Automatic winder fault detection method supporting image recognition

Abstract

The invention discloses an automatic winder fault detection method supporting image recognition, and relates to the technical field of intelligent detection and control, and the method comprises the steps: S1, obtaining the data of a yarn tension sensor during the operation of winding equipment, collecting a tension fluctuation signal through a real-time monitoring module, extracting a peak value feature from a signal strength feature and fluctuation frequency distribution, and carrying out the detection of the peak value feature; adopting a support vector machine algorithm to carry out pattern recognition on the anomaly classification label to obtain an anomaly type identifier containing a timestamp sequence and a sensor number; s2, according to the timestamp sequence and the equipment operation state in the abnormality type identifier, calculating the longitudinal height reference of abnormality occurrence in combination with the operation speed correlation parameter, querying the spindle position height mapping table through the spindle position number index, obtaining a corresponding offset compensation coefficient, and obtaining a corrected height reference calibration coordinate; according to the automatic winder fault detection method supporting image recognition, closed-loop control of fault detection, positioning, marking and verification is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent detection and control, and particularly relates to an automatic winding machine fault detection method supporting image recognition. BACKGROUND

[0002] As a core link in the textile production chain, the winding process directly affects the quality stability and production continuity of the subsequent weaving process. In this critical process, yarn tension abnormalities, broken ends, and poor joints frequently occur, and any minor quality defects can be magnified indefinitely in subsequent processing, causing quality problems in the entire batch of products.

[0003] The current winding equipment fault detection system generally faces the problem of balancing detection accuracy and response speed. Although traditional detection methods can identify fault occurrence, they lack spatial accuracy and time response in fault positioning, often providing only a rough area range and failing to meet the strict requirements of modern high-speed winding equipment for precise positioning. More critically, existing systems have a significant information conversion delay in the fault information transmission process, and the time difference between detection and manual intervention leads to the continuous expansion of the fault impact range. Precise positioning of the fault location in the winding process faces the technical challenge of synchronization of detection signals and physical markers. Due to the high speed of winding equipment and the complex working environment, the fault information obtained by the detection system needs to be converted into accurate spatial coordinates in real time, and the accuracy of this coordinate conversion directly determines the effectiveness of the subsequent physical markers. When the detection signal cannot accurately correspond to the actual physical position of the equipment, the marker position may be offset. For example, when the system detects a yarn tension anomaly, if the specific spool position and height position of the anomaly cannot be accurately calculated, the laser indicator or indicator light may point to the wrong area, and the operator needs to spend a lot of time searching in adjacent areas, not only delaying the fault handling opportunity, but also missing the real fault point due to inaccurate positioning. SUMMARY

[0004] The present application aims to provide an automatic winding machine fault detection method supporting image recognition to solve the problems existing in the prior art.

[0005] To achieve the above object, the application provides the following technical scheme: An automatic winding machine fault detection method supporting image recognition. It comprises S1, obtaining yarn tension sensor data in the running of the winding equipment, collecting tension fluctuation signals through a real-time monitoring module, extracting peak value features from signal intensity features and fluctuation frequency distribution, using a support vector machine algorithm to perform pattern recognition on abnormal classification labels, and obtaining abnormal type identification containing a timestamp sequence and a sensor number; S2, calculating the longitudinal height reference of abnormal occurrence according to the timestamp sequence in the abnormal type identification and the equipment running state, combining the running speed correlation parameter, querying the spindle position height mapping table through the spindle position number index, obtaining the corresponding offset compensation coefficient, and obtaining the corrected height reference calibration coordinates; S3, if the corrected height reference calibration coordinates exceed the height layering range, starting the height error correction program through a dynamic adjustment factor, adjusting the longitudinal distance measurement parameter, verifying the calibration accuracy through a calibration error threshold comparison, and obtaining the accurate height position that passes the verification; S4, calling a laser power adjustment module according to the accurate height position and the spindle position number index, setting the beam angle control parameter in combination with the running speed correlation parameter, calculating the laser marking instruction containing the spot diameter adjustment information through the pulse frequency setting and the marking duration, and generating the laser marking instruction; S5, using a laser wavelength selection algorithm to match the marking brightness requirement of the abnormal classification label, combining the indicator response delay parameter to optimize the execution timing of the marking instruction, transmitting the optimized marking instruction to the laser indicator equipment through a real-time synchronization protocol, and obtaining the physical marking signal in the activated state; S6, monitoring the change of the marking coverage range of the fault spindle position according to the feedback data of the physical marking signal, comparing the marking coverage range with the abnormal trigger threshold through a range reduction detection algorithm, judging whether the marking effect meets the expected precision requirement, and obtaining the optimized spindle abnormality detection cycle process.

[0006] Preferably, S1 comprises obtaining yarn tension sensor data and sensor number identification from the winding equipment, obtaining tension fluctuation signals containing a timestamp sequence; extracting signal intensity features and fluctuation frequency distribution from the tension fluctuation signals, wherein the signal intensity features are obtained by calculating signal amplitude peaks, and the fluctuation frequency distribution is obtained by Fourier transform to obtain a frequency spectrum, and peak features are obtained; using a support vector machine algorithm to perform pattern recognition on the peak features, wherein the support vector machine algorithm input is the peak features, and the output is an abnormal classification label, and the abnormal classification label is obtained; obtaining an abnormal type identification according to the abnormal classification label in combination with the sensor number identification.

[0007] Preferably, the S2 comprises obtaining a timestamp sequence and a device running state from the abnormal type identification, calculating a longitudinal height reference of the abnormal occurrence through a product relationship of the timestamp sequence and the running state data combined with a running speed associated parameter, obtaining the longitudinal height reference; matching the longitudinal height reference with the spindle position number index to obtain a corresponding offset compensation coefficient from the spindle position height mapping table, wherein the spindle position height mapping table is established based on a preset corresponding relationship between sensor position coordinates and yarn tension distribution, obtaining the offset compensation coefficient; correcting the longitudinal height reference by multiplying the offset compensation coefficient to determine a height reference calibration coordinate, obtaining the height reference calibration coordinate; and according to the height reference calibration coordinate combined with the abnormal type identification, if the height reference calibration coordinate exceeds a preset threshold, judging that the yarn tension distribution is offset, obtaining the yarn tension distribution offset.

[0008] Preferably, the S3 comprises obtaining a part exceeding the height stratification range from the corrected height reference calibration coordinate, calculating an offset value based on a product relationship of an initial value of the longitudinal distance parameter and a difference value of the stratification range using a dynamic adjustment factor, obtaining the offset value; starting a height error correction program through the offset value to adjust the matching relationship of the longitudinal distance parameter and the calibration error threshold, determining the adjusted parameter; performing threshold comparison verification on the adjusted parameter, judging the result of the calibration verification, obtaining the verification result; and according to the verification result combined with the process of accurate position acquisition, if the preset threshold is exceeded, distance measurement adjustment is adopted, obtaining the accurate height position of the verification passing.

[0009] Preferably, the S4 comprises obtaining a spindle position number index from the accurate height position, calculating a light beam angle control parameter using a running speed associated parameter combined with a height position index, obtaining a preliminary angle value; calling a laser power adjustment module for the preliminary angle value, calculating through a pulse frequency setting combined with a mark duration and a laser power regulation to determine an adjusted power level.

[0010] Preferably, the S4 further comprises generating an intermediate instruction containing spot diameter adjustment information according to the adjusted power level, obtaining a matching relationship of the spot diameter and the angle value through spot diameter optimization; if the matching relationship exceeds a preset threshold, verifying the optimized running speed linkage parameter through an error calibration mechanism combined with a parameter threshold, judging the stability of the optimized parameter, obtaining a stable parameter set; and generating a final laser marking instruction by fusing the position accuracy improvement attribute and the instruction generation fusion using the stable parameter set.

[0011] Preferably, the S5 comprises obtaining a marking brightness requirement from the abnormal classification label, matching the marking brightness requirement with a wavelength parameter using a preset wavelength matching rule, obtaining a preliminary wavelength value; optimizing the execution timing combined with a response delay parameter for the preliminary wavelength value, determining an optimized timing set through a preset synchronization protocol adjusting the timing sequence.

[0012] Preferably, the S5 further comprises transmitting the transmission instruction sequence to the laser pointer device by adopting the optimized timing set fusion mark instruction transmission attribute; when the laser pointer device receives the transmission instruction sequence, the device state is activated, the activation stability is checked by the preset verification rule, and a verification signal group is obtained; and the physical mark generation attribute is integrated according to the verification signal group, the abnormality detection linkage is fused, and the physical mark signal of the activated state is generated.

[0013] Preferably, the S6 comprises obtaining the coverage range change from the mark signal feedback, calculating the range boundary shrinkage ratio for the coverage range change by adopting the shrinkage detection algorithm, and obtaining a range shrinkage value; the abnormality triggering threshold is fused for the range shrinkage value, the range shrinkage value and the preset threshold limit are compared through the threshold comparison mechanism, the signal data is integrated, and the effect precision judgment is determined.

[0014] Preferably, the S6 further comprises adjusting the detection cycle process according to the effect precision judgment, obtaining an optimization completion state, and judging whether the optimization completion state meets the expected precision requirement; if the optimization completion state meets the expected precision requirement, a cycle process is generated, and the optimized spool position abnormality detection cycle process is obtained.

[0015] From the above technical solutions, the present application has the following beneficial effects: The automatic winder fault detection method supporting image recognition can realize real-time identification of abnormal types in the winding process by collecting yarn tension sensor signals and combining frequency characteristics and timestamp information, and realize accurate positioning of the fault position based on spool position numbers and chain position coding indexes, thereby overcoming the problems of low fault detection precision and large response delay in the prior art. Through dynamic error correction and laser mark control logic linkage, the present application further improves the accuracy and execution efficiency of the space mark, and combines image recognition feedback to verify the mark result, effectively avoids the mark deviation and missed detection phenomenon, realizes the closed-loop control of fault detection, positioning, marking and verification, and significantly improves the intelligent level and stability of the winding process. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The automatic winder fault detection method supporting image recognition of the present application is shown in the flowchart. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0018] As Figure 1As shown, the present application provides a technical solution: an automatic cone machine fault detection method supporting image recognition, comprising S1, acquiring yarn tension sensor data in the operation of the coning equipment, collecting tension fluctuation signals through a real-time monitoring module, extracting peak features from signal strength characteristics and fluctuation frequency distribution, using a support vector machine algorithm to perform pattern recognition on abnormal classification labels, and obtaining an abnormal type identifier containing a timestamp sequence and a sensor number; S2, according to the timestamp sequence in the abnormal type identifier and the equipment operation state, combining the running speed correlation parameter to calculate the longitudinal height reference of the abnormal occurrence, querying the spindle position height mapping table through the spindle position number index to obtain the corresponding offset compensation coefficient, and obtaining the corrected height reference calibration coordinates; S3, if the corrected height reference calibration coordinates exceed the height layering range, starting the height error correction program through a dynamic adjustment factor, adjusting the longitudinal distance measurement parameter, verifying the calibration accuracy through a calibration error threshold comparison, and obtaining the accurate height position that passes the verification; S4, according to the accurate height position and the spindle position number index, calling a laser power adjustment module, setting the beam angle control parameter in combination with the running speed correlation parameter, calculating through the pulse frequency setting and the marking duration, generating a laser marking instruction containing spot diameter adjustment information; S5, using a laser wavelength selection algorithm to match the marking brightness requirement of the abnormal classification label, combining the indicator response delay parameter to optimize the execution timing of the marking instruction, transmitting the optimized marking instruction to the laser indicator equipment through a real-time synchronization protocol, and obtaining the physical marking signal in the activated state; S6, according to the feedback data of the physical marking signal, monitoring the change of the marking coverage range of the fault spindle position, comparing the marking coverage range with the abnormal trigger threshold through a range reduction detection algorithm, judging whether the marking effect meets the expected precision requirement, and obtaining the optimized spindle abnormality detection cycle process.

[0019] The core principle of the method is to combine the dynamic characteristics of yarn tension signal with intelligent recognition algorithm, and realize automatic detection and spatial positioning of faults through signal pattern recognition. First, the tension sensor collects the tension fluctuation signal in real time during the running process of the yarn, and establishes a tension fluctuation model using the signal strength peak value and fluctuation frequency distribution characteristics. The system uses support vector machine (SVM) classification algorithm to train and identify the collected signal, extracts abnormal patterns and generates abnormal labels with time stamp. Subsequently, the system calculates the abnormal occurrence position by associating the time stamp with the device running parameters, combines the preset spindle position height mapping table and offset compensation coefficient, and obtains the accurate coordinates of the abnormal point in the longitudinal direction of the device. If the coordinates exceed the layered range, the system automatically calls the dynamic correction algorithm, compares the error threshold value and calibrates the longitudinal distance measuring parameters in real time to realize self-verification of the coordinate accuracy. Further, the system sets the laser beam angle, pulse frequency and duration according to the calibrated height and spindle number to form a suitable marking track. The laser wavelength selection algorithm dynamically matches the appropriate wavelength according to the marking brightness requirements of different fault types, and combines the synchronous timing protocol to ensure accurate execution of laser instructions. Finally, the system detects the marking coverage range through the optical feedback mechanism, and judges the marking accuracy using the range change analysis algorithm to realize closed-loop detection and visual confirmation of automatic winding machine faults.

