Acoustic emission detection method and system for pitting corrosion damage of gear tooth surface

Abstract

The invention discloses a gear tooth surface pitting damage acoustic emission detection method and system, and relates to the technical field of mechanical fault diagnosis, and the method comprises the following steps: obtaining an attachment pressure sequence and an acoustic emission waveform sequence of a target object; and performing morphological consistency judgment on the attachment pressure sequence according to a sliding window to generate a trusted window set, and establishing hyperbolic mapping for the acoustic waveform sequence from the head end and the tail end and the attachment pressure sequence of the corresponding trusted window in the trusted window set to generate a hysteresis mapping table. According to the scheme, the pressure representative value is subjected to bias expansion according to the hysteresis mapping table, the bias expansion result changing along with the hysteresis of the pressing force is formed, threshold mismatch caused by coupling drift is restrained, boundary judgment is performed on the bias expansion result and the preset base line proportion to generate the boundary judgment mark, and the threshold mismatch is restrained. Therefore, the time that the threshold needs to be adjusted is quickly identified during proportion crossing, and the strong impact traction threshold is prevented from being lifted.

Description

Technical Field

[0001] This invention relates to the field of mechanical fault diagnosis technology, specifically to a method and system for detecting pitting damage on gear tooth surfaces using acoustic emission. Background Technology

[0002] Gear transmission systems are widely used in equipment in industries such as metallurgy, cement, and chemicals. Especially in external gearboxes that are used in high-temperature kilns, gears operate under high load, thermal radiation, and temperature cycling conditions for a long time, and the tooth surfaces are prone to contact fatigue damage. To achieve early identification and condition assessment of pitting on gear tooth surfaces, acoustic emission detection technology is often used for gear damage monitoring because it is sensitive to transient high-frequency elastic wave responses such as microcrack initiation and localized material spalling.

[0003] In existing technologies, acoustic emission detection schemes for gear pitting typically use fixed thresholds or adaptive thresholds based on waveform statistics to trigger events, count the triggered events, and supplement them with sensors to form a rotational reference when phase positioning is required. This allows for the discrimination of abnormal event density in specific meshing phase segments. The advantage of these schemes is their relatively simple structure and ability to reflect transient impact changes during gear meshing to a certain extent. However, in the scenario of an external gearbox in a high-temperature kiln, the gearbox housing and fixture are subjected to continuous thermal radiation and temperature cycling. The sensor attachment state changes over time, and the coupling transmission of acoustic emission signals exhibits overall scaling and local distortion. This causes the triggering results of fixed thresholds or conventional adaptive thresholds to fluctuate with the coupling state, leading to both missed and false triggers. Furthermore, if the rotational reference relies on sensors, it is difficult to establish a stable rotational reference. Ultimately, this results in a systematic drift of the phase aggregation characteristics with changes in coupling, making it difficult to distinguish between phase shifts caused by changes in coupling state and phase aggregation enhancement caused by the development of pitting on the tooth surface. Consequently, it is difficult to achieve stable output for pitting detection. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for detecting acoustic emission of pitting damage on gear tooth surfaces, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: In a first aspect, the present invention discloses an acoustic emission detection method for pitting damage on gear tooth surfaces, applicable to the acoustic emission detection of pitting damage on gear tooth surfaces inside an external gearbox of a high-temperature kiln, comprising the following steps: Obtain the attachment pressure sequence and acoustic waveform sequence of the target object; The attachment pressure sequence is subjected to morphological consistency discrimination by sliding window to generate a set of reliable windows. Within the set of reliable windows, a hyperbola mapping is established for the acoustic waveform sequence from the beginning and end and the attachment pressure sequence of the corresponding reliable window, and a hysteresis mapping table is generated. The attachment pressure sequence is biased and expanded according to the hysteresis mapping table, and boundary discrimination and threshold update are performed on the bias expansion result and the preset baseline ratio to generate a threshold update set. The acoustic waveform sequence is compared with the threshold update set, and the threshold comparison results are grouped, counted, and the extreme number is taken to generate a cycle candidate period. Then, the threshold comparison results are extrapolated at equal intervals according to the cycle candidate period to generate a cycle reference sequence. The rotation reference sequence is subjected to windowing and offset averaging processing according to the set of reliable windows to generate a position centroid sequence. The position centroid sequence and the attachment pressure sequence are then piecewise fitted to generate a phase offset function. The rotation reference sequence is compensated according to the phase offset function to generate pitting detection results.

[0006] Secondly, this invention discloses an acoustic emission detection system for pitting damage on gear tooth surfaces, comprising: The data acquisition module is used to acquire the attachment pressure sequence and acoustic waveform sequence of the target object; The acoustic signal processing module is used to perform morphological consistency judgment on the attachment pressure sequence according to the sliding window, generate a set of reliable windows, and establish a hyperbola mapping for the acoustic waveform sequence from the beginning and end and the attachment pressure sequence of the corresponding reliable window within the set of reliable windows, and generate a hysteresis mapping table. The threshold update module is used to perform bias expansion on the attachment pressure sequence according to the hysteresis mapping table, and perform boundary discrimination and threshold update on the bias expansion result and the preset baseline ratio to generate a threshold update set. The benchmark generation module is used to perform threshold comparison on the acoustic waveform sequence according to the threshold update set, and to group and count the threshold comparison results and take the extreme number to generate a cycle candidate period. Then, based on the cycle candidate period, the threshold comparison results are extrapolated at equal intervals to generate a cycle benchmark sequence. The function fitting module is used to perform windowing and offset averaging processing on the rotation reference sequence according to the set of reliable windows to generate a position centroid sequence, and to perform piecewise fitting on the position centroid sequence and the attachment pressure sequence to generate a phase offset function. The detection result generation module is used to compensate the rotation reference sequence according to the phase offset function to generate pitting detection results.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This scheme extends the pressure representative value bias based on the hysteresis mapping table, forming a bias extension result that varies with the hysteresis of the clamping force, in order to suppress threshold mismatch caused by coupling drift. By performing boundary discrimination on the bias extension result and the preset baseline ratio, a boundary discrimination mark is generated, thereby quickly identifying the timing when the threshold needs to be adjusted when the ratio exceeds the limit, avoiding the threshold rise caused by strong impact. The current threshold data is updated based on the boundary discrimination mark to generate a threshold update set, so that the threshold is only updated in lockstep when the ratio crosses the interval, thereby reducing missed detections and false alarms.

[0008] 2. This scheme calculates the ratios of the energy sorting results from the beginning and end separately to generate the proportions of high-energy segments and low-energy segments. The relative proportions replace the absolute amplitudes to reduce errors caused by overall scaling and local distortion and improve the comparability across time windows. The pressure representative value and the proportion of high-energy segments, as well as the pressure representative value and the proportion of low-energy segments, are fitted with hyperbolic curves to generate a hysteresis mapping table, quantifying the differences in the rising and falling paths. The coupling correction parameterization under the clamping force hysteresis condition is realized to reduce over-correction and under-correction. Attached Figure Description

[0009] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein: Figure 1 A flowchart illustrating the steps of the acoustic emission detection method for pitting damage on gear tooth surfaces provided by the present invention. Figure 2 A schematic diagram of the process for generating pressure direction indicators provided by the present invention; Figure 3 This is a schematic diagram of the process for generating a hysteresis mapping table provided by the present invention; Figure 4 This is a schematic diagram of the process for generating a threshold update set provided by the present invention; Figure 5 This is a schematic diagram of the module functions of the acoustic emission detection system for pitting damage on gear tooth surfaces provided by the present invention. Detailed Implementation

[0010] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0011] Application Overview: In acoustic emission detection applications of external gearboxes in high-temperature kilns, thermal radiation and temperature cycling cause changes in sensor attachment over time. The coupling and transmission of acoustic emission signals exhibit overall scaling and local distortion, resulting in fluctuations in event triggering mechanisms based on fixed thresholds or waveform statistics, with both missed and false triggering occurring. Furthermore, the establishment of rotational references is difficult to maintain stability due to sensor dependence, and the phase aggregation characteristics undergo systematic drift with changes in coupling state. It is difficult to effectively distinguish between the enhanced phase aggregation caused by pitting corrosion development and the phase shift caused by changes in sensor coupling state, thus affecting the reliability of pitting damage detection.

[0012] For example, in the monitoring scenario of an external gearbox in a cement production rotary kiln, the gearbox housing is exposed to high-temperature thermal radiation for a long time. The adhesion pressure between the sensor and the housing fluctuates periodically with temperature cycles, and the amplitude and waveform characteristics of the acoustic emission signal change nonlinearly. Under these conditions, the event trigger threshold is frequently adjusted due to changes in the coupling state, resulting in unstable statistical results of abnormal event density in the gear meshing phase segment, and a shift in the phase positioning reference. The pitting corrosion discrimination method based on phase aggregation is difficult to provide consistent detection output.

[0013] If the above problems are not addressed, the early identification of pitting damage will be compromised, which may lead to the continued development of pitting on the tooth surface without timely detection, eventually causing gear contact fatigue failure and resulting in unplanned equipment downtime.

[0014] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0015] Example 1: Please see Figure 1 An acoustic emission detection method for pitting damage on gear tooth surfaces, applied to the acoustic emission detection of pitting damage on gear tooth surfaces inside external gearboxes of high-temperature kilns, includes the following steps: Obtain the attachment pressure sequence and acoustic waveform sequence of the target object; The morphological consistency of the attachment pressure sequence is judged by sliding window, a set of reliable windows is generated, and a hyperbola mapping is established for the acoustic waveform sequence from the beginning and end and the attachment pressure sequence of the corresponding reliable window within the set of reliable windows, generating a hysteresis mapping table. The attachment pressure sequence is biased and extended according to the hysteresis mapping table, and boundary discrimination and threshold update are performed on the bias extension result and the preset baseline ratio to generate a threshold update set. The acoustic waveform sequence is compared with the threshold update set, and the threshold comparison results are grouped, counted and the extreme number is taken to generate the candidate cycle. Then, the threshold comparison results are extrapolated at equal intervals according to the candidate cycle to generate the cycle reference sequence. Based on the set of reliable windows, the rotation reference sequence is subjected to windowing and offset averaging to generate a position centroid sequence. The position centroid sequence and the attachment pressure sequence are then piecewise fitted to generate a phase offset function. The rotation reference sequence is compensated based on the phase shift function to generate pitting detection results.

[0016] Among them, the attachment pressure sequence refers to the set of pressure sampling values ​​arranged in chronological order to characterize the attachment state between the acoustic emission sensor and the external gearbox housing of the high-temperature kiln. Acoustic waveform sequence refers to the set of continuous digital waveform data formed by the transient elastic waves generated by the gear tooth surface material under load during the meshing operation of gear teeth, which are then propagated through the structure, collected in real time by acoustic emission sensors and arranged in chronological order. A reliable window set refers to a set of window indices that, after processing the attachment pressure sequence with a sliding window, are selected from all sliding windows and can truly reflect the attachment status of the acoustic emission sensor and are suitable for subsequent acoustic emission data processing. Hysteresis mapping table refers to the set of correspondences between the proportion of adhesive pressure and the energy distribution of acoustic waveform, respectively, when the adhesive pressure is in the upward and downward change process, under the same adhesive pressure level. The offset expansion result refers to a set of data within an allowable ratio range obtained by calculating the restricted offset of the pressure representative value in the hysteresis mapping space in the vertical direction for each confidence window corresponding to the pressure representative value in the attachment pressure sequence. The preset baseline ratio refers to the comparison benchmark used for subsequent threshold updates and pressure bias expansion. It can be obtained by extracting reference waveform segments from the acoustic waveform sequence within the trusted window set and performing a ratio mapping from the beginning based on the energy sorting results of the reference waveform segments. The threshold update set refers to the set of threshold change records formed when the acoustic emission detection threshold is updated under the constraint of the confidence window set, based on the hysteresis mapping relationship between the attachment pressure sequence and the acoustic emission waveform sequence. Threshold comparison result refers to the comparison sequence and its continuous segment expression result obtained after performing amplitude comparison operation point by point on the acoustic waveform sequence within the same time range, with the current valid threshold value in the threshold update set corresponding to the confidence window as the comparison benchmark. The candidate cycle refers to the set of interval values ​​that appears most frequently and has the highest stability among the time intervals between multiple acoustic emission triggering results obtained by threshold comparison in the acoustic emission waveform sequence under the threshold update set constraint. The rotation reference sequence refers to a set of time position sequence data derived from the triggering result obtained by threshold comparison in the acoustic waveform sequence, which is used to characterize the repeated position of a single rotation cycle of a gear on the time axis; The position centroid sequence refers to a numerical sequence that characterizes the overall offset center position of a sound emission event within a confidence window in a rotational coordinate system. The phase offset function refers to a deterministic mapping relationship established within a time range defined by a set of confidence windows, with the numerical change of the attachment pressure sequence as the independent variable and the position centroid offset of the acoustic event in the rotational reference sequence as the dependent variable. Pitting detection results refer to a comprehensive set of data used to characterize whether pitting damage exists on the tooth surface of a target gear and its stable spatial distribution characteristics.

[0017] This scheme generates a set of reliable windows by judging the consistency of the morphology of the attached pressure sequence and establishes a hyperbolic hysteresis mapping table within the reliable window. This can distinguish the rising and falling paths and eliminate the coupling drift caused by thermal cycling, making the acoustic emission characteristics comparable across windows. The attached pressure sequence is biased and extended according to the hysteresis mapping table, and the threshold is updated by judging the extension result and the baseline ratio boundary, generating a threshold update set. This ensures that the threshold only changes as the pressure enters a new interval, thereby suppressing missed detections and false detections caused by strong impact traction. The threshold of the acoustic emission waveform sequence is compared with the threshold of the threshold update set, and the extreme number is obtained by grouping and counting to obtain the candidate cycle period. The cycle reference sequence is generated by extrapolation at equal intervals, so that a stable cycle framework can be obtained without external rotation speed to improve phase consistency. The cycle reference sequence is windowed and averaged according to the reliable window set to generate a position centroid sequence, and a phase offset function is generated by segmenting and fitting it with the attached pressure sequence. This can quantify the pressure-related time delay drift and realize a correctable phase reference. The pitting detection result is output based on the phase offset function to compensate for the cycle reference sequence, and the phase shift caused by coupling drift is removed to improve the reliability of pitting accumulation identification.

[0018] The above describes a complete scheme for acoustic emission detection of pitting damage on gear tooth surfaces. The following section details the acquisition of the attachment pressure sequence and acoustic waveform sequence of the target object, specifically including: The attachment pressure sequence of the target object is obtained through a pressure sensing device; the attachment pressure sequence includes, but is not limited to, the clamping force sampling value, the sampling time corresponding to each sampling value, and the arrangement information of the sampling values ​​in the order of sampling time, etc. Acoustic emission waveform sequence of target object is obtained by acoustic emission sensor; acoustic emission waveform sequence includes but is not limited to waveform amplitude sampling point set, sampling time corresponding to each waveform amplitude sampling point, and arrangement information of waveform amplitude sampling points in the sampling time order, etc. The morphological consistency of the attached pressure sequence is determined using a sliding window, and a set of reliable windows is generated. The attached pressure sequence is divided into sliding windows on the time axis, and the pressure sample values ​​in each sliding window are combined into a window pressure sequence according to the sampling order. Within each sliding window, the difference between two adjacent sampled values ​​in the window pressure sequence is calculated, and the continuous monotonic segment is determined based on the sign continuity of the difference result. Specifically, the length of the segment with consecutive positive elements and the length of the segment with consecutive negative elements in the difference result are statistically analyzed. Within the same sliding window, the position where the first sign of the difference result changes from positive to negative or from negative to positive is further located as the bounce position. After the bounce position, the absolute value of the difference result is statistically analyzed to form a plateau discrimination metric. Specifically, the average value of the absolute value of the difference result after the bounce position is calculated, and a continuous interval where the absolute value of the difference result is continuously less than the average value is found near the average value to determine the plateau interval. When a continuous monotonic segment, a rebound position, and a plateau interval exist simultaneously within the same sliding window, and these three appear sequentially in time, the sliding window is determined to satisfy the morphological consistency. Sliding windows that are determined to have morphological consistency are then regarded as reliable windows. The reliable windows are then aggregated in chronological order to generate a reliable window set.

[0019] The above describes how to obtain the attachment pressure sequence and acoustic waveform sequence of the target object. The following section describes how, after generating the trusted window set, the generation of pressure direction indicators is also included. Please refer to [link / reference needed]. Figure 2 , Figure 2 This is a schematic diagram of the process for generating a pressure direction identifier provided in an embodiment of this application. Generating the pressure direction identifier specifically includes: Within the set of trustworthy windows, the difference between each pair of adjacent attached pressure sequences is calculated to generate a pressure difference sequence, and the sign count of the pressure difference sequence is performed. The sign counting includes counting the number of elements greater than zero in the pressure difference sequence to obtain a positive count and counting the number of elements less than zero to obtain a negative count. Calculate the proportion of positive and negative counts in the non-zero elements of the pressure difference sequence, determine the size relationship of the proportions, and generate pressure direction indicators.

[0020] Among them, the pressure difference sequence refers to the numerical sequence obtained by performing a difference operation on two adjacent pressure sample values ​​in the attached pressure sequence in chronological order within the time range corresponding to each confidence window; Positive count refers to the total number of pressure difference results that are judged to be greater than zero and classified into the positive change set among all pressure difference results within the confidence window; Negative sign counting refers to the total number of pressure difference results within the confidence window that are judged to be less than zero and classified into the negative change set. The proportion result refers to the proportion of various directional differences in the pressure difference sequence obtained by calculating the differences between adjacent adjacent pressure sequences within the same confidence window.

[0021] The above content will be described in detail below: Within the set of trustworthy windows, the difference between each pair of adjacent attached pressure sequences is calculated to generate a pressure difference sequence, and the sign count of the pressure difference sequence is performed. The sign counting includes counting the number of elements greater than zero in the pressure difference sequence to obtain a positive count, and counting the number of elements less than zero to obtain a negative count. Within the set of trustworthy windows, the difference between each pair of adjacent pressure sequences is calculated to generate a pressure difference sequence. Each difference element in the pressure difference sequence is read and compared with zero. When the difference element is greater than zero, the positive count is incremented by one while the negative count remains unchanged. When the difference element is less than zero, the negative count is incremented by one while the positive count remains unchanged. When the difference element is equal to zero, both the positive and negative counts remain unchanged. This process continues until all difference elements have been counted, thus obtaining the positive and negative counts respectively. Calculate the proportions of positive and negative counts among the non-zero elements in the pressure difference sequence, determine the magnitude of the proportions, and generate pressure direction indicators. Calculate the proportion of positive counts in the non-zero difference set and the proportion of negative counts in the non-zero difference set. The proportion of positive counts is calculated by dividing the positive count by the total number of elements in the non-zero difference set, and the proportion of negative counts is calculated by dividing the negative count by the total number of elements in the non-zero difference set. The proportions of positive and negative signs are compared and their relative values ​​are determined. Specifically, when the proportion of positive signs is greater than that of negative signs, a pressure direction indicator representing the upward direction is generated. When the proportion of negative signs is greater than that of positive signs, a pressure direction indicator representing the downward direction is generated. When the proportions of positive and negative signs are equal, a pressure direction indicator representing an unclear direction is generated, and then the pressure direction indicator is output.

[0022] This scheme generates a pressure difference sequence by calculating the difference between pairs of adjacent attached pressure sequences within a reliable window set. This explicitly transforms the minute increases and decreases in clamping force within the reliable time window into a numerical sequence with consistent direction, thereby suppressing the interference of absolute pressure drift on direction judgment. By performing sign counting on the pressure difference sequence to obtain positive and negative counts, short-term fluctuation noise can be separated from the overall trend, so that the direction information is reflected by a large number of difference sign statistics rather than determined by individual jumps. The proportion results of positive and negative counts in non-zero elements are calculated separately, which normalizes the count results under different sampling densities or different fluctuation amplitudes to a comparable scale, reducing the direction misjudgment caused by changes in sampling conditions. The magnitude relationship of the proportion results is judged to generate pressure direction identifiers, which can stably output the upward or downward direction anchor points within the reliable window, providing a consistent direction basis for subsequent hysteresis mapping grouping and reducing the correction deviation caused by aliasing of rising and falling segments.

[0023] As described above, after generating the trusted window set, the process also includes generating pressure direction identifiers. The following describes how, within the trusted window set, hyperbolic mappings are established for the acoustic waveform sequence from both the beginning and end points to the corresponding attached pressure sequences of the trusted windows, generating a hysteresis mapping table. Please refer to [reference needed]. Figure 3 , Figure 3 This is a schematic diagram of the process for generating a hysteresis mapping table provided in an embodiment of this application. Generating the hysteresis mapping table specifically includes: Within the set of confidence windows, the acoustic waveform sequence is segmented to generate a set of candidate waveform segments. The set of candidate waveform segments is then sorted by amplitude and descending order to generate an energy ranking result. The energy sorting results are compared with the candidate waveform segment set from the beginning and end, respectively, to generate the proportion of high-energy segments and the proportion of low-energy segments. Hyperbolic fitting was performed on the pressure representative value and the proportion of high-energy segments, and on the pressure representative value and the proportion of low-energy segments, respectively, to generate a hysteresis mapping table; The pressure representative value is obtained by statistical analysis of the attached pressure sequence within a set of trustworthy windows.

[0024] Among them, the candidate waveform segment set refers to the set of multiple waveform segments continuously extracted from the acoustic waveform sequence according to a unified segmentation rule within the time range corresponding to each confidence window; The energy ranking result refers to the ranking structure data obtained by calculating the amplitude energy of each candidate waveform segment and ranking all candidate waveform segments according to the magnitude of the amplitude energy. The high-energy segment ratio refers to the ratio between the number of high-energy segments at the beginning of the energy sorting result and the total number of candidate waveform segments within the confidence window; The low-energy segment ratio refers to the ratio between the number of low-energy segments at the end of the energy sorting result and the total number of candidate waveform segments within the confidence window; The pressure representative value refers to a single numerical value extracted and calculated from the attachment pressure sequence within a single reliable time window defined by the reliable window set, which is used to characterize the overall level of sensor attachment and clamping state within that reliable time window.

[0025] The above content will be described in detail below: Within a trustworthy window set, the acoustic waveform sequence is segmented to generate a candidate waveform segment set. The candidate waveform segment set is then sorted by amplitude and descending order to generate an energy ranking result. The start and end positions of each time window corresponding to the set of reliable windows are obtained, and the acoustic waveform sequence is continuously truncated within the start and end positions of each time window to obtain window waveform data. Then, the window waveform data is segmented sequentially according to a fixed segment length to generate a set of candidate waveform segments within the time window. The fixed segment length is represented by the number of sampling points. The segmentation method is to start from the starting sampling point of the window waveform data, take consecutive sampling points according to the fixed segment length to form a segment, until the end of the window waveform data is reached. If the number of remaining sampling points at the end is less than the fixed segment length, the remaining sampling points and the sampling points adjacent to the end of the previous segment are combined to form the last segment to ensure the integrity of the segment. The amplitude statistics of the candidate waveform segments are performed on each segment, and segment statistics are generated. The amplitude statistics include squaring and summing the amplitudes of each sampling point within each candidate waveform segment to obtain the energy value of the segment, and averaging the absolute values ​​of the amplitudes of each sampling point within the candidate waveform segment to obtain the average amplitude of the segment. The energy value and the average amplitude are combined to obtain the comprehensive amplitude statistics of the segment. The specific calculation formula is as follows: ; In the formula, Indicates the first The energy value of each candidate waveform segment Indicates the first The average amplitude of each candidate waveform segment This represents the combined weighting coefficients. All the above data have been normalized during the calculation. Each candidate waveform segment is paired with its segment identifier and its comprehensive amplitude statistic. The segment statistical pairs are then sorted in descending order based on the comprehensive amplitude statistic to generate the energy ranking result. The energy ranking results are compared with the candidate waveform segment set from both the beginning and end, respectively, to generate the proportions of high-energy segments and low-energy segments. Candidate waveform segments are extracted from the beginning and end of the energy sorting result to form a beginning segment set and an end segment set, respectively. The beginning segment set is the set of the N candidate waveform segments with the highest sorting position in the energy sorting result, and the end segment set is the set of the M candidate waveform segments with the lowest sorting position in the energy sorting result. The generation process of N and M is as follows: Calculate the energy difference between adjacent sorted positions of the energy sorting result to obtain an energy difference sequence, and take the average value of the energy difference sequence to obtain the mean difference. Then, select segments to expand into the queue from the beginning and end of the energy sorting result respectively. The expansion rule is to continue expanding when the energy difference between the current expanded segment and its adjacent segments is greater than the mean difference, and to stop expanding when the energy difference is not greater than the mean difference. Thus, a range of high-energy segments with significant energy differences is obtained at the beginning, and the candidate waveform segments in this range are summarized into the beginning segment set. A range of low-energy segments with significant energy differences is obtained at the end, and the candidate waveform segments in this range are summarized into the end segment set. Therefore, N specifically refers to the number of segments contained in the beginning segment set automatically determined by the above energy difference comparison operation, and M specifically refers to the number of segments contained in the end segment set automatically determined by the above energy difference comparison operation. The total number of segments in the candidate waveform segment set is obtained by counting the total number of segments. The proportion of high-energy segments and the proportion of low-energy segments are calculated separately. The calculation process for generating the proportion of high-energy segments is as follows: divide N with the total number of segments as the numerator and perform a division operation to obtain the proportion of high-energy segments. The calculation process for generating the proportion of low-energy segments is as follows: divide M with the total number of segments as the numerator and perform a division operation to obtain the proportion of low-energy segments. Hyperbolic fitting was performed on the pressure representative value and the proportion of high-energy segments, and on the pressure representative value and the proportion of low-energy segments, respectively, to generate a hysteresis mapping table; The pressure representative value is obtained through statistical analysis of the attachment pressure sequence within a set of confidence windows. Statistical analysis is performed on the attachment pressure sequence within each reliable time window defined by the reliable window set to obtain the representative pressure value. Specifically, the corresponding pressure sample value set is extracted within the reliable time window. The pressure sample value set is summed and divided by the number of sample values ​​to obtain the pressure mean of the reliable time window, which is then used as the representative pressure value. Construct a first mapping sample set of "pressure representative value - high energy segment ratio" and a second mapping sample set of "pressure representative value - low energy segment ratio". The first mapping sample set consists of a pair of pressure representative values ​​of each confidence time window and the high energy segment ratio of that confidence time window, and the second mapping sample set consists of a pair of pressure representative values ​​of each confidence time window and the low energy segment ratio of that confidence time window. Hyperbolic fitting is performed on the first mapped sample set to generate the first proportional curve parameters. The calculation process is as follows: the representative pressure value is used as the independent variable, the proportion of high-energy segments is used as the dependent variable, the fitting residuals are calculated for all sample pairs and their sum of squares is obtained, and then the hyperbolic parameters are determined under the principle of minimizing the sum of squares to obtain the first proportional curve parameters. The specific calculation formula is as follows: ; In the formula, This represents the parameter vector of the hyperbola model. This represents the number of sample pairs in the first mapped sample set. Indicates the first The proportion of high-energy segments in each sample pair (dependent variable). Indicates the first The stress representative value (independent variable) for each sample pair. This represents the bias term of the hyperbola. This represents the scaling factor of the hