Encoder-based line array camera synchronous triggering method and system

Abstract

The application discloses a kind of line array camera synchronous trigger method and system based on encoder, high-resolution pulse sequence is generated by hardware logic acquisition and quadrature frequency decoding generation to encoder quadrature difference input, deviation minimum phase interval positioning trigger time is identified in the pulse interval fluctuation caused by mechanical vibration, and hierarchical interval correction is applied according to the intensity and recovery characteristics of multiple skid conditions, and line trigger pulse is generated;Through the trigger edge of speed mutation section is screened out to establish delay correction reference, combined with the phase difference change rate early prediction acceleration master-slave cascade phase-locked convergence, and continuously monitor multi-stage drift trend to evaluate synchronization locking state;Accordingly, exposure window pulse count deviation is classified and evaluated according to cause, and frame quality record is collected, and breakpoint state is archived when pause;When continue transmission, the correction confidence interval is established by using the deviation history before breakpoint, the multi-channel breakpoint alignment line trigger instruction is output by displacement accumulation scheduling in accelerated section, and the geometric continuity of continue transmission is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of industrial vision inspection technology, and in particular to a method and system for synchronous triggering of a linear scan camera based on an encoder. Background Technology

[0002] Line scan cameras in industrial production line inspection use encoder pulses as the trigger reference for equal displacement, driving the camera to acquire image rows at equal displacement intervals to obtain geometrically uniform images of the inspected object. However, inherent eccentricity and vibration of the transmission mechanism superimpose mechanical periodic fluctuations on the pulse spacing. When triggered with a fixed phase, the spacing between adjacent rows fluctuates periodically within each mechanical cycle, resulting in striped quality defects in uniform texture areas. When the transmission slips, the pulse count becomes disconnected from the scanned displacement, causing geometric compression or stretching in the corresponding image area. Existing solutions lack targeted identification and compensation mechanisms for these problems.

[0003] Meanwhile, the discrete parameters of each channel device during parallel acquisition by multiple cameras cause inherent time delay differences. During long-term operation, clock thermal drift causes continuous deviation in the trigger timing of each channel. Existing synchronization solutions lack quantitative tracking and level evaluation of this accumulated drift. After the production line is paused and restarted, existing solutions lack systematic archiving of the acquisition status before the breakpoint, lack statistical estimation basis for the position positioning of the first frame of the resumed transmission, and geometric discontinuities are easily generated between the resumed frame and the last frame of the pause, which restricts the production efficiency in high-precision detection scenarios. Summary of the Invention

[0004] This invention discloses a method and system for synchronous triggering of linear array cameras based on an encoder. It aims to accurately generate line trigger pulses through optimal phase selection of mechanical vibration and slippage classification compensation, and achieve long-term synchronization of trigger timing of multiple cameras by combining steady-state spacing screening and phase difference advance prediction. Then, based on the quality assessment of cause classification and the archiving of breakpoint status, the correction amount of the first frame is estimated by the deviation statistical confidence interval during the transmission and the multi-channel aligned line trigger command is output through the cumulative scheduling of displacement in the acceleration segment. This provides a complete trigger management solution for linear array industrial inspection systems from signal acquisition to transmission continuity.

[0005] The first aspect of this invention proposes a method for synchronous triggering of a linear scan camera based on an encoder, comprising the following steps: The quadrature phase signal is obtained by real-time hardware logic acquisition of the quadrature differential input of the encoder. Based on the quadrature phase signal, quadrature frequency multiplication decoding and edge-by-edge hardware latching are performed to generate a high-resolution pulse sequence. For the high-resolution pulse sequence, the spacing fluctuation is tracked to determine the trigger spacing parameter. The trigger spacing parameter is used to identify the spacing step pattern to obtain the slippage event marker. After spacing compensation is applied to the slippage event marker, the pulse-by-pulse timing is scheduled to generate the row trigger pulse. The line trigger pulse is subjected to speed change segment jitter deviation screening to form a delay correction benchmark. Based on the delay correction benchmark, master-slave cascade triggering edge phase-locked correction is performed to obtain a global synchronization time base. The first and last stage drift is measured according to the global synchronization time base to generate a synchronization lock state. Based on the synchronous locking state, the exposure window pulse count deviation is verified to obtain a fuzzy risk mark. The fuzzy risk mark is used to collect the out-of-limit frame sequence number and the spacing fluctuation amplitude to form a frame quality record. The frame quality record is archived for pre-pause fluctuation to establish a breakpoint latch record. The recovery start parameter is obtained from the breakpoint latch record location trigger coordinates and frame quality compensation parameters. The recovery start parameter is then used to perform acceleration segment metering correction and resume transmission scheduling to generate a resume transmission trigger sequence. Based on the resume transmission trigger sequence, the breakpoint sequence number continuity deviation is tracked and a multi-path trigger command is output.

[0006] A second aspect of this invention provides an encoder-based linear scan camera synchronization triggering system, comprising: The signal acquisition module is used to acquire orthogonal phase signals in real time from the orthogonal differential input of the encoder using hardware logic, and to perform orthogonal frequency multiplication decoding and edge-by-edge hardware latching based on the orthogonal phase signals to generate high-resolution pulse sequences. The trigger scheduling module is used to determine the trigger spacing parameter by tracking the spacing fluctuation of the high-resolution pulse sequence, to obtain the slippage event marker by using the trigger spacing parameter to perform spacing step pattern recognition, and to generate the row trigger pulse by pulse timing scheduling after performing spacing compensation on the slippage event marker. The clock synchronization module is used to screen the jitter deviation of the speed change segment of the row trigger pulse to form a delay correction reference, perform master-slave cascade triggering edge phase-locked correction based on the delay correction reference to obtain a global synchronization time base, and perform first and last stage drift measurement based on the global synchronization time base to generate a synchronization lock state. The quality control module is used to verify the pulse count deviation of the exposure window based on the synchronous locking state to obtain a fuzzy risk mark, use the fuzzy risk mark to collect the out-of-limit frame sequence number and the spacing fluctuation amplitude to form a frame quality record, and archive the frame quality record for fluctuation before pause to establish a breakpoint latch record. The breakpoint resume module is used to obtain the recovery start parameter from the breakpoint latch record location trigger coordinates and frame quality compensation parameters, perform acceleration segment metering correction on the recovery start parameter to generate a resume trigger sequence, and output multiple line trigger commands based on the resume trigger sequence to track the continuity deviation of the breakpoint sequence number.

[0007] The beneficial effects of this invention are reflected in the following points: First, the inventiveness of this invention lies in the shift in perspective regarding the mechanical vibration of the transmission mechanism: traditional solutions treat the periodic fluctuations of the pulse spacing as interference to be eliminated, while this invention identifies them as predictable phase patterns and actively locates the trigger moment in the interval with the smallest spacing deviation, transforming the source of interference into an optimization constraint for trigger accuracy; in terms of slippage compensation, traditional solutions correct speed jumps with a single amplitude, while this invention uses the step amplitude of the differential rate of change and its duration to jointly characterize the slippage intensity and recovery mode, applying corresponding levels of correction according to the operating condition category, so that the compensation strategy matches the operating condition characteristics. Second, this invention actively identifies and excludes the trigger edge of the speed abrupt segment, extracting the spacing statistics only in the stable speed interval to establish a delay correction benchmark, eliminating the bias of the benchmark estimation caused by the jitter of the abrupt segment from the root; traditional pure feedback phase-locked loops produce tracking lag when the phase changes rapidly, while this invention introduces the advance prediction of the phase difference rate of change, enabling the correction command to respond in advance to the current phase change trend, transforming passive tracking into a combination of feedforward and feedback; and continuously quantifies the phase drift trend of multi-level cascades, providing an online evaluation basis for the synchronization quality of long-term acquisition. Finally, a cause identification step is added before the deviation exceeding the limit judgment. First, the causes of vibration jitter and speed drift are distinguished. Then, the corresponding threshold set is matched according to the cause, which solves the problem that a uniform threshold cannot distinguish the differences in causes. The positioning of the first frame of the resume transmission is to establish a corrected confidence interval by statistically analyzing the deviation of multiple frames before the pause, instead of directly using the end value of the breakpoint as the coordinate of the resume transmission. Displacement accumulation correction is implemented during the acceleration phase to constrain the start position of the resume transmission to the vicinity of the breakpoint coordinates. Statistical interval estimation is used instead of single-point estimation, which improves the ability to ensure the geometric continuity of the resume transmission. Attached Figure Description

[0008] Figure 1 This is a flowchart of a method for synchronous triggering of a linear array camera based on an encoder, according to the present invention.

[0009] Figure 2 This is a schematic diagram of the resonant wave phase interval positioning and trigger point selection of the present invention.

[0010] Figure 3 This is a structural block diagram of a linear array camera synchronous triggering system based on an encoder according to the present invention.

[0011] Wherein: 1- Spacing fluctuation curve; 2- Resonance compensation amount curve; 3- Optimal phase interval; 4- Interval start boundary line; 5- Interval end boundary line; 6- Compensation amount threshold mark; 7- Trigger phase selection point; 8- Theoretical mean baseline. Detailed Implementation

[0012] 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.

[0013] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0014] 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.

[0015] The technical solutions of the embodiments of this application will be described below.

[0016] like Figure 1 As shown, this embodiment of the invention provides a method for synchronous triggering of a linear scan camera based on an encoder, including the following steps S101-S105: Step S101: The quadrature phase signal is obtained by real-time hardware logic acquisition of the quadrature differential input of the encoder. Based on the quadrature phase signal, quadrature frequency multiplication decoding and edge-by-edge hardware latching are performed to generate a high-resolution pulse sequence.

[0017] Specifically, the quadrature phase signal is obtained in real time by hardware logic acquisition of the encoder's quadrature differential input. The encoder's quadrature differential input consists of two differential signal pairs, A and B, which are nominally 90° out of phase, and are the original level source for reconstructing the quadrature phase signal. The differential receiver samples the positive and negative terminals of each channel separately, and determines the logic high or low based on the potential difference between the two terminals rather than the absolute potential of a single terminal. This ensures that the common-mode interference induced in the same direction on both lines cancels each other out before the determination, which is a prerequisite for reliably reconstructing the quadrature phase signal even when the signal line is long. In high-speed scanning by a linear scan camera, the length of the traces, distributed capacitance, and line resistance cause the differential pair's rising edge slope to degrade and drift at the threshold. The acquisition unit uses a Schmitt trigger circuit with hysteresis at the input stage to transform the degraded slope into a steep digital transition. The hysteresis voltage is set according to a certain proportion of the expected minimum swing, filtering out jitter near the threshold without missing the true transition. Furthermore, the terminal matching resistor and receiving threshold are adaptively configured according to the encoder line drive model to suppress reflections caused by impedance mismatch in long lines, ensuring that the signal falls within the optimal judgment zone under different line lengths. After shaping, the edge jitter of the quadrature phase signal is reduced to within one sampling period. When the amplitude is asymmetrical due to inconsistent trace lengths or terminal loads between the two paths, the differential noise margin of the lower path narrows. The acquisition unit continuously monitors the difference between the two paths. If the difference exceeds the threshold and remains above a set count, a single-ended imbalance is judged. During this period, the quadrature phase signal validity flag is pulled low, and output is paused. It is automatically reset to valid only after the difference between the two paths returns to symmetry. After the two channels are further adjusted to compensate for the fixed phase error of the discrete devices through inter-channel delay correction, they are aligned and combined into orthogonal phase signals. The time period when the measured phase difference exceeds the 90° tolerance is recorded as the phase deviation, and the timing confidence of the subsequent orthogonal frequency doubling results is reduced.

[0018] In some embodiments, the step of generating a high-resolution pulse sequence by performing orthogonal frequency multiplication decoding and edge-by-edge hardware latching based on the orthogonal phase signal includes: performing dual-edge timing detection and pulse edge polarity classification on the orthogonal phase signal to obtain a pulse edge sequence; performing real-time latching of the pulse edge sequence using a hardware counter between adjacent edges to obtain an edge-by-edge spacing count; performing count truncation remainder accumulation across edges based on the edge-by-edge spacing count to generate a pulse subdivision compensation amount; and performing overflow prediction and advance zeroing correction on the pulse subdivision compensation amount to form a high-resolution pulse sequence.

