A shaver linear motor adaptive control method and system

Abstract

The application discloses a shaving razor linear motor adaptive control method and system, collects control input data and response data, analyzes envelope attenuation end section associated characteristics, and outputs early warning identification of drift; overshoot phase difference detection is carried out on input and output data, combined with matching of sub-working condition spectrum library to establish error checking matrix; gain rule groups are formed by implementing early phase section priority gain distribution on the error checking matrix according to the early warning identification of drift, and abnormal characteristic points are positioned by multi-working condition phase time difference analysis and deviation overrun judgment; an excitation sequence is obtained by activating the gain rule groups, an amplitude stability index is output by pseudo-convergent oscillation detection and cross-load reverse correlation strength evaluation; a control characteristic atlas is constructed by combining the amplitude stability index and excitation sequence deviation interval characteristics, adaptive control parameters are output by parameter deviation mapping to complete linear motor control parameter configuration, and adaptive configuration of control parameters of the linear motor under variable working conditions is realized.

Description

Technical Field

[0001] This invention relates to the field of automation control technology, and in particular to an adaptive control method and system for a linear motor of an electric shaver. Background Technology

[0002] Linear motors in shavers face complex and variable load environments during actual use. The contact pressure, beard density, and skin curvature of the shaving head continuously change when shaving in different facial areas, causing dynamic deviations in the control input and vibration response. Current mainstream control strategies primarily rely on fixed parameter configurations, making it impossible to perceive phase shifts and amplitude fluctuations between the control input signal and the actual response in real time. This is especially true during rapid switching between operating conditions, where the tracking error between motor amplitude attenuation and control parameters is difficult to correct automatically using preset parameters. Furthermore, the characteristic drift caused by the combined effects of coil thermal effects and mechanical component wear after long-term operation is even more difficult to identify in advance.

[0003] Existing linear motor control schemes lack quantitative detection methods for the overshoot phase difference between the control input signal and the response signal, making it impossible to distinguish the causes of deviations under different load conditions. They also lack mechanisms to identify latent deviation patterns such as pseudo-convergent oscillations. When the motor response enters a state of apparent stability but actual alternating oscillations, fixed parameter configurations cannot detect the continuous accumulation of deviations, ultimately leading to decreased shaving performance and even motor protection activation. Summary of the Invention

[0004] This invention discloses an adaptive control method and system for a linear motor in an electric shaver. The method involves performing envelope attenuation analysis on the control input and response data to output an early warning indicator for drift. An error verification matrix is ​​established by combining overshoot phase difference detection and reference spectrum matching. Based on the drift warning, a priority gain allocation of the leading phase segment is implemented to form a gain rule set. Abnormal characteristic points are located through phase shift characteristic spectrum analysis. The gain rule set is activated to generate an excitation sequence, and the amplitude stability index is evaluated through pseudo-convergent oscillation detection and inverse correlation. Adaptive control parameters are output by combining deviation interval analysis and control characteristic spectrum mapping.

[0005] The first aspect of this invention provides an adaptive control method for a linear motor in a shaver, comprising:

[0006] Acquire control input data and response data, perform envelope attenuation terminal correlation analysis on the control input data and response data, and output an early drift warning indicator;

[0007] Input-output overshoot phase difference detection is performed on the control input data and the response data to obtain the deviation phase intensity distribution. The response data is classified according to operating conditions to construct a reference spectrum library. The deviation phase intensity distribution is matched with the reference spectrum library to establish an error verification matrix.

[0008] Based on the early warning drift indicator, the error verification matrix is ​​given a priority gain allocation for the leading phase segment to form a gain rule group. Phase offset analysis is performed on the gain rule group to output the phase offset feature spectrum. Deviation over-limit judgment is performed on the phase offset feature spectrum to locate abnormal characteristic points.

[0009] The excitation sequence is obtained by activating the gain rule group based on the abnormal characteristic points. The deviation density sequence is obtained by performing pseudo-convergent oscillation detection on the excitation sequence. The amplitude stability index is output by performing reverse correlation detection on the deviation density sequence and the excitation sequence.

[0010] By combining the amplitude stability index with the excitation sequence to perform deviation interval analysis, a deviation interval sequence is obtained. The deviation interval sequence and the gain rule group are used to perform deviation high-incidence section calibration to construct a control characteristic map. The control characteristic map is compared with the control input data to perform parameter deviation mapping and output adaptive control parameters.

[0011] A second aspect of the present invention provides an adaptive control system for a shaver linear motor, comprising:

[0012] The signal acquisition module is used to acquire control input data and response data, and to perform envelope attenuation terminal correlation analysis on the control input data and response data to output an early drift warning indicator;

[0013] The matrix construction module is used to perform input-output overshoot phase difference detection on the control input data and the response data to obtain the deviation phase intensity distribution, classify the response data according to the operating conditions to construct a reference spectrum library, and match the deviation phase intensity distribution with the reference spectrum library to establish an error verification matrix.

[0014] The rule generation module is used to perform advance phase segment priority gain allocation on the error verification matrix according to the early drift warning identifier to form a gain rule group, perform phase shift analysis on the gain rule group to output a phase shift feature spectrum, and perform deviation over-limit judgment on the phase shift feature spectrum to locate abnormal characteristic points.

[0015] The density detection module is used to activate the gain rule group according to the abnormal characteristic points to obtain the excitation sequence, perform pseudo-convergent oscillation detection on the excitation sequence to obtain the deviation density sequence, and perform reverse correlation detection based on the deviation density sequence and the excitation sequence to output the amplitude stability index.

[0016] The graph output module is used to combine the amplitude stability index and the excitation sequence to perform deviation interval analysis to obtain a deviation interval sequence, perform deviation high-incidence section calibration on the deviation interval sequence and the gain rule group to construct a control characteristic graph, and compare the control characteristic graph with the control input data to perform parameter deviation mapping and output adaptive control parameters.

[0017] The beneficial effects of this invention are reflected in the following points: 1. Addressing the problem of frequent switching of operating conditions and difficulty in predicting load fluctuations in shaving scenarios, this invention uses graded detection of the slope change rate at the end of the envelope attenuation segment to output early warning indicators of drift, shifting the perception of characteristic drift to before the deviation becomes apparent. Combined with overshoot phase difference detection under different operating conditions and cross-operating condition deviation direction analysis using a multi-temperature node reference spectral library, the causes of deviations between drive commands and response signals are classified into three categories: drive deviation type, operating condition drift type, and composite type, improving the accuracy of identifying deviation sources under different load conditions. 2. The gain allocation of the error verification matrix is ​​reordered according to the drift early warning indicator level and the type of advanced phase segment, allowing the gain to adjust synchronously with the degree of drift and phase direction rather than being a fixed output. The multi-operating condition phase time difference sequence is identified through high volatility and low mean dual-condition recognition and steady-state segment merging, separating intermittent phase anomalies from random jitter and locating them as abnormal characteristic points, improving the detection accuracy of phase shift anomalies under complex operating conditions. 3. The latent failure of the motor under pseudo-convergent oscillation state, which appears stable but has a continuous accumulation of deviations, is included in the deviation density statistics by continuously alternating the signs of the continuous cycle. The amplitude stability index is constructed by combining the cross-load reverse correlation rate and the intensity uniformity to realize the quantitative assessment of the amplitude stability of the motor. Based on the high-risk window of the amplitude stability index and the deviation interval characteristics of the excitation sequence, the high-incidence section of the deviation is calibrated. The adaptive control parameters are output through the characteristic compensation quantity mapping, realizing the autonomous parameter correction in the scenario of cumulative degradation of motor characteristics. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an adaptive control method for a linear motor in a shaver according to the present invention.

[0019] Figure 2 This is a structural block diagram of an adaptive control system for a linear motor in a shaver according to the present invention. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0022] The technical solutions of the embodiments of this application will be described below.

[0023] like Figure 1 As shown, this embodiment of the invention provides an adaptive control method for a shaver linear motor, including the following steps S101-S105:

[0024] Step S101: Obtain control input data and response data, perform envelope attenuation terminal correlation analysis on control input data and response data, and output early drift warning indicator.

[0025] Specifically, control input and response data during the operation of the shaver's linear motor are acquired. Control input data is read in real-time from the control signal register at a sampling rate of 1kHz, covering three control quantities: control duty cycle, operating frequency, and target displacement. The nominal operating frequency is 200Hz, the peak-to-peak value of the target displacement is set to 1.2mm, and the normal operating range of the control duty cycle is 45%-75%. The rated peak value U0 of the control signal is defined as the amplitude of the control signal corresponding to a 75% control duty cycle, serving as the criterion for subsequent rising edge triggering; exceeding the 45%-75% range triggers an abnormal acquisition flag. Response data is acquired in parallel by a Hall displacement sensor and a response current sensor under the same hardware time base, with an actual displacement resolution of 0.01mm and a response sensing voltage sampling accuracy of 1mV. Vibration acceleration is acquired by a triaxial MEMS sensor, collecting the combined vector components. The three response quantities are packaged and written into a dual-port buffer with a period of 0.5ms. The rated value of the vibration acceleration is the steady-state combined vector mean. During shaving, as the razor head transitions from the sparse beard area on the cheek to the dense area on the chin, the duty cycle in the control input data increases by 8%-12% within 50ms to compensate for the increased load. Simultaneously, the peak-to-peak displacement drops from 1.2mm to below 0.9mm. If the displacement drop rate exceeds 0.015mm / ms, it is marked as a high-load switching event. This event interval is weighted twice by the abrupt change point in the final slope change detection. The difference between the target displacement and the actual displacement in the control input data continues to increase within 3-5 vibration cycles after the beard cutting impact. Once the deviation increase rate exceeds 0.05mm / cycle, it indicates that the motor dynamic response has entered the nonlinear decay range. This threshold is determined by jointly calibrating the motor's rated spring stiffness of 18N / m and the oscillator mass of 3.2g. The two data streams use the rising edge of the control signal as the hardware synchronization anchor point alignment time reference, with an anchor point trigger pulse width of 10μs and a timestamp alignment deviation constrained to within 0.1ms.

[0026] In some embodiments, the step of performing envelope attenuation terminal correlation analysis on the control input data and the response data to output an early drift warning identifier includes: performing time-domain synchronization processing on the control input data and the response data to construct an input-output synchronization sequence; performing envelope curve fitting on the actual displacement field in the input-output synchronization sequence to construct a response envelope; performing terminal slope change detection on the envelope line of the response envelope to determine the slope change distribution; and performing slope attenuation rate classification based on the slope change distribution to output an early drift warning identifier.

[0027] A time-domain synchronization process is performed on the control input data and response data to construct an input-output synchronization sequence. The two timestamps are aligned using the rising edge of the control signal as the hardware synchronization anchor point. The anchor trigger pulse width is set to 10μs, and the trigger signal is simultaneously written to both the control input acquisition channel and the sensor acquisition channel, with an initial deviation constraint within 0.1ms. The control input data sampling period is 1ms, and the response data sampling period is 0.5ms. When both are mapped to a common time axis with a resolution of 0.5ms, the control input data side uses linear interpolation of adjacent original sampling points to complete the intermediate nodes. The shaver head commutation cycle is nominally 5ms. The phase jump of the control input data is most concentrated at the commutation moment. Therefore, the input-output synchronization sequence is refined twice within a 1ms interval before and after the commutation moment, using a 0.1ms sub-window. After refinement, the alignment error of the fields on both sides does not exceed 0.05ms, and the original 0.5ms resolution is maintained outside the refinement interval. The common time axis uses the first valid rising edge of the control signal as the zero point, and subsequent nodes are addressed in multiples of 0.5ms. If a sensor loses a point in a certain sampling period, it is supplemented by secondary interpolation of the previous and next valid nodes, and a lost point mark is added. If the cumulative number of lost points in a single sub-window exceeds 2, that window exits secondary refinement and only retains the original resolution to avoid interpolation amplification error. At this point, the control input data and response data are time-domain aligned and merged into an input-output synchronization sequence. Each node stores three control input fields: control duty cycle, operating frequency, and target displacement, and three response fields: actual displacement, response induced voltage, and vibration acceleration. The typical duration is 180-300 seconds, corresponding to 360,000 to 600,000 sequence nodes.

[0028] Envelope curve fitting is performed on the actual displacement field in the input-output synchronization sequence to construct the response envelope. The response envelope consists of two curves: an upper envelope and a lower envelope. Specifically, the actual displacement field in the input-output synchronization sequence is gradually fitted with a sliding window of 20ms and a step size of 0.5ms. Within each window, the local waveform is fitted with a quadratic function to extract positive and negative peak values. The peak value extraction accuracy is better than 50% of the original sampling resolution of 0.01mm. The positive peak sequence is fitted with a third-order spline curve to form the upper envelope of the response envelope, and the negative peak sequence is fitted in the same way to form the lower envelope of the response envelope. The fitting node spacing is 20ms and interpolated to a resolution of 0.5ms. The spacing between the upper and lower envelopes of the response envelope is defined as the instantaneous width W(t) (mm). When W(t) drops below 60% of the steady-state average W0 within 5 seconds after the zero point, the moment is marked as the decay node. Decay nodes with a continuous duration of more than 50ms are merged into decay segments. When the razor is in close contact with the curved surface of the neck during shaving, the back pressure on the skin increases periodically. The narrowing amplitude of W(t) in a single instance does not exceed 15% of W0, and the rebound time is less than 30ms. The temporal correlation coefficient between this regular narrowing and the synchronous increase of the control duty cycle exceeds 0.85. Based on this, the envelope fluctuation caused by the adaptive load adjustment is distinguished from the drift decay. Only the narrowing segment with a correlation coefficient lower than 0.85 is included in the decay segment statistics. The last segment of the response envelope is defined as the segment of the envelope line on the response envelope within the last 20% of the duration of a single shaving task, which serves as the direct input for subsequent slope change detection in the last segment. The continuous decay segment within the last segment with a duration exceeding 300ms is marked with a continuous decay indicator. The node corresponding to this marked segment is weighted 1.5 times in the slope change detection of the last segment, while the weight of non-continuous decay segments remains at 1.0. The initial steady-state mean W0 is locked after sliding to a stable value every 0.5 seconds within 5 seconds before the task starts, and is no longer refreshed with load fluctuations; when the root mean square of the quadratic fitting residual of a certain window exceeds twice the peak extraction accuracy, its peak value is replaced by linear extrapolation of the peak value of the adjacent reliable window; if 3 consecutive windows are unreliable, the response envelope segment of that segment is removed from the attenuation statistics.

[0029] The slope change distribution is determined by detecting the slope change at the end of the envelope on the response envelope. For every 5 adjacent nodes (2.5 ms time span) at the end of the envelope on the response envelope, the local slope k(t) is calculated using the difference quotient formula, in μm / ms. When k(t) is negative and |k(t)| monotonically increases within 10 consecutive calculation windows (25 ms in total), it is identified as an accelerated decay segment, corresponding to the thermal instability state where the actuator's vibration output capability decreases due to thermal effects after continuous high-load shaving for more than 8 minutes. The slope sequences of the sustained decay segment and the non-sustained decay segment of the response envelope are separately calculated, with their respective mean values ​​k_d and k_n. Once the difference between the two exceeds 0.3 μm / ms, a local instability marker is added to the slope change distribution, and the weighting coefficient for abrupt changes within this period is increased to 1.8 times. Nodes where |k(t)| exceeds twice the mean of the last segment of k(t) are marked as abrupt changes, and their amplitude and trigger time are written into the slope change distribution. When the shaver head is momentarily unloaded during a cleaning and rinsing operation, a peak exceeding 3 μm / ms and lasting less than 2 ms appears in the slope sequence of the response envelope. The mean rate of slope regression before and after the peak exceeds 1.5 μm / ms². Based on these two criteria, the corresponding slope peak is removed from the mutation point statistics and does not participate in the subsequent decay rate classification. The slope change distribution is constructed with the final time axis as the horizontal axis and k(t) as the vertical axis: the time period with a mutation point density of more than 5 per second is a high fluctuation zone. The high fluctuation intensity index HI (unit μm) is defined as the product of the time period length (in ms) and the peak amplitude of the mutation point (in μm / ms). A high intensity fluctuation indicator is triggered when HI reaches 0.8 μm. If the total number of mutation points in the entire segment is less than 3, it is directly determined that there is no drift and the classification calculation is locked.

[0030] Based on the slope change distribution, the slope decay rate is graded and an early drift warning indicator is output. If the previous stage has determined that there is no drift and the grading calculation is locked, the weighted total score S of this stage is directly set to 0, the early drift warning indicator outputs a no-drift level, and the following calculations are skipped; otherwise, the grading is performed according to the following steps. The ratio of the amplitude difference ΔA between adjacent abrupt change points in the slope change distribution to the time interval Δt is defined as the decay rate R: R = ΔA / Δt (ΔA is in μm / ms, Δt is in ms, and R is in μm / ms²). The R value is divided into three levels: low (R < 0.5R0), medium (0.5R0 ≤ R < 1.5R0), and high (R ≥ 1.5R0) according to 0.5 times and 1.5 times the mean R0 of the abrupt change points in the entire slope change distribution. The corresponding trigger weights are 1, 2, and 3, respectively. The cumulative sum of the number of triggers and the weights of each level is the weighted total score S: S below 15 indicates no drift, 15-30 indicates mild, 30-50 indicates moderate, and above 50 indicates severe. If S exceeds 50, it indicates severe triggering, and no further subdivision is required. After continuous use of the shaver for more than 10 minutes and the actuator temperature rises above 65°C, the vibration output decreases, causing high-level R values ​​to persist and HI to exceed 0.8μm. In this scenario, the S value typically falls within the 120-180 range. The early drift warning indicator directly maps to the heavy setting, and the continuous duration of high-level triggers is recorded simultaneously to distinguish between transient impacts and continuous thermal drift. Within the high fluctuation zone, the R value is not constrained by absolute amplitude and is directly classified as a high level. In low-temperature environments (below 5°C), the viscosity of the lubricating grease increases, and the motor starting resistance is systematically higher by about 12%-18%. Therefore, the overall grading threshold shifts upward by 10%, with the lower limit of the light setting increasing from 15 to 16.5, the medium setting from 30 to 33, and the heavy setting from 50 to 55. To prevent the S value from fluctuating near the setting boundaries and causing grade jumps, a one-way hysteresis is set between adjacent settings: when upgrading, the threshold in the table is used as the standard, and when downgrading, the threshold is additionally lowered by 8%, meaning that upgrading is sensitive and downgrading is conservative.

[0031] Step S102: For the control input data and response data, the input-output overshoot phase difference is detected to obtain the deviation phase intensity distribution. The response data is classified according to the working conditions and a reference spectrum library is constructed based on the multi-temperature node test bench calibration data. The deviation phase intensity distribution is matched with the reference spectrum library to establish an error verification matrix.

[0032] In some embodiments, the step of detecting the input-output overshoot phase difference between the control input data and the response data to obtain the deviation phase intensity distribution includes: locating the overshoot peak time of the control input data to obtain the control peak timing; locating the overshoot peak time of the response data to obtain the response peak timing; performing successive time difference analysis on the control peak timing and the response peak timing to determine the phase difference distribution; and constructing the deviation phase intensity distribution by weighting the phase difference distribution according to the phase difference abrupt change rate.

[0033] The control input data is used to locate the overshoot peak time and obtain the control peak timing. The target displacement signal in the control input data is first preprocessed by a second-order Butterworth low-pass filter with a cutoff frequency of 800Hz to filter out control signal switching noise. Then, the second-order difference is calculated point by point using 5 sampling points as a differential window. The zero-crossing point of the differential is marked as the candidate extreme value time. Among them, the amplitude exceeding the peak value by 120%, i.e., 1.44mm, is confirmed as the overshoot peak time and written into the control peak timing. If multiple adjacent zero-crossings occur near the second-order differential zero-crossing point due to quantization noise, the point with the highest amplitude after filtering is taken as the unique extreme value, and the others are suppressed as debouncing points. There is also a negative overshoot in the return direction of the target displacement. Its time is extracted synchronously and listed separately with negative amplitude. It is only used for commutation symmetry verification and does not participate in the successive time difference pairing of positive overshoot. When the interval between two consecutive overshoot peaks is less than 4ms, the one with the larger amplitude is retained and the one with the smaller amplitude is discarded to prevent the double peaks of the reversing overshoot from being mistakenly counted as two independent overshoot events. After merging, the peak with the larger amplitude is used as the valid time to update the control peak time sequence. If the difference in amplitude between the two peaks is less than 0.03mm, both peaks are marked with a low-distinction label, and their weight is reduced to 0.6 in the weighted calculation of successive time difference analysis. After extracting the working frequency level from the control input data, the overshoot lasts for about 1.5ms and the amplitude is about 128% of the peak-to-peak value at the high speed setting (working frequency 220Hz), and about 2.2ms and the amplitude is about 122% at the low speed setting (180Hz). The inherent overshoot characteristics of the two settings are marked on each record according to the working frequency level. During successive time difference analysis, the inherent phase offset is corrected by looking up a table according to the setting (high speed setting -0.15ms, low speed setting +0.12ms, and linear interpolation for intermediate frequency settings).

[0034] The response data is analyzed to pinpoint the overshoot peak time and obtain the response peak timing sequence. Before extreme value extraction, the actual displacement signal output by the Hall displacement sensor in the response data is denoised using a low-pass filter with a cutoff frequency of 500Hz. This cutoff frequency is lower than the 800Hz on the control input side to compensate for signal hysteresis caused by sensor bandwidth limitations. After filtering, extreme value candidates are extracted using a 5ms differential window. Candidate times must simultaneously meet two criteria: the actual displacement amplitude exceeds 1.44mm and the resultant vector of vibration acceleration in the response data exceeds 115% of the rated value. These are then confirmed as overshoot peak times and written into the response peak timing sequence. This dual-criteria mechanism reduces the false overshoot misjudgment rate caused by sensor drift from approximately 12% to less than 2%. Moments where only the displacement amplitude exceeds the limit but the acceleration does not are marked as suspected overshoot. These entries have a weight reduced to 0.5 in successive time-difference analysis. A typical scenario is when the blade briefly leaves the skin surface, causing displacement readings to drift without a corresponding impact response. The magnitude and duration of the displacement exceeding the limit are recorded together for subsequent attribution. When the contact angle between the razor foil and the skin deviates by more than 15°, the actual displacement overshoot duration is extended to 1.3-1.6 times the normal value. This type of entry is marked with a contact angle deviation indicator. The trigger condition is that the transverse component of the vibration acceleration in the response data exceeds 35% of the resultant vector and lasts for no less than 8ms. Before time difference pairing, the deviation indicator entry corrects the inherent delay by about 0.3ms, and the compensation is added proportionally as the deviation angle increases to avoid the introduction of systematic time difference inflation due to posture differences. The deviation indicator entry has a weight of 0.8 in the weighted calculation of successive time difference analysis. The Hall displacement sensor experiences zero drift as the actuator temperature rises. If the zero drift exceeds twice the actual displacement resolution within 500ms, the extreme value candidate amplitude is recalculated using the average displacement of the same segment as the dynamic baseline to prevent the zero drift from causing a systematic shift in the overshoot amplitude criterion of 1.44mm. When the interval between adjacent overshoot peak times within the same segment is less than half of the commutation cycle, the larger amplitude is selected according to the double-peak merging rule consistent with the control input side to update the response peak timing, so as to ensure that the overshoot counting calibers of the two channels are aligned.

[0035] Successive time-difference analysis is performed on the control peak timing and response peak timing to determine the phase difference distribution. Using each trigger moment in the control peak timing as the main index, the nearest trigger moment within a 5ms window is searched in the response peak timing. When a match is found, the time difference τ(n) is calculated as: Response peak timing time - Control peak timing time (in milliseconds; τ>0 indicates a response lag behind the control, normally stable at 0.5-0.8ms). τ(n) is included in the phase difference distribution. Moments in the control peak timing without a match within 5ms are marked as isolated points and skipped. When the impact exceeds sensor saturation and persists for more than 3ms, the phase difference distribution is estimated using linear interpolation of the adjacent paired time differences, and the weight is reduced to 0.4. When an entry in the response peak timing falls into multiple control peak timing search windows simultaneously, it is uniquely matched according to the principle of minimum |τ|. The missing control peak timing moment is searched for the next response entry; if it still exceeds the window after extension, it is classified as an isolated point. The impact reaction force when cutting coarse, hard whiskers causes the oscillator to decelerate instantaneously. When the pairing time difference suddenly increases to 2.5-3.0ms, and τ(n) exceeds 2.0ms for 3 consecutive times, it is recorded as a high phase difference cluster segment. When the proportion of isolated points exceeds 15%, the hole marker is used instead of the time difference value and included in the phase difference distribution. Holes wider than 50ms are removed from the weighted calculation, and the effective τ(n) average value at the boundary is used as a conservative fill value and retained in the phase difference distribution. When τ(n) is lower than 0.3ms for more than 5 consecutive times, it indicates that the sensor is installed off-center. A continuous segment with negative τ(n) and an absolute value exceeding 0.8ms is marked as a leading segment and is not included in the normal phase difference distribution statistics but is transferred to the leading phase segment priority gain allocation processing. All effective pairings of the control peak timing and response peak timing are arranged according to the trigger time. The τ(n) value and the quality marker together constitute the phase difference distribution. The phase difference distribution reflects the phase relationship between the control end and the response end at each time with the time axis as the horizontal axis and τ(n) as the vertical axis.

