. Then, it calculates the failure risk probability by combining the boundary of the resonant cascade poles and the limit tolerance distance. This enables the system to identify the potential risks of the servo driver triggering derating protection and mechanical hysteresis response amplification before the machining error has deteriorated significantly. 5. This invention adaptively switches between high-frequency tracking compensation commands and dimension-reducing damping control commands based on the comparison results of system control entropy and danger threshold, and uses the failure risk probability as the compensation step size adjustment parameter or control gain adjustment parameter respectively, ensuring that the system can achieve a bounded and hierarchical dynamic trade-off between maintaining processing accuracy and protecting overall stability. 6. This invention improves the compensation adaptability under complex nonlinear cutting disturbances by implementing bandwidth reduction, introducing virtual mechanical dead zone, and limiting maximum output torque and acceleration in the damping defense operation, and by using a deep reinforcement learning network model to output dynamic correction instructions that take into account both machining error and system control entropy constraints in the dynamic compensation operation. This also suppresses ineffective high-frequency energy injection and hardware life consumption. Attached Figure Description

[0012] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the module of the real-time dynamic compensation system for the machining error of the flat-top drill bit provided in the embodiments of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Please see Figure 1 A real-time dynamic compensation system for machining errors of flat-top drill bits is applied to the servo actuator that drives the feed of the flat-top drill bit. It includes: a signal acquisition module for acquiring the dynamic feedback signal of the servo actuator, which includes the micro-cutting force mutation signal caused by the cutting of hard agglomerate blanks by the flat-top drill bit; The control entropy quantization module is used to input the dynamic feedback signal into a preset real-time control theory model, extract the high-frequency jitter features in the dynamic feedback signal, and calculate the system control entropy of the servo actuator based on the high-frequency jitter features. The risk prediction module is used to predict the failure probability of the servo actuator triggering the underlying derating protection based on the system control entropy. The underlying derating protection is the mechanical hysteresis response state of the servo actuator triggered by the overheating of the servo driver. The adaptive decision-making module compares the system control entropy with a preset control entropy danger threshold and generates control strategy instructions based on the comparison results and failure risk probability. The instruction execution module responds to the control strategy instructions, adjusts the control weight parameters of the servo actuator, and performs dynamic compensation or damping defense operations.

[0015] This embodiment provides a real-time dynamic compensation mechanism for the machining error of flat-top drill bits; specifically, this mechanism is deployed on a fully automated flexible production line in a deep-earth drilling equipment manufacturing base, and serves the servo feed system of CNC machine tools. In this production line, the machine tools operate under continuous production conditions with high load and high feed for a long time, and the spindle is already close to the chatter boundary under normal conditions; When a blank containing high-hardness tungsten carbide agglomerates is mixed in the upstream material, random and short-term sudden changes in cutting force will occur at the drilling interface. If the control system tracks based solely on instantaneous errors, the servo drive is prone to accumulating thermal load due to frequent forward and reverse adjustments, which can further trigger derating protection and cause feed response lag. This lag, combined with the original vibration of the machine tool body, may induce spindle resonance cascade. Therefore, this embodiment does not simply pursue the minimum error, but takes the physical loss and heat accumulation state of the control system itself as one of the decision objects and incorporates them into the control along with the processing error; Specifically, the signal acquisition module collects dynamic feedback signals from the servo actuator and its associated sensing unit; the dynamic feedback signals here include not only conventional quantities such as position, speed, current, vibration and acoustic emission, but also high-frequency cutting disturbance traces formed at the moment of material hard point cutting. The control entropy quantization module feeds the acquired dynamic feedback signal into the real-time control model to extract high-frequency jitter features. This high-frequency jitter characteristic physically reflects whether the servo command is in a state of frequent reversal due to being pulled by small disturbances; once this state continues, it means that the system is no longer just compensating for processing errors, but is aggravating the physical wear and thermal degradation of the actuator to combat unsustainable transient shocks. Based on this, the module further forms the system control entropy as a representation quantity to comprehensively reflect the command jitter trend and the thermal load evolution trend; The risk prediction module maps the system control entropy and real-time dynamic feedback to the stability boundary of the servo system and outputs the failure risk probability. The failure risk probability here does not mean that the tool has been damaged, but rather represents the possibility that the servo drive will enter overheating derating and cause mechanical hysteresis response when the current control state continues to be maintained. The adaptive decision-making module then determines whether the system should continue to perform active compensation or actively switch to defensive damping control based on the position of the system control entropy relative to the danger threshold. Based on this decision, the instruction execution module adjusts the control weight parameters and switches between dynamic compensation and damping defense operations to prioritize moving the system away from the resonance cascade boundary. As an anomaly handling mechanism, if there is a missing key channel in the dynamic feedback signal, such as temporary disconnection of temperature sampling, loss of vibration sampling package, or abnormal feedback of servo encoder, the system will not directly enter the high-frequency compensation mode. Instead, it will first switch the control weight to a conservative configuration, retain only the basic position closed loop and low bandwidth speed closed loop operation, and suspend the high-frequency compensation action that relies on incomplete data. If the sampling returns to normal, the system control entropy quantization and risk prediction will be re-enabled. If the anomaly persists for more than the preset duration, a maintenance alarm will be output and the damping defense mode will be maintained to avoid erroneous compensation in the event of incomplete information. When a batch of blanks is continuously processed on the production line, the cutting state of the first thirty blanks is basically stable, the system control entropy is maintained within a safe range, and the system adopts dynamic compensation operation to maintain feed efficiency. The newly replaced batch of blanks contained unevenly distributed high-hardness agglomerates. After the drill bit cut in, dense short pulse jumps appeared simultaneously in the acoustic emission sensor and servo current signal. The system identified that these jumps were not caused by changes in macroscopic cutting parameters, but were closer to microscopic impacts caused by hard points inside the material. At this point, the control entropy quantization module determines that high-frequency jitter has begun to accumulate significantly, and the risk prediction module simultaneously provides an increased probability of failure risk. The adaptive decision module reduces the tracking intensity of transient errors accordingly and switches to damping defense operation when necessary, thereby avoiding the servo drive from triggering derating protection due to continuous temperature rise caused by high-frequency regulation. The purpose of this step / mechanism is to combine machining error control with the health status management of servo actuators or equipment reliability maintenance, so that the system has the ability to actively yield in high-risk cutting scenarios and does not cross the resonance cascade boundary in pursuit of local instantaneous compensation.