[0019] The quadrature phase signal undergoes dual-edge timing detection and pulse edge polarity classification to obtain a pulse edge sequence. Quadrature frequency multiplication decoding captures all rising and falling edges of both paths A and B. The dual-edge detection module performs an XOR operation between the current sampled value of the quadrature phase signal and the sampled value of the previous cycle in each sampling period. A non-zero XOR indicates a transition, and the direction is determined by the high / low relationship between the two cycles. To avoid metastability introduced by asynchronous sampling clock and encoder transitions, the quadrature phase signal is first synchronously timed by two stages of D flip-flops before entering the XOR comparison. The sampling clock frequency is set to a multiple higher than the highest expected edge frequency, ensuring that even the narrowest effective edge is stably captured in at least one cycle and written into the pulse edge sequence. Pulse edge polarity is determined by the relative levels of the two paths at the transition moment: a high level on path B during a rising edge of path A indicates positive direction, and a low level indicates negative direction. A transition on path B is symmetrically determined by the current level of path A. The four combinations each map to forward and reverse rotation. If the transition times of paths A and B are close and occur simultaneously within the same sampling period, the module prioritizes path A before determining the direction to avoid incorrect direction. Furthermore, the two paths of the quadrature phase signal undergo a de-jitter confirmation before entering the pulse edge sequence generation. The state is only updated when the levels are consistent for two consecutive beats, further reducing the proportion of pseudo-edges written into the pulse edge sequence. Due to crosstalk, a single path may occasionally jump within one or two adjacent beats and then immediately recover, forming an extremely short pseudo-jump. This can cause the quadrature frequency multiplication count to be artificially high. Based on this, the module sets a minimum effective pulse width threshold. Pulses shorter than this threshold are discarded as pseudo-jumps and not written into the pulse edge sequence. Jump events during the phase deviation period are recorded as low confidence at the corresponding position in the pulse edge sequence, indicating that the phase difference between the two paths deviates from the nominal value and there is a risk to the timing accuracy during this period.

[0020] The edge-by-edge interval count is obtained by real-time latching of adjacent edges using a hardware counter. The system master clock drives a freely running hardware counter to continuously increment. Upon arrival of each transition in the pulse edge sequence, the current counter value is instantaneously latched into a register. The latching is completed within one clock cycle after the transition confirmation, completely decoupled from processor software scheduling. Therefore, the edge-by-edge interval count is unaffected by task load fluctuations. The difference between two adjacent latched values ​​is the clock interval between those adjacent edges, denoted as dC_n = C_n - C_(n-1), where C_n and C_(n-1) are the latched values ​​for the nth and (n-1)th edges, respectively. The counter bit width determines the maximum measurable interval for edge-by-edge interval counting: taking a 32-bit counter and a master clock in the hundreds of megabits range as an example, a single count takes approximately 10 nanoseconds, and the full scale can cover tens of seconds, sufficient to accommodate long intervals at extremely low encoder speeds. When the encoder crawls near its stop, the interval between adjacent edges may exceed the full scale, causing the counter to wrap around. After wrapping, the current latched value is less than the previous value, and the direct subtraction is negative. At this time, the negative difference is added to the full scale value using two's complement to restore the true clock count. Wrapping can be identified solely by the sign of the difference, without the need for additional flag bits. Although the pulse edge sequence has filtered out false transitions, residual crosstalk may still cause some intervals to be abnormally short. Those shorter than the set minimum threshold are moved to an independent archive partition and do not participate in subsequent interval statistics, to avoid a few abnormal points skewing the statistical benchmark of the edge-by-edge interval counting. The capacity of this independent archive partition is reserved according to the worst crosstalk rate to ensure that the main sequence of edge-by-edge interval counting remains continuous and without missing points in high-interference batches.

[0021] The pulse subdivision compensation is generated by truncating the count remainder across edges based on the edge-by-edge interval count. The core of the pulse subdivision compensation is to divide each edge-by-edge interval into several sub-slots and then allocate an integer number of clock cycles to each sub-slot: Take the subdivision number N (fixed to the configuration register during initialization), divide the nth edge-by-edge interval count dC_n by N, and obtain the quotient q_n and remainder R_n. q_n is the reference clock cycle number for each sub-slot of that edge. The larger the subdivision number N, the denser the sub-slots within a single edge. With orthogonal frequency multiplication, the total system resolution can reach N times the number of edges. The value of N is determined during initialization based on a tradeoff between the required row resolution and the master clock frequency. The challenge lies in handling the remainder R_n. If it is discarded directly, the truncation error will accumulate repeatedly between edges, causing the trigger pulse position to deviate linearly from the encoder's true displacement as the acquisition time progresses. This solution adds R_n edge-by-edge to the cross-edge remainder register C_acc, forming a modulo-N accumulator: for each edge, the remainder is accumulated. Once C_acc reaches or exceeds N, an additional clock is inserted into the sub-slot sequence of the current edge, and C_acc is subtracted from N, effectively converting the fractional part into an integer clock. In this way, the long-term average of the number of clocks in each sub-slot (i.e., the pulse subdivision compensation) is precisely equal to dC_n / N. The truncation error is evenly distributed to subsequent sub-slots, and long-range offset converges, similar in mechanism to the error diffusion of linear interpolation. The remainder of the edge-by-edge spacing count entries restored by two's complement is directly calculated on the restored value; the suspicious spacing moved into the archive partition is skipped from accumulation, C_acc remains unchanged, the pulse subdivision compensation amount of the corresponding edge is estimated and backfilled using the result of the previous valid edge, and when contiguous sections appear, the segment is expected to be passed downstream with lower timing accuracy.

[0022] A high-resolution pulse sequence is formed by implementing overflow prediction and advance zeroing correction for the pulse subdivision compensation amount. The clock counts of each edge sub-slot given by the pulse subdivision compensation amount are sequentially loaded into a countdown counter inside the high-resolution pulse generator. The countdown counter loads the value at the beginning of each sub-slot and outputs a pulse when it reaches zero. The trigger segment of this edge ends after all sub-slots have passed, and the start and end segments of each edge form a continuous high-resolution pulse sequence. The problem is that if timing competition occurs between loading and zeroing at the boundary of adjacent sub-slots, the countdown counter may reach zero prematurely at the boundary and trigger an extra time, resulting in a short-interval glitch pulse mixed into the high-resolution pulse sequence. To address this, overflow prediction is introduced: before each loading, the difference between the remaining value of the countdown counter and the clock count of the current slot is compared. Once it is determined that the current slot will naturally reach zero, the advance zeroing correction forces the countdown counter to zero and restarts from zero at the nominal clock count minus the overflow margin, ensuring that only one valid pulse is generated in that slot, thus guaranteeing a glitch-free output of the high-resolution pulse sequence. The peak value of the leading zero-reset amount is tracked in real time during operation. Its magnitude directly reflects the maximum fluctuation of the estimation error of the pulse subdivision compensation amount in this batch, and can be used as an online indicator of output quality. To balance jitter and throughput, the countdown counter bit width is designed with a margin of one bit based on the maximum number of sub-slot clocks, so that the high-resolution pulse sequence can accommodate the instantaneous peak value of the pulse subdivision compensation amount without overflowing at the highest line frequency. For sub-slots filled with estimated values, the zero-reset time itself carries an error, and the timing accuracy of the pulse subdivision compensation amount corresponding to that pulse is degraded accordingly, and the accuracy degradation is marked and propagated downstream with that pulse.

[0023] Step S102: For the high-resolution pulse sequence, the spacing fluctuation is tracked to determine the trigger spacing parameter. The spacing step pattern is identified using the trigger spacing parameter to obtain the slippage event marker. After spacing compensation is applied to the slippage event marker, the pulse-by-pulse timing is scheduled to generate the row trigger pulse.

[0024] In some embodiments, the step of determining the trigger spacing parameter by tracking the spacing fluctuation of the high-resolution pulse sequence includes: forming a spacing sliding window sequence by arranging the high-resolution pulse sequence in real time with pulse-by-pulse spacing sliding windows; identifying the spacing deviation of equally spaced repeating segments in the spacing sliding window sequence to generate resonance fluctuation characteristics; performing in-phase trigger point spacing deviation prediction and correction based on the resonance fluctuation characteristics to obtain resonance compensation amount; and using the resonance compensation amount to perform resonance optimal phase interval positioning to determine the trigger spacing parameter.

[0025] A pulse-by-pulse spacing sliding window sequence is formed by real-time arrangement of pulses in the high-resolution pulse sequence. The window width W is initialized to an integer multiple of the typical number of pulses corresponding to the encoder's mechanical cycle, typically 2-4 mechanical cycles. A window that is too narrow results in incomplete resonant waveforms and broadened autocorrelation peaks within a single window; a window that is too wide leads to sluggish response to instantaneous speed changes and significant lag in the estimation of the resonant cycle during acceleration and deceleration. Using an integer multiple of the mechanical cycle also ensures cleaner alignment of the autocorrelation main peak and reduces sidelobes truncated by non-integer cycles. With each pulse advance, a new spacing is added from the end of the high-resolution pulse sequence while the earliest one is removed. The number of pulses within the window remains constant at W. Arranging the pulses according to their inclusion order yields the spacing sliding window sequence, which slides and updates with the high-resolution pulse sequence, always reflecting the most recent motion state. The historical portion of the spacing sliding window sequence at the batch boundary comes from the previous batch. The current batch waits until W new pulses have accumulated before forming the first complete window. During this period, the resonant parameters of the previous batch are used as a transitional fill to avoid a cold start with an empty initial window. In high-resolution pulse sequences, spacing marked with precision degradation is included with low weight in the corresponding window. When the proportion of low-weight spacing in a window exceeds the upper limit, the entire window is treated as low-confidence, and its resonance period estimation is replaced by the previous valid window. This maintains the continuity of the spacing sliding window sequence and avoids a few low-quality spacings from skewing the resonance estimation. When the scanning speed changes across levels, the switching between multiple window widths is completed in the overlapping area of ​​the two windows. The new window runs parallel to the old window for one mechanical cycle before handing over, so that the spacing sliding window sequence does not have statistical discontinuities when the speed changes, and continues to advance step by step with the high-resolution pulse sequence.

[0026] Resonance fluctuation characteristics are generated by identifying the spacing deviation of equally spaced repeating segments in the pitch sliding window sequence. The autocorrelation of the spacing sequence in each window of the pitch sliding window sequence is calculated. A significant peak appears at the resonance period T_r. The peak correlation coefficient reflects the energy proportion of the component in that period. Periods with a proportion exceeding the threshold are judged as the resonant main frequency, constituting the fundamental parameter of the resonance fluctuation characteristics. This threshold is generally obtained by first acquiring a baseline correlation coefficient during equipment idling, and then taking approximately 30% to 50% of it as a lower limit, ensuring that low-amplitude true resonances can still be identified while suppressing noise misjudgments. To suppress spectral leakage caused by non-integer period truncation, a slowly varying window function, i.e., a Hanning window, is added to the pitch sliding window sequence before calculating the autocorrelation for end weighting. The width of the autocorrelation main peak further reflects the resonance stability and is written into the resonance fluctuation characteristics. A narrow and high peak indicates frequency stability within the batch, while a wide and low peak indicates frequency drift, often corresponding to encoder shaft thermal expansion or local loosening of the transmission chain, resulting in lower amplitude estimation accuracy. In the spacing sliding window sequence, low-confidence windows participate with zero contribution, and low-weighted spacings are weighted according to coefficient reduction in the weighted autocorrelation, ensuring that the accuracy degradation entries do not skew the peak positions. To reduce the computational load per window, the spacing sliding window sequence uses sliding recursive autocorrelation, updating only the two spacing increments that are included and removed, allowing the resonance fluctuation characteristics to be refreshed in real time with each pulse arrival without having to recalculate the entire window. Some transmission systems exhibit multi-order resonances, with multiple peaks appearing in the autocorrelation at integer multiples of T_r. When the harmonic amplitude accounts for a high proportion of the dominant frequency, the resonance fluctuation characteristics indicate strong high-order nonlinear vibrations in the transmission chain. Fixed-frequency interferences such as power supply ripple are identified as interferences and excluded because their peak values ​​do not fall within integer multiples of the machine period.