[0020] The method realizes the intelligentization and automation of winding machine fault detection by fusing signal analysis and image recognition technology. First, the dynamic characteristics of tension signal and the classification ability of SVM algorithm can effectively identify various abnormal patterns, improving the accuracy of detection. Second, the abnormal height position is calculated by time stamp and running parameters, and combined with offset compensation and dynamic calibration, the accuracy of spatial positioning is significantly improved. The visual prompt in the form of laser marking enables maintenance personnel to quickly locate the abnormal spindle, reducing manual inspection time and improving maintenance efficiency. At the same time, the adaptive wavelength control and timing optimization mechanism of the system ensures that the brightness and response speed of the marking signal are optimal. Overall, the invention can realize automatic detection, positioning and visual marking of fault spindles, reduce downtime, and improve the continuity and reliability of the production line.

[0021] S1 includes obtaining yarn tension sensor data and sensor number identification from the winding device to obtain tension fluctuation signals containing a time stamp sequence; signal strength characteristics and fluctuation frequency distribution are extracted from the tension fluctuation signals, wherein the signal strength characteristics are obtained by calculating the signal amplitude peak value, and the fluctuation frequency distribution is obtained by Fourier transform to obtain the frequency spectrum; the peak value characteristics are identified by support vector machine algorithm, wherein the support vector machine algorithm input is the peak value characteristics, and the output is the abnormal classification label, and the abnormal classification label is obtained; the abnormal type identification is obtained according to the abnormal classification label combined with the sensor number identification.

[0022] In the embodiment, after the winding device starts, the tension readings are obtained from the yarn tension sensor at a fixed sampling frequency, and the sensor number identification and timestamp are read synchronously. The sampling frequency is determined by the principle that it is twice the maximum mechanical vibration frequency corresponding to the maximum working speed of the device before starting. First, several candidate frequencies are collected in the trial run, and the false alarm rate and the missed alarm rate are calculated. When both indicators are not higher than the preset target, the minimum frequency that meets the conditions is taken as the final sampling frequency. The timestamp accuracy is fixed at the millisecond level to ensure the complete recording of the time interval between adjacent samples. The system writes the tension readings obtained by the same sensor within a continuous time into the cache in timestamp order to form a tension fluctuation signal containing a timestamp sequence and triggers subsequent processing in a windowed manner. The length of the time window depends on the production rhythm of the device and the non-stationary degree of tension fluctuation. First, at least three candidate lengths are verified offline, and the classification accuracy and calculation delay are compared. Under the condition that the delay does not exceed the preset upper limit and the classification accuracy is the highest, an integer second is selected as the final time window length. The time window sliding step is half of the time window length to balance the response speed and overlapping coverage. When entering the time domain feature extraction, the system first completes the calibration of the peak value determination threshold. This threshold is obtained by collecting tension data for at least 30 seconds in two states: device stationary and empty load low speed. The representative level and dispersion range are calculated from the stable section of the data to give the noise reference. Then, during the trial run phase, the threshold increment is gradually adjusted, and the sum of the false alarm rate and the missed alarm rate is minimized to lock a fixed value as the peak value determination threshold. Subsequently, the system identifies valid peaks in each time window by the point-by-point comparison method. That is, when the current sample amplitude is strictly greater than the threshold value and strictly greater than the amplitudes of its adjacent samples, the point is recorded as a valid peak and the amplitude and occurrence time are recorded. In a time window, the maximum peak amplitude, average peak amplitude, total number of peaks, and average value of adjacent peak intervals are counted and stored in a determined order as the time domain signal strength features. In the frequency domain feature extraction, the system performs Fourier transform on the tension fluctuation signal within the same time window to obtain the discrete frequency spectrum. The frequency search range is from zero to half of the sampling frequency. The frequency resolution is uniquely determined by the time window length and does not need to be specified manually. The main peak is defined as the spectrum line with the maximum amplitude in the search range. The main peak frequency position and main peak amplitude are recorded. After excluding a fixed width around the main peak that is proportional to the frequency resolution, the secondary peak is searched and the secondary peak frequency position and secondary peak amplitude are recorded. When the frequency interval between the main peak and the secondary peak is less than twice the frequency resolution, only the main peak is retained to avoid repeated counting caused by adjacent leakage, and the secondary peak-related quantities are set to zero. The amplitude ratio of the main peak and the secondary peak is calculated and recorded as a determined value. The above frequency domain main peak frequency, main peak amplitude, secondary peak frequency, secondary peak amplitude, and amplitude ratio are concatenated in a fixed order with the aforementioned time domain quantities to form the peak value feature set. Thus, the entire calculation process of "extracting signal strength features and fluctuation frequency distribution from the tension fluctuation signal and obtaining peak value features" is completed.When entering the pattern recognition phase, the system inputs the peak feature set into the support vector machine classification model to complete anomaly recognition. The kernel type parameter, regularization strength parameter, and kernel width parameter of the support vector machine are determined through cross-validation grid search with no less than 5 folds before going online. The search criteria are that the classification accuracy of the validation set is maximum, and the false alarm rate and the missed alarm rate are not higher than the preset target. Once the above three parameters are determined, they are written into the read-only configuration and remain unchanged in the field. During online inference, the system strictly operates according to the parameter combination. The classification output is a clear anomaly classification label, which has a unique number and a unique name in the pre-established class list. The class list is confirmed by maintenance personnel with field experience for normal working conditions and common abnormal working conditions, and the class number and name are frozen. They are not modified during field operation. The system combines the obtained anomaly classification label with the sensor number identifier recorded in the time window to form a complete anomaly type identifier, including sensor number, timestamp start value, timestamp end value, anomaly class number, and anomaly class name, thereby achieving the realization of "obtaining an anomaly type identifier according to an anomaly classification label combined with a sensor number identifier".