[0016] In a preferred embodiment of the present invention, the signal acquisition module acquires the dynamic feedback signal of the servo actuator, specifically for: acquiring the initial operating signal of the servo actuator; acquiring the transmission delay parameter and electromagnetic noise parameter of the initial operating signal; performing phase compensation processing on the initial operating signal based on the transmission delay parameter to acquire a phase alignment signal; and performing frequency domain filtering processing on the phase alignment signal based on the electromagnetic noise parameter to generate a dynamic feedback signal.

[0017] This embodiment provides a mechanism for acquiring dynamic feedback signal preprocessing; specifically, in the aforementioned production line scenario, simply acquiring the raw signals of the servo system is insufficient to support stable control. The reason is that there are strong electromagnetic interference sources on site, and the wiring distance between the machine tool working area and the control cabinet exceeds the preset signal lossless transmission threshold. In addition, the high temperature gradient environment makes it easy for vibration, acoustic emission and drive feedback signals to be delayed and noise coupled. If these problems are not addressed first, the subsequent control module will incorrectly identify the phase delay error caused by the lag as real vibration, resulting in an incorrect compensation direction. Specifically, the signal acquisition module collects the initial operating signal; this initial operating signal may come from the position feedback, speed feedback, driver current feedback, spindle proximity vibration sensor, acoustic emission sensor, and driver housing temperature sensor, etc.; the module acquires the corresponding transmission delay parameters and electromagnetic noise parameters; The transmission delay parameter is used to describe the time misalignment between various feedback quantities from the physical point of origin to the controller input. For example, when the vibration sensor is located near the spindle and the controller is located in a remote cabinet, its sampling packet often arrives at the controller later than the data from the servo internal encoder. Electromagnetic noise parameters are used to characterize the frequency band pollution features introduced by field frequency converters, high-power spindle drives, and switching power supplies; based on transmission delay parameters, the system performs phase compensation processing on the initial operating signal to make signals from different physical locations and different transmission links as consistent as possible in terms of time reference. After phase compensation, the timing of the sudden changes in cutting force in multiple channels is no longer misaligned, making it easier to determine whether these disturbances come from actual cutting events or from occasional interference in a single channel. Then, based on the electromagnetic noise parameters, frequency domain filtering is applied to the phase alignment signal to filter out non-cutting components related to field power supply, drive switching frequency or communication noise, and finally generate a dynamic feedback signal for control judgment. In the process of multi-channel signal alignment: assuming that a material hard spot cutting event occurs at the same time, the encoder feedback signal is recorded as S1, the vibration signal is recorded as S2, and the acoustic emission signal is recorded as S3. If S1 arrives almost instantly when the controller receives the data, while S2 and S3 are relatively delayed, the system first moves S2 and S3 forward along the time axis to a reference time consistent with S1 based on their respective delay parameters, thus obtaining aligned S1', S2', and S3'. Subsequently, the narrowband spikes in S2' and S3' that match the characteristics of strong electromagnetic interference in the field are suppressed, while the broadband response components consistent with the cutting impact are retained; the subsequent modules obtain the consistent response characteristics of the same physical event in multiple channels, rather than disordered asynchronous noise. As an anomaly handling mechanism, if a certain channel cannot stably estimate the transmission delay parameter, such as when the sensor intermittently goes offline or the timestamp drifts abnormally, then that channel will not participate in the joint determination of high-frequency cutting abrupt changes and will only be retained as an auxiliary observation. If the noise parameters change drastically in a short period of time, making the original filtering configuration no longer applicable, the system enters the self-checking branch, performs a short-period background noise scan on the newly sampled segment, and resumes normal output after the noise boundary converges again; if multiple key channels are abnormal at the same time, the dynamic feedback signal is marked as low confidence and the system switches to conservative control. During the aforementioned processing, due to the spindle drive operating under high load for an extended period, strong electromagnetic radiation was generated near the cabinet, and pulse noise of a fixed frequency band was superimposed in the acoustic emission channel. At the same time, the start and stop of the cooling system at the other end of the production line caused a sudden change in the temperature of the working area, which caused the increase in the transmission delay of the remote vibration acquisition link to exceed the preset delay fluctuation range. At this point, the signal acquisition module first aligns the phase of the vibration and acoustic emission channels, then removes the noise components related to the drive switching frequency, and finally separates the real micro-cutting force mutation from the noise background. The purpose of this step / mechanism is to provide a dynamic feedback signal with consistent time and clear physical meaning for subsequent control entropy quantification and risk prediction, so as to prevent the controller from misjudging transmission lag and electromagnetic noise as real cutting disturbances.

[0018] In a preferred embodiment of the present invention, the control entropy quantization module calculates the system control entropy of the servo actuator based on the high-frequency jitter characteristics, specifically used for: obtaining a preset time sliding window; performing frequency statistics on the high-frequency jitter characteristics within the time sliding window to obtain the high-frequency jitter rate of the signal; and obtaining the real-time temperature data corresponding to the servo actuator. If the real-time temperature data is greater than the preset temperature safety threshold, the thermal fatigue accumulation degree is quantified based on the magnitude and duration of the real-time temperature data exceeding the temperature safety threshold; if the real-time temperature data is less than or equal to the temperature safety threshold, the thermal fatigue accumulation degree is determined to be zero; the system control entropy is obtained based on the high-frequency jitter rate of the signal and the thermal fatigue accumulation degree.

[0019] This embodiment provides a system control entropy quantization mechanism; specifically, after the aforementioned signal has completed phase alignment and noise processing, if the magnitude of the processing error is still used as the control basis, the system is prone to fall into a hidden instability state where the processing error converges but the risk of thermal fatigue and structural instability of the actuator is aggravated. Especially when hard spots appear continuously in the material, the servo control loop will repeatedly generate small and high-speed correction actions. The amount of change of the single displacement command of these actions is lower than the preset conventional correction threshold, but they are highly concentrated in time, which will cause the driver and motor windings to be in a state of frequent energy exchange. Traditional temperature rise monitoring can only obtain lagging steady-state parameters, which are difficult to reflect the evolution mechanism of transient heat load in real time. Therefore, this embodiment introduces system control entropy to quantify in advance the trend of disordered control activities, physical losses and deterioration of heat accumulation state. Specifically, the control entropy quantization module first obtains a preset time sliding window; this time sliding window is used to extract the high-frequency flip features of control commands under continuous time series to characterize the stability of the system's dynamic control state. Within this window, the system performs frequency statistics on high-frequency jitter characteristics to obtain the signal high-frequency jitter rate; in engineering, this high-frequency jitter rate corresponds to the activity level of rapid direction changes, rapid jumps, or short-cycle reciprocating corrections in servo commands or feedback within a unit observation period. The module extracts real-time temperature data of the servo actuator; if the real-time temperature data exceeds the temperature safety threshold, it further forms the thermal fatigue accumulation degree; the thermal fatigue accumulation degree here is not a simple instantaneous temperature value, but a combination of the over-temperature amplitude and duration, reflecting whether the power devices, winding insulation and connectors inside the driver are under continuous thermal stress. The high-frequency jitter rate and thermal fatigue accumulation are fused according to preset weights to obtain the system control entropy; the thermal fatigue accumulation is the time integral characteristic of the portion of real-time temperature data that exceeds the temperature safety threshold; the system control entropy is obtained by linearly weighting and fusing the signal high-frequency jitter rate and thermal fatigue accumulation with preset weight coefficients. The resulting system control entropy can reflect both whether there is overly sensitive compensation behavior and whether such compensation behavior has been transformed into a perceptible thermal risk at the hardware level. During the evolution of the dynamic control strategy, two sets of features are extracted within the same sliding window. The first set comes from the instruction repetitive correction behavior, and the second set comes from the temperature evolution behavior. If there are only a few high-frequency corrections in a certain window and the temperature is always below the safe threshold, the system control entropy remains low, indicating that although the system has compensation actions, it is still in a sustainable healthy control zone. If dense high-frequency corrections occur in the next window, and the temperature continuously exceeds the threshold, the high-frequency jitter rate and thermal fatigue accumulation will increase simultaneously, and the system control entropy will increase significantly, indicating that the control activity begins to pose a substantial threat to the health of the actuator. The focus here is not on showing specific calculated values, but on illustrating that control entropy has a dual-channel meaning: one comes from the control behavior itself, and the other comes from the physical state of the actuator hardware. As an anomaly handling mechanism, if the temperature sensor malfunctions briefly but the high-frequency jitter rate increases significantly, the system can first provide an early warning level control entropy based on the high-frequency jitter rate, and at the same time reduce the temperature-related weights to avoid the overall judgment being interrupted due to a single hardware failure. If the high-frequency jitter rate is low but the temperature is high, it is necessary to further check whether there are uncontrollable factors such as deterioration of external heat dissipation conditions, excessively high ambient temperature or blockage of cooling air ducts. In this situation, the system can retain the conservative entropy value and trigger an equipment maintenance alarm instead of directly attributing it to overcompensation; if both types of signals are unreliable, the system will suspend adaptive compensation and revert to the basic control mode. During the batch processing of the aforementioned problematic blanks, the initial few hard point impacts only caused sporadic corrections in the servo system, and the temperature of the driver housing remained stable, so the system control entropy only fluctuated slightly; as the distribution of hard points inside the same blank increased, the servo actuator began to frequently issue forward and reverse compensation actions, and the drive current pulsation became more intense. If the driver temperature continues to be higher than the safety threshold thereafter, it indicates that these high-frequency corrections are no longer short-term, negligible events, but are continuously consuming the hardware's lifespan; the control entropy quantization module marks this operating condition as a high-entropy state accordingly. The purpose of this step / mechanism is to transform the judgment of whether high-frequency compensation is excessive from mere experience into a quantifiable control status indicator, thereby enabling proactive identification of health risks in servo actuators.

[0020] In a preferred embodiment of the present invention, the risk prediction module predicts the failure risk probability of the servo actuator triggering the underlying derating protection, specifically used for: constructing a state space equation for the servo actuator; inputting the system control entropy and dynamic feedback signal into the state space equation to map the current system poles of the servo actuator; Obtain the preset resonant cascade pole boundary; calculate the phase space distance between the current system pole and the resonant cascade pole boundary; obtain the failure risk probability based on the phase space distance and the preset limit tolerance distance.

[0021] This embodiment provides a failure risk prediction mechanism; specifically, system control entropy alone is insufficient to determine whether a dangerous boundary will actually be crossed; because in some cases, although the actuator may have a thermal load temperature rise exceeding the steady-state set value, the overall dynamics of the machine still have sufficient stability margin. In other cases, even if the temperature has not risen significantly, if the difference between the coupling frequency of the current vibration mode and the structure's natural mode is less than the preset safety band margin, it may quickly evolve into a resonance cascade. Therefore, this embodiment further constructs the state space equation of the servo actuator to unify the increase in control entropy and the degradation of machine tool dynamic stability into the same risk representation framework; Specifically, the state-space equation is used to describe the position, velocity, acceleration, drive response, mechanical hysteresis, and vibration energy transfer relationship of the servo feed system during the machining process; The state variables here can be understood as a comprehensive expression of the current dynamic attitude of the system, rather than looking at the value of a single sensor in isolation; After the system inputs the dynamic feedback signal and the system control entropy into the equation, the current system poles can be mapped. In practice, the pole mapping here, that is, finding the poles, is essentially an online identification and feature solving process. The system uses dynamic feedback signals as observation inputs and system control entropy as time-varying environmental parameters characterizing equipment thermal degradation. It dynamically updates the system matrix, i.e. the A matrix, in the state space equation through online identification algorithms such as recursive least squares method. Solve for the eigenvalues ​​of the updated system matrix. The distribution of these eigenvalues ​​in the complex plane is then mapped to the current system poles. The current system poles reflect the dynamic convergence or divergence trend formed by the control loop and mechanical structure at this time: if the poles are far from the danger boundary, it means that the system still has sufficient damping and recovery ability even if it is subjected to material hard point impact. If the pole approaches the preset boundary of the resonant cascade pole, it indicates that the phase margin, damping margin or equivalent recovery capability of the system has decreased significantly. The system calculates the phase space distance between the current system pole and the boundary of the resonant cascade pole, and combines it with the limit tolerance distance to form a normalized risk quantity, thus obtaining the failure risk probability between zero and one. In the specific conversion calculation of distance and probability, this normalization process has a clear reverse mapping logic: when the phase space distance is greater than or equal to the limit tolerance distance, the failure risk probability is reduced to zero. When the phase space distance is less than the limit tolerance distance, the failure risk probability is calculated by subtracting the phase space distance from the limit tolerance distance and then dividing by the limit tolerance distance. The closer this probability is to one, the more likely the current operating condition is to induce servo drive derating, mechanical hysteresis amplification, and resonance cascading. In the process of dynamic control strategy evolution; assuming that the state space mapping result forms poles P1 and P2 in two consecutive time periods, pole P1 is located inside the safe zone and the phase space distance between pole P1 and the boundary of the resonant cascade pole is greater than the preset safety isolation constant, while pole P2 is significantly closer to the preset boundary B; if P1 corresponds to a low control entropy and low vibration coupling condition, then the output risk probability is low. If P2 corresponds to a working condition with high control entropy and continuous high-frequency cutting impact, then its distance from the boundary B will shorten, and the output risk probability will increase accordingly. If in the next period, servo response slows down, vibration peak expands, and temperature rise continues, then the pole will get closer to the boundary. The system interprets this trend as approaching the underlying derating protection trigger zone, thus providing a basis for subsequent decision modules. As an anomaly handling mechanism, if some feedback quantities used for state space mapping are temporarily missing, such as a decrease in the reliability of the vibration channel, the system can adopt a reduced-order state model, retaining only core states such as position, velocity, current, and temperature for risk estimation, and correspondingly increasing the conservatism of the risk output. If the calculated current system poles change abnormally in a short time and are inconsistent with the physical conditions, such as isolated points caused by sensor spikes, the system can require multiple consecutive sampling windows to maintain a consistent trend before confirming high risk, in order to avoid a one-time false alarm causing frequent degradation of the production line. If the boundary parameters change due to machine tool maintenance, the resonant cascade pole boundaries and limit tolerance distances are allowed to be recalibrated to ensure that the risk probability remains consistent with the actual equipment condition. During the aforementioned abnormal batch processing, after a certain blank entered the middle section of cutting, although the processing error was still controlled within an acceptable range, the system control entropy continued to rise, and the vibration feedback showed that the structural response began to concentrate on the spindle sensitive mode. After mapping these two types of information to the state space, the risk prediction module finds that the current system poles have moved from the center of the stable region to the boundary of the resonance cascade, thus outputting a high probability of failure risk. This result indicates that the real danger at this time is not the error that has already occurred, but that if the system continues to operate according to the original active compensation strategy, it is very likely to enter the irreversible instability region due to driver overheating and phase misalignment. The purpose of this step / mechanism is to translate the health deterioration trend of control behavior into a predictable risk of overall machine instability boundary, thereby enabling the control system to have the ability to intervene in advance towards the failure boundary.