[0027] Based on the characteristics of resonant waves, the spacing deviation of the in-phase trigger points is predicted and corrected to obtain the resonant compensation amount. If the spacing deviation within the resonant period is left uncorrected, the phase taken by the row trigger will be random within the period, and the actual spacing of different rows will fluctuate with the phase. This will result in bright and dark stripes appearing along the scanning direction, which are in sync with the mechanical cycle and are particularly glaring on a uniform surface. This is the motivation for introducing the resonant compensation amount. Let the phase of a certain pulse position within the resonant period be φ, which is determined by the resonant order and pulse number given by the characteristics of resonant waves. φ = 2 × π × m × n / N_period, where m is the resonant order to be compensated with a dominant frequency of 1 and is a dimensionless integer, n is the pulse number, N_period is the number of pulses in a single mechanical cycle, and φ is the zero point of the phase with the peak of the spacing deviation. Figure 2The horizontal axis is consistent; the phase processing spacing deviation is estimated by the resonance amplitude A in the resonance wave characteristics as A×cos(φ) (φ=90°, i.e., the deviation at point 7 is zero, which is cosine-shaped with curve 1, and the deviation at point 7 is the smallest). Taking its opposite sign is used as the resonance compensation amount at this position, so that the compensation is equal in magnitude and opposite in direction to the deviation. When the encoder speed changes, N_period is updated accordingly to maintain the phase accuracy of the speed change segment. There is a fixed hardware pipeline delay in the compensation execution, so when calculating the resonance compensation amount, several pulses are extrapolated in advance to offset this delay. The extrapolation step size is measured by the trigger timing calibration during the calibration stage. For those with significant second harmonics, the resonance compensation amount is superimposed with the harmonic according to the main frequency to make the compensation more in line with the actual waveform. When the resonance amplitude A given by the resonance wave characteristics is far beyond the historical normal range, and the resonance compensation amount estimated based on it exceeds the reasonable range, the resonance compensation amount is limited to the set upper limit and recorded as saturation; frequency components judged as interference are not included in the compensation, and only the main frequency is included in the resonance compensation amount sequence of the current batch.

[0028] The trigger spacing parameters are determined by using the resonance compensation amount to perform optimal phase interval positioning for resonance. For example... Figure 2 As shown, with the resonant phase within one mechanical cycle as the horizontal axis, the theoretical mean baseline 8 marks the reference zero point of the target pulse spacing. The spacing fluctuation curve 1 depicts the deviation of the actual spacing from the baseline 8 at each phase, exhibiting a cosine-shaped periodic oscillation. The peaks and troughs correspond to the maximum positive and negative disturbances of the eccentric transmission within one mechanical cycle. The resonant compensation curve 2 plots the spacing correction amount, i.e., the resonant compensation amount, applied by the system at each phase prediction. It is a mirror image of curve 1 and leads the phase by a fixed angle to compensate for pipeline delay. The resonant compensation amount crosses zero where curve 2 crosses the baseline 8. The compensation amount threshold mark 6 defines the acceptable upper limit of the resonant compensation amount with positive and negative Δ horizontal dashed lines. The phase segment of curve 2 running between the two marks 6 forms the optimal phase interval 3. Within the interval, the absolute value of the resonant compensation amount is lower than this limit, and the spacing deviation is minimized within the cycle. The left and right boundaries of interval 3 are marked by the interval start boundary line 4 and the interval end boundary line 5, with two symmetrical segments appearing within one cycle. Two optimal phase intervals provide two available triggering opportunities per mechanical cycle. The trigger phase selection point 7 is marked at the center of the first interval 3. Here, curve 2 intersects the baseline 8, resulting in zero compensation and minimal deviation of curve 1. Triggering at this point can constrain the trigger spacing parameter deviation within the tolerance of interval 3. If the trigger phase shifts outside interval 3, the trigger spacing parameter deviation increases with the increase of the resonance compensation. Therefore, the trigger spacing parameter is taken as the target spacing corresponding to this selection point. When the absolute value of the resonance compensation exceeds the compensation threshold mark 6 within a certain mechanical cycle, and there is no optimal phase interval falling within mark 6, the trigger spacing parameter is marked as interval degradation. When the resonance parameter of this window is used to replace the result of the previous effective window due to low confidence, and the optimal phase interval located accordingly is actually the previous window interval, the trigger spacing parameter is marked as interval replacement.

[0029] In some embodiments, the step of using the trigger spacing parameter to perform spacing step pattern recognition to obtain a slippage event marker includes: using the trigger spacing parameter to calculate the pulse-by-pulse rate of change of the spacing difference before and after to form a spacing difference sequence; performing joint screening of the step amplitude duration segment of the spacing difference sequence to generate slippage feature data; performing fusion matching of the slippage feature data with the steady-state features before the step and the grading coefficient to determine the grading compensation parameters; and performing grading labeling of the slippage segment and event level assessment according to the grading compensation parameters to obtain a slippage event marker.

[0030] A pulse-by-pulse calculation of the rate of change of the difference between preceding and following pulses is performed using the trigger spacing parameter to form a spacing difference sequence. The rate of change of the difference of the k-th pulse in the spacing difference sequence is defined as the ratio of the difference between adjacent pulses to the baseline pulse, e_k=(dT_k-dT_(k-1)) / T_ref, where dT_k and dT_(k-1) are the pulse spacings between adjacent pulses, and T_ref is the core value of the trigger spacing parameter. After normalization using the core value of the trigger spacing parameter, e_k is a dimensionless ratio, with positive values ​​indicating an increase in spacing and negative values ​​indicating a decrease. Normalization makes the same set of judgment thresholds applicable at different line frequencies without the need for recalibration with speed. During uniform speed operation, the e_k of the spacing difference sequence fluctuates slightly and randomly near zero, with an amplitude usually within 0.5%, forming the baseline for judging slippage. This baseline statistic also serves as the basis for subsequent over-threshold calibration. Once a sudden change in the slippage speed relationship occurs, the absolute value of e_k in the spacing difference sequence at the initial pulse rises sharply to several times or even tens of times the baseline, becoming a significant indicator of the slippage initiation. The two types of slippage are distinctly different in waveform: the instantaneous impact type has a high peak e_k value but only lasts for a few pulses, appearing as a sharp pulse; while the continuous drift type has a low peak e_k value but can last for tens or even hundreds of pulses, appearing as a trapezoidal plateau. These two types are easily distinguishable in the time domain, laying a two-dimensional feature foundation for the subsequent joint screening of step amplitude and sustained segments. To suppress quantization noise, the interval difference sequence is pre-filtered with the median of three points on adjacent intervals before calculation to eliminate single-point distortions, resulting in a flatter baseline and a lower false alarm rate for step detection. When the trigger interval parameter itself is labeled as interval degradation or interval substitution, the reliability of its core value decreases, and the interval difference sequence calculated using it as the denominator will carry a systematic bias. For this type of batch, the judgment threshold should be relaxed accordingly during step screening to avoid misjudging normal fluctuations as slippage due to a low baseline.

[0031] For example, the step amplitude continuous segment joint screening of the interval difference sequence to generate slippage feature data includes: performing pre-stability and exit slope directional calibration on the interval difference sequence to obtain segment boundary amplitude markers; tracking the maximum offset amplitude and continuous pulse number within the segment based on the segment boundary amplitude markers to obtain the segment offset; calculating the ratio of step recovery slope to continuous offset duration to determine the slippage recovery coefficient; and performing instantaneous impact type and continuous drift type feature classification based on the slippage recovery coefficient to generate slippage feature data.

[0032] The segment boundary amplitude markers are obtained by performing pre-stability calibration and exit slope calibration on the inter-segment difference sequence for the overthreshold segment. The segment boundary amplitude markers use the e_k of the inter-segment difference sequence as the only input and delineate the slippage segment boundary in three steps: First, pulse segments in the inter-segment difference sequence whose absolute value of e_k continuously exceeds a set threshold are identified as overthreshold segments. The threshold is set to 3 to 5 times the standard deviation of normal e_k. To avoid repeated crossings of e_k near the threshold causing boundary jitter, a pair of hysteresis-based double thresholds are used for the entry and exit segments—entry is only considered when the higher entry threshold is exceeded, and exit is only considered when the threshold falls below the lower exit threshold. This prevents brief overthreshold spikes from being cut into multiple fragmented segments. The hysteresis is set to 2 to 3 times the background noise amplitude. Secondly, stability is calibrated for the pre-stable segment consisting of several pulses before the entry segment. Its length is approximately half the minimum expected number of pulses for slippage duration to ensure it does not overlap with the slippage segment. A near-zero mean e_k value and small dispersion in this segment indicate stable entry, while large dispersion indicates unstable entry. Thirdly, the slope of the e_k drop at the end of the over-threshold segment is calibrated: a monotonically decreasing return to below the threshold after multiple pulses is considered smooth recovery, while a sudden drop within a few pulses is considered rapid recovery. The entry segment number, exit segment number, entry stability, and exit recovery type are all written into the segment boundary amplitude marker for downstream use in identifying slippage segments. When the interval between two adjacent over-threshold segments is very short, the pre-stable segment of the subsequent segment may fall within the recovery period of the preceding segment. In this case, the segment is marked as contaminated, and its mean reference is changed to an earlier, clean, stable segment to avoid including the residual slippage from the previous segment in the subsequent baseline, and the segment boundary amplitude marker is corrected accordingly.

[0033] The segment offset is obtained by tracking the maximum offset amplitude and the number of continuous pulses within the segment based on the segment boundary amplitude markers. Within the range defined by the entry and exit segment numbers given by the segment boundary amplitude markers, the maximum absolute value of e_k is the maximum offset amplitude, and the number of continuous over-threshold pulses is the number of continuous pulses. The two are combined to form the segment offset: the former reflects the peak value of slippage intensity, and the latter reflects the range of slippage wave and row number. Their product can roughly measure the total disturbance of slippage on the row spacing sequence. For segments marked as unstable entry by the segment boundary amplitude markers, the mean baseline of the preceding stable segment is subtracted before calculating the maximum offset amplitude. If the maximum offset amplitude falls below the threshold after subtraction, it indicates that it was originally just a pseudo-over-threshold superimposed on a non-zero baseline, and the segment offset is recorded as a pseudo-over-threshold segment. Two typical types of slippage are complementary in this two-dimensional feature: belt drive slippage results in a sharp drop in the coefficient of friction, which recovers slowly as the surface temperature rises or the relative slippage decreases, manifested as a relatively large number of continuous pulses with a moderate amplitude; coupling slippage, when subjected to an instantaneous impact, causes the speed to surge to a peak within a very small number of pulses and then quickly return to zero, manifested as an extremely short number of continuous pulses with an extremely high amplitude. For suspected drift segments with extremely long continuous pulse numbers, the segment offset is also noted on whether its amplitude increases monotonically with the pulses to distinguish it from slow-changing interferences such as sensor zero drift. These notes are transmitted downstream along with the segment boundary amplitude markers. Because one falls in a long and gentle angular region and the other in a short and steep angular region, they can be clearly separated by the plane spanned by the maximum offset amplitude and the continuous pulse number. The measured cluster boundaries of the segment offsets are distinct with little overlap, providing stable features for subsequent segment-by-segment classification based on the segment boundary amplitude markers.

[0034] The slippage recovery coefficient is determined by calculating the ratio of the step recovery slope to the duration of continuous slippage in the segment offset. Starting from the recovery starting point marked by the segment offset's outgoing segment number, the absolute value of e_k decreases within a few consecutive pulses, which is the step recovery slope κ, measured as the amount of e_k decrease per pulse. Segments marked as rapidly recovering have a large slope, corresponding to a rapid rebound of the elastic transmission component, with the speed returning to normal within a few pulses. The continuous pulse count of the segment offset is divided by the current pulse frequency to convert it into a continuous slippage duration D, defining the slippage recovery coefficient ρ = D / κ. Here, ρ is only used as a relative ranking indicator of the severity of slippage between segments; its dimension is not for physical interpretation. A larger ρ indicates a longer slippage duration and slower recovery, requiring subsequent compensation to extend to more pulses to cover the recovery segment. Segments with smooth recovery have a gentle slope and a larger slippage recovery coefficient, commonly seen in transmission mechanisms dominated by high inertia or viscous resistance, where the kinetic energy accumulated in the deceleration segment is released slowly, and the compensation extension range is significantly longer than that of rapidly recovering segments. When continuously collecting data on a surface with a low coefficient of friction, the batch average of the slip recovery coefficient will be generally higher, which can serve as a warning of surface condition, prompting cleaning of the contact surface or replacement of transmission components. To avoid noise exaggerating the recovery slope, the recovery segment of the segment offset is first fitted with a least-squares straight line, and then the slope is taken as the step recovery slope. If the fitting residual is too large, the recovery segment is judged as unreliable, and the corresponding slip recovery coefficient is reduced in weight, making it more robust to short and rapid jumps. Segments with pseudo-overthreshold labels are not included in the slip recovery coefficient calculation. The gap is filled with the average coefficient of other valid segments in the same batch, so that these segments still have reasonable substitution values ​​during classification and that the mapping from segment offset to slip recovery coefficient remains complete across the entire segment.