[0036] The bias phase intensity distribution is constructed by weighting the phase difference distribution according to the phase difference mutation rate. The mutation rate V(n) is defined as the ratio of the time difference between adjacent pairs |τ(n) - τ(n-1)| to the pairing interval Δt: V(n) = |τ(n) - τ(n-1)| / Δt (τ and Δt are in ms, and V(n) is the dimensionless mutation rate). The weighting coefficient is 3.0 when V(n) exceeds 1.5 times the mean of the entire sequence, 1.5 when it falls between 0.5 and 1.5 times, and 1.0 when it is below 0.5 times. The bias phase intensity distribution slides along the time axis with a sliding window of 500 ms and a step size of 10 ms. The cumulative sum of each pair's V(n) and the weighting coefficient within the window, divided by the total number of effective pairs, yields the intensity value I(t) at that moment. Hole marker segments in the phase difference distribution are filled with linear interpolation of the effective τ(n) on both sides before participating in the V(n) calculation, with each interpolation entry weighted by 0.4. Adjacent sliding windows overlap by 490ms within a 10ms step. The I(t) in the overlapping area takes the result of the later window to ensure priority for the latest pairing. No abrupt smoothing is applied at window switching points to preserve the true phase transition. The I(t) of a single task is normalized to the mean of the stable shaving segment, allowing for horizontal comparison of intensity values ​​across different tasks and individuals without being affected by absolute amplitude differences. In the high-frequency switching scenario of razor reversal, V(n) is densely triggered near the phase transition moment, and the peak value of I(t) can reach 4-6 times that of the stable shaving stage. In the deviation phase intensity distribution, the interval where I(t) continuously exceeds twice the mean of the entire time domain and is not less than 200ms is marked as the excess segment. If the half-peak width of the V(n) time distribution in the over-limit section is greater than 80ms, it is judged as broadband phase diffusion caused by motor characteristic drift. If it is less than 30ms, it is judged as narrowband phase abrupt change caused by commutation impact. The gain reference values ​​of the two types of features correspond to different column indices of the error check matrix. The reference value of the broadband diffusion type is about 0.15 higher than that of the narrowband abrupt change type, so as to strengthen the correction of persistent drift.

[0037] A reference spectral library was constructed by classifying the response data according to operating conditions. First, the response data was classified into three categories based on the control duty cycle with dual thresholds of 55% and 65%: light load (<55%), medium load (55%-65%), and heavy load (>65%). Then, bench calibration data was collected for each category to construct the reference spectral library. Response data for each operating condition were collected at three temperature nodes: 5°C, 20°C, and 40°C. Samples were taken from 50 new linear motors from the same batch, with each motor and each node collecting at least 10 steady-state oscillation segments, each lasting at least 300ms. The actual displacement signal in the response data was low-pass filtered at a cutoff frequency of 500Hz and then subjected to a short-time Fourier transform with a window of 200ms, a step size of 20ms, and Hanning window weighting. The frequency domain resolution was 0.5Hz, outputting a phase response curve in the range of 150-250Hz. The average of the 50 samples was used as the standard spectral line for each operating condition temperature node. The reference spectral library organizes standard spectra according to three dimensions: operating condition level, temperature node, and deviation direction. For each operating condition level, positive and negative deviation spectra are stored separately for each temperature node: positive deviation spectra are extracted from data of the actual displacement phase-leading control signal, while negative deviation spectra are extracted from phase-lag data. These two types are stored separately to preserve deviation direction information, which is then used for subsequent spectral interval positioning and comparison based on operating condition level and temperature node. When the measured temperature falls between two adjacent calibration nodes, a virtual reference spectrum is generated by weighted interpolation of the spectra of those two adjacent temperature nodes according to the proximity of the measured temperature to the two nodes. The closer the measured temperature is to a node, the greater the weight of that node's spectrum. When the temperature exceeds the 5°C-40°C calibration range, the spectrum of the nearest endpoint is substituted with an extrapolation indicator. During the bench data acquisition phase, outliers are removed from each of the 50 samples. If the spectral line of a certain temperature node on a single instrument deviates from the mean of that node by more than three times the standard deviation, the data for that node of the entire instrument is discarded and reacquired. Once the reference spectral library is calibrated, it is fixed as a read-only reference and is not updated with user data throughout the entire lifespan of the instrument.

[0038] In some embodiments, the step of matching the deviation phase intensity distribution with the reference spectrum library to establish an error verification matrix includes: dividing the deviation phase intensity distribution into high-frequency switching segments under operating conditions to obtain segmented intensity groups; performing cross-operating condition deviation direction analysis on the segmented intensity groups and the reference spectrum library to construct a directional deviation spectrum; determining layered deviation parameters by performing deviation level stratification based on the directional deviation spectrum; and establishing an error verification matrix by allocating matrix intervals according to the layered deviation parameters.

[0039] The deviation phase intensity distribution is divided into segmented intensity groups by high-frequency switching segments under operating conditions. On the time axis of the deviation phase intensity distribution, a continuous interval with a control duty cycle jump amplitude exceeding 8% and a jump interval of less than 200ms is defined as a high-frequency switching segment. Its starting point is the rising edge of the control duty cycle jump, and its ending point is the moment when the control duty cycle return mean ±2% is stably exceeded for more than 150ms. Each high-frequency switching segment extracts the corresponding I(t) segment of the deviation phase intensity distribution according to the start and end times, forming an entry of the segmented intensity group. The switching intensity level is distinguished by four indicators: intensity mean, intensity standard deviation, control duty cycle level, and number of switching: those with more than 4 switching times are classified into the high-frequency multi-hop category, and those with less than 2 switching times are classified into the low-frequency single-hop category. Different frequency domain weighting strategies are adopted when matching the reference spectrum library between the two. The frequency domain weight of the high-frequency multi-hop category is tilted towards the 200Hz main frequency band, and the concentration coefficient is 1.4 times that of the low-frequency single-hop category. When the razor switches from the cheek to the upper lip area, the load increases sharply for a short period. The duty cycle changes 3-4 times within 100ms. The standard deviation of I(t) in this interval is about 2.8 times that of the stable segment. The corresponding entries are classified into the high-frequency multi-jump category and then subdivided into single-jump sub-entries for matching one by one. The sub-entry I(t) is divided by the rising edge of each duty cycle jump. The start and end ranges of each sub-entry I(t) are taken 50ms before and after the jump time. If the time ranges of adjacent sub-entries overlap, they are distributed to both sides by 50% of the average intensity of the overlapping segment. After distribution, the corresponding entries in the segment intensity group are updated. The stable segment other than the high-frequency switching segment is extracted as the control intensity segment, and its mean is used as the normalization benchmark. The segment intensity group is divided into three levels: light load, medium load, and heavy load, and is indexed independently. The level is determined by the mode of the duty cycle within the switching segment. Switching segments where the duty cycle fluctuates within ±1% of the boundary value between two levels are marked with a cross-level mark and are included in the matching of the two adjacent levels.

[0040] For example, the step of performing cross-condition deviation direction analysis on the segmented intensity group and the reference spectral library to construct a directional deviation spectrum includes: locating spectral intervals between the segmented intensity group and the reference spectral library to obtain a matching interval set; performing cross-condition intensity gradient difference analysis on the matching interval set to obtain a difference sequence; identifying deviation segments with low differences and high frequencies based on the difference sequence to obtain a persistent deviation feature group; and constructing a directional deviation spectrum based on a deviation direction consistency assessment of the persistent deviation feature group.

[0041] The segmented intensity groups and the reference spectral library are used to locate spectral intervals to obtain a matching interval set. The spectral interval location process is carried out one by one for each entry in the segmented intensity group: first, according to the operating condition level of the entry, standard spectral lines of the same level and temperature nodes are retrieved from the reference spectral library as candidate comparison objects to narrow down the comparison range. When the measured temperature is between two calibration nodes, virtual spectral lines generated by interpolation of spectral lines of adjacent nodes in the reference spectral library are used as candidates; then, the Euclidean distance between the frequency domain intensity sequence of the entry and each candidate spectral line is calculated along the frequency axis point by point. The frequency interval with the smallest Euclidean distance and a continuous width of not less than 5Hz is determined as the spectral matching interval of the entry, and its level, center frequency and mean Euclidean distance are written into the matching interval set. After all entries in the segmented intensity group have been processed, the matching interval set is a collection of the best aligned frequency bands of each entry in the reference spectral library. Intervals in the matching interval set whose mean Euclidean distance is less than 50% of the total mean are marked as high-confidence matches and weighted 1.3 times in subsequent gradient difference analysis. When multiple candidate intervals exist for the same entry, the one with the smallest Euclidean distance is selected as the main matching interval and written into the matching interval set. If an entry has no continuous low-distance intervals with a width of 5Hz or more within the search range, it is determined that it does not match any of the reference spectral libraries, is marked with a spectral mismatch flag, and is removed from the gradient difference analysis. If the number of such entries exceeds 20% of the total number of groups in a single run, it indicates that the operating condition classification threshold deviates from the actual load distribution. When the operating frequency of the shaver shifts to 198Hz or 202Hz due to the temperature rise of the actuator, the peak value of the segmented intensity group in the frequency domain drifts accordingly, while the peak value of the standard spectral line is fixed at 200Hz. This causes the minimum Euclidean distance interval to shift by ±2-3Hz as a whole. This shift is recorded as the frequency offset. Entries with an absolute frequency offset exceeding 3Hz are marked with a temperature drift indicator. During gradient difference analysis, their frequency coordinates are first shifted to the nearest standard frequency point before being included in the calculation.

[0042] A difference sequence is obtained by performing cross-load condition intensity gradient difference analysis on the matching interval set. After aligning the light, medium, and heavy load matching intervals in the matching interval set with the same frequency axis, the intensity gradient difference Δg(f) between adjacent loads is extracted point by point: the light-to-medium gradient difference is the difference between the medium load matching strength and the light-to-heavy load matching strength, and the medium-to-heavy gradient difference is the difference between the heavy load matching strength and the medium load matching strength. These two constitute the light-to-medium and medium-to-heavy difference sequences, respectively. Under normal motor characteristics, the ratio of the two sets of Δg(f) should be stable in the range of 0.8-1.2. This ratio reflects the linear gain characteristics of the motor across the load range. A ratio deviating from the range indicates a nonlinear response at the corresponding load inflection point. Such frequency points are marked with nonlinearity markers, and when the density reaches more than two per 5 Hz, the frequency band is marked as a nonlinearity concentration segment. The nonlinearity concentration segment triggers nonlinear compensation lookup table, and the compensation amount is selected from the built-in three-level coefficients according to the degree of nonlinearity. High-confidence matching entries are included in the statistics with a weighting of 1.3. Frequency points where |Δg(f)| exceeds 1.5 times the mean of the corresponding gradient level are marked as gradient over-limit points. When the continuous distribution width on the frequency axis exceeds 8Hz, high gradient anomaly segments are marked. The anomaly intensity is quantified by combining the gradient difference extreme value and the distribution width: Intensity score = (extreme value / mean of the baseline gradient) × (width / Δf0), where Δf0 = 5Hz is the reference frequency width, and the intensity score is dimensionless. An intensity score exceeding 1.8 triggers a key attention marker. The threshold for this marker frequency band is lowered from 70% to 55% when extracting continuous deviation feature groups to improve sensitivity. When the three matching intervals within the matching interval set are aligned on the frequency axis, the peak frequencies of each interval may be slightly misaligned due to temperature drift. Therefore, the peak frequencies of the main matching intervals of each interval are first resampled to a common 0.5Hz frequency grid and then the gradient difference is calculated. The resampling interpolation error is included in the reliability of the difference sequence. If a frequency point has only a valid match in two intervals (the third interval is an isolated input point or a hole), only the available single gradient difference is calculated and the single interval is marked as missing. The missing points are not included in the linearity criterion of the light-medium and medium-heavy ratios.

[0043] Based on the difference sequence, low-difference, high-frequency deviation segments are identified to obtain a persistent deviation feature group. Frequency points in the difference sequence where |Δg(f)| is below 50% of the overall sequence mean are defined as low-difference points. These low-difference segments are merged when their continuous frequency distribution width is not less than 3Hz. The lower the mean |Δg(f)| within a low-difference segment, the higher the consistency between the actual characteristics of that frequency band and the reference spectral library. The segment with the highest consistency is usually the motor's most stable vibration frequency window. The difference sequence is statistically analyzed using a 50ms sliding window to count the frequency of deviations at each frequency point. Points with a frequency exceeding 70% of the total calculations within the window are marked as high-frequency deviation points. When the spatial intersection width between low-difference points and high-frequency deviation points is not less than 2Hz, they are included in the persistent deviation feature group. If it is less than 2Hz, it is downgraded to a suspected segment, and the frequency is re-counted in the next 50ms window for a second judgment. If it is still less than 2Hz, it is removed. The frequency statistics window for the secondary judgment is staggered by 25ms from the primary judgment to introduce independent samples. Those that meet the secondary standard but have an intersection width in the critical range of 2-3Hz are included in the continuous deviation feature group and given a narrow band label. They are also included in the directional consistency assessment with a weight of 0.7. The center frequency of the low difference segment with the highest consistency is used as the output of the motor's current stable vibration window for phase calibration to select the calibration frequency band with the least interference. When the shaver's long-term use causes the oscillator spring stiffness to decrease by about 8%-12%, the difference sequence will show low difference and high frequency deviation points in the 190-195Hz frequency band. The deviation frequency of items in this frequency band usually exceeds 85%, and the consistency with the new product reference spectrum is abnormally high. At this time, the difference between the center frequency of the item frequency band and the new product reference frequency of 200Hz is the natural frequency shift caused by the spring stiffness drift. Every 1Hz shift in the natural frequency corresponds to a 2%-3% decrease in spring stiffness. When the natural frequency shifts down by more than 6 Hz, a spring failure warning is triggered, corresponding to a stiffness decrease of about 13.8%. If this is exceeded, the heavy-load amplitude will deviate from the rated tolerance band.

[0044] A directional deviation spectrum is constructed based on the consistency assessment of deviation direction using a persistent deviation feature group. The intensity difference sign within each frequency band of the persistent deviation feature group is grouped and statistically analyzed according to operating conditions: when the positive deviation ratio in the same frequency band exceeds 60% across light, medium, and heavy loads, it is considered a consistent positive deviation across all operating conditions. The negative consistency criterion is symmetrical; if neither is met, it is considered a directional split between operating conditions. For directional split segments, the three values ​​representing the positive deviation ratio for each load are retained separately to describe their non-uniformity. Those with light and medium loads in the same direction but heavy loads in the opposite direction are marked as high-load reversible segments, corresponding to motors entering the nonlinear saturation region in the heavy load range. The center frequency and mean deviation intensity of the consistent positive deviation segments across all operating conditions from the persistent deviation feature group are written into the positive segment of the directional deviation spectrum, and the negative consistency segments are written into the negative segment. The directional deviation spectrum is constructed with the frequency axis as the horizontal axis and the intensity difference as the vertical axis. Positive, negative, and split segments are color-coded. Adjacent segments with a distance of less than 2Hz in the same direction are merged into a single continuous segment to simplify subsequent layering boundaries. The center frequency after merging is the weighted average of the number of frequency points in the two segments. After more than 6 months of use, the frictional resistance of the razor foil increases. Under heavy load conditions, the 200Hz frequency band shows a consistent positive trend, with the average positive deviation intensity approximately 18% higher than that under medium load conditions. A stable positive intensity peak forms in the directional deviation spectrum at this frequency. When the ratio of this peak amplitude to the heavy load baseline of the reference spectrum exceeds 1.25, a wear characteristic indicator is identified. If the wear characteristic indicator and spring failure warning appear simultaneously, it is jointly determined to be a combined aging state of increased foil friction and decreased spring stiffness. In this case, the deviation level threshold for the corresponding frequency band is lowered by one level to proactively detect the gain mismatch risk caused by the combined degradation.

[0045] The deviation level is determined by stratifying the deviation parameters based on the directional deviation spectrum. The absolute value of the intensity difference |D(f)| at each frequency point of the directional deviation spectrum is divided into four levels according to 0.5 times, 1.0 times, and 1.5 times the mean μD of the entire frequency band: Level 1 is |D(f)| below 0.5μD (slight deviation, characteristics still within the tolerance band of the reference spectrum), Level 2 is 0.5μD to 1.0μD (moderate deviation, requiring moderate intensity correction), Level 3 is 1.0μD to 1.5μD (significant deviation, corresponding to moderate wear or thermal effect accumulation to a high level), and Level 4 is above 1.5μD (severe deviation, corresponding to significant stiffness decay or composite aging). As the service time increases, the level of each frequency point generally shifts towards a higher level, which can be used to track the aging process of the motor. The number of frequency points covered by each level and the center frequency of the frequency band are statistically extracted point by point, and the center frequency is located by the weighted average of |D(f)| of each point within the level. The positive and negative segments of the directional deviation spectrum are independently layered. The layered deviation parameters are constructed into an 8-row two-dimensional parameter table (four levels × positive and negative directions) with the deviation level as the first dimension and the deviation direction as the second dimension. The mean deviation amplitude of each row of the layered deviation parameters is weighted by the absolute value of the intensity difference of the corresponding frequency points according to the number of frequency points. The contribution of each level is proportional to the number of frequency points. When the difference between the center frequency of the fourth-level deviation band and the peak frequency of the standard spectrum of 200Hz exceeds 5Hz, a frequency drift indicator is added to indicate that the inherent frequency has shifted as a whole rather than simply gain mismatch. The corresponding entry triggers an independent frequency correction channel when allocating in the matrix interval. The gain reference value is superimposed with a first-order compensation term converted at 0.02 / Hz. The frequency points of the directional split frequency band between operating conditions are not included in any level statistics to avoid the segment with chaotic direction interfering with the allocation of the gain matrix.

[0046] An error verification matrix is ​​established by allocating matrix intervals based on the hierarchical deviation parameters. The center frequency, mean deviation amplitude, and weighting coefficient of each level of the hierarchical deviation parameters are mapped to the row and column coordinates of the error verification matrix: the row index corresponds to the operating condition type (two rows each for light, medium, and heavy loads, and one row each for positive and negative deviation directions), and the column index corresponds to the deviation level (one column for each of levels one to four). The matrix dimension is 6 rows × 4 columns, with a total of 24 units. Each unit stores the phase correction gain reference value G(i,j), which is obtained by linearly normalizing the mean deviation amplitude of the corresponding entry of the hierarchical deviation parameters. The normalization reference is the maximum allowable phase deviation value of 4.0ms for all operating conditions. If the corresponding frequency band is marked as broadband diffusion type in the deviation phase intensity distribution, then 0.15 is added to the normalized result of G(i,j) in this column (if it exceeds 0.9 after adding, it is clamped); no additional adjustment is made for narrowband abrupt change type. The upper limit of the fourth-level unit G(i,j) is clamped at 0.9 to prevent overcompensation and oscillation. The lower limit of the first-level unit is no less than 0.1 to ensure the minimum correction response for small deviations. The second and third-level units are linearly distributed in the range of 0.3-0.7 with the mean deviation amplitude. When the shaver is used in a high-humidity environment (relative humidity exceeding 85%), the signal-to-noise ratio of the Hall sensor decreases by about 3dB, and the mean amplitude of each level of deviation is generally higher by about 8%. Therefore, the humidity correction version of the error check matrix is ​​multiplied by an environmental correction factor of 0.92 on the standard G(i,j). In a low-temperature environment (below 5°C), the increased viscosity of the lubricating grease causes the mean amplitude of each level to be lower by about 6%. The low-temperature correction version is multiplied by 1.06 accordingly. The three versions are automatically switched based on real-time environmental temperature and humidity data. When both conditions are met, the correction factors are multiplied together and then applied. When there is no corresponding frequency point for a certain deviation level in either the positive or negative direction, the unit G(i,j) in that column is filled with linear extrapolation of the adjacent levels in the same row and an interpolation label is attached.

[0047] Step S103: Based on the early warning drift indicator, the error verification matrix is ​​given priority gain allocation for the advanced phase segment to form a gain rule group. Phase offset analysis is performed on the gain rule group to output the phase offset feature spectrum. The deviation exceeds the limit and abnormal characteristic points are located based on the phase offset feature spectrum.

[0048] Based on the early drift warning indicator, a gain rule group is formed by prioritizing the gain allocation of the leading phase segment on the error check matrix. The early drift warning indicator divides the degree of motor characteristic deviation into four levels: no drift, mild, moderate, and severe. The adjustment intensity of the four-level deviation column G(i,4) of the error check matrix increases sequentially for each level: severe level moves up by 0.15, moderate level moves up by 0.08, mild level remains unchanged, and no drift level shrinks the lower limit of G(i,4) to 0.05. After moving up, if it exceeds 0.9, it is clamped to maintain a stable margin. The higher the degree of drift, the stronger the response gain of the higher deviation column, and the compensation capability is automatically strengthened with the degree of deviation. The leading phase segment is determined by τ(n) being continuously below -0.8ms for more than 5 paired cycles. This is usually caused by installation misalignment or loose fasteners, and is opposite in direction to the hysteresis bias caused by thermal drift, requiring an independent gain allocation strategy. Within the leading phase segment, each row of the error check matrix is ​​rearranged into a gain allocation vector with a fourth-level column of 1.8, a third-level column of 1.4, a second-level column of 1.0, and a first-level column of 0.6. This vector is then element-wise multiplied with the original G(i,j) and normalized to the 0.1-0.9 range, concentrating greater weight on the higher-level bias columns to strengthen the correction of the frequency band with the largest leading amplitude. The shaver continuously shaves under heavy load for more than 10 minutes, and the actuator temperature rise exceeds 6... At 5°C, the drift increases to the severe level, and the phase lag of the oscillator exceeds the upper limit of 4.0ms. If the positive deviation row G(5,4) of the heavy load exceeds 0.75, the weight of the fourth-level column is further increased to 2.2 to cope with the high-risk scenario of thermal drift and heavy load deviation superimposed. The gain rule group constructs 24 gain allocation combinations of 3×4×2 according to the three-dimensional index of the operating condition level, the early warning indicator level of drift and the phase segment type, covering all operating condition cross-scenarios of light, medium and heavy load × four-level drift × normal and advanced phase. When the level is downgraded, each column G(i,j) linearly backs down by 0.01 per step and at an interval of 50ms to prevent the rapid alternation of warning indicators from causing the gain rule group to jitter.

[0049] In some embodiments, performing phase shift analysis on the gain rule group to output a phase shift feature spectrum includes: calibrating the driving phase time of each operating condition in the gain rule group to obtain a multi-operating condition phase time group; performing adjacent operating condition time difference analysis on the multi-operating condition phase time group to obtain a phase time difference sequence; identifying abnormal time difference segments by performing high fluctuation and low mean segment identification on the phase time difference sequence; and merging and integrating the phase shift based on the steady-state segment in the abnormal time difference segment to output a phase shift feature spectrum.

[0050] The driving phase timing of each operating condition is calibrated using the gain rule group to obtain a multi-operating condition phase timing group. The calibration of the driving phase timing of each operating condition is performed step by step according to the operating condition level index of the gain rule group: the linear motor is driven sequentially with the gain allocation vectors of the corresponding index layers of light load, medium load, and heavy load. Under each level of drive, the control signal is collected and the driving phase timing of that level is extracted. This calibration is based on the zero-crossing detection of the rising edge of the control signal. The digital comparator scans in real time with a resolution of 0.1ms. A rising edge trigger moment is confirmed when the amplitude of three consecutive sampling points exceeds 50% of the rated peak value U0 of the control signal. The length of the calibration trigger moment sequence for each level is not less than 100 consecutive oscillation cycles, and the acquisition time for a single level is about 500ms (based on 200Hz). The phase average of all rising edge trigger moments within the same level is taken to obtain the driving phase timing reference for that level. The sampling window for the heavy load index layer of the gain rule group is set to 200ms, and 150ms for light and medium loads. The heavy load window is longer to cover the phase jitter convergence process caused by load fluctuations. When the razor is continuously shaving in the dense beard area of ​​the chin, the heavy load phase moment jitters continuously by ±0.3ms within 50ms. Therefore, the heavy load index layer must wait for the jitter to converge (the amplitude of 5 consecutive cycles is less than ±0.1ms) and then take the average value as the effective phase moment to avoid the transient contamination of statistics by the load impact. After the gain rule group is calibrated, the effective phase time averages of the light, medium, and heavy levels are arranged in the order of the operating condition index to form a multi-operating condition phase time group. Each entry in the multi-operating condition phase time group synchronously records the operating condition level, the measured average operating frequency within the calibration window, and the drift early warning indicator level. Entries with a measured average value that differs from the nominal 200Hz by more than ±3Hz are marked with a temperature drift indicator. When analyzing the time difference between adjacent operating conditions, the systematic offset caused by the inherent frequency difference (converted at 0.005ms / Hz) is first deducted from the measured time difference to ensure that each entry in the multi-operating condition phase time group reflects the true operating condition phase characteristics.