[0022] In a preferred embodiment of the present invention, the adaptive decision module generates control strategy instructions based on the comparison results and failure risk probability, specifically: if the system control entropy is greater than or equal to the control entropy danger threshold, then the failure risk probability is used as the control gain adjustment parameter to generate a dimension-reduced damping control instruction, and the dimension-reduced damping control instruction is used as the control strategy instruction. If the system control entropy is less than the control entropy danger threshold, the failure risk probability is used as the compensation step size adjustment parameter to generate a high-frequency tracking compensation command, and the high-frequency tracking compensation command is used as the control strategy command.

[0023] This embodiment provides an adaptive decision-making mechanism; specifically, based on the above, even if high control entropy and high failure risk can be identified, if there is no clear strategy switching principle, the system may still fall into two types of imbalance: one is being too conservative, triggering a global frequency reduction defense mechanism when receiving a small disturbance signal below the preset amplitude threshold, resulting in limited processing cycle and degraded feed continuity. Another type is overly aggressive, continuing to perform high-frequency tracking even when the system is close to the depreciation boundary, ultimately amplifying the risk; therefore, this embodiment establishes a dynamic trade-off mechanism between tracking accuracy and system robustness by comparing the system control entropy with the danger threshold and embedding the failure risk probability into the strategy generation process. Specifically, when the system control entropy is less than the danger threshold, it indicates that although the current compensation behavior has a certain level of activity, it has not yet significantly damaged the health of the implementing agency. At this point, the adaptive decision-making module prioritizes maintaining processing efficiency, uses the failure risk probability as the compensation step size adjustment parameter, and generates high-frequency tracking compensation commands. The compensation step size here is not infinitely amplified, but rather converges adaptively with the risk probability: the lower the risk, the larger the step size is allowed to track micro-errors. When the risk increases but has not yet exceeded the threshold, the compensation step size is automatically reduced to make the tracking smoother and avoid excessive control overshoot near the safety boundary; conversely, when the system control entropy is greater than or equal to the danger threshold, it indicates that the system has entered a state of impaired health from an active compensation state. At this point, even if the processing error has not yet increased, we no longer pursue a complete correction of all transient deviations. Instead, we use the failure risk probability as a control gain adjustment parameter to generate a dimension-reduced damping control command. Dimensionality reduction here refers to reducing the degrees of freedom for tracking high-frequency details, and narrowing the control objectives to core tasks such as maintaining overall stability, limiting energy injection, and preventing resonant expansion. In the process of dynamic control strategy evolution; assuming that the system control entropy is below the threshold and the risk probability is low during a certain period, the generated strategy is high-frequency tracking compensation, and the compensation step size is a relatively aggressive configuration. If the system control entropy rises to near the threshold but does not cross the line in the next period due to the dense occurrence of hard points, the step size will automatically decrease, indicating that the system begins to restrain high-frequency correction. If the control entropy crosses the danger threshold in the next time period, the strategy will no longer output the tracking step size, but instead output the damping control gain. The control logic will switch from trying to fit the error as closely as possible to prioritizing the suppression of the unstable trend. This will enable a continuous transition of the control state rather than discrete step cutoff. As an anomaly handling mechanism, if the system control entropy is near the threshold and the risk probability fluctuates frequently, in order to avoid the strategy from switching between the two modes at high frequency, a hysteresis zone or a minimum holding time can be set. That is, once the damped control mode is entered, it will not immediately return to the high-frequency tracking mode before the recovery conditions are met. If the risk probability increases but the reliability of the control entropy data is insufficient, the system will prioritize a conservative compensation of intermediate strength rather than directly reducing the dimensionality completely. If the control entropy is low but abnormal heat dissipation of critical hardware is detected, a defensive strategy can be forced to reflect the priority protection of the actual state of the hardware. In the first half of the aforementioned abnormal batch processing, although the system detected short-term cutting fluctuations caused by hard points, the control entropy had not yet reached the dangerous threshold. Therefore, the adaptive decision module still maintained high-frequency tracking compensation, but as the risk probability increased, the compensation step size was gradually tightened. As subsequent hard impacts accumulate, the driver temperature rises and the state space prediction results show that the stability margin continues to decrease, and the control entropy exceeds the danger threshold. At this point, the adaptive decision module immediately abandons further enhancing the compensation sensitivity and instead outputs a dimension-reduced damping control command, causing the system to return from a high-response state to a steady-state defense state. The purpose of this step / mechanism is to establish a feasible control switching criterion, enabling the system to implement hierarchical adaptive control based on threshold constraints between error tracking and stability maintenance, according to its own physical losses, heat accumulation state and instability trend.