[0023] The sampling frequency is determined by the two-fold principle and offline comparison and is fixed as a single value. The timestamp precision is fixed at the millisecond level and is used as an immutable constant in the program. The time window length is a single integer second value determined by the accuracy and time delay joint optimization criteria. The sliding step is fixed at half of the time window length. The peak determination threshold is a single fixed increment above the noise reference and is determined by the criterion of minimizing the sum of false alarm rate and missed alarm rate. The upper limit of frequency search is fixed at half of the sampling frequency. The selection rule of the main peak and the secondary peak is that the amplitude is the largest and the second largest, and the distance between them is not less than twice the frequency resolution. If not, only the main peak is retained. The amplitude ratio is the main peak amplitude divided by the secondary peak amplitude, which is stored as a small number in the implementation. When there is only a main peak, the amplitude ratio is set to a predetermined maximum value, which is a constant. The kernel type parameter, regularization strength parameter, and kernel width parameter of the support vector machine are uniquely determined through cross-validation and recorded as specific values, which do not change automatically with field data. The anomaly classification label set is frozen before deployment and contains a fixed number of class items. The sensor number identifier is checked one by one based on the physical location during installation and debugging, marked with adhesive, and registered as a determined integer number in the system. S2 includes obtaining a timestamp sequence and a device running state from the anomaly type identification, calculating a longitudinal height reference of the occurrence of the anomaly through a product relationship of the timestamp sequence and the running state data combined with a running speed correlation parameter, obtaining the longitudinal height reference; matching the longitudinal height reference with the spindle position number index to obtain the corresponding offset compensation coefficient from the spindle height mapping table, wherein the spindle height mapping table is established based on the corresponding relationship between the preset sensor position coordinates and the yarn tension distribution, obtaining the offset compensation coefficient; correcting the longitudinal height reference by multiplying the offset compensation coefficient to determine the height reference calibration coordinates, obtaining the height reference calibration coordinates; according to the height reference calibration coordinates combined with the anomaly type identification, if the height reference calibration coordinates exceed the preset threshold, it is judged that the yarn tension distribution is offset, obtaining the yarn tension distribution offset. In the embodiment, firstly, the timestamp sequence and the device running state are read from the abnormal type identification, the timestamp is recorded with millisecond precision, the device running state contains the current speed gear and whether it is in one of the three states of acceleration, constant speed or deceleration, the running speed correlation parameter is a set of fixed numbers indexed by the speed gear, each number is determined by calibration test before online, the calibration method is to let the device run continuously and stably under constant speed condition for not less than 3 speed gears, collect not less than 300 seconds of tension and displacement reference data at each gear, pair the longitudinal displacement given by the displacement reference device with the time increment, calculate the average longitudinal displacement increment per unit time and fix the average as the running speed correlation parameter of the gear, if there is an acceleration or deceleration state, record the average displacement increment per unit time of the acceleration segment and the deceleration segment respectively during the calibration stage and fix them as two independent parameters; then enter the longitudinal height reference calculation, the system pairs each pair of time difference and corresponding running speed correlation parameter to obtain the longitudinal displacement increment in the time interval by multiplication operation, and then adds each increment in the window in time sequence to obtain the longitudinal height reference, the addition starts from the zero reference corresponding to the abnormal time segment starting timestamp, the zero reference is determined as a fixed number by a no-load alignment process after each device startup and is written into the read-only configuration; after obtaining the longitudinal height reference, the system performs spindle position matching, the spindle position number index is a list arranged in ascending order, each number in the list corresponds to a longitudinal reference height, the reference height is measured from the ground reference to each spindle position during the installation and debugging stage of the device and recorded as a fixed number, the system compares the longitudinal height reference with the list item by item, selects the number with the smallest absolute value difference from the longitudinal height reference as the matching result, if there are two numbers with the same difference, select the smaller number to ensure uniqueness; in order to eliminate structural and installation errors, the system obtains the offset compensation coefficient according to the spindle height mapping table, the spindle height mapping table is established offline before online, the establishment method is to collect not less than 300 seconds of data for each spindle under the condition of constant speed and stable operation of the device, record the sensor position coordinates and the stable interval of tension distribution at the same period under the premise of maintaining qualified yarn quality, correspond the sensor position coordinates and the longitudinal position of the tension distribution peak one by one according to the spindle, calculate the systematic offset of the spindle, then convert the offset into a proportional coefficient that makes the height reference converge to the real position and fix it as the offset compensation coefficient of the spindle, the mapping table takes the spindle number as the index item, each item only contains one determined offset compensation coefficient number, and is not automatically updated on site; after obtaining the offset compensation coefficient, the system multiplies the coefficient with the longitudinal height reference to obtain the height reference calibration coordinate, which is the compensated longitudinal positioning result;The system then uses the height reference calibration coordinates together with the anomaly type identification for offset judgment, which uses a single threshold comparison method, with a preset threshold being a fixed number obtained by playing back historical normal samples and historical abnormal samples in the offline stage. The determination method is to select continuous production data of not less than 7 days to construct a sample set, respectively count the maximum deviation of the height reference calibration coordinates in the normal samples and the minimum deviation in the abnormal samples according to the anomaly type, take the intermediate value between the two as the threshold value of the anomaly type, and if only a single threshold is configured, take the minimum of the above intermediate values of all anomaly types as the unified threshold and fix it. During online operation, the system performs a comparison for each anomaly type identification, and if the height reference calibration coordinates exceed the corresponding threshold, the result is determined as yarn tension distribution offset, and the offset conclusion is immediately output, and the determined numbers such as spindle position number, time stamp start value and end value, longitudinal height reference, offset compensation coefficient, height reference calibration coordinates and threshold value used are recorded for subsequent step calling. If the threshold is not exceeded, the non-offset conclusion is output and the same recording is completed.

[0024] S3 includes obtaining the part exceeding the height stratification range from the corrected height reference calibration coordinates, calculating the offset value based on the product relationship of the initial value of the longitudinal distance parameter and the difference value of the stratification range using a dynamic adjustment factor, obtaining the offset value; starting the height error correction program through the offset value, adjusting the matching relationship of the longitudinal distance parameter and the calibration error threshold value, determining the adjusted parameter; performing threshold comparison verification on the adjusted parameter, determining the result of the standard verification, obtaining the verification result; according to the verification result combined with the process of obtaining the accurate position, if the preset threshold is exceeded, distance measurement adjustment is used, and the accurate height position that passes the verification is obtained.