[0051] Phase time difference sequences are obtained by performing adjacent time difference analysis on multi-condition phase time groups. The adjacent time difference T_lm(n) between light and medium loads in the multi-condition phase time group is defined as the time difference between the medium and light load phases at time n. The adjacent time difference T_mh(n) between medium and heavy loads is defined as the time difference between the heavy load phase and the medium load phase at time n. The two are concatenated to form a complete phase time difference sequence, with a total length twice that of a single-condition time sequence. The phase time difference sequence entries with added frequency offset indicators have had systematic offsets deducted before time difference calculation; the normal range for T_lm(n) is 0.4-0.6 ms, and the normal range for T_mh(n) is 0.6-0.9 ms. After prolonged use, the spring stiffness of the shaver decreases, causing T_mh(n) to become systematically large. The heavy-load phase moment drifts backward by 0.3-0.5ms. In this scenario, T_mh(n) continuously exceeds the normal upper limit of 0.9ms and remains there for more than 20 time points. When the consecutive time points exceeding the limit exceed 30, a spring decay warning is triggered. The threshold corresponds to a stiffness decay of about 11%. After the warning is triggered, each column G(i,j) of the heavy-load index layer of the gain rule group is automatically increased by 0.01 in the next 50ms step cycle (clamping occurs when the increase exceeds 0.9), and the spring decay characteristic is written into the current task gain record for additional risk correction during the amplitude stability index mapping stage. The point-by-point sum of T_lm(n) and T_mh(n) is defined as the phase extension amount E_ext(n) across three gears, with a normal range of 1.0-1.5ms. An E_ext(n) higher than 1.5ms indicates an abnormal phase accumulation and extension across gears, while a value lower than 1.0ms indicates phase compression in the middle gear. These are marked with expansion and compression abnormality indicators, respectively. When the density of expansion abnormality indicators reaches more than three per 10ms, a cross-gear aggregation abnormality indicator is triggered. This indicator, along with its density value, is marked in the corresponding time period of the phase time difference sequence. Because the number of effective phase moments in each gear of the multi-condition phase time group may be unequal, the other two gears are downsampled proportionally based on the shortest gear sequence length before splicing to ensure that T_lm and T_mh can be paired point-by-point. Expansion abnormalities often correspond to the gradual accumulation of phase across gears and are commonly seen when the overall spring stiffness decreases. Compression abnormalities often correspond to overcompensation of gain in the middle gear, causing the phase to be artificially pulled closer. Both types of indicators are incorporated into the phase time difference sequence along with the corresponding time.

[0052] Anomalous time difference segments are identified by performing high-fluctuation and low-mean segment identification on phase time difference sequences. For phase time difference sequences T_lm(n) and T_mh(n), the mean μ(w) and standard deviation σ(w) within each window are calculated using a sliding window of 10 time points (approximately 50ms). σ(w) exceeding 1.5 times the standard deviation of the entire sequence is marked as a high-fluctuation window, while μ(w) below the corresponding lower limit of the normal range (0.4ms for T_lm group and 0.6ms for T_mh group) is marked as a low-mean window. When both labels coexist within the same window, the segment is included in the candidate set of anomalous time difference segments. A valid anomalous time difference segment is confirmed when the high-fluctuation window and low-mean window overlap for at least 3 sliding windows (150ms). If the overlap is less than 150ms, it is downgraded to a suspected segment. A reassessment is performed after 200ms; if the overlap is still less than 150ms, the segment is removed from the candidate set. When a shaver switches from wet to dry shaving mode, liquid enters the shaving head, causing a temporary increase in the resistance of the oscillator. T_mh(n) exhibits a V-shaped characteristic of first dropping sharply and then rising within 3-5 seconds after the switch. The low-mean value of the sharp drop segment typically lasts less than 100ms, failing to meet the 150ms threshold and thus not forming a valid segment. This filters out transient disturbances during wet shaving switching, preventing misjudgment of environmental switching as motor characteristic degradation. The duration threshold for the cross-gear aggregation anomaly identification segment is lowered to 100ms, with a merging weight of 1.8. The normal anomaly time difference segment is weighted at 1.0 to ensure that cross-gear composite phase anomalies obtain spectral peak intensities commensurate with the risk level in the phase shift characteristic spectrum. Those that meet the criteria of high volatility but not low mean are marked as single high volatility segments, and vice versa, they are marked as single low mean segments. These two types of single segments do not form abnormal time difference segments, but are marked with dashed reference contours in the phase shift characteristic spectrum to distinguish them from solid abnormal time difference segments. The ratio of the amplitudes of the two contours reflects the relative proportions of systematic components and random disturbance components in the phase time difference sequence.

[0053] Phase shift feature spectra are output by merging and integrating the steady-state segments within the abnormal time difference segments. The steady-state segments within the abnormal time difference segments are those where σ(w) / μ(w) is below 0.5 in both T_lm and T_mh groups. Adjacent segments are merged within a 50ms merging window; segments with a gap of less than 50ms are merged into a single merged segment. Within each segment, T_lm and T_mh are weighted averages based on the duration of the sub-segments. When cross-level aggregation segments and ordinary segments enter steady-state merging simultaneously, the former's weighting coefficient is 1.8. The ratio of the mean σ(w) to the mean μ(w) of the merged section, R_norm, is converted into the phase shift estimate Φ(f) at each frequency point using the following formula: Φ(f) = arctan(R_norm(f) × k_scale) × (180 / π), where k_scale is a dimensionless conversion coefficient determined by the mean gain distribution vector of the current operating condition (18 for light load, 22 for medium load, and 28 for heavy load), Φ(f) is in degrees, and f is the frequency axis of the phase time difference sequence transformation. When R_norm exceeds 1.0, Φ(f) exceeds 83° in the heavy load section, and the merged section is directly confirmed as a severely abnormal time difference section and skips the suspected stage. The phase shift characteristic spectrum uses the merged frequency distribution as the skeleton, with the center frequency of the merged section corresponding to the peak frequency. The peak width (in Hz) = 1000 / the time span of the merged section (in ms). Adjacent peaks with a spacing of less than 1 Hz are merged into a broad peak, and the amplitude of the broad peak is the area-weighted average of the two peaks. In the phase shift characteristic spectrum, frequency bands with solid line peak amplitudes exceeding 45° directly identify abnormal characteristic points. Single high-fluctuation segments outside the steady-state segment are marked with dashed outlines in the phase shift characteristic spectrum. When a shaver operates under rated load for a long time, causing the spring stiffness and actuator thermal resistance to degrade synchronously, the solid line peak and the dashed reference area expand synchronously, and the center frequency drifts to lower frequencies. A drift exceeding 5Hz indicates that the composite degradation has exceeded the upper limit of the single-term correction capability.

[0054] Anomaly detection is performed to determine the location of abnormal characteristic points based on deviation exceeding limits in the phase shift characteristic spectrum. The offset Φ(f) at each frequency point of the phase shift characteristic spectrum is compared point-by-point with the corresponding allowable deviation upper limit (2.5° for light load, 3.5° for medium load, and 5.0° for heavy load). The allowable deviation upper limit increases with the load to accommodate the larger inherent phase tolerance under heavy load conditions, reflecting a graded design of parameter correction sensitivity under different loads. When the solid line peak Φ(f) of the phase shift characteristic spectrum exceeds the corresponding upper limit and the continuous exceeding frequency band is not less than 3Hz, it enters the candidate judgment. This 3Hz threshold reduces the probability of random noise triggering at a single frequency point from approximately 18% to less than 2%, ensuring that the abnormal characteristic point represents a true systemic phase shift. Candidate points exceeding the allowable upper limit by more than 20% are directly confirmed as formal abnormal characteristic points; those less than 20% are initially marked as suspected, and a second comparison is performed with the latest calculation results after 500ms. If the deviation still exceeds the limit, it is converted to a formal abnormal characteristic point. Otherwise, the process is cancelled. The secondary confirmation mechanism filters out transient over-limits caused by short-term load impacts. Different frequency bands correspond to different physical fault modes for abnormal characteristic points: in the low frequency band (below 20Hz), over-limits indicate that the spring stiffness has decayed, causing low-frequency mechanical resonance. For every 1° increase in the over-limit amplitude, the stiffness has decayed by about 2%. In the high frequency band (above 80Hz), dense over-limits are a typical feature of hair entanglement in the blade mesh. When the density of abnormal characteristic points in the high frequency band reaches more than three per 10Hz, a hair entanglement warning is triggered. The entanglement intensity is divided into three levels: mild (3-4 times), moderate (4-5 times), and severe (more than 5 times). The corresponding weights of the four levels increase sequentially from 1.8 to 2.2. Abnormal characteristic points are arranged in descending order of over-limit amplitude. The first 5 are given priority to enter the activation process, and the remaining formal abnormal characteristic points are temporarily reserved as candidates and will be filled in order after the first 5 are processed.

[0055] Step S104: Obtain the excitation sequence by activating the gain rule group based on the abnormal characteristic points, perform pseudo-convergent oscillation detection on the excitation sequence to obtain the deviation density sequence, and perform reverse correlation detection on the deviation density sequence and the excitation sequence to output the amplitude stability index.

[0056] Specifically, the excitation sequence is obtained by activating the gain rule group based on the abnormal characteristic points. The first 5 abnormal characteristic points are located one by one in the 3D index of the gain rule group according to the descending order of the over-limit amplitude, corresponding to the working condition level and the drift early warning indicator level. After location, the gain vector of the index combination is output to the control parameter configuration interface in the parameter writing mode, with a writing delay of no more than 0.5ms. A timing excitation pulse lasting 200ms is started from the trigger time of the abnormal characteristic point. When the shaver triggers the warning due to hair entanglement, the abnormal characteristic points appear densely in the 100-120Hz high frequency band, and the high frequency index layer G(i,4) is activated synchronously. In this scenario, the excitation sequence contains 3-5 dense pulses with a phase interval of less than 20ms. Pulses with an interval of less than 15ms are merged into a single wide pulse to prevent over-excitation oscillation. The duration of the merged wide pulse is the sum of the two pulses, with an upper limit of 400ms. When the abnormal characteristic point exceeds the allowable upper limit by 50%, the corresponding index combination level four column G(i,4) adds an additional 0.05 single-increment compensation on top of the standard pulse. If the added value exceeds 0.9, clamping occurs. This is applied only once per pulse without accumulation to prevent overcompensation. If the average actual displacement within 50ms after an excitation pulse application does not reach 85% of the target displacement, it is judged as under-response and marked with an under-response flag. If the under-response flag appears three times consecutively, the corresponding gain rule group index layer G(i,4) is automatically increased by 0.02 before the next pulse trigger. If the value exceeds 0.9, clamping occurs. The pulses are arranged in the order of triggering time to form an excitation sequence. The total number of pulses equals the effective count of activated abnormal characteristic points after pulse merging. When the maximum silence interval between pulses exceeds 500ms, a sustain pulse is inserted at the midpoint. Its gain vector is the linear average of the preceding and following pulses to maintain continuous tracking capability of motor characteristic drift.

[0057] A deviation density sequence was obtained by performing pseudo-convergent oscillation detection on the excitation sequence. During the application of the excitation sequence, the actual displacement was compared with the target displacement periodically, using a single oscillation period (5ms) as the statistical unit. The absolute value of the period deviation |e(k)| being less than 0.05mm was used as the convergence criterion. If the convergence criterion was met for three consecutive oscillation periods but the sign of e(k) alternated between positive and negative, it was determined to be a pseudo-convergent oscillation. The alternation determination required at least two sign flips within three periods. Random single flips caused by pure noise were not triggered. During the underresponse indicator pulse coverage period, the convergence threshold of |e(k)| was relaxed from 0.05mm to 0.08mm to compensate for the omissions caused by the large basic deviation in the underresponse scenario. After relaxation, the pseudo-convergent oscillation capture rate increased from about 65% to 82%. The local deviation density D_local is obtained by summing |e(k)| in 10ms windows (approximately 2 oscillation cycles). Arranged according to the trigger time, this forms a complete sequence. The mean D_local value in the pseudo-convergent oscillation segment is typically 2.5-4 times that of the non-oscillation segment. Under heavy load conditions, after triggering gain compensation, the oscillator displacement overshoots. When the overshoot exceeds 10% of the target displacement, e(k) changes from negative to positive and then quickly rebounds to negative. This overshoot rebound is highly similar to the alternating sign of the pseudo-convergent oscillation. The difference lies in the fact that the |e(k)| of the overshoot rebound decays to within the convergence threshold after the second oscillation cycle, while the |e(k)| of the pseudo-convergent oscillation fluctuates near the threshold without decaying for more than 3 cycles. Therefore, the overshoot rebound segment D_local in the deviation density sequence drops rapidly after the second window, while the pseudo-convergent oscillation segment remains high for more than 30ms. Based on this, overshoot and pseudo-convergent oscillation indicators are marked and written into each pulse record of the excitation sequence.

[0058] In some embodiments, the step of performing reverse correlation detection based on the deviation density sequence and the excitation sequence to output an amplitude stability index includes: grouping the excitation sequence by load type to obtain a load response group; performing reverse correlation analysis on the load response group and the deviation density sequence to form a load correlation deviation group; performing cross-load reverse correlation strength evaluation based on the load correlation deviation group to determine the reverse correlation strength parameter; and mapping the reverse correlation strength parameter according to the priority of low reverse correlation strength intervals to output an amplitude stability index.

[0059] The excitation sequence is grouped by load type to obtain load response groups. The control duty cycle corresponding to the trigger time of each pulse in the excitation sequence is extracted, and the pulse is classified into one of three categories according to the dual thresholds of 55% and 65%: light load (<55%), medium load (55%-65%), or heavy load (>65%). Pulses of the same category and their corresponding segments in the deviation density sequence are grouped into the same level of the load response group and indexed independently. This ensures that each level of the load response group contains all the excitation pulses in that level and carries the corresponding deviation density sequence segments for pulse-by-pulse retrieval in subsequent pre-deviation time series baseline analysis. Pulses in the excitation sequence whose trigger moment control duty cycle falls within the boundary band of 55%±2% or 65%±2% are marked with a cross-gear indicator. Adjacent gears are also included in the reverse correlation analysis, with each gear weighted at 0.6 to avoid double-weighting. When the number of such pulses exceeds 15% of the total number of groups, a gear boundary blurring alarm is triggered. The reverse correlation rate (CR) for each gear is defined as the proportion of pulses in the gear's deviation density sequence that show an inverse change after excitation triggering relative to before triggering. When the boundary is blurred, the CR result is marked with a ±5% confidence interval. When shaving in the upper lip area, the load of the shaver periodically switches between light and medium loads, with the two gear groups alternating, each lasting approximately 1-2 seconds. The pulse quantity ratio is approximately 1:1.3. The mean D_local value of the medium load group is about 22% higher than that of the light load group, reflecting a relative decrease in motor amplitude stability under medium load in this area. When this difference exceeds 35%, the medium load group is marked with a high-load sensitivity indicator, and this group is weighted 1.3 times in the cross-load reverse correlation strength assessment. Pulses within each load response group are arranged according to their trigger time. Pulses with under-response markings retain their markings and participate in reverse correlation analysis with a weight of 1.5. Groups with fewer than 3 pulses are marked with insufficient samples, and their analysis results have a confidence coefficient reduced to 0.5 and are archived separately. Load type groups are updated incrementally in real time with the excitation sequence. When a new pulse in the excitation sequence is triggered, its duty cycle is controlled according to its trigger time and it is incorporated into the load response group without waiting for the entire segment to end.

[0060] For example, the step of performing reverse correlation analysis on the load response group and the deviation density sequence to form a load-related deviation group includes: performing pre-deviation time-series baseline analysis based on the load response group to obtain a load time-series baseline set; performing cross-load time-series misalignment analysis on the load time-series baseline set and the deviation density sequence to obtain a time-series deviation sequence; performing high-misalignment and reverse-changing time periods identification based on the time-series deviation sequence to obtain reverse correlation features; and performing deviation quantification and summarization based on the reverse correlation features to form a load-related deviation group.

[0061] Based on load response grouping, pre-deviation time-series baseline analysis is performed to obtain the load time-series baseline set. The pre-deviation time-series baseline analysis is carried out pulse-by-pulse within each level of the load response group: the deviation density sequence segment 100ms before the trigger of each excitation pulse in that level is taken as the pre-deviation time-series baseline of that pulse. The 100ms window covers about 20 oscillation cycles to ensure the statistical stability of the baseline steady-state estimation. The mean D_local value within the segment is recorded as the pre-deviation baseline value μ_pre of that pulse. The μ_pre of all pulses in the same level are arranged according to the trigger time to form the pre-deviation baseline subsequence of that level. The subsequences of the light, medium and heavy levels are summarized to form the load time-series baseline set. Each entry in the load time-series baseline set records the level, the corresponding pulse index and the μ_pre value. The normal range of μ_pre under light load is 0.02-0.08 mm / cycle, under medium load 0.05-0.12 mm / cycle, and under heavy load 0.08-0.18 mm / cycle. The normal range increases with the load, reflecting that the heavier the load, the higher the natural background level of motor vibration deviation. If μ_pre continuously and monotonically increases within a single task at the same load level, it usually corresponds to a slow rise in the deviation base caused by the accumulation of actuator thermal effects. If there is a pseudo-convergent oscillation segment within the 100ms window before triggering, μ_pre excludes this segment and takes the average of the remaining sampling points. If the exclusion ratio exceeds 50%, an unstable baseline indicator is added. This type of pulse is weighted 0.6 times in the timing misalignment analysis. In high humidity environments (relative humidity exceeding 85%), the water film on the shaver head increases damping, and the overall μ_pre of the heavy load group is about 15%-20% higher. When the proportion of pulses with μ_pre exceeding the normal upper limit in the heavy load subsequence exceeds 60%, a high damping condition indicator is triggered. After activation, the corresponding D_local pseudo-convergence judgment threshold is raised by 12%, and the normal upper limit of μ_pre under heavy load in the load timing baseline set is temporarily extended to 0.22mm / cycle to adapt to wet shaving deviation characteristics. Pulses in each gear sequence whose μ_pre standard deviation exceeds 30% of the load response group mean are marked with a high fluctuation baseline indicator. When the trigger frequency of this indicator exceeds 20% of the total number of pulses in the group, it indicates that the pulse trigger time of this gear is systematically close to the reversal time.

[0062] A time series deviation sequence is obtained by performing cross-load time series misalignment analysis on the load time series baseline set and the deviation density sequence. For each gear in the load timing baseline set, the mean value of the baseline before the sequence deviation μ_pre is paired with the mean value of the deviation density sequence μ_post within a 100ms window after the corresponding pulse excitation trigger. The misalignment δ(n) = μ_post - μ_pre (unit: mm / cycle). A negative δ(n) indicates that the deviation density decreases from the baseline level after excitation and the gain compensation is effective. A positive δ(n) indicates that the deviation density increases after excitation and the compensation direction fails. The larger the difference between the two, the more significant the excitation's effect on the deviation density control. Pulses with |δ(n)| below 0.005 mm / cycle are considered zero-bias pulses and are not included in subsequent high misalignment identification to avoid quantization errors interfering with statistical conclusions. The δ(n) values ​​of the light, medium, and heavy gears are sequentially spliced ​​according to the trigger time to form a complete timing deviation sequence. A separator is inserted at the gear switching boundary of the timing deviation sequence. Adjacent differences are not calculated on both sides of the separator to prevent numerical jumps of different load characteristics from being misjudged as high misalignment events. The mean value of |δ(n)| of the three gears of a normal motor should increase with the load. The gain increases in a stepwise manner, reflecting the strong control capability of gain compensation over high load deviations. If the mean value of |δ(n)| in the heavy load setting is lower than that in the medium load setting, it indicates that the heavy load compensation is saturated or the direction is wrong. This setting needs to be focused on when extracting reverse correlation features. When a razor cuts coarse beard, the impact reaction force causes D_local to rise sharply within 2-3 vibration cycles and then fall back. This kind of artificially high μ_post caused by external impact will make δ(n) positive and misjudged as output in the same direction. Therefore, the impact interference flag pulse is changed to the mean value of the deviation density sequence after the sharp rise ends and μ_post is recalculated. When the difference between the corrected δ(n) and the original value exceeds 0.02mm / cycle, an impact correction amplitude flag is added and the confidence coefficient is reduced to 0.85. The load time series baseline set baseline instability flag, high fluctuation baseline flag and impact interference flag (corrected) pulses are weighted by 0.6, 0.7 and 0.8 times respectively, for identification and differential use during high misalignment periods of the time series deviation sequence, to ensure that low-quality pulses do not excessively affect the statistical conclusions.

[0063] Reverse correlation features are obtained by identifying high-misalignment and reverse-change periods in the time-series deviation sequence. When the proportion of high-misalignment pulses in each subsequence of the time-series deviation sequence exceeds 40%, that subsequence is marked as a high-misalignment concentration subsequence. Pulses whose |δ(n)| exceeds 1.5 times the mean of the entire sequence are defined as high-misalignment pulses (the mean calculation does not include zero-misalignment pulses). When two adjacent pulses have opposite δ(n) signs and both meet the high-misalignment condition, the pulse pair is determined to have reverse-change features. Reduced-weighted entries (weights below 0.8) must have |δ(n)| exceeding twice the mean of the entire sequence to participate in the judgment, in order to compensate for the slack threshold caused by the weight reduction. When three or more pairs of reverse-change pulses appear consecutively, a reverse-change segment is formed. Its start and end pulse indices and δ(n) sign sequence are recorded together as an entry of the reverse correlation feature. If the proportion of reduced-weighted entries in the segment exceeds half, a low-confidence reverse segment label is added. This entry is weighted down to 0.6 when the deviation quantification is summarized. The time-series deviation sequence requires at least three consecutive pairs of reverse change segments. This is because single or double pairs of reverse changes are easily caused by random noise, and three consecutive pairs can reduce the false acceptance probability to below 5%. If a load range does not form any reverse change segments in the entire statistical period, the reverse correlation feature of that range is recorded as empty, and conservatively filled with the lower limit of the historical average of that range during quantization and summarization. A "no reverse segment" label is attached for cross-load strength assessment to identify its lack of suppression capability. When a shaver switches from light load to heavy load, the δ(n) of the last segment of light load is usually negative (gain compensation is effective), and it turns positive at the beginning of heavy load due to the sudden increase in load. If the difference of the reverse change pulse pairs |δ(n)| at the light-to-heavy switch boundary exceeds 50% of the average of the light load subsequence, it is marked as a high-intensity switch type. This item is weighted by 1.6 times during deviation quantization and summarization. The reverse correlation features are arranged in descending order of the number of reverse change segment pulse pairs for each item. When the number is the same, the one with the larger average |δ(n)| is given priority.

[0064] Based on the reverse correlation characteristics, deviation quantification and summation are performed to form a load-related deviation group. Within each segment of the reverse correlation characteristics, |δ(n)| is weighted and summed according to pulse identifier weights (1.6 for high-intensity switching, 0.6 for low-confidence reverse segments, 1.5 for pseudo-convergent oscillation, 0.8 for overshoot, and 1.0 for none; the highest weight is used when a pulse has multiple identifiers simultaneously), and then divided by the total number of pulse pairs to obtain the weighted average reverse deviation Q_rev. The larger Q_rev is, the more severe the reverse response of the deviation density after the excitation is triggered, meaning the gain rule group currently assigned to this segment has a stronger deviation suppression capability. Each item of the reverse correlation characteristics is categorized according to the load segment. Within the same segment, Q_rev is weighted by the number of reverse change segment pulse pairs to obtain the segment-representative reverse deviation Q_avg. Q_avg and the standard deviation σ_δ of the corresponding segment's δ(n) together constitute the deviation quantification result for that segment. The higher Q_avg / σ_δ is, the more stable the reverse correlation response; below 0.5 indicates a critically unstable state. The Q_avg and σ_δ values ​​for light, medium, and heavy loads, along with the corresponding CR values, constitute three entries in the load correlation deviation group. For loads with insufficient samples, Q_avg is filled with 50% of the lower limit of the normal range for that load, and σ_δ is filled with 120% of the upper limit of the normal range, conservatively reflecting the impact of insufficient samples on the amplitude stability index mapping. After the shaver runs continuously under rated load for 6 minutes, the actuator thermal effect increases, and the number of pulse pairs in the reverse change segment of the heavy load range decreases from the normal 8-12 pairs to 3-5 pairs. The Q_avg value in the heavy load range of the load correlation deviation group decreases from 0.12-0.18 mm / cycle to below 0.06 mm / cycle. When Q_avg is below 0.06 mm / cycle and Q_avg / σ_δ is below 0.5, the joint judgment indicates a significant degradation in the reverse correlation capability of the heavy load range, and the heavy load index layer G(i,4) of the gain rule group is automatically increased by 0.01 in the next 50ms step cycle.

[0065] The reverse correlation strength parameters are determined by evaluating the cross-load reverse correlation strength based on the load correlation deviation group. The three CR levels are weighted averaged according to the total number of excitation pulses of each level to obtain the overall reverse correlation rate CR_total for the entire working condition. A CR_total higher than 55% indicates that the current gain allocation has a strong reverse suppression capability against load changes. A CR_total lower than 25% indicates that the gain allocation is highly consistent with the actual load response and the suppression is insufficient. The overall allocation direction of the gain rule group needs to be re-evaluated. 25%-55% is a transition state. In the transition state, the corresponding working condition line G(i,3) is increased by 0.01 in the next cycle for gradual correction. The standard deviation σ_CR of the three CR levels quantifies the uniformity of cross-load suppression. σ_CR lower than 5% indicates that the suppression capabilities of the three levels are comparable and the gain allocation is stable and consistent on the load axis. A σ_CR higher than 15% usually indicates that the CR is higher under light and medium loads and lower under heavy loads. This indicates that the deviation under heavy load conditions has exceeded the current gain coverage range. The gain allocation is effective for light and medium loads but ineffective for heavy loads. The CR deviation of the high load sensitivity group is weighted by 1 in the σ_CR calculation. The value is multiplied by 3 to amplify its impact on uniformity assessment, preventing its degradation from being masked by the good performance of other groups; the comprehensive index IS=CR_total / (1+σ_CR) reflects the suppression uniformity (substituting the decimal). The larger the IS, the stronger and more uniform the reverse correlation ability of each level. When IS is below 0.2, the upper limit of the fourth-level column gain of the overloaded index layer is expanded from 0.9 to 0.95 to release more compensation space. When IS is above 0.55, the fourth-level column clamp value is contracted from 0.9 to 0.85 to prevent overcompensation from causing oscillation. When IS is between the two, the upper limit of gain remains unchanged. After a long shave in a high-density beard area, the IS of the shaver continues to drop from the normal 0.4. It usually triggers a transition state first and then enters the low IS region, which is an early signal of thermal drift accumulation. The three items CR, σ_CR and IS of each level constitute the core value of the reverse correlation strength parameter. The CR of the spring decay characteristic mark level is separately marked to distinguish the contribution source of structural decay and random bias.