[0024] In a preferred embodiment of the present invention, the instruction execution module performs a damping defense operation, specifically for: responding to a dimension reduction damping control instruction, obtaining the current instruction bandwidth of the servo actuator; multiplying the current instruction bandwidth by a preset damping attenuation coefficient to perform frequency reduction processing, and obtaining the target low-frequency bandwidth; A preset virtual mechanical dead zone parameter is introduced into the control loop corresponding to the servo actuator. The virtual mechanical dead zone parameter is the absolute value threshold of the deviation between the command control position and the feedback actual position. When the absolute value of the deviation is less than the threshold, the output command of the shielding system tracking compensation torque is executed. Under the constraints of the target low-frequency bandwidth and the virtual mechanical dead zone parameter, the servo actuator is subjected to drive operations that limit the maximum output torque and reduce acceleration, thus completing the damping defense operation.

[0025] This embodiment provides a damping defense operation mechanism; specifically, in the previous embodiment, the system was already able to determine when to switch from active compensation to defense mode. However, if only an abstract command to reduce compensation is issued at the logical level without a clear execution path, the control system may still continue to inject high-frequency energy into the structure due to excessive bandwidth of the original control loop, excessive output torque, or ineffective utilization of the mechanical dead zone. Therefore, this embodiment implements the defense strategy as three types of coordinated behaviors: bandwidth reduction, introduction of virtual machine dead zones, and limit of drive capabilities. Specifically, after receiving the dimension-reduced damping control command, the command execution module obtains the current command bandwidth of the servo actuator. The current command bandwidth reflects the control loop's ability to respond to rapid error changes. The higher the value, the better the system is at tracking small and rapid disturbances, but it is also more likely to be induced to generate unnecessary high-frequency actions in high-noise environments and under mechanical lag conditions. The system performs frequency reduction processing on the current command bandwidth based on the damping attenuation coefficient to obtain the target low-frequency bandwidth. This target low-frequency bandwidth does not simply reduce the response rate, but sets the control loop to prioritize filtering out high-frequency non-sustained impacts, retain the response gain to low-frequency steady-state evolution, and adopt a strategy of ignoring or weakly responding to short-time, random, non-sustained micro-impacts. The system introduces virtual mechanical dead-zone parameters into the control loop. The virtual mechanical dead zone can be understood as a small range set by humans where high-frequency, small-amplitude reciprocating corrections are not executed immediately. Its business logic is as follows: when the absolute value of the deviation between the command control position and the actual feedback position is less than the virtual mechanical dead zone parameter, the system does not output additional tracking compensation torque commands. The system only applies corrective driving force to the actuator when the deviation exceeds the dead zone boundary; this can significantly reduce the chance of the actuator being repeatedly driven by high-frequency noise. Under the combined constraints of the target low-frequency bandwidth and the virtual mechanical dead zone, the system limits the maximum output torque and acceleration to prevent the servo actuator from impacting the structural modes with high energy during the critical risk stage, thus completing the damping defense operation. During the evolution of the dynamic control strategy, it is assumed that the original control loop can quickly make bidirectional corrections to small disturbances, which is beneficial to improving dynamic following performance during normal processing without hard point disturbances. When entering the high-entropy and high-risk stage, the system changes the loop to a lower bandwidth state and sets a virtual interval for small-amplitude corrections that are not executed immediately. In this way, if scattered micro-hard point impacts occur again, the controller will not issue dense reversal commands continuously as before, but will only mitigate and correct deviations that are truly persistent and sufficient to affect the overall cutting stability. Meanwhile, limiting the maximum output torque and acceleration is equivalent to setting a hard constraint on the driving energy of the actuator, preventing it from entering a high-load operating state due to overcompensation of transient errors; As an anomaly handling mechanism, if the processing error rises too quickly and exceeds the allowable range of the process after entering the damping defense mode, the system can perform graded defense: firstly, the control bandwidth is increased to the preset intermediate threshold; if the risk probability is still higher than the safety threshold, the frequency reduction state is maintained, and the feed speed is reduced or the process cycle is re-arranged. If the virtual mechanical dead zone is set too large, resulting in a significant lack of low-speed tracking capability, the system can dynamically reduce the dead zone according to the continuous error trend to avoid completely losing the necessary correction capability. If the risk probability continues to rise even in defense mode, it means that the current status of the blank or equipment is approaching the uncontrollable boundary. At this time, the system can trigger shutdown protection or manual review process instead of continuing to try online compensation. In the latter half of the aforementioned abnormal batch, the hard points inside a certain blank were abnormally densely distributed, and the system control entropy had exceeded the dangerous threshold. Based on this, the instruction execution module switched the servo feed control loop from high response bandwidth to target low frequency bandwidth, and introduced a virtual mechanical dead zone within the small position correction range. Even if there are still frequent short pulses in the acoustic emission channel, the servo system will no longer make large reverse compensation for each pulse. Instead, it will maintain continuous cutting with limited torque and low acceleration to avoid further heating of the driver and triggering derating protection. The purpose of this step / mechanism is to transform the proactive yielding control concept into explicit execution actions, thereby suppressing high-frequency energy injection and providing the system with recovery space away from the resonant cascade boundary.

[0026] In a preferred embodiment of the present invention, the instruction execution module performs a dynamic compensation operation, specifically for: responding to a high-frequency tracking compensation instruction, inputting a dynamic feedback signal as a state space input to a preset deep reinforcement learning network model; and under the constraint of reducing processing error and system control entropy as the reward function, outputting the corresponding compensation amount as a dynamic correction instruction through the policy network in the deep reinforcement learning network model. The dynamic correction command is sent to the servo actuator, which then performs the dynamic compensation operation.

[0027] This embodiment provides a dynamic compensation operation mechanism; specifically, in the aforementioned scheme, when the system control entropy is lower than the danger threshold, the system still allows high-frequency tracking compensation to be performed; However, if the compensation still uses fixed rules or rigid parameters, it is difficult to cope with the complex and nonlinear cutting force changes in the machining of flat-top drill bits, especially when there is local inhomogeneity in the material structure. Traditional controllers often have to compromise between response lag or overcompensation. Therefore, this embodiment introduces a deep reinforcement learning network model so that dynamic compensation not only takes into account the instantaneous error, but also the long-term control consequences of system robustness under control entropy constraints. Specifically, after receiving the high-frequency tracking compensation instruction, the instruction execution module uses the dynamic feedback signal as the state space input to the deep reinforcement learning network model. The state space input here can comprehensively include position deviation trends, velocity changes, drive current fluctuations, vibration response, acoustic emission disturbances, and temperature-related health signals; It needs to be clarified that, in this application context, using dynamic feedback signals as input to the state space does not refer to the global infinite set of all possible system states in reinforcement learning control theory. Rather, it refers to mapping the aforementioned multi-dimensional dynamic feedback signals, after dimensionality reduction and sequence concatenation, to a specific feature coordinate in the network state space, that is, instantiating it as the input state feature vector of the policy network at the current moment. This structured decomposition ensures that heterogeneous physical signals are aligned on the same vector dimension; the policy network outputs corresponding compensation amounts based on this state information, forming dynamic correction instructions. The reward function consists of a comprehensive penalty term constructed from the processing position error magnitude and the system control entropy according to preset penalty weight coefficients; Unlike compensation methods that only aim to minimize processing error, the reward constraint in this embodiment considers both reducing processing error and suppressing the increase of system control entropy. Therefore, the policy network will autonomously optimize the adaptive control strategy during training and operation. Effective compensation is provided for deviations that can stably improve the processing condition, while the response is weakened for micro-perturbations that can only induce high-frequency oscillations and cannot form an effective convergent output. In the process of dynamic control strategy evolution, it is assumed that a certain state input consists of four types of segments: position deviation trend segment E1, vibration intensity change segment E2, driving current disturbance segment E3, and temperature rise trend segment E4. After reading segments E1 to E4, the policy network outputs a compensation amount U1. If the processing error decreases and the system control entropy does not increase significantly after executing U1, then this type of decision is considered better. If the network output compensation amount U2 in another state can instantly reduce the error, but causes significantly higher high-frequency jitter and temperature rise, then this type of decision will be weakened in subsequent optimization; thus, the network output is not a single error tracking result, but a dynamic compensation strategy that takes into account the physical boundaries of the actuator. As an anomaly handling mechanism, if the current dynamic feedback signal is not reliable enough, or an abnormal state that exceeds the model's empirical boundary occurs, such as simultaneous distortion of multiple channels, interruption of temperature sampling, or abnormally severe mechanical impact, the system will not force the policy network to be called for compensation, but will instead revert to a more conservative rule-based control until the input state is restored to reliability. If the compensation amount output by the policy network exceeds the device's allowable range, it will be safely pruned before being sent to the servo actuator to ensure that the torque, speed, or acceleration limits are not exceeded. If the compensation effect is poor for several consecutive cycles and the control entropy increases instead, the system can reduce the policy network weight and hand it over to the damping defense logic. In the early stage of the aforementioned abnormal batch processing, a blank just began to show scattered hard point impacts. The system control entropy was still below the danger threshold, so the instruction execution module still adopted dynamic compensation operation. After receiving state input consisting of position deviation, servo current fluctuation, vibration response and acoustic emission abrupt change, the deep reinforcement learning network model outputs a set of amplitude-controlled and direction-continuous correction commands, enabling the feed servo to correct machining deviations without significantly increasing high-frequency oscillations. When subsequent similar disturbances become more frequent and begin to increase the system's control entropy, although the model can still provide compensation suggestions, its output magnitude is adaptively limited because the reward constraint includes a control entropy term; once the control entropy further exceeds the danger threshold, the system is then taken over by damping defense logic. The purpose of this step / mechanism is to enable dynamic compensation to adapt to complex nonlinear cutting disturbances, while avoiding one-sided control behavior that only pursues the minimum short-term error but ignores the thermal load and stability boundary of the actuator.

[0028] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A real-time dynamic compensation system for machining errors of flat-top drill bits, applied to a servo actuator that drives the feed of a flat-top drill bit, characterized in that, include: The signal acquisition module is used to acquire the dynamic feedback signal of the servo actuator, the dynamic feedback signal including the micro-cutting force change signal caused by the flat-top drill bit cutting the hard agglomerate blank; The control entropy quantization module is used to input the dynamic feedback signal into a preset real-time control theory model, extract the high-frequency jitter features in the dynamic feedback signal, and calculate the system control entropy of the servo actuator based on the high-frequency jitter features. The risk prediction module is used to predict the failure probability of the servo actuator triggering the underlying derating protection based on the system control entropy. The underlying derating protection is the mechanical hysteresis response state of the servo actuator triggered by the overheating of the servo driver of the servo actuator. An adaptive decision-making module is used to compare the system control entropy with a preset control entropy danger threshold, and generate control strategy instructions based on the comparison result and the failure risk probability. The instruction execution module is used to adjust the control weight parameters of the servo actuator in response to the control strategy instruction, and to perform dynamic compensation operation or damping defense operation. The signal acquisition module acquires the dynamic feedback signal of the servo actuator, specifically including: acquiring the initial operating signal of the servo actuator; acquiring the transmission delay parameter and electromagnetic noise parameter of the initial operating signal; performing phase compensation processing on the initial operating signal based on the transmission delay parameter to acquire a phase alignment signal; and performing frequency domain filtering processing on the phase alignment signal based on the electromagnetic noise parameter to generate the dynamic feedback signal. The control entropy quantization module calculates the system control entropy of the servo actuator based on the high-frequency jitter characteristics, specifically including: obtaining a preset time sliding window; performing frequency statistics on the high-frequency jitter characteristics within the time sliding window to obtain the signal high-frequency jitter rate; obtaining real-time temperature data corresponding to the servo actuator; if the real-time temperature data is greater than a preset temperature safety threshold, then quantifying and obtaining the thermal fatigue accumulation degree based on the magnitude and duration of the real-time temperature data exceeding the temperature safety threshold; if the real-time temperature data is less than or equal to the temperature safety threshold, then determining that the thermal fatigue accumulation degree is zero; and obtaining the system control entropy based on the signal high-frequency jitter rate and the thermal fatigue accumulation degree. The risk prediction module predicts the failure risk probability of the servo actuator triggering the underlying derating protection, specifically including: constructing a state space equation for the servo actuator; inputting the system control entropy and the dynamic feedback signal into the state space equation to map the current system pole of the servo actuator; obtaining a preset resonant cascade pole boundary; calculating the phase space distance between the current system pole and the resonant cascade pole boundary; and obtaining the failure risk probability based on the phase space distance and a preset limit tolerance distance. The adaptive decision-making module generates control strategy instructions based on the comparison results and the failure risk probability. Specifically, it includes: if the system control entropy is greater than or equal to the control entropy danger threshold, then the failure risk probability is used as a control gain adjustment parameter to generate a dimension-reduced damping control instruction, and the dimension-reduced damping control instruction is used as the control strategy instruction; if the system control entropy is less than the control entropy danger threshold, then the failure risk probability is used as a compensation step size adjustment parameter to generate a high-frequency tracking compensation instruction, and the high-frequency tracking compensation instruction is used as the control strategy instruction.

2. The real-time dynamic compensation system for machining errors of flat-top drill bits according to claim 1, characterized in that, The instruction execution module performs damping defense operations, specifically including: In response to the dimension reduction damping control command, the current command bandwidth of the servo actuator is obtained; The current command bandwidth is multiplied by a preset damping attenuation coefficient to perform frequency reduction processing, thereby obtaining the target low-frequency bandwidth; A preset virtual mechanical dead zone parameter is introduced into the control loop corresponding to the servo actuator. The virtual mechanical dead zone parameter is the absolute value threshold of the deviation between the command control position and the feedback actual position. Under the constraints of the target low-frequency bandwidth and the virtual mechanical dead zone parameters, the servo actuator performs drive operations to limit the maximum output torque and reduce acceleration, thereby completing the damping defense operation.

3. The real-time dynamic compensation system for machining errors of flat-top drill bits according to claim 1, characterized in that, The instruction execution module performs dynamic compensation operations, specifically including: In response to the high-frequency tracking compensation command, the dynamic feedback signal is used as the state space input and input to the preset deep reinforcement learning network model; Under the constraint of reducing processing error and the system control entropy as reward functions, the corresponding compensation amount is output by the policy network in the deep reinforcement learning network model as a dynamic correction instruction. The dynamic correction command is sent to the servo actuator, which then performs the dynamic compensation operation.

Images