[0035] Based on the slippage recovery coefficient, slippage feature data is generated by classifying instantaneous impact and persistent drift features. The instantaneous impact type is defined as a slippage recovery coefficient below a low-coefficient threshold and a continuous pulse count below a short-duration threshold. This reflects a rapid regression after a large jump in spacing within a very short time, often caused by mechanical impact or shaft collision. On the image, this typically manifests as only one or a few lines of local compression or stretching, with limited impact on the overall geometric uniformity; it is classified as an impact type in the slippage feature data. The persistent drift type is defined as a slippage recovery coefficient exceeding a high-coefficient threshold and a continuous pulse count exceeding a long-duration threshold. This corresponds to scenarios of belt slippage or a persistently low friction coefficient. Without compensation, this will distort the geometric proportions of the area and significantly amplify dimensional measurement errors. The segment where both the slippage recovery coefficient and the continuous pulse count fall between the two thresholds is classified as a mixed type in the slippage feature data. Mild intermittent vibrations on hard surfaces and temporary belt mis-engagement are most likely to fall into this category. The initial classification threshold is determined during debugging based on the statistical distribution of known slippage events. After a period of production, it is periodically recalibrated based on the measured event database to ensure that the threshold adapts to equipment wear and environmental changes. To handle boundary samples between the two classes, the slippage feature data retains the normalized distance from the slippage recovery coefficient to each threshold while recording the class. This allows for soft weighting based on distance rather than hard switching during classification, reducing classification jitter near the threshold. Segments with estimated coefficients are recorded as low-confidence in the slippage feature data, reducing their impact weight when determining subsequent classification compensation parameters. Each segment also retains the determination of whether the entry segment is stable. For unstable entry segments, the pre-entry disturbance bias needs to be additionally deducted when determining the compensation baseline.

[0036] The slippage characteristic data is fused and matched with the steady-state characteristics before the step jump and the graded coefficients to determine the graded compensation parameters. The slippage type and coefficient value of each segment in the slippage characteristic data jointly determine the compensation level, which is divided into three levels from light to heavy and written into the graded compensation parameters: the light level corresponds to the segment with low coefficient and short duration. After the slippage ends, the line spacing within a few pulses naturally converges. The compensation amount is 30% to 50% of the peak amplitude and is only superimposed in a small number of pulses at the beginning of the event. The heavy level corresponds to the segment with high coefficient and long duration. The compensation amount is 80% to 90% of the peak amplitude and spans the entire event interval. In the recovery segment, it extends backward for dozens of pulses with exponential decay. The reason for using exponential decay instead of a nonlinear ramp is that the actual velocity recovery is closer to the first-order process. Its time constant can be deduced from the recovery slope of the segment. Using linear cutoff would leave a reverse deviation at the end of the recovery segment. The compensation amount and extension range of the medium level are linearly interpolated between the light and heavy levels. In the slippage feature data, the classification coefficients for mixed-type segments are obtained by weighting the coefficients of instantaneous impact and continuous drift types according to their distance ratios from the thresholds of the two types, with the closer ones having higher weights. To prevent abrupt changes in compensation amounts at event boundaries between adjacent gears, the classification compensation parameters insert a linear transition of several pulses at gear shifts. The transition length adapts to the recovery slope given by the slippage feature data, ensuring a smooth transition of compensation amounts along the event path without leaving new steps on the image. For segments with unstable entry points, a steady-state correction is superimposed on the classification compensation parameters to align the compensation starting point with the true zero-bias baseline before slippage. Segments classified as low-confidence in the slippage feature data are treated with a reduction ratio in the classification compensation parameters to prevent them from being over-recognized in the classification.

[0037] The slippage event is marked by classifying and labeling the slippage segment according to the graded compensation parameters and evaluating the event level. The level and compensation coefficient of each segment in the graded compensation parameters determine its event level: mild corresponds to level 1, moderate to level 2, and severe to level 3. The higher the level, the greater the spacing correction applied by the scheduling triggered by the slippage event mark in the corresponding interval. If adjacent segments in the same batch have the same level and the interval is less than the set merging window, they are merged into a continuous event. The merging window is taken as the minimum distinguishable slippage interval, about 5 to 10 pulses. The reason for merging is that adjacent events with extremely short intervals will force the scheduling to frequently switch the compensation amount at the two boundaries. The timing jitter introduced by the switching itself may induce new line spacing periodic fluctuations. Merging eliminates this hidden danger, so that the slippage event mark will not induce secondary fluctuations. The compensation amount after merging is determined by weighted average of the number of continuous pulses of the two segments. When the scheduling processes merged events, a more conservative switching slope is used to prevent the weighted compensation amount from being locally overcompensated in a certain original segment. Adjacent segments that exceed the merging window are each treated as independent items. To facilitate direct execution downstream, the slippage event marker also records three key fields for each event: start and end pulse sequence number, peak compensation amount, and recovery segment decay time constant. The decay time constant is provided by fitting the recovery segment of the hierarchical compensation parameters, allowing scheduling to reconstruct the compensation envelope without backtracking the original differential, significantly reducing the computational burden and latency of real-time scheduling. For segments with reduction labels in the hierarchical compensation parameters, the compensation amount is capped in the slippage event marker by the reduced coefficient. For segments with steady-state correction amounts, the corresponding correction amount is added to the slippage event marker, ensuring that each marker carries a directly executable compensation magnitude.

[0038] After spacing compensation is applied to slippage event markers, pulse-by-pulse timing scheduling generates row trigger pulses. Trigger scheduling uses the core value of the trigger spacing parameter as the target inter-row spacing, accumulating the spacing pulse-by-pulse on a high-resolution pulse sequence. Once the target value is reached, a row trigger pulse is output and the sequence is reset and recalculated. Without slippage, this mechanism ensures that the physical spacing between adjacent row triggers is strictly equal to the target row spacing. After optimal phase interval positioning via resonance, row spacing fluctuations can be reduced to the theoretical row spacing level of 0.1%, and the geometric cumulative error approaches zero when tens of thousands of rows are continuously acquired. When a slippage event falls within the event marker interval, scheduling dynamically corrects the target spacing according to the event level and compensation amount. Smaller compensations are superimposed on first-level intervals, and larger compensations are superimposed on third-level intervals. The compensation amount extends exponentially backward from the event in the recovery segment, conforming to the true convergence curve of the fitting speed, bringing the row trigger pulse position back to the correct displacement. For slippage event markers with reduction labels, the compensation amount does not exceed the upper limit of the reduction coefficient, and the corresponding row trigger pulse position is recorded as conservative compensation, indicating that the spacing deviation at that location has not been fully corrected. When resonance compensation and slippage compensation are applied simultaneously, they are independently superimposed in the time domain. The timing accuracy of the line trigger pulse depends on the premise that the two are orthogonal—the frequency bands and triggering mechanisms are different, and the residual coupling is very small. To verify the compensation effect, after each slippage event, the scheduler compares the actual cumulative displacement of the line trigger pulse segment with the theoretical displacement. If the residual exceeds the limit, the event is marked as insufficient compensation, and the compensation is automatically increased by one level for the next similar operating condition, so that the geometric accuracy of the line trigger pulse is self-tuned during operation. When the slippage event marking boundary is updated, the scheduler switches the compensation amount from the next pulse, which is applied to the line trigger pulse timing without hysteresis. The switching ratio in the high-speed segment is extremely small, and the additional timing deviation is negligible.

[0039] Step S103: Screen the jitter deviation of the speed change segment of the row trigger pulse to form a delay correction benchmark. Based on the delay correction benchmark, perform master-slave cascade triggering and phase-locked loop correction to obtain the global synchronization time base. Based on the global synchronization time base, perform first and last stage drift measurement to generate a synchronization lock state.

[0040] Specifically, a delay correction benchmark is established by screening the jitter deviation of the line trigger pulse during the speed abrupt change segment. In a multi-channel cascaded system, each module has an inherent propagation delay from receiving the line trigger pulse to initiating exposure. Differences in device parameters cause slight variations in the delays of each channel. When the delays are inconsistent, the actual exposure times of each camera under the same line trigger pulse correspond to different physical positions on the scanned object. This results in blurred edges at the misalignment points during multi-channel stitching. This misalignment is particularly critical when the size of the inspected defect is close to the level of the line spacing, thus requiring the establishment of a delay correction benchmark for correction. Within the speed abrupt change segment, the spacing of the line trigger pulses changes non-uniformly with the speed. Random jitter of the trigger path causes much greater spacing disturbance in this segment than in the stable segment. Directly using the spacing of the abrupt change segment to estimate the inherent delay will inevitably introduce a large deviation. The screening process first identifies the boundaries of each velocity abrupt change segment in the horizontal trigger pulse sequence, delineating the corresponding trigger edges as exclusion intervals, and only taking the effective spacing in the stable segments. Then, median and interquartile range (IMR) analyses are performed on the effective spacing within each batch. The median and IMR are used instead of the mean and standard deviation because they are less sensitive to outliers and can prevent distorted delay correction benchmarks. Spacing exceeding a set multiple of the IMR is considered outlier and removed. The mean of the remaining effective spacing is the core value of the delay correction benchmark. To adapt to gradual changes in horizontal frequency, the core value of the delay correction benchmark is updated exponentially with a slow time constant between batches, filtering out single-batch jitter and keeping up with device thermal drift, ensuring consistent delay compensation from the horizontal trigger pulse to exposure over the long term. Trigger edges with conservative compensation labels have lower reliability in spacing and are given additional weight in the statistics. When the number of effective samples is below the lower limit, the core value of the delay correction benchmark is temporarily replaced by the result of the previous batch.

[0041] In some embodiments, the step of obtaining a global synchronization time base by performing master-slave cascaded trigger edge phase-locked correction based on the delay correction reference includes: capturing the master unit trigger edge time and periodically latching to establish a trigger edge timestamp based on the delay correction reference; measuring the phase deviation between clock reference edges to obtain the trigger edge phase difference based on the trigger edge timestamp; performing phase difference change rate advance correction step size prediction based on the trigger edge phase difference and outputting a phase-locked correction command; and performing master-slave cascaded clock phase closed-loop adjustment based on the phase-locked correction command to form a global synchronization time base.

[0042] The main unit trigger edge timing is captured and periodically latched to establish trigger edge timestamps based on a delay correction reference. The system clock latches the current count and forms a trigger edge timestamp upon the arrival of each trigger edge in the row trigger sequence. Latching is completed within one clock cycle after trigger edge confirmation. Typical linear array system clocks are in the hundreds of megahertz to gigahertz range, with a time resolution in the nanosecond to sub-nanosecond range, sufficient to resolve sub-microsecond jitter in high-speed scenarios. The resolution must be a trade-off between counter bit width and system power consumption. The difference between two adjacent latched trigger edge timestamps represents the trigger edge interval. This is compared with the delay correction reference core value. Adjacent edge pairs falling within the positive and negative tolerances of the delay correction reference core value are considered valid periods, and their difference is recorded as the period latch value for that edge. Trigger edges falling within the exclusion range marked by the delay correction reference are excluded and do not participate in subsequent phase measurements to avoid contaminating phase comparisons due to non-uniformity in the interval between abrupt changes. To prevent the system clock's own phase noise from infiltrating, the latch count for the trigger edge timestamp is taken from the master clock purified by the phase-locked loop. A clock quality flag is added to each latch. When the master clock jitter exceeds the threshold, this flag is set, and the corresponding trigger edge timestamp entry is downweighted in subsequent phase measurements. This ensures that both the delay correction reference and the trigger edge timestamp are established on a clean time reference. If the adjacent spacing exceeds a multiple of the core value setting, it is considered an abnormal spacing. The corresponding time is still recorded but marked separately in the trigger edge timestamp. Consecutive occurrences of abnormal spacing often indicate intermittent open circuits in the trigger link or abnormal trigger source, requiring a return to the row trigger generation stage for investigation. This ensures that the trigger edge timestamp provides a clean and reliable time reference for subsequent phase measurements.

[0043] The trigger edge phase difference is obtained by measuring the phase deviation between the clock reference edges and the trigger edge timestamp. The clock reference edge is generated by the master unit clock source according to a set period and serves as a reference for phase alignment of each slave unit. The phase deviation P_n = t_n - t_ref of each valid trigger edge latch time in the trigger edge timestamp is obtained by subtracting the most recent reference edge time from the effective trigger edge latch time in the trigger edge timestamp. Here, the phase deviation is measured in time, with a positive value indicating that the trigger edge lags behind the reference and a negative value indicating that it leads. Trigger edges marked with exclusion in the trigger edge timestamp are not included in the measurement; the trigger edge phase difference of edges marked with abnormal spacing is still calculated but recorded as low confidence. Low confidence entries are not included in the subsequent phase difference change rate calculation to avoid nonlinear phase jumps caused by trigger link anomalies from distorting the lead prediction. The measurable range of the trigger edge phase difference is only up to half the period of the reference edge. Exceeding this range results in phase rollover. Rollover is identified by a sudden change in phase difference between adjacent edges exceeding half a period; once identified, the phase difference returns to the main value range. To improve phase resolution under weak signals, the trigger edge phase difference undergoes a least-squares regression based on the trigger edge timestamp after filtering. The regression slope is used to estimate the average phase drift rate, reflecting both instantaneous deviation and trend. In high-frequency triggering scenarios, the reference edge period is very short, and timing noise accounts for a larger proportion of the phase deviation. Therefore, a sliding window mean filter is applied to the trigger edge phase difference of each edge within the batch. The filtered deviation better represents the steady-state phase difference of the system. Both the original and filtered values ​​are derived from and retained from the trigger edge timestamps; the former is used for rollover detection, and the latter for rate of change calculation.

[0044] The phase difference is predicted based on the phase difference at the trigger edge, using a lead correction step size to output the phase-locked loop (PLL) correction command. Pure feedback PLL lags when the phase difference changes rapidly at the trigger edge, causing the correction command to always fall behind the actual phase difference change. This can even lead to continuous phase difference oscillations and difficulty in convergence during acceleration and deceleration. Lead correction step size prediction, on the other hand, calculates the rate of change of the phase difference between adjacent trigger edges to predict the phase position of the next edge, and adds the corresponding compensation step size to the PLL correction command in advance. This is equivalent to adding a feedforward path in addition to the feedback, significantly accelerating convergence and enabling locking within the few seconds of acceleration from startup to rated speed. The phase difference rate is calculated by dividing the difference in deviation between adjacent valid entries after filtering by the corresponding edge spacing. A larger rate of change indicates a faster phase shift, and the lead step size is increased proportionally. Low-confidence entries are not included in the rate of change calculation to avoid misinterpreting abnormally spaced phase jumps as actual drift. The direction of the phase-locked loop (PLL) correction instruction is opposite in sign to the phase difference of the current trigger edge. The step size is given by U_n = k_p × |P_n| + c_lead, where U_n is the correction step size for the nth edge, k_p is the proportional coefficient, P_n is the current phase difference, and c_lead is the lead compensation amount calculated by multiplying the phase difference change rate by the lead gain. k_p is calibrated according to the clock frequency and the maximum correction step size during debugging; too large a step size can lead to overshoot, while too small a step size can result in slow convergence. To prevent overcorrection, a lower limit dead zone is set for the single-step size of the PLL correction instruction. No correction is issued when the phase difference of the trigger edge falls within the dead zone, avoiding jitter near zero after locking. When the phase difference change rate continuously exceeds the threshold, the step size of the PLL correction instruction is limited to the maximum allowable value and recorded as tracking over-limit. This limitation also serves to resist integral saturation and prevent the compensation amount from reaching an uncontrollable amplitude during sudden error changes.

[0045] A global synchronization time base is formed by closed-loop adjustment of the master-slave cascaded clock phase based on phase-locked loop correction commands. In the cascaded structure, the master unit holds the reference clock, and the clock frequency and phase of each slave unit are adjusted in real time by a digitally controlled oscillator driven by the phase-locked loop correction commands. The oscillator updates the phase offset once per trigger cycle, continuously locking onto the phase of the master unit's trigger edge, and converges towards the global synchronization time base cycle by cycle. The locking window width is determined according to the system timing tolerance during initialization. If it is too narrow, the convergence is slow; if it is too wide, the actual phase error may exceed the imaging tolerance. A compromise must be made based on the measured error distribution during debugging. Convergence is judged when the phase difference of several consecutive trigger edges falls into the locking window. After convergence, the global synchronization time base is output in a locked state; before convergence, it is in a tracking state. When the phase-locked loop correction command has a tracking over-limit label, it means that the external disturbance exceeds the capability of the phase-locked loop. The closed loop switches to fast tracking mode, and the upper limit of the correction step size is widened to twice the normal level. In this mode, the phase adjustment amplitude of each slave unit is larger, and sub-pixel misalignment may occur briefly in multi-channel splicing. The corresponding global synchronization time base is continuously marked as tracking state to distinguish it from normal locking. If the tracking exceeds the set number of batches consecutively, a lockout alarm is triggered. To shorten the relocking time, the global synchronization time base retains the most recent valid phase offset as the initial value for relocking after a lockout alarm, ensuring that the phase-locked correction command starts from near the convergence point rather than searching from zero. In cascaded circuits, each stage of the slave unit serially accesses the master unit's trigger path, and the synchronization residuals of the preceding stage are superimposed on the subsequent stage. The more stages there are, the greater the accumulation in the final stage. Taking an eight-channel cascade as an example, when the residual of the first stage is on the order of nanoseconds, the residual of the final stage can accumulate to tens or even twenty nanoseconds. The global synchronization time base records the phase residual sequences of each stage, and by comparing them stage by stage, the weakest stage in synchronization can be identified.

[0046] Synchronization lock state is generated by measuring the first and last stage drift based on the global synchronization time base. First and last stage drift refers to the total phase difference formed after accumulating multiple stages of phase residuals between the first and last stage slave units in a cascaded system. Its increase over time stems from the different clock temperature coefficients of each slave unit: during several hours of large-format acquisition, the temperature rise of each stage is inconsistent. Even if the single-stage residual is only a few picoseconds, the last stage can accumulate to tens or even hundreds of picoseconds after eight stages. Once this exceeds the line image timing tolerance, the imaging quality significantly deteriorates. Therefore, it is continuously measured using the global synchronization time base and classified into a synchronization lock state. During measurement, the difference sequence between the first and last stages is taken from the residual sequence of each stage of the global synchronization time base. The moving average provides the drift trend component, and the moving standard deviation provides the random component; both together constitute the drift description of the synchronization lock state. When the trend component continuously exceeds the warning threshold, a drift warning is marked in the synchronous lock state, indicating that the current drift rate may exceed the imaging tolerance within the next few hours, and recalibration should be arranged in advance. For batches with unlock alarms, the drift measurement for the corresponding time period is unreliable, and the trend analysis for that time period is obtained by linear interpolation of reliable time periods before and after. To distinguish between trend drift and occasional disturbances, the synchronous lock state sets a dual time window for the drift trend component: a short window to capture sudden changes and a long window to characterize slow drift. The combination of the two is used to jointly determine whether immediate shutdown and calibration are required by the residual trend of the global synchronization time base. The lock level of the synchronous lock state is comprehensively evaluated based on the proportion of locked and unlock alarm batches in the global synchronization time base: a very low unlock ratio indicates complete lock, a medium ratio indicates partial lock, and a ratio exceeding the set proportion indicates unlock. The lower the level, the worse the consistency of multi-channel inter-line stitching.

[0047] Step S104: Based on the synchronous lock state, verify the pulse count deviation of the exposure window to obtain the fuzzy risk mark. Use the fuzzy risk mark to collect the out-of-limit frame sequence number and the spacing fluctuation amplitude to form a frame quality record. Archive the frame quality record before pause to establish a breakpoint latch record.

[0048] In some embodiments, the step of verifying the exposure window pulse count deviation based on the synchronization lock state to obtain a fuzzy risk label includes: measuring the real-time pulse count within the exposure window of each channel based on the synchronization lock state to generate an exposure period pulse count; performing pulse differential-assisted hierarchical deviation tracking on the exposure period pulse count to establish a deviation hierarchical record; identifying the inter-frame deviation sign flip frequency and deviation amplitude distribution on the deviation hierarchical record to obtain a fuzzy cause category; and using the fuzzy cause category to perform classification risk level threshold matching and out-of-limit frame labeling to obtain a fuzzy risk label.

[0049] The exposure period pulse count is generated by measuring the real-time pulse count within the exposure window of each channel based on the synchronization lock status. The synchronization lock status serves as both the time base for measurement and the basis for deviation correction and validity determination: before measurement, the lock level of the synchronization lock status and the phase residual of each channel are read to calibrate the reference and eliminate unreliable time periods. The exposure window of each channel is opened by the rising edge of the row trigger and closed by the falling edge. The window width is fixed according to the target exposure duration during initialization. The number of high-resolution pulses arriving during the window period is latched at the end of each frame's exposure, i.e., the actual pulse count N_act for that frame. The reference pulse count N_ref is estimated by removing the corresponding target position of the exposure window and using the average pulse interval. The difference between the two is the frame deviation ΔN_f = N_act - N_ref given by the exposure period pulse count. A positive deviation indicates that the frame covers more physical rows than expected, while a negative deviation indicates fewer rows. For batches in the sync-locked state that are partially locked, the corresponding channel N_ref must be corrected according to the current phase residual of that channel. Otherwise, systematic phase deviation will misjudge truly zero-deviation frames as having deviations, contaminating the pulse count during the exposure period. For batches in the sync-locked state marked with unreliable time periods, the corresponding frames are marked as unreliable and excluded from the deviation distribution statistics. There is also a sensitivity issue that varies with speed: at high speeds, the number of pulses in a single exposure window can reach hundreds to thousands, and missing one pulse has almost no impact on the deviation ratio; at low speeds, there may only be a few to dozens of pulses in the window, and missing one pulse has a significantly amplified relative impact. Therefore, the deviation threshold for pulse counting during the exposure period at low speeds is adaptively tightened according to the pulse frequency, so that the same absolute deviation is not easily missed at low speeds. This tightening strategy is also dynamically fine-tuned according to the sync-locked state's lock level, and the tightening range has an upper limit to prevent normal counts from being misjudged as deviations.

[0050] Pulse differential counting during the exposure period is used to establish a layered deviation record for deviation tracking. The absolute value of the deviation in each frame given by the pulse count during the exposure period determines its layer assignment in the deviation layering record. The layer boundaries are determined during initialization according to the historical blur rate distribution, generally set to three layers: slight, moderate, and severe. The pulse deviation rate threshold corresponding to each layer is confirmed by comparing known clear images with blurred images during debugging. After arranging the deviations of each frame by the pulse count during the exposure period in frame order, the difference between adjacent frames forms a pulse differential sequence, which is used to identify the direction of deviation change: a positive difference indicates that the deviation has increased compared to the previous frame, and a negative difference indicates that it has contracted. A large difference amplitude indicates that the deviation is changing rapidly, and a stable amplitude close to zero indicates that the deviation is in a plateau state. These two forms often have different causes of blurring, so the deviation layering record includes this direction information in addition to the layer level. The layer level of frames marked as unreliable is replaced by linear interpolation of the effective frames before and after. Extreme frames with deviations exceeding three times the upper limit of the severe layer are marked separately. These extreme values ​​are mostly caused by mechanical shock or temporary encoder signal failure and are not included in the layer distribution statistics to avoid distorting the overall distribution of the deviation layering record. To balance sensitivity and stability, the layer determination in the deviation layering record incorporates a hysteresis step. The layer is only updated when the deviation crosses the boundary of the adjacent layer for several consecutive frames, preventing frequent layer jumps caused by single-frame noise in the pulse count during the exposure period. During the acceleration phase, the pulse interval changes rapidly, and the difference in the number of pulses within the exposure window of adjacent frames is significant. The pulse difference sequence exhibits continuous unidirectional change in this phase, and the layer crossing rate of the deviation layering record is significantly higher than in the constant-speed phase. This is a normal reflection of the speed change and is confirmed by the synchronous change trend of the pulse count during the exposure period.

[0051] For example, the step of identifying the inter-frame deviation symbol flipping frequency and deviation amplitude distribution of the deviation layered record to obtain the fuzzy cause category includes: performing frame-by-frame annotation of the deviation symbols and identification of the flipping event period in the deviation layered record to obtain a flipping frequency sequence; determining the deviation amplitude distribution partition by judging the boundaries of the autocorrelation oscillation segment and decay segment of the flipping frequency sequence; classifying the fuzzy cause weights by assigning vibration jitter type and velocity drift type cause weights according to the deviation amplitude distribution partition; and generating the fuzzy cause category by performing a comprehensive rating and optimization judgment of the cause category based on the fuzzy cause weights.

[0052] The deviation stratification record is used to generate a frame-by-frame labeling of deviation symbols and identification of flip event cycles to obtain a flip frequency sequence. First, the positive and negative signs of the deviations in each frame of the deviation stratification record are extracted into sign bits. If adjacent frames have different signs, a flip event is recorded, along with its frame number and the layer before and after the flip. The number of frames between two consecutive flips is the flip period, and its reciprocal is the flip frequency for that segment. These segments are then compiled into a flip frequency sequence. Extreme deviation frames marked in the deviation stratification record are not included in the flip determination to prevent abnormal sign jumps caused by mechanical shocks from being treated as vibration flips. A flip event must occur at least twice consecutively to be included in the valid period statistics. A single isolated flip is recorded as a zero frequency. These isolated flips typically involve a brief spike in speed followed by a rapid return to the mean, and their image quality risk is lower than that of continuous vibration segments; therefore, they are included with a low weight in the flip frequency sequence. In terms of frame sequence, flipping exhibits two types of patterns: segments with highly consistent periods correspond to the positive and negative oscillations caused by mechanical vibration, where the relative vibration frequency between the scanned object and the camera directly determines the flipping period; the faster the vibration, the more frequent the flipping. Segments where the flipping period monotonically increases with frame sequence correspond to the transition from high or low velocity to normal, where the deviation sign approaches zero as the velocity nears normal, and the flipping stops. To avoid truncation and omission of low-frequency, long-period flipping, the statistical window length of the flipping frequency sequence is set to two periods of the expected lowest vibration frequency. If the window contains less than one complete flipping period, the judgment is suspended until the data is supplemented before being filled back in, ensuring no distortion in the initial segment of the batch. In the flipping frequency sequence, the former appears as a stable value, while the latter appears as a monotonically decreasing value. Combined with the hierarchical trend of the deviation stratification record, they can be distinguished temporally by their slope.

[0053] The deviation amplitude distribution is partitioned by identifying the boundaries of the autocorrelation oscillation and decay sections in the flip frequency sequence. If the vibration-driven segment and the velocity convergence segment are treated together without distinction, appropriate out-of-limit parameters cannot be selected for each cause during threshold matching, leading to a significant increase in misjudgments. Therefore, the deviation amplitude distribution partition is first determined based on the flip frequency sequence. Autocorrelation is performed on the frequency values ​​of the flip frequency sequence: in the vibration-driven segment, the flip frequency remains near a fixed value, and the autocorrelation shows a significant peak at the corresponding delay; in the velocity convergence segment, the flip frequency changes monotonically, and the autocorrelation shows no significant peak but decays power-law. The difference between these two types of autocorrelation patterns is the boundary criterion for the deviation amplitude distribution partition. The boundary of the oscillation section is defined by the frame range where the confidence level of the autocorrelation peak continuously exceeds the threshold. Static segments without obvious flips in the flip frequency sequence are separately classified as static segments in the deviation amplitude distribution partition. This corresponds to situations where the deviation is stable at a non-zero plateau, often seen when the scan speed is consistently low or the systematic delay is not corrected. When the number of frames in the static segment exceeds half of the entire batch, the reliability of the dynamic cause analysis for this batch decreases. The autocorrelation peak at the junction of the oscillation and decay segments gradually decreases, which is denoted as boundary ambiguity, and a transition coefficient is applied to the subsequent weight allocation. To improve the stability of the partition boundaries, the boundary determination of the deviation amplitude distribution partition is subject to a minimum segment length constraint. Fragmented partitions shorter than this length are merged into adjacent main partitions, so that the partitions obtained from the flip frequency sequence are not fragmented by local spikes. Entries in the flip frequency sequence whose frequencies exceed the set upper limit are separately classified into high-frequency oscillation segments in the deviation amplitude distribution partition. The vibrational ambiguity at the edge of the corresponding motion direction is the most severe. Under the same pulse deviation rate, high-frequency vibrations give a significantly stronger sense of ambiguity than low frequencies.

[0054] Based on the deviation amplitude distribution partition, the weights for vibration-jitter and velocity-drift causes are allocated sequentially to form the fuzzy cause weights. The basic allocation of fuzzy cause weights is determined by the proportion of frames in each segment of the deviation amplitude distribution partition: the higher the proportion of oscillating segments, the greater the weight for vibration-jitter; the higher the proportion of attenuating segments, the greater the weight for velocity-drift, and the sum of the two is equal to one. The number of frames in stationary segments in the deviation amplitude distribution partition is deducted from the total number of frames before allocation to avoid diluting the relative proportion of the two types of dynamic causes. The number of frames in high-frequency oscillating segments in the deviation amplitude distribution partition is counted at 1.5 times when calculating the weight for vibration-jitter, because high-frequency vibration contributes more to fuzziness than low-frequency vibration, and not weighting it would underestimate the true share of vibration causes, thereby improving the representativeness of the vibration side in the fuzzy cause weights. When the effective dynamic frames are zero, the fuzzy cause weights are directly inherited from the previous batch. The weights of the two types of boundary-fuzzy frames are each taken in half and then weighted and merged with the boundary-clear frames and written into the fuzzy cause weights. To reflect the migration of causes over time, the weights of fuzzy causes are updated in a sliding manner between batches. The weights of the current batch are merged with the historical weights according to their reliability, so that occasional fluctuations in the distribution of deviation amplitudes do not cause sudden jumps in the weights of fuzzy causes. There is also a special case that needs to be handled: when the scanned object itself has periodic textures and its texture frequency is close to the system vibration frequency, the flip frequency will couple with the texture frequency, causing the vibration jitter type weights to be artificially high. For this reason, a texture frequency exclusion check is performed on such batches. If the check is successful, the vibration jitter type weights of the corresponding frame segment are reduced to half of the normal value and written back to the fuzzy cause weights. Frame segments that fail the check are marked as suspected textures and left for manual review.

[0055] Based on the fuzzy cause weights, a comprehensive rating and optimization judgment of cause categories are performed to generate fuzzy cause categories. When the weight of vibration jitter exceeds the set dominant threshold (usually 0.7), vibration jitter is judged as the dominant cause of this batch and written into the fuzzy cause category. If the weight of velocity drift exceeds the dominant threshold, velocity drift is judged as dominant. If neither exceeds the threshold, it is classified as a mixed type. The mixed type corresponds to vibration and velocity alternating to dominate within the batch, which is common in high-speed scanned surfaces with slight frictional chatter. For batches with similar weights and no significant difference, the dominant threshold is relaxed to 0.6. If the dominant cause still cannot be determined after relaxation, it is recorded as cause uncertain. For these batches, the most conservative set of thresholds is used in subsequent threshold matching to prevent overcompensation or undercompensation due to incorrect cause judgment. The comprehensive rating score is calculated by multiplying each of the two weights by its benchmark risk coefficient and then summing the results: G = w_v × a_v + w_d × a_d, where G is the comprehensive rating score, w_v and w_d are the two weights with w_v + w_d = 1, and a_v and a_d are the benchmark risk coefficients. A higher score indicates a more destructive cause of the deviation; exceeding the production benchmark value by a certain percentage triggers an equipment status warning. The fuzzy cause category also outputs a confidence level, defined as the difference between the dominant and secondary weights divided by the dominant threshold; the larger the difference, the higher the confidence level. This confidence level directly adjusts the acceptance strength of the cause category—if it is higher than the set value, it is directly classified as the dominant cause; if it is lower than the set value, the category is downgraded to provisional, and the comprehensive rating score is multiplied by a reduction factor less than one, preventing over-acceptance of batches with ambiguous boundaries, and automatically biasing subsequent threshold matching towards more conservative values. If the dominant cause of adjacent batches changes abruptly, it is recorded as a cause change at the batch switch point. Repeated occurrences often indicate potential faults in the transmission system and should be investigated from a mechanical perspective.

[0056] A fuzzy risk label is obtained by classifying risk levels and matching thresholds based on fuzzy cause categories and labeling out-of-limit frames. Different sets of out-of-limit thresholds are assigned to different fuzzy cause categories: when vibration jitter is dominant, the threshold is determined based on vibration amplitude characteristics; when velocity drift is dominant, the threshold is determined based on velocity deviation tolerance. During debugging, the two sets of thresholds are determined by subjective scoring of fuzzy images and a comparison experiment of deviation rates. For mixed types, the two sets of thresholds are interpolated, and the interpolation coefficients are given by the relative proportions of the two categories' weights. Based on this, fuzzy risk labels are labeled for each frame. The level of each frame in the deviation stratification record is compared with the level of the matched out-of-limit threshold. Frames exceeding the threshold are considered out-of-limit frames and are classified into three levels according to the degree of exceedance in the fuzzy risk label: Level 1 is slightly out-of-limit with some blurred edges but still discernible features; Level 3 is severely out-of-limit with object size distortion exceeding 1%, and is directly rejected in precision size detection. When acquiring data synchronously across multiple channels, the distribution of out-of-limit frames within the same batch may differ. Each channel independently generates out-of-limit labels, which are then merged into the corresponding channel field. The highest risk level extreme value for each channel is then used as the overall risk level for the entire batch and written into the fuzzy risk marker. Considering multi-frame correlation, the fuzzy risk marker treats isolated single-frame out-of-limit and consecutive multi-frame out-of-limit differently. Isolated out-of-limit are often occasional noise, and their risk level is lowered by one level. Consecutive out-of-limit are adjusted upwards based on the destructiveness given by the fuzzy cause category, making it closer to the true assessment of image usability. When the proportion of out-of-limit frames within a batch exceeds a set percentage, the entire batch is downgraded, indicating that the overall acquisition quality of this segment is substandard. For batches with an uncertain fuzzy cause category, their overall risk level is further increased by one level in the fuzzy risk marker to compensate for potential missed detections due to inaccurate cause determination.

[0057] A frame quality record is formed by using fuzzy risk markers to collect out-of-limit frame numbers and spacing fluctuation amplitudes. First, the risk levels and out-of-limit labels of each frame in the fuzzy risk markers are compiled into a frame-level index arranged in frame order. Each frame number is recorded along with its risk level and the associated cause type; this forms the core of the frame quality record. Consecutive out-of-limit frames are merged into consecutive out-of-limit segments in the index based on segment boundaries, recording the first and last frame numbers and the highest risk level within the segment. The merging window is set to the smallest frame granularity for subsequent recovery operations to avoid excessively long indexes from densely packed out-of-limit frames. For batches with batch downgrade labels in the fuzzy risk markers, the entire range of frame numbers is included in the index, rather than being listed frame by frame. The spacing fluctuation amplitude is calculated by taking the standard deviation of the pulse spacing within the exposure window of each frame in the high-resolution pulse sequence of this batch. A larger standard deviation indicates more unstable pulse spacing and worse acquisition motion during the exposure of that frame. It corresponds one-to-one with each out-of-limit frame and is included in the frame quality record, characterizing the image quality risk of that frame along with the risk level. To support cross-batch trend analysis, the frame quality record also retains a compact feature summary for each batch, including three items: the dominant cause, the proportion of out-of-limit frames, and the average amplitude of spacing fluctuations. This allows for comparison without expanding frame-by-frame indexing during long-term backtracking and facilitates quick extraction of frame quality records from recent batches during breakpoint archiving. The frame quality record is organized by batch, with each batch entry containing the set of out-of-limit frames, the amplitude of spacing fluctuations between each frame, and the batch's comprehensive score. The comprehensive score is obtained by weighting the proportion of out-of-limit frames and the highest risk level. The index-weighted average of the comprehensive scores from the most recent batches is then used as the quality trend. When the trend continues to decline and the rate of decline exceeds a set value, it means that the acquisition quality is continuously deteriorating and a full calibration should be initiated. This determination is based entirely on the frame quality record.

[0058] A breakpoint latch record is created by archiving the frame quality records before the pause. Archiving is triggered when the acquisition system receives a pause command or detects a line-triggered stop, and is completed within a fixed time window after pause confirmation. The most recent batches of quality data in the frame quality records are retrieved and latched into the breakpoint latch record within this window. The number of batches retrieved depends on the historical analysis window required for recovery calibration, typically 3-5. The core of the breakpoint latch record is the breakpoint frame number, which locates the last valid frame with a risk level lower than level one before the pause in the frame quality records—that is, the last moment before the acquisition quality degraded. Resuming from this frame ensures geometric continuity with the last frame of the pause. If the comprehensive scores of the last few batches in the frame quality records are all lower than the minimum acceptable benchmark, the breakpoint frame number is rolled back to the last valid frame of an earlier batch, and the breakpoint rollback is marked in the breakpoint latch record, indicating that the recovery start point is earlier than the nominal pause time. The mean and standard deviation of the interval fluctuation amplitude of the several batches before the pause in the frame quality records are also included in the breakpoint latch record to characterize the overall stability of the acquisition state before the pause; the more stable the data, the more reliable the breakpoint quality. To address the risk of data not being written to disk during pauses, breakpoint latch records employ dual buffering and alternating writing. A new record is only replaced after it has been archived and verified, ensuring that a complete and usable breakpoint latch record is always available for recovery. During temporary production line shutdowns for material changes or cleaning of surfaces, encoder temperature and mechanical tension may fluctuate. The greater the difference between the fluctuation characteristics of the breakpoint record and the initial measured fluctuation after recovery, the greater the change in mechanical condition. If the difference exceeds the set upper limit, the operator should redo the complete initialization before restarting.

[0059] Step S105: Obtain the recovery start parameter from the breakpoint latch record location trigger coordinate and frame quality compensation parameter; perform acceleration segment meter correction on the recovery start parameter to generate a resume transmission trigger sequence; and output a multi-path trigger command based on the continuity deviation of the breakpoint sequence number tracked by the resume transmission trigger sequence.

[0060] In some embodiments, obtaining the recovery start parameter from the breakpoint latch record by locating the trigger coordinates and frame quality compensation parameters includes: verifying the breakpoint frame number and the last value of the interval from the breakpoint latch record to obtain the breakpoint trigger final state; measuring the pre-pause deviation of the breakpoint trigger final state frame by frame to form the breakpoint coordinate deviation average; determining the resume correction amount by measuring the confidence interval of the resume first frame correction based on the breakpoint coordinate deviation average; and encapsulating the resume correction amount with coordinate and frame quality compensation parameters to obtain the recovery start parameter.

[0061] The final state of the breakpoint trigger is obtained by verifying the breakpoint frame number and the final value of the interval from the breakpoint latch record. The verification starts from the breakpoint position marked in the breakpoint latch record, confirming that the frame number has a valid record in the high-resolution pulse sequence and is not marked as an untrusted frame. Based on this, the final state of the breakpoint trigger is established. If the breakpoint latch record already has a breakpoint rollback mark for the sequence number, the verification starting point is automatically switched to the last valid frame of an earlier batch to ensure that the position pointed to by the final state of the breakpoint trigger is physically reliable. The final value of the interval is the end value of the interval sequence 3 to 5 pulses before the trigger pulse corresponding to the breakpoint frame. If the window is too short, the single pulse noise will have a large impact on the final value; if it is too long, the final value will regress to the average speed and deviate from the true state at the moment of pause. This end value represents the last valid observation of the encoder's running state at the moment of pause and is written into the final state of the breakpoint trigger for reference benchmarks in the acceleration phase speed curve modeling after recovery. If there are extreme deviation markers near the breakpoint frame, or if the breakpoint latch record gives a comprehensive score significantly lower than the batch average, the final value of the interval is replaced by the average of the final values ​​of the previous few valid frames, and the weight of that frame is halved in subsequent deviation measurements to prevent the estimated final value from leading the deviation average in a distorted direction. To improve the representativeness of the final value, the velocity within the window is confirmed to be stable before taking the endpoint value in the breakpoint-triggered final state. If the final window falls on a small fluctuation, the window is shifted forward to make the running state recorded in the breakpoint-triggered final state as close as possible to the true steady-state final value. The breakpoint-triggered final state also records the comprehensive score of the batch in which the breakpoint frame belongs. When the score is lower than the benchmark, the final state is downgraded. The deviation average measurement of downgraded batches will expand the sampling range and appropriately relax the outlier judgment, so that batches with poor quality can still obtain enough valid samples without insufficient samples due to overly strict filtering.

[0062] The mean value of the breakpoint coordinate deviation is calculated frame-by-frame by measuring the deviation before the pause in the final state of the breakpoint trigger. The measurement traces back historical frames from the breakpoint frame number marked at the final state of the breakpoint trigger, typically within 10-30 frames. Too short a range results in occasional jitter having a significant impact on the mean; too long a range indicates a large difference between the velocity state of distant frames and the pause time. The difference between the actual pulse interval and the core value of the trigger interval parameter is recorded frame-by-frame. This series of differences reflects the most recent history of the motion state of the scanned surface before the pause, and the sum of these differences forms the sample basis for the mean value of the breakpoint coordinate deviation. Frames marked with unreliable or extreme deviations within the tracing range are excluded from the measurement. The mean value of the breakpoint coordinate deviation is obtained by weighting the effective frame differences after removing the first and last extreme values, with higher weights for near frames and lower weights for far frames. When the number of effective frames is below the lower limit, the mean value of the historical interval fluctuation amplitude of the batch in which the breakpoint trigger final state is located is used instead, and the sample is marked as insufficient. If the difference sequence shows a monotonically increasing or decreasing trend, it indicates that the motion state before the pause was already in a state of drift; this drift is then marked. To prevent trend terms from contaminating the mean, the mean of breakpoint coordinate deviations is first tested for a significant linear trend in the difference sequence using first-order differencing before weighting. If a trend is found, it is deducted before calculating the mean, and the trend is recorded separately. The final mean of breakpoint coordinate deviations provides more than just a mean: the mean and standard deviation characterize the central tendency and dispersion of the velocity state before the pause, while the deviation drift annotation adds directional information that the mean alone cannot express. Together, these three factors constrain the width and direction of the confidence interval for the continuation correction, ensuring that the interval reflects the statistical characteristics of the breakpoint state in this batch rather than relying solely on a single point estimate. This is also a key statistic for the transmission of the final state triggered by the breakpoint to subsequent continuation stages.

[0063] The follow-up correction amount is determined by measuring the confidence interval of the first frame correction based on the mean deviation of the breakpoint coordinates. If the mean deviation of the breakpoint coordinates is used directly as the correction amount without establishing a confidence interval to account for the uncertainty of the follow-up correction amount, a systematic directional bias will be introduced in batches with small sample sizes or trending difference sequences. The actual landing point of the first frame after follow-up will deviate unilaterally from the breakpoint coordinates, causing the line spacing of the first few frames after follow-up to converge periodically rather than in a steady state. The confidence interval is constructed around the mean of the breakpoint coordinate deviation, with the half-width determined by H = k_c × s, where H is the half-width, k_c is the confidence factor, and s is the standard deviation of the mean deviation of the breakpoint coordinates. k_c is calibrated during initialization according to the system's allowable range for the first frame's displacement deviation. When the mean deviation of the breakpoint coordinates is annotated with deviation drift, the half-width is further widened because the representativeness of the drift batch mean is poor, and the actual uncertainty of the follow-up correction amount is greater than the range described by the standard deviation. The intersection of the confidence interval and the historical deviation allowable range yields the effective candidate interval for the re-transmission correction. The midpoint of the candidate interval is used as the nominal correction value, and the width is used as a quantitative indicator of correction accuracy. To balance robustness and responsiveness, an engineering hard limit is set for the re-transmission correction outside the confidence interval, ensuring that it does not exceed the maximum correctable deviation in a single frame regardless of statistical results, thus preventing excessive jumps in the first frame even in abnormal batches. An empty candidate interval indicates that the deviation fluctuation in this batch has exceeded the normal tolerance. In this case, the nominal value of the re-transmission correction is changed to the conservative value of the midpoint between the two nearest endpoints, and the guarantee of geometric continuity of re-transmission decreases accordingly. When there is insufficient sample labeling, the re-transmission correction is simply output as zero correction, allowing the position tracking of the accelerated transition segment to naturally converge the deviation back to within the tolerance, which is more reliable.

[0064] The recovery start parameters are obtained by encapsulating the continuation correction amount with coordinates and frame quality compensation parameters. Coordinate encapsulation adds the nominal value of the continuation correction amount to the breakpoint trigger coordinates according to the mean deviation, obtaining the trigger position that the continuation system considers most likely to achieve geometric continuity, and writes it into the recovery start parameters. If the nominal value of the continuation correction amount exceeds the set maximum allowable trigger offset, it is truncated to the upper limit and marked as a correction truncation. The coordinate deviation caused by the truncation is recorded in the correction uncertainty field of the recovery start parameters; the larger the deviation, the more significant the residual difference between the continuation start point and the theoretical optimal position. Frame quality compensation parameter encapsulation takes the historical correction mean of various periodic errors and the speed stability reference from the historical batch data stored in the breakpoint latch record, and merges them into the compensation initialization field of the recovery start parameters. Historical parameters are preset to ensure that the initial compensation level of the first batch of continuation is close to the steady-state level before the pause, thus significantly shortening the initial compensation convergence time of the continuation. Multiple cameras share the same compensation initialization field across all channels, unifying the physical compensation benchmark and laying the foundation for consistent image quality across channels during the initial continuation phase. To support multiple consecutive pauses, the recovery start parameter includes a timestamp and the cumulative number of pauses for each pause. The confidence level of the resumption correction gradually tightens as the number of pauses increases, ensuring that the recovery start parameter does not drift due to the accumulation of historical errors under frequent start-stop operations and that the resumption start point remains near the breakpoint. When the quantitative index of the correction accuracy of the resumption correction is low, i.e., the candidate interval is wide, the recovery start parameter is generally marked as conservative recovery. In this case, the convergence judgment threshold of the acceleration segment is tightened accordingly, prioritizing stability over speed, to ensure that the recovery start parameter does not push the resumption start point to a dangerous position in low-confidence scenarios.

[0065] The acceleration phase meter-counting correction is applied to the recovery start parameters to generate the follow-up trigger sequence. During the encoder's recovery from standstill to scanning speed, the clock spacing between adjacent pulses gradually converges from wide to narrow. If the triggering is still performed according to a fixed line spacing target during the acceleration phase, ignoring the non-uniform spacing, the trigger position will continuously deviate from the breakpoint coordinates in terms of physical displacement. By the end of the acceleration phase, the deviation can accumulate to several times the line spacing. Therefore, the follow-up trigger sequence must be meter-counting corrected during the acceleration phase. The acceleration phase meter-counting correction uses the breakpoint trigger coordinates in the recovery start parameters as the physical target. Standard or relaxed speed curve modeling parameters are selected based on whether there is a conservative recovery annotation in the recovery start parameters. The speed curve is fitted according to the servo system's rated parameters, and is mostly an S-shaped or trapezoidal curve with jerk limitations to avoid impact during the start phase. The expected speed at each moment is calculated pulse by pulse and converted into a predicted clock spacing. Then, displacement values ​​are integrated pulse by pulse with the predicted spacing as the step size. When the accumulated displacement reaches the target displacement corresponding to the breakpoint coordinates, the first trigger point of the follow-up trigger sequence is output. After the initial output, the scheduling switches back to normal line spacing, and subsequent outputs continue the trigger sequence with the core value of the trigger spacing parameter as the target. At the end of the acceleration segment, if the difference between the predicted spacing and the measured spacing is less than a set threshold, the acceleration segment is considered to have ended, and the corresponding trigger point is marked as the end of the transition segment. To prevent integral drift, the displacement values ​​of the continued trigger sequence are integrated using the trapezoidal method, and the cumulative value is periodically corrected with measured pulses, so that the displacement error at the end of the acceleration segment does not accumulate with the number of integration steps. The trigger points within the acceleration segment are marked separately with the acceleration segment meter; if the recovery start parameter includes a correction truncation or conservative recovery label, the corresponding segment is marked as low confidence in the continued trigger sequence, leaving room for downstream expectations of the trigger accuracy of that segment.

[0066] The multi-path trigger command is output based on the continuity deviation of the breakpoint sequence number tracking in the resume trigger sequence. The difference between the last frame number recorded by each channel before the pause and the breakpoint frame number is defined as the continuity deviation of the breakpoint sequence number for that channel: if the deviation is zero, the channel can seamlessly resume from the breakpoint frame after recovery; if the deviation is non-zero, an alignment correction is added to the channel based on the resume trigger sequence, with the correction amount being a multiple of the trigger interval corresponding to the number of deviation frames. This results in the multi-path trigger command for that channel. This deviation stems from the different queuing depths of the line buffers of each channel during the pause—when the pause command arrives, some channels have already sent the last frame into the buffer queue but have not yet confirmed it, while others have already confirmed the last frame. These two types of channels naturally differ by one to several frames, so each channel must independently apply alignment correction to smooth out the difference. Trigger points marked with low confidence in the resume trigger sequence have their corresponding channel trigger times simultaneously marked as low confidence in the multi-path trigger command, indicating that the accuracy of this alignment correction is limited. To ensure consistent multi-channel startup, a synchronization alignment window is inserted before the first frame of the multi-channel trigger command. Each channel completes sequence number alignment within the window and is then released uniformly, ensuring that the commands output according to the resume trigger sequence are in the same logical frame upon recovery of the first frame. After alignment, the trigger timing of each channel converges to the same logical frame number. Using the logical frame number as the primary key, the trigger times of each channel are aggregated into a joint trigger table, i.e., the multi-channel trigger command. If the trigger time of any channel in the table deviates from the average by more than a set tolerance, the frame indicates a channel timing deviation. If such deviation occurs in multiple consecutive frames for a channel, it is likely due to a problem with its clock synchronization link, requiring a return to the global synchronization time base level for tracing the source and revising the channel component of the resume trigger sequence accordingly.

[0067] To implement the encoder-based linear scan camera synchronization triggering method corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects. See also Figure 3 , Figure 3 This diagram illustrates a structural block diagram of an encoder-based line scan camera synchronization triggering system 300 according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The encoder-based line scan camera synchronization triggering system 300 provided in this embodiment includes: Signal acquisition module 301 is used to acquire orthogonal phase signals in real time from the orthogonal differential input of the encoder using hardware logic, and to perform orthogonal frequency multiplication decoding and edge-by-edge hardware latching based on the orthogonal phase signals to generate high-resolution pulse sequences. Trigger scheduling module 302 is used to perform spacing fluctuation tracking to determine trigger spacing parameters for the high-resolution pulse sequence, use the trigger spacing parameters to perform spacing step pattern recognition to obtain slippage event markers, and perform spacing compensation on the slippage event markers and then generate row trigger pulses by pulse-by-pulse timing scheduling. The clock synchronization module 303 is used to screen the jitter deviation of the speed change segment of the row trigger pulse to form a delay correction reference, perform master-slave cascade triggering edge phase-locked correction based on the delay correction reference to obtain a global synchronization time base, and perform first and last stage drift measurement based on the global synchronization time base to generate a synchronization lock state. The quality control module 304 is used to verify the pulse count deviation of the exposure window based on the synchronous locking state to obtain a fuzzy risk mark, use the fuzzy risk mark to collect the out-of-limit frame sequence number and the spacing fluctuation amplitude to form a frame quality record, and archive the frame quality record for fluctuation before pause to establish a breakpoint latch record. The breakpoint resume module 305 is used to obtain the recovery start parameter from the breakpoint latch record location trigger coordinate and frame quality compensation parameter, perform acceleration segment meter correction on the recovery start parameter to generate a resume trigger sequence, and output a multi-path trigger command based on the resume trigger sequence to track the continuity deviation of the breakpoint sequence number.

[0068] The encoder-based line scan camera synchronization triggering system 300 described above can implement the encoder-based line scan camera synchronization triggering method of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0069] 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. A method for synchronous triggering of a linear scan camera based on an encoder, characterized in that, include: The quadrature phase signal is obtained by real-time hardware logic acquisition of the quadrature differential input of the encoder. Based on the quadrature phase signal, quadrature frequency multiplication decoding and edge-by-edge hardware latching are performed to generate a high-resolution pulse sequence. For the high-resolution pulse sequence, the spacing fluctuation is tracked to determine the trigger spacing parameter. The trigger spacing parameter is used to identify the spacing step pattern to obtain the slippage event marker. After spacing compensation is applied to the slippage event marker, the pulse-by-pulse timing is scheduled to generate the row trigger pulse. The line trigger pulse is subjected to speed change segment jitter deviation screening to form a delay correction benchmark. Based on the delay correction benchmark, master-slave cascade triggering edge phase-locked correction is performed to obtain a global synchronization time base. The first and last stage drift is measured according to the global synchronization time base to generate a synchronization lock state. Based on the synchronous locking state, the exposure window pulse count deviation is verified to obtain a fuzzy risk mark. The fuzzy risk mark is used to collect the out-of-limit frame sequence number and the spacing fluctuation amplitude to form a frame quality record. The frame quality record is archived for pre-pause fluctuation to establish a breakpoint latch record. The recovery start parameter is obtained from the breakpoint latch record location trigger coordinates and frame quality compensation parameters. The recovery start parameter is then used to perform acceleration segment metering correction and resume transmission scheduling to generate a resume transmission trigger sequence. Based on the resume transmission trigger sequence, the breakpoint sequence number continuity deviation is tracked and a multi-path trigger command is output.

2. The method according to claim 1, characterized in that, The process of generating a high-resolution pulse sequence by performing orthogonal frequency multiplication decoding and edge-by-edge hardware latching based on the orthogonal phase signal includes: The orthogonal phase signals are subjected to dual-edge timing detection and pulse edge polarity classification to obtain a pulse edge sequence; The pulse edge sequence is latched in real time by a hardware counter between adjacent edges to obtain the edge-by-edge interval count; Based on the edge-by-edge counting, the count truncation remainder is accumulated across edges to generate a pulse subdivision compensation amount; The pulse subdivision compensation amount is subjected to overflow prediction and advance zero correction to form a high-resolution pulse sequence.

3. The method according to claim 1, characterized in that, The step of determining the trigger spacing parameter by tracking the spacing fluctuations of the high-resolution pulse sequence includes: The high-resolution pulse sequence is arranged in real time using pulse-by-pulse interval sliding windows to form an interval sliding window sequence; The spacing deviation of the equally spaced repeating segments in the aforementioned sliding window sequence is identified to generate resonant wave characteristics; Based on the resonant wave characteristics, the resonance compensation amount is obtained by predicting and correcting the deviation between in-phase trigger points. The trigger spacing parameter is determined by using the resonant compensation amount to perform optimal phase interval positioning of the resonance.

4. The method according to claim 1, characterized in that, The step-pattern recognition of the spacing using the trigger spacing parameter to obtain the slippage event marker includes: The interval difference sequence is formed by pulse-by-pulse calculation of the rate of change of the interval difference before and after the trigger interval parameter; Joint screening of continuous segments of step amplitude is performed on the interval difference sequence to generate slippage feature data; The slippage feature data is fused and matched with the pre-step steady-state features and the graded coefficients to determine the graded compensation parameters. Based on the aforementioned graded compensation parameters, the slippage section is graded and labeled, and the event level is assessed to obtain the slippage event label.

5. The method according to claim 1, characterized in that, The step of obtaining a global synchronization time base by performing master-slave cascaded triggering edge-locked correction based on the delay correction benchmark includes: Based on the aforementioned delay correction benchmark, the main unit trigger edge time is captured and periodically latched to establish the trigger edge timestamp; The phase difference of the trigger edge is obtained by measuring the phase deviation between clock reference edges of the trigger edge timestamp; Based on the trigger, a phase difference change rate advance correction step size prediction is performed along the phase difference to output a phase-locked correction command; Based on the phase-locked correction instruction, the master-slave cascaded clock phase closed-loop adjustment is performed to form a global synchronization time base.

6. The method according to claim 1, characterized in that, The step of verifying the exposure window pulse count deviation based on the synchronous lock state to obtain the fuzzy risk label includes: Based on the synchronous locking state, the real-time pulse count within the exposure window of each channel is measured to generate an exposure period pulse count; The pulse counts during the exposure period are used to perform pulse differential-assisted layered deviation tracking to establish a layered deviation record; The ambiguity cause category is obtained by identifying the inter-frame deviation sign flip frequency and deviation amplitude distribution of the deviation layered records; The fuzzy risk label is obtained by performing threshold matching of classification risk level and annotation of over-limit frames using the aforementioned fuzzy cause categories.

7. The method according to claim 1, characterized in that, The step of obtaining the recovery start parameters from the breakpoint latch record location trigger coordinates and frame quality compensation parameters includes: The final state of the breakpoint trigger is obtained by verifying the breakpoint frame number and the last value of the interval from the breakpoint latch record; The average value of the breakpoint coordinate deviation is formed by measuring the frame-by-frame deviation of the final state before the pause. The correction amount for resuming transmission is determined by measuring the confidence interval of the first frame of the resumed transmission based on the average deviation of the breakpoint coordinates. The recovery start parameter is obtained by encapsulating the coordinate and frame quality compensation parameters on the resume correction amount.

8. The method according to claim 4, characterized in that, The step amplitude persistence segment joint screening of the interval difference sequence to generate slippage feature data includes: The interval difference sequence is subjected to pre-stability and out-of-segment slope calibration of the over-threshold segment to obtain the segment boundary amplitude mark; The segment offset is obtained by tracking the maximum offset amplitude and the number of continuous pulses within the segment based on the segment boundary amplitude markers. The slippage recovery coefficient is determined by calculating the ratio of the step recovery slope to the duration of continuous offset for the aforementioned segment offset. Based on the slip recovery coefficient, instantaneous impact type and continuous drift type feature classification is performed to generate slip feature data.

9. The method according to claim 6, characterized in that, The process of identifying the fuzziness cause category by analyzing the inter-frame deviation sign flip frequency and deviation amplitude distribution of the deviation-layered records includes: The deviation layer records are then used to perform frame-by-frame annotation of deviation symbols and identification of flip event cycles to obtain a flip frequency sequence. The deviation amplitude distribution is partitioned by identifying the boundaries between the autocorrelation oscillation section and the decay section of the flipped frequency sequence. Based on the deviation amplitude distribution partition, the causal weights for vibration jitter type and velocity drift type are allocated sequentially to form fuzzy causal weights. Based on the fuzzy cause weights, a comprehensive rating and optimization determination of cause categories are performed to generate fuzzy cause categories.

10. A synchronous triggering system for a linear scan camera based on an encoder, characterized in that, include: The signal acquisition module is used to acquire orthogonal phase signals in real time from the orthogonal differential input of the encoder using hardware logic, and to perform orthogonal frequency multiplication decoding and edge-by-edge hardware latching based on the orthogonal phase signals to generate high-resolution pulse sequences. The trigger scheduling module is used to determine the trigger spacing parameter by tracking the spacing fluctuation of the high-resolution pulse sequence, to obtain the slippage event marker by using the trigger spacing parameter to perform spacing step pattern recognition, and to generate the row trigger pulse by pulse timing scheduling after performing spacing compensation on the slippage event marker. The clock synchronization module is used to screen the jitter deviation of the speed change segment of the row trigger pulse to form a delay correction reference, perform master-slave cascade triggering edge phase-locked correction based on the delay correction reference to obtain a global synchronization time base, and perform first and last stage drift measurement based on the global synchronization time base to generate a synchronization lock state. The quality control module is used to verify the pulse count deviation of the exposure window based on the synchronous locking state to obtain a fuzzy risk mark, use the fuzzy risk mark to collect the out-of-limit frame sequence number and the spacing fluctuation amplitude to form a frame quality record, and archive the frame quality record for fluctuation before pause to establish a breakpoint latch record. The breakpoint resume module is used to obtain the recovery start parameter from the breakpoint latch record location trigger coordinates and frame quality compensation parameters, perform acceleration segment metering correction on the recovery start parameter to generate a resume trigger sequence, and output multiple line trigger commands based on the resume trigger sequence to track the continuity deviation of the breakpoint sequence number.

Images