[0066] The amplitude stability index is output by mapping the inverse correlation strength parameter according to the priority of the low inverse correlation strength interval. Among the inverse correlation strength parameters, IS is divided into three levels according to the bench calibration range (mean 0.42, lower limit of 3σ 0.28): high (IS≥0.42), medium (0.28≤IS<0.42), and low (IS<0.28). The mapping priority of the low level is higher than that of the medium and high levels: when IS falls into the low level, the amplitude stability index is directly mapped to the 0-0.4 interval without the interference of other statistics. The mapping function is IS / 0.28×0.4 (0.0 when IS is 0, and 0.4 when IS reaches exactly 0.28). The mid-range and high-range are linearly mapped to their respective interval endpoints (mid-range mapped to 0.4-0.7, high-range mapped to 0.7-1.0). The piecewise functions for the three ranges are continuous at the boundaries to avoid abrupt changes in the amplitude stability index during range switching. Before the final output, an exponential moving average with a time constant of 200ms is used for smoothing to suppress exponential jitter caused by instantaneous IS fluctuations. When the inverse correlation strength parameter σ_CR exceeds 15%, the mid-range and high-range mapping intervals shrink by 10% towards the low-range (mid-range upper limit decreases from 0.7 to 0.63, high-range upper limit decreases from 1.0 to 0.9). If the spring decay characteristic indicates an abnormal range CR, it is further shifted down by 0.05 to reflect the additional risk of structural decay to overall amplitude stability. After the shaver runs continuously under rated load for 8 minutes, the spring stiffness and actuator thermal resistance degrade synchronously, and the IS value remains below 0.28. The amplitude stability index is mapped to the 0.15-0.25 range. When it falls below 0.2, the positive deviation rows G(5,3) and G(5,4) of the gain rule group are each increased by 0.05 to compensate for the degradation immediately. If the value exceeds 0.9 after the increase, clamping occurs. If this automatic increase is triggered more than 5 times in a single run, the amplitude stability index mapping range shifts down by one level. The amplitude stability index is finally output with a resolution of 0.001 and is archived along with the IS, σ_CR, and spring attenuation characteristic indicator level.

[0067] Step S105: Combine the amplitude stability index and the excitation sequence to perform deviation interval analysis to obtain the deviation interval sequence. Perform deviation high-incidence section calibration on the deviation interval sequence and the gain rule group to construct the control characteristic map. Compare the control characteristic map with the control input data to perform parameter deviation mapping and output adaptive control parameters.

[0068] Specifically, a deviation interval sequence is obtained by combining the amplitude stability index and the excitation sequence deviation interval analysis. Periods with an amplitude stability index below 0.3 are defined as high-risk windows. Within these windows, the interval between the trigger times of adjacent excitation pulses is extracted in pairs at a resolution of 0.1ms (including sustain pulses), and arranged according to trigger times to form high-risk segments of the deviation interval sequence. For low-risk windows with an amplitude stability index above 0.3, the interval between adjacent pulses is extracted simultaneously to form low-risk segments. These two types of segments are alternately spliced ​​according to the actual timing to form a complete deviation interval sequence. Boundary identifiers are inserted at the splicing boundaries, and adjacent differences are not calculated on either side of the identifiers. After the shaver switches to heavy-load operation, the actuator temperature rise triggers thermal drift. The amplitude stability index rapidly drops from above 0.5 to below 0.2 within 10-15 seconds, and the interval between adjacent pulses shrinks from the normal 120-150ms to 40-60ms for densification compensation. At this time, the average interval of the high-risk segments of the excitation sequence shrinks to 35%-45% of that of the low-risk segments. A shrinkage rate exceeding 60% indicates that thermal drift has exceeded the boundary of the gain rule group's conventional compensation capability. When the amplitude stability index frequently crosses the threshold between 0.28 and 0.32, the excitation triggering mode repeatedly switches between dense and sparse, and the standard deviation of the corresponding segment spacing exceeds 50% of the mean. This type of high-fluctuation segment is labeled with a threshold jitter indicator, and its weight is reduced to 0.6 when labeling segments with high deviation rates. Pulse pairs with an absolute spacing of less than 20ms are labeled with an ultra-dense indicator. When three or more consecutive pairs appear, the corresponding index layer triggers over-excitation protection, forcibly raising the minimum trigger spacing of subsequent excitation sequences to 25ms.

[0069] In some embodiments, the step of performing deviation high-incidence segment calibration and constructing a control characteristic map by the deviation interval sequence and the gain rule group includes: statistically analyzing the high-incidence periods of deviation for each operating condition in the gain rule group to obtain a high-incidence density distribution; performing intermittent high-density segment location on the high-incidence density distribution to determine a benchmark high-incidence segment; performing difference compensation on the benchmark high-incidence segment and the deviation interval sequence to determine a characteristic compensation amount; and performing control characteristic interval mapping based on the characteristic compensation amount to construct a control characteristic map.

[0070] The high-incidence density distribution of deviations under various operating conditions was obtained by statistically analyzing the high-incidence periods of the gain rule group. The activation time of each activation combination (operating condition gear × drift early warning indicator gear × phase segment type) in the gain rule group was extracted from the excitation sequence by index. The number of activations within the statistical window was calculated using a 100ms window and a 10ms step size. The resulting sequence reflects the activation density of this combination on the time axis. The heavy load × severe drift × advanced phase segment combination (index (3,4,1)) typically had a much higher activation density than the other 23 combinations during continuous high load phases. When its mean was more than three times the mean of the entire index, it was marked as extremely high-incidence and given priority in locating the baseline high-incidence section. The activation density sequences of the 24 combinations were normalized within the operating condition level dimension (the baseline being the average activation density of the three levels below the corresponding level). Normalized values ​​were compared within the range of 0-3, with periods exceeding 1.5 marked as high-incidence periods. High-incidence periods for each combination were superimposed along the time axis, but the combinations were not weighted equally: combinations containing heavily drifted layers or leading phase segments were counted at 1.5 times the weight, while those with light or no drift layers were counted at 1.0 times the weight, to ensure that the contribution of higher-risk index layers was commensurate with their physical hazard. Combinations with zero activations within the statistical window were not included in the superposition at that moment to avoid diluting the high-incidence characteristics of long-term inactive combinations. Higher superposition values ​​indicate more index layers simultaneously exhibiting high incidence at that moment. The high-incidence density distribution was constructed with the time axis as the horizontal axis and the normalized density superposition value as the vertical axis: superposition values ​​exceeding 4.0 marked multi-index simultaneous occurrence areas, with a width exceeding 200ms indicating a cross-operating condition systematic deviation in that period; superposition values ​​below 0.5 marked low-incidence areas, corresponding to the sparse segment contour filling position of the control characteristic spectrum. The superimposed values ​​at each time point are stored at a resolution of 0.1 in the high-incidence density distribution for use in the location of intermittent high-incidence segments. Once the extremely high-incidence flag is triggered, in addition to participating in the location of the baseline high-incidence segment first, it is also written back to the index layer corresponding to the gain rule group, so that the back-off rate of G(i,j) in the subsequent 50ms step is halved, in order to extend the gain maintenance time of high-risk combinations.

[0071] Intermittent high-density segment positioning is used to determine the benchmark high-density segment for high-density distribution. A continuous period with a superposition value exceeding 2.5 is defined as a high-density segment. Within a segment, a single drop in superposition value below 2.0 lasting no more than 50ms is considered a brief, uninterrupted fluctuation. Drops exceeding 50ms constitute independent segments on either side of the dividing point. This 50ms threshold is set 10 times the motor commutation cycle to filter out brief density fluctuations caused by commutation impacts without breaking down continuous deviation concentration areas into fragments. The alternating pattern of high-density segments and low-density areas (superposition value below 0.5) in the high-density distribution is called intermittent high density, corresponding to the alternating process of periodic contact between the shaver head and the skin, triggering of deviation concentration, and natural recovery. When the shaver reciprocates on the flat area of ​​the cheek, the contact cycle is approximately 1-2 seconds, with the superposition value rising sharply to 3.0-4.5 upon contact and dropping back to below 0.3 upon lifting. Non-intermittent high density (high-density segments are continuous)... If there is no interruption in the low-frequency area for more than 3 seconds, it indicates continuous motor instability. The judgment threshold is increased from 2.5 to 3.5 to focus on the most serious deviation concentration section and avoid misclassifying the entire continuous high-frequency section into many fragmented benchmarks, thus dispersing compensation and correction resources. When the razor is turned to the chin arc or mandible, the contact surface contour is more complex, the duration of the high-density section is significantly extended, and the peak value of the superposition value is also higher. This can be used to distinguish the deviation concentration characteristics of two scenarios: planar shaving and continuous contour shaving, and trigger different intensities of compensation and correction accordingly. The benchmark high-frequency section is included in the list in order of the longest duration among the high-density sections. The multi-index homonymous area participates in the subsequent difference compensation calculation with a weight of 1.8 times. The list is arranged in descending order of the product of duration × weight. The first 10 are used as the main benchmark high-frequency sections to participate in the characteristic compensation calculation, and the rest are downgraded to reference benchmarks to control the expansion of the characteristic map boundary.

[0072] The characteristic compensation amount is determined by differentiating between the benchmark high-incidence segment and the deviation interval sequence. The difference ΔI between the mean interval of the deviation interval sequence corresponding to the time period in the benchmark high-incidence segment and the mean interval of the global low-risk sub-segment is defined as the segment interval deviation: a negative ΔI (narrowing interval) indicates that the excitation in this segment is forcibly densified, while a positive ΔI indicates that it is still under sparse excitation, and the compensation directions are opposite in the two cases. If the deviation interval sequence |ΔI| exceeds 30% of μ_ref, it is determined that there is a significant interval deviation. For such segments, the mean vector of the corresponding activation combination G(i,j) is extracted from the excitation sequence snapshot, and its product with the mean of the corresponding segment D_local is defined as the segment deviation intensity index BP (BP is in mm / cycle, the mean of G(i,j) is dimensionless, and the mean of D_local is in mm / cycle); the larger the BP, the stronger the combined effect of the gain configuration and deviation density in this segment, and the more urgent the compensation requirement. When the shaver is working under heavy load at high speed (220Hz), the average BP value in the high-incidence area is about 28% higher than that at low speed (180Hz). Because the high speed has a higher reversal impact frequency, the spacing narrows more drastically. Therefore, the characteristic compensation amount of the high speed component is multiplied by a frequency correction factor of 1.2 on the standard calculated value. The weighted average of the top 10 high-incidence sections of the main benchmark (BP) is obtained by normalizing the duration and weights to obtain the comprehensive BP mean BP_avg. Based on this, the control duty cycle compensation component ΔDC_comp and the operating frequency compensation component ΔF_comp of the characteristic compensation amount are calculated: ΔDC_comp = BP_avg × α_dc, ΔF_comp = BP_avg × α_f, where α_dc is in the unit of %·cycle / mm, determined by interpolation of the grid density of the current landing point of the control characteristic map (0.3-0.8%·cycle / mm), and α_f = 0.15Hz·cycle / mm. Thus, ΔDC_comp is in the unit of % and ΔF_comp is in the unit of Hz. The upper limits of the two components are ±5% for the control duty cycle and ±3Hz for the operating frequency, respectively. Exceeding the limits results in clamping. The BP of the reference high-incidence section (after the 10th) is not included in BP_avg, but is only used to check its representativeness. When the deviation between the two exceeds 30%, a low representativeness mark is added to BP_avg. When triggered, ΔDC_comp and ΔF_comp are each reduced to 70% of the standard value.

[0073] A control characteristic map is constructed by mapping control characteristic intervals based on characteristic compensation amounts. The control duty cycle compensation component ΔDC_comp and the operating frequency compensation component ΔF_comp of the characteristic compensation amount are defined as compensation vectors on the two-dimensional characteristic plane of the control characteristic map. These vectors point from the current operating condition's landing point coordinates to the target correction coordinates. The vector direction reveals the dominant type of the current deviation: a large ΔDC_comp indicates that load fluctuation is the main cause of the deviation; a large ΔF_comp indicates that overall resonant frequency drift is the main cause; and a simultaneous large ΔF_comp corresponds to a composite deviation caused by thermal drift and load accumulation. The main control characteristic interval (±2% control duty cycle × ±1Hz operating frequency) and a buffer control characteristic interval extending 50% outwards are labeled with the compensation vector coverage area as the core. The main interval corresponds to a high-precision correction zone, and the buffer interval corresponds to a medium-risk transition correction zone. These two types are stored in the control characteristic map with different line types for hierarchical retrieval during parameter deviation mapping. (The last sentence appears to be unrelated and refers to a razor on the mandibular arc surface.) After continuous shaving for more than 5 minutes, ΔDC_comp increased from 1.2% to 2.8%, and the main control characteristic range expanded to a higher level in the control duty cycle dimension, with the coverage range increasing from the initial 55%-65% to 60%-72%. The offset direction and magnitude of the graph centroid recorded the joint shaping trajectory of thermal drift and load accumulation on the parameter space. The switching pattern of different shaving areas left a quantifiable distribution of deviations in the graph. When the low representativeness flag is triggered, the boundary of the main control characteristic range shrinks inward by 10% in the direction of the buffer zone, constraining the correction target to a region with higher confidence, and preventing the low confidence characteristic compensation amount from leading the parameters to remote corners. The version number of the control characteristic graph is bound to the time range of the deviation interval sequence corresponding to the generation time. It is archived when a single shaving task ends, and the center coordinates of the main range are directly written into the initial G(i,j) vector of the corresponding working condition gear index layer when the next task is initialized.

[0074] The control characteristic graph is compared with the control input data to perform parameter deviation mapping and output adaptive control parameters. The measured values ​​of the control duty cycle and operating frequency at the current moment of the control input data are marked on the two-dimensional characteristic plane of the control characteristic graph as the landing point of the current operating condition. The shortest Euclidean distance from the landing point to the boundary of the high-density grid region in the control characteristic graph is used to judge the parameter risk level. When the landing point enters the boundary of the high-density grid, parameter deviation mapping is triggered. The adjustment amount of the control duty cycle is determined by ΔDC_comp superimposed with the landing point grid density coefficient (strong in high-density areas, conservative in sparse areas) and the amplitude stability index (the lower the index, the larger the adjustment amount). The adjustment amount of the operating frequency is determined by ΔF_comp superimposed with the overall gain level of the currently activated gain rule group. The two adjustment amounts are superimposed on the measured values ​​of the current control duty cycle and operating frequency of the control input data, respectively, and both are clamped with upper limits to ensure that the parameter deviation does not exceed the safe operating boundary and to prevent overcompensation from introducing new oscillations. When the shaver is within the sparse segment of the control characteristic graph, both adjustment values ​​are reduced to 30% of the standard value for conservative correction, avoiding unnecessary parameter fluctuations in low-risk areas. When the shaver enters the threshold jitter indicator segment, the adjustment value is output with a progressive smoothing coefficient to prevent the amplitude stability index from repeatedly crossing the threshold and driving rapid oscillation of the adjustment value, which would affect the continuity of the shaving experience. When the shaver moves from the flat area of ​​the cheek to the jawbone contour, the shaver quickly moves into the high-density area, and the duty cycle adjustment value is increased stepwise with the sudden increase in load. The operating frequency adjustment value compensates for the resonant frequency drift in sync. The two parameters work together to achieve a smooth transition when switching operating conditions. The adaptive control parameters consist of two parts: the target value of the control duty cycle and the target value of the operating frequency. The adaptive control parameters are written to the control parameter register with a delay of no more than 1ms, completing the control parameter configuration of the linear motor.

[0075] To implement the adaptive control method for a shaver linear motor corresponding to the above method embodiments, in order to achieve the corresponding functions and technical effects. See also Figure 2 , Figure 2 This paper shows a structural block diagram of an adaptive control system 200 for a shaver linear motor according to an embodiment of this application, including:

[0076] Signal acquisition module 201 is used to acquire control input data and response data, and perform envelope attenuation terminal correlation analysis on the control input data and response data to output an early drift warning indicator;

[0077] The matrix construction module 202 is used to perform input-output overshoot phase difference detection on the control input data and the response data to obtain the deviation phase intensity distribution, classify the response data according to the operating conditions to construct a reference spectrum library, and match the deviation phase intensity distribution with the reference spectrum library to establish an error verification matrix.

[0078] Rule generation module 203 is used to perform advance phase segment priority gain allocation on the error verification matrix according to the early drift warning identifier to form a gain rule group, perform phase shift analysis on the gain rule group to output phase shift feature spectrum, and perform deviation over-limit judgment to locate abnormal characteristic points based on the phase shift feature spectrum.

[0079] Density detection module 204 is used to activate the gain rule group according to the abnormal characteristic point to obtain an excitation sequence, perform pseudo-convergent oscillation detection on the excitation sequence to obtain a deviation density sequence, and perform reverse correlation detection based on the deviation density sequence and the excitation sequence to output an amplitude stability index.

[0080] The spectrum output module 205 is used to combine the amplitude stability index and the excitation sequence to perform deviation interval analysis to obtain a deviation interval sequence, perform deviation high-incidence section calibration on the deviation interval sequence and the gain rule group to construct a control characteristic spectrum, and compare the control characteristic spectrum with the control input data to perform parameter deviation mapping to output adaptive control parameters.

[0081] The aforementioned adaptive control system 200 for a shaver linear motor can implement an adaptive control method for a shaver linear motor according to the above-described method embodiments. The options described in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining contents of this application's embodiments can be referred to the contents of the above method embodiments, and will not be repeated in this embodiment.

[0082] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

Claims

1. An adaptive control method for a linear motor in a shaver, characterized in that, include: Acquire control input data and response data, and perform envelope attenuation terminal correlation analysis on the control input data and response data to output an early drift warning indicator, including: performing time-domain synchronization processing on the control input data and response data to construct an input-output synchronization sequence; performing envelope curve fitting on the input-output synchronization sequence to construct a response envelope; performing terminal slope change detection on the response envelope to determine the slope change distribution; and outputting an early drift warning indicator based on the slope attenuation rate classification according to the slope change distribution. The process involves detecting the input-output overshoot phase difference between the control input data and the response data to obtain a deviation phase intensity distribution. The response data is then categorized by operating condition to construct a reference spectrum library. Finally, the deviation phase intensity distribution is matched with the reference spectrum library to establish an error verification matrix. This includes: dividing the deviation phase intensity distribution into segmented intensity groups based on high-frequency switching segments; performing cross-operating condition deviation direction analysis on the segmented intensity groups and the reference spectrum library to construct a directional deviation spectrum; determining stratified deviation parameters based on the directional deviation spectrum; and establishing an error verification matrix by allocating matrix intervals according to the stratified deviation parameters. Based on the early drift warning identifier, the error verification matrix is ​​subjected to a leading phase segment priority gain allocation to form a gain rule group. Phase offset analysis is then performed on the gain rule group to output a phase offset feature spectrum. This includes: calibrating the driving phase time of each operating condition within the gain rule group to obtain a multi-operating condition phase time group; performing adjacent operating condition time difference analysis on the multi-operating condition phase time group to obtain a phase time difference sequence; identifying abnormal time difference segments by performing high-fluctuation and low-mean segment identification on the phase time difference sequence; merging and integrating the phase offset based on the steady-state segments in the abnormal time difference segments to output a phase offset feature spectrum; and performing deviation exceeding limit judgment on the phase offset feature spectrum to locate abnormal characteristic points. The excitation sequence is obtained by activating the gain rule group based on the abnormal characteristic points. The deviation density sequence is obtained by performing pseudo-convergent oscillation detection on the excitation sequence. The amplitude stability index is output by performing reverse correlation detection on the deviation density sequence and the excitation sequence. By combining the amplitude stability index with the excitation sequence to perform deviation interval analysis, a deviation interval sequence is obtained. The deviation interval sequence and the gain rule group are used to perform deviation high-incidence section calibration to construct a control characteristic map. The control characteristic map is compared with the control input data to perform parameter deviation mapping and output adaptive control parameters.

2. The method according to claim 1, characterized in that, The step of detecting the input-output overshoot phase difference between the control input data and the response data to obtain the deviation phase intensity distribution includes: The control input data is used to locate the overshoot peak time and obtain the control peak timing. The overshoot peak time is located by analyzing the response data to obtain the response peak time sequence; Successive time difference analysis is performed on the control peak timing sequence and the response peak timing sequence to determine the phase difference distribution; The bias phase intensity distribution is constructed by weighting the phase difference distribution according to the phase difference abrupt change rate.

3. The method according to claim 1, characterized in that, The step of performing reverse correlation detection based on the deviation density sequence and the excitation sequence to output an amplitude stability index includes: The excitation sequence is grouped by load type to obtain load response groups; A reverse correlation analysis is performed between the load response grouping and the deviation density sequence to form a load-correlated deviation grouping; Based on the load correlation deviation group, the cross-load reverse correlation strength assessment is performed to determine the reverse correlation strength parameters; The amplitude stability index is output by mapping the reverse correlation strength parameter according to the priority of the low reverse correlation strength interval.

4. The method according to claim 1, characterized in that, The step of calibrating high-incidence segments of deviations and constructing control characteristic maps by comparing the deviation interval sequence with the gain rule group includes: The high-incidence density distribution of deviations under various operating conditions is obtained by statistically analyzing the high-incidence periods of the gain rule group. Intermittent high-density segment location was performed on the high-incidence density distribution to determine the benchmark high-incidence segment; The characteristic compensation amount is determined by performing difference compensation between the benchmark high-incidence segment and the deviation interval sequence; A control characteristic map is constructed by mapping the control characteristic intervals based on the aforementioned characteristic compensation amount.

5. The method according to claim 1, characterized in that, The step of performing cross-condition deviation direction analysis on the segmented intensity group and the reference spectral library to construct a directional deviation spectrum includes: The segmented intensity groups and the reference spectral library are used to locate spectral line intervals to obtain a matching interval set; A difference sequence is obtained by performing cross-condition intensity gradient difference analysis on the matching interval set; Based on the difference sequence, low-difference and high-frequency deviation segments are identified to obtain a continuous deviation feature group; Based on the persistent deviation feature group, a deviation direction consistency assessment is performed to construct a directional deviation spectrum.

6. The method according to claim 3, characterized in that, The step of performing a reverse correlation analysis between the load response grouping and the deviation density sequence to form a load-correlation deviation group includes: Based on the load response grouping, a load time series baseline set is obtained by performing pre-deviation time series baseline analysis. A time-series misalignment analysis is performed on the load time-series baseline set and the deviation density sequence to obtain the time-series deviation sequence; Based on the time-series deviation sequence, perform high-misalignment and reverse-changing time period identification to obtain reverse correlation features; Based on the aforementioned reverse correlation features, deviation quantification and summarization are performed to form a load correlation deviation group.

7. An adaptive control system for a linear motor of a shaver, characterized in that, include: The signal acquisition module is used to acquire control input data and response data, and to perform envelope attenuation terminal correlation analysis on the control input data and response data to output an early drift warning indicator. This includes: performing time-domain synchronization processing on the control input data and response data to construct an input-output synchronization sequence; performing envelope curve fitting on the input-output synchronization sequence to construct a response envelope; performing terminal slope change detection on the response envelope to determine the slope change distribution; and performing slope attenuation rate classification based on the slope change distribution to output an early drift warning indicator. The matrix construction module is used to perform input-output overshoot phase difference detection on the control input data and the response data to obtain the deviation phase intensity distribution, classify the response data according to operating conditions to construct a reference spectrum library, and match the deviation phase intensity distribution with the reference spectrum library to establish an error verification matrix. This includes: dividing the deviation phase intensity distribution into segmented intensity groups based on high-frequency switching segments of the operating conditions; performing cross-operating condition deviation direction analysis on the segmented intensity groups and the reference spectrum library to construct a directional deviation spectrum; determining layered deviation parameters based on the directional deviation spectrum; and establishing an error verification matrix by allocating matrix intervals according to the layered deviation parameters. The rule generation module is used to perform advance phase segment priority gain allocation on the error verification matrix according to the early drift warning identifier to form a gain rule group, and perform phase shift analysis on the gain rule group to output a phase shift feature spectrum. This includes: calibrating the driving phase time of each operating condition in the gain rule group to obtain a multi-operating condition phase time group; performing adjacent operating condition time difference analysis on the multi-operating condition phase time group to obtain a phase time difference sequence; identifying abnormal time difference segments by performing high fluctuation and low mean segment identification on the phase time difference sequence; merging and integrating the phase shift based on the steady-state segment in the abnormal time difference segment to output a phase shift feature spectrum; and performing deviation exceeding judgment on the phase shift feature spectrum to locate abnormal characteristic points. The density detection module is used to activate the gain rule group according to the abnormal characteristic points to obtain the excitation sequence, perform pseudo-convergent oscillation detection on the excitation sequence to obtain the deviation density sequence, and perform reverse correlation detection based on the deviation density sequence and the excitation sequence to output the amplitude stability index. The graph output module is used to combine the amplitude stability index and the excitation sequence to perform deviation interval analysis to obtain a deviation interval sequence, perform deviation high-incidence section calibration on the deviation interval sequence and the gain rule group to construct a control characteristic graph, and compare the control characteristic graph with the control input data to perform parameter deviation mapping and output adaptive control parameters.

Images