[0025] In the embodiment, firstly, the upper and lower boundaries of the height stratification range and the current value of the corrected height reference calibration coordinate are read, the part exceeding the height stratification range is calculated, if the calibration coordinate is between the upper and lower boundaries, the exceeding amount is zero, if it is higher than the upper boundary, the exceeding amount is the difference between the calibration coordinate and the upper boundary, if it is lower than the lower boundary, the exceeding amount is the difference between the lower boundary and the calibration coordinate, and the exceeding amount is the stratification range difference value; then the offset value calculation is entered, the initial value of the longitudinal distance parameter is a single fixed number obtained by calibration under the standard tool before the upper line, and the dynamic adjustment factor is a determined number set corresponding to the exceeding degree, and the determination method of the two is that, under the condition that the equipment is in stable normal working condition, not less than three representative speeds are collected for not less than three hundred seconds of positioning verification data, a set of discrete factors from small exceeding to large exceeding is given to minimize the correction iteration times and not to produce over-adjustment, and the initial value of the longitudinal distance parameter is multiplied by the stratification range difference value and then multiplied by the dynamic adjustment factor to obtain the offset value; when the offset value is zero, the threshold comparison verification is directly entered, and when the offset value is greater than zero, the height error correction program is started with the offset value, the matching relationship of the longitudinal distance parameter and the calibration error threshold is adjusted in a single step, specifically, without changing the physical meaning of the longitudinal distance parameter, the longitudinal distance parameter is updated by a certain step size in the direction of the offset value, and the calibration error threshold is updated by a certain amplitude in the same direction to keep the matching of the ranging sensitivity and the allowed error unchanged, the certain step size and the certain amplitude are obtained by playing back the data during offline calibration and recorded as a single fixed number; after parameter updating, the threshold comparison verification is entered, the system calculates the difference between the current calibration coordinate and the target stratification reference position, the target stratification reference position is fixed as the stratification center or the stratification boundary during deployment, and is globally applicable in a single way, the comparison rule is that if the difference is less than or equal to the current calibration error threshold, the calibration verification is passed, and if the difference is greater than the calibration error threshold, the verification is failed and returns to the correction program to continue the next update in the same direction and the same rule; in order to avoid long fine adjustment, the system sets a preset threshold as the trigger threshold of distance measurement adjustment, the determination method of the preset threshold is to extract the verified normal samples and abnormal samples from the historical production data of not less than seven consecutive days, to obtain the calibration coordinate error distribution by using the same calculation link as online, to take the minimum fixed number that can control the convergence step within an acceptable range as the preset threshold under the premise that the false alarm rate and the false negative rate are not higher than the online target, and to write it into the read-only configuration;When the verification fails and the difference exceeds the preset threshold value, the system performs distance measurement adjustment, which is a larger step update in one time to quickly narrow the distance between the calibration coordinates and the target hierarchical reference position. After the update, the threshold value comparison verification is performed again. If it passes, the accurate height position of the verification is output, and the hierarchical range difference, the initial value of the longitudinal distance parameter, the dynamic adjustment factor, the offset value, the calibration error threshold value, the preset threshold value, and the final accurate height position are recorded for subsequent steps. If it fails, the distance measurement adjustment and single-step correction are alternately performed until it passes.

[0026] S4 includes obtaining the ingot position index from the accurate height position, calculating the beam angle control parameter using the running speed associated parameter combined with the height position index to obtain the preliminary angle value; calling the laser power adjustment module for the preliminary angle value, calculating through the pulse frequency setting combined with the mark duration and laser power regulation to determine the adjusted power level; generating intermediate instructions containing spot diameter adjustment information according to the adjusted power level, obtaining the matching relationship of spot diameter and angle value through spot diameter optimization; if the matching relationship exceeds the preset threshold value, the error calibration mechanism combined with the parameter threshold value verification optimization running speed linkage parameter is used to judge the stability of the optimized parameter to obtain the stable parameter set; using the stable parameter set to fuse the position precision improvement attribute and the instruction generation fusion to generate the final laser marking instruction.

[0027] In the embodiment, firstly, the ingot position number index and the height position index are read from the accurate height position, the height position index is fixed as a continuous integer after being measured by a standard gauge in the installation and debugging stage, and each value uniquely corresponds to a longitudinal geometric position; then, the light beam angle control parameter is calculated by the running speed correlation parameter and the height position index, the calculation process is to establish an angle reference table and a speed linkage table in the offline calibration stage, the angle reference table is a one-to-one mapping of the height position index to the angle reference value, and the speed linkage table is a one-to-one mapping of the device speed gear to the angle correction amount, both tables are fixed as fixed numbers, and the generation method is to measure the angle value of the laser incidence reaching the minimum deviation of the vertical projection under the condition of not less than three speed grades and record it as the angle reference value, and at the same time, record the stable offset of the incidence angle caused by the speed change under the same working condition and take the stable middle as the angle correction amount; in the online stage, the system takes the angle reference value from the angle reference table according to the ingot position number index and takes the angle correction amount from the speed linkage table according to the current running speed correlation parameter, adds the two to obtain the preliminary angle value and records it to the intermediate buffer in degrees; after obtaining the preliminary angle value, the system calls the laser power adjustment module to perform power calculation, the power calculation takes the pulse frequency setting, the mark duration and the power step table as the input, the pulse frequency setting is selected from the candidate frequency set in the offline stage for each speed gear and angle interval to form the minimum frequency that can form a continuous readable mark and is fixed as a single number, the mark duration is the shortest time length that can form a complete mark at the same surface material and the same running speed and is fixed as a single number, and the power step table includes two fixed numbers of single step increment and single maximum increment, in the online stage, the system calculates the energy target required for this mark according to the pulse frequency setting and the mark duration, and gradually increases or decreases according to the power step table until the energy target is reached or first exceeded, and when it is exceeded, it is backed up by one single step increment to avoid overshoot, and finally the adjusted power level is determined; the system generates the intermediate instruction including the spot diameter adjustment information according to this, and the intermediate instruction field includes six determined numbers of ingot position number index, preliminary angle value, pulse frequency setting, mark duration, adjusted power level and spot diameter target, wherein the determination of the spot diameter target depends on two read-only tables of the spot diameter target table and the angle incidence distance table, the spot diameter target table is measured and fixed in the offline stage for different incidence distances to form the minimum spot diameter, and the angle incidence distance table is measured and fixed in the offline stage according to the device geometry and the optical path installation distance; in the online stage, the system finds the incidence distance in the angle incidence distance table according to the preliminary angle value, and finds the target diameter in the spot diameter target table and writes it into the intermediate instruction; then, the spot diameter optimization is performed to obtain the matching relationship of the spot diameter and the angle value, the optimization process is to calculate the absolute difference value between the spot diameter target in the intermediate instruction and the actual reachable spot diameter, and the difference value is the matching error amount;The system compares the matching error amount with a preset threshold value, which is a single fixed number determined in the offline stage as a standard for marking edge continuity and no underlap, and if the matching error amount does not exceed the preset threshold value, it is considered that the matching relationship meets the requirements and enters the stability evaluation; if the matching error amount exceeds the preset threshold value, the error calibration mechanism is triggered and the parameter threshold value verification is combined to optimize the running speed linkage parameter, the execution order of the error calibration mechanism is three steps of angle fine tuning, power fine tuning and spot diameter fine tuning, all of which are restricted by a single maximum limit, the passing condition of the parameter threshold value verification is that the matching error amount is less than or equal to the preset threshold value and the continuous verification times are not less than 3 times, the continuous verification times threshold is a fixed number, if the continuous two rounds still fail and the error direction is consistent, the angle correction amount of the current gear in the speed linkage table is updated by a fixed fine tuning amount once and recorded as a shadow parameter, which only takes effect in the current process and does not change the formal configuration; when passing the parameter threshold value verification, the system converges the final angle value, the final power level, the final pulse frequency, the final mark duration and the final spot diameter target in the current process to form a stable parameter set, and the formation condition of the stable parameter set includes the passing of the preset threshold value and the continuous verification passing; after obtaining the stable parameter set, the system performs position accuracy improvement attribute and instruction generation fusion, the position accuracy improvement attribute is a one-time fixed weight historical deviation compensation for the preliminary angle value to eliminate residual system error, and the fixed weight is a single fixed number determined in the offline stage as a target for minimizing repeated mark displacement, and the instruction generation fusion is to combine the five numbers in the stable parameter set and the angle value after position accuracy improvement in a fixed field order to form a final laser marking instruction, the final instruction field includes six items of spindle position number index, angle value, power level, pulse frequency, mark duration and spot diameter target, and the values are all from the above determination process.

[0028] S5 includes obtaining a marking brightness requirement from the anomaly classification label, matching the marking brightness requirement with the wavelength parameter using a preset wavelength matching rule to obtain a preliminary wavelength value; for the preliminary wavelength value, combining a response delay parameter to optimize the execution timing, adjusting the timing sequence through a preset synchronization protocol to determine an optimized timing set; using the optimized timing set to fuse the marking instruction transmission attribute to generate a transmission instruction sequence and transmit it to the laser pointer device; when the laser pointer device receives the transmission instruction sequence, the device state is activated, the activation stability is checked through a preset verification rule to obtain a verification signal group; according to the verification signal group, the physical mark generation attribute is integrated, the anomaly detection linkage is fused, and the physical mark signal of the activated state is generated.

[0029] In the embodiment, first, the marked brightness requirement is read from the abnormal classification label, the marked brightness requirement is offline graded into several fixed levels before deployment and represented by an integer, each level corresponds to a minimum readable illuminance number and a target contrast number, both of which are determined by repeating the measurement of the sample under the same surface material and the same incident condition for no less than 30 times and taking the stable median, after reading, the table lookup calculation of the preset wavelength matching rule is entered, the preset wavelength matching rule is solidified in the form of read-only table, each brightness level in the table uniquely corresponds to a wavelength parameter, the wavelength parameter is obtained by point-by-point testing of multiple candidate wavelengths under the premise that the power is not higher than the safety upper limit of the device and the illuminance is not lower than the minimum readable illuminance of the level, and the target contrast is maximum, so that the preliminary wavelength value is directly indexed according to the brightness level; the system then jointly optimizes the execution time sequence with the preliminary wavelength value and the response delay parameter, the response delay parameter is defined as the fixed time from receiving the starting instruction to outputting the stable light beam of the laser pointer device, the single number is determined by performing no less than 30 trigger measurements according to each wavelength and three power level combinations and taking the stable median, the calculation process of the optimized execution time sequence is to align and correct the fixed time with the time slot structure of the preset synchronization protocol, the preset synchronization protocol consists of four digital elements: fixed time slot length, fixed starting mark position, fixed confirmation feedback time and fixed verification sequence, the four elements are determined offline to minimize packet loss rate and jitter, and are solidified in the offline playback process, in the online stage, the system moves the instruction initiation time forward or backward to the nearest starting mark allowed position according to the response delay parameter, and accordingly extends the time length and the confirmation feedback time, forming an optimized time sequence set; after obtaining the optimized time sequence set, the system generates a transmission instruction sequence by fusing the optimized time sequence set with the marked instruction transmission attribute, the marked instruction transmission attribute includes four fixed numbers: frame header length, load field sequence, verification field length and initiation interval, the four numbers are determined and solidified by offline stress testing under the condition that the device bandwidth upper limit is not exceeded and the verification coverage rate and processing margin are guaranteed, the specific steps of generating the transmission instruction sequence are as follows: write the preliminary wavelength value, the angle value in the laser mark instruction generated in the previous step, the power level, the pulse frequency and the mark duration in sequence according to the load field sequence, then write the three time parameters in the optimized time sequence set, finally append the verification field and align the starting mark according to the initiation interval, and then send it to the laser pointer device through the preset synchronization protocol.The laser pointer device switches to an active device state when receiving a transmission instruction sequence, and the system immediately checks the activation stability according to preset verification rules and generates a verification signal group. The preset verification rules are composed of three fixed numbers, including a rise time threshold, a power stability threshold, and a timing jitter threshold. The rise time threshold is determined as a single number by taking the maximum allowed value of the rise time distribution of consecutive 7-day qualified samples. The power stability threshold is determined as a single number by taking the maximum allowed value of the power fluctuation range of the stable period of the qualified samples. The timing jitter threshold is determined as a single number by taking the maximum allowed value of the deviation between the confirmed reply and the expected reply in the playback test under the preset synchronization protocol. The online verification process is to sample the activation process in millisecond granularity within a fixed monitoring window, calculate the rise time, the stable period power fluctuation range, and the reply deviation respectively, and compare them with the three thresholds one by one. If all of them do not exceed the thresholds, write a pass identifier in the verification signal group and record the three measured numbers. If any of them exceeds the threshold, write a fail identifier and record the exceeding item and the exceeding amplitude. After obtaining the verification signal group, the system generates a physical marker signal in the active state by integrating the physical marker generation attributes and the abnormality detection association. The physical marker generation attributes include three fixed number sets, i.e., beam opening sequence, duty cycle, and movement beat. The three sets are selected with the smallest error from multiple combinations in offline calibration, with the goals of edge continuity, no blank breakpoint, and minimum thermal influence. The integration calculation process is to compare the rise time, power stability, timing jitter, and the above three fixed sets recorded in the verification signal group for consistency. If the identifier is pass, the fixed set is directly used, and the abnormality classification label and the ingot position number index are written into the additional field of this physical marker task as abnormality detection association information. Then, the beams are turned on at the starting time of the optimized timing set according to the beam opening sequence, the pulse switch is controlled according to the duty cycle, and the beam movement is driven according to the movement beat. The continuous output is output to the end of the marker duration, generating a physical marker signal in the active state. If the identifier is fail, the verification is performed again at a delayed initiation time in the same optimized timing set with unchanged wavelength parameters. The retry is limited to a single time. If the retry result is still fail, the current marker is stopped and a fault entry is recorded.

[0030] S6 includes obtaining the coverage range change from the marker signal feedback, calculating the range reduction value by using the reduction detection algorithm for the coverage range change; fusing the abnormal trigger threshold for the range reduction value, comparing the range reduction value with the preset threshold limit through the threshold comparison mechanism, integrating the signal data, and determining the effect precision judgment; adjusting the detection loop process according to the effect precision judgment, obtaining the optimization completion state, and judging whether the optimization completion state meets the expected precision requirement; if the optimization completion state meets the expected precision requirement, generating a loop process, and obtaining the optimized ingot position abnormality detection loop process.

[0031] In the embodiment, first, the coverage range data is read from the marker signal feedback, the coverage range data is determined by 4 items as the upper boundary, the lower boundary, the left boundary and the right boundary of the coverage, and the corresponding 4 items of numbers of the last period are read at the same time, the coverage range change calculation is entered, specifically, the difference between the current period and the last period is calculated item by item, the difference of the inward shrinkage is kept as a positive value, the difference of the outward expansion is set to zero, and then the 4 positive values are added to obtain the total boundary shrinkage amount, and the coverage height and the coverage width of the last period are calculated to obtain the perimeter and the diagonal of the coverage area of the last period, and then the range boundary shrinkage ratio is obtained by dividing the total boundary shrinkage amount by the larger one of the perimeter and the diagonal, the range boundary shrinkage ratio is expressed as a percentage and directly written into the result buffer as the range reduction value; the system then fuses the abnormal trigger threshold and the preset threshold limit according to the range reduction value, and performs the threshold comparison mechanism, the determination method of the abnormal trigger threshold is to calculate the maximum stable value of the range boundary shrinkage ratio in the normal samples for not less than 7 consecutive days, and calculate the minimum abnormal value of the range boundary shrinkage ratio in not less than 3 typical abnormal samples, and take the intermediate value between the two as a single fixed number and write it into the read-only configuration; the determination method of the preset threshold limit is to gradually reduce the acceptable range boundary shrinkage ratio step by step in the offline playback environment with the goal of edge continuity and no under-coverage, without changing the marker process parameters, and select the minimum number that can meet the false alarm rate and the false negative rate not higher than the online acceptance target and solidify it as the read-only configuration; the threshold comparison mechanism in the online stage includes two deterministic comparisons, the first comparison is the comparison between the range reduction value and the abnormal trigger threshold, if the range reduction value is greater than or equal to the abnormal trigger threshold, it is recorded as triggered, otherwise it is recorded as not triggered; the second comparison is the comparison between the range reduction value and the preset threshold limit, if the range reduction value is less than or equal to the preset threshold limit, it is recorded as passed, otherwise it is recorded as not passed; the system integrates the 4 coverage boundary numbers, the range reduction value, the first comparison result and the second comparison result into signal data in the order of fixed fields and enters the effect accuracy judgment, the effect accuracy judgment rule is that when the first comparison result is not triggered and the second comparison result is passed, it is judged as meeting the accuracy, and when any result does not meet the accuracy, it is judged as not meeting the accuracy and the reason item and the corresponding number are recorded; the system adjusts the detection cycle process according to the effect accuracy judgment, the detection cycle process includes two fixed parameters of sampling interval and coverage range extraction times per period, both are solidified as read-only numbers with default values when deployed, when the effect accuracy judgment of the current period is not meeting the accuracy, the sampling interval of the next period is shortened by a certain multiple to half of the original value and the coverage range extraction times is increased by a certain multiple to twice of the original value, while keeping the two thresholds unchanged to ensure the consistency of the judgment standard; when the effect accuracy judgment is meeting the accuracy, the sampling interval and the coverage range extraction times are restored to the default values in the next period;The system calculates the optimization completion state after the above adjustment is performed, the determination rule of the optimization completion state is that the range reduction value of the continuous 3 periods is less than or equal to the preset threshold limit, and the first comparison result is not triggered, and there is no continuous single-side trend of the single period difference of the four coverage boundaries, all the three conditions are met, and then the optimization completion state is recorded, otherwise the iteration is continued according to the shortening or recovery strategy until the conditions are met; the system performs consistency check based on the optimization completion state and the expected accuracy requirement at the end of each period, the expected accuracy requirement is a single digital set fixed before going online, the set includes three fixed numbers of the maximum allowed range reduction percentage, the maximum allowed iteration period number and the minimum readable boundary margin, the compliance judgment is that the range reduction value is less than or equal to the maximum allowed range reduction percentage, the cumulative iteration period is less than or equal to the maximum allowed iteration period number, and the minimum distance between the current coverage boundary and the target region boundary is greater than or equal to the minimum readable boundary margin, and the three conditions are met at the same time, which is the expected accuracy requirement; when the optimization completion state meets the expected accuracy requirement, the system performs the loop process generation, the calculation process of the loop process generation is that 7 determined numbers of the final sampling interval of the current period, the final coverage range extraction times, the final range reduction value, the abnormal trigger threshold, the preset threshold limit, the effect accuracy judgment result and the optimization completion state are written in the loop process template in the order of fixed fields, and the spindle position number and the time stamp are used as the template index for direct reuse next time, so that the optimized spindle abnormality detection loop process is obtained.

[0032] In the present application, the input includes tension data (including time stamp and sensor number), equipment running state and running speed parameter, spindle position number and height mapping and offset compensation, longitudinal layering range and each threshold, angle reference and speed linkage, pulse frequency and mark duration and power step, wavelength matching rule and response delay, synchronization protocol and stability verification rule; the output includes abnormal type identification, accurate height position, stable parameter set, final laser marking instruction (angle value, power level, pulse frequency, mark duration, spot diameter target), activated physical mark signal, optimized spindle abnormality detection loop process.

[0033] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, the scope of the present application being defined by the appended claims and their equivalents.

Claims

1. A fault detection method for an automatic winding machine supporting image recognition, characterized in that, include: S1. Acquire yarn tension sensor data during the operation of the winding equipment, collect tension fluctuation signals, extract peak features from signal intensity characteristics and fluctuation frequency distribution, and use support vector machine algorithm to perform pattern recognition on anomaly classification labels to obtain anomaly type identifiers containing timestamp sequences and sensor numbers. S2. Based on the timestamp sequence and equipment operating status in the anomaly type identifier, and combined with the operating speed related parameters, calculate the longitudinal height reference for the occurrence of the anomaly. Query the spindle height mapping table through the spindle number index to obtain the corresponding offset compensation coefficient and obtain the corrected height reference calibration coordinates. S3. If the corrected height reference calibration coordinates exceed the height layer range, the height error correction procedure is initiated by dynamically adjusting the factor, adjusting the longitudinal distance measurement parameters, verifying the coordinate accuracy by comparing the calibration error threshold, and obtaining the verified accurate height position. S4. Based on the precise height position and spindle number index, and combined with the running speed related parameters, set the beam angle control parameters, and generate a laser marking instruction containing beam diameter adjustment information by setting the pulse frequency and calculating the marking duration. S5. The laser wavelength selection algorithm is used to match the marking brightness requirements of the abnormal classification label. The execution timing of the marking command is optimized by combining the indicator response delay parameter. The optimized marking command is transmitted to the laser indicator device through the real-time synchronization protocol to obtain the physical marking signal in the active state. S6. Based on the feedback data of the physical marker signal, monitor the change in the marker coverage of the faulty spindle position, compare the marker coverage with the abnormal trigger threshold through the range reduction detection algorithm, determine whether the marking effect meets the expected accuracy requirements, and obtain the optimized spindle position abnormality detection loop process.

2. The automatic winding machine fault detection method supporting image recognition according to claim 1, characterized in that: S1 includes: The yarn tension sensor data and sensor number identifier are obtained from the winding equipment to obtain a tension fluctuation signal containing a timestamp sequence; Signal intensity features and fluctuation frequency distribution are extracted from the tension fluctuation signal. The signal intensity features are obtained by calculating the peak value of the signal amplitude, and the fluctuation frequency distribution is obtained by Fourier transform to obtain the frequency spectrum and peak features. The peak features are used to perform pattern recognition, wherein the peak features are the input of the support vector machine algorithm and the anomaly classification label is the output, thus obtaining the anomaly classification label. An anomaly type identifier is obtained by combining the anomaly classification label with the sensor number identifier.

3. The automatic winding machine fault detection method supporting image recognition according to claim 1, characterized in that: S2 includes: The timestamp sequence and equipment operating status are obtained from the anomaly type identifier. The vertical height benchmark of the anomaly is calculated by combining the operating speed correlation parameter with the product relationship between the timestamp sequence and the operating status data. By matching the spindle position number index with the longitudinal height reference, the corresponding offset compensation coefficient is obtained from the spindle position height mapping table. The spindle position height mapping table is established based on the preset correspondence between the sensor position coordinates and the yarn tension distribution to obtain the offset compensation coefficient. The offset compensation coefficient is multiplied by the longitudinal height reference for correction, and the height reference calibration coordinates are determined to obtain the height reference calibration coordinates. Based on the height reference calibration coordinates and the anomaly type identifier, if the height reference calibration coordinates exceed a preset threshold, the yarn tension distribution deviation is determined, and the yarn tension distribution deviation is obtained.

4. The automatic winding machine fault detection method supporting image recognition according to claim 1, characterized in that: S3 includes: The portion exceeding the height stratification range is obtained from the corrected height reference calibration coordinates. The offset value is calculated using a dynamic adjustment factor based on the product of the initial value of the longitudinal distance parameter and the difference in stratification range. The height error correction procedure is initiated by using the offset value to adjust the matching relationship between the longitudinal distance parameter and the calibration error threshold, and to determine the adjusted parameters. The adjusted parameters are compared and verified using thresholds to determine the accuracy of the coordinate verification and obtain the verification results. Based on the verification results and the precise location acquisition process, if the location exceeds a preset threshold, distance measurement is used for adjustment to obtain the verified precise height location.

5. The automatic winding machine fault detection method supporting image recognition according to claim 1, characterized in that: S4 includes: The spindle position number index is obtained from the precise height position, and the beam angle control parameters are calculated by combining the running speed correlation parameter with the height position index to obtain the preliminary angle value; The laser power adjustment module is invoked based on the initial angle value. The adjusted power level is determined by calculating the pulse frequency setting combined with the mark duration and laser power control.

6. The automatic winding machine fault detection method supporting image recognition according to claim 5, characterized in that: S4 further includes: Based on the adjusted power level, an intermediate instruction containing beam diameter adjustment information is generated, and the matching relationship between beam diameter and angle value is obtained through beam diameter optimization. If the matching relationship exceeds the preset threshold, the running speed linkage parameter is optimized by combining the error calibration mechanism with the parameter threshold verification, the stability of the optimized parameter is judged, and a stable parameter set is obtained. A stable parameter set is used to fuse position accuracy improvement attributes and command generation to generate the final laser marking command.

7. The automatic winding machine fault detection method supporting image recognition according to claim 1, characterized in that: S5 includes: Obtain the required brightness of the marker from the anomaly classification label, and match the required brightness with the wavelength parameter using a preset wavelength matching rule to obtain the preliminary wavelength value; Based on the initial wavelength value and response delay parameters, the execution timing is optimized by adjusting the timing sequence through a preset synchronization protocol to determine the optimized timing set.

8. The automatic winding machine fault detection method supporting image recognition according to claim 7, characterized in that: The S5 also includes: An optimized time-series set is used to fuse the marked instruction transmission attributes to generate a transmission instruction sequence, which is then transmitted to the laser pointer device. When the laser pointer device receives the transmission command sequence, it activates the device status, checks the activation stability through preset verification rules, and obtains a verification signal group. Based on the verification signal group, the physical marker generation attributes are integrated and the anomaly detection linkage is fused to generate the physical marker signal in the active state.

9. A fault detection method for an automatic winding machine supporting image recognition according to claim 1, characterized in that: S6 includes: Changes in coverage area are obtained from the feedback of the marked signal. A shrinkage detection algorithm is used to calculate the shrinkage ratio of the coverage boundary for the changes in coverage area, and the shrinkage value is obtained. For the abnormal trigger threshold of the range reduction value fusion, the range reduction value is compared with the preset threshold boundary through a threshold comparison mechanism, and the signal data is integrated to determine the accuracy of the effect.

10. A fault detection method for an automatic winding machine supporting image recognition according to claim 9, characterized in that: S6 further includes: Adjust the detection loop based on the accuracy of the effect, obtain the optimization completion status, and determine whether the optimization completion status meets the expected accuracy requirements. If the optimized state meets the expected accuracy requirements, a loop process is generated, resulting in the optimized spindle position anomaly detection loop process.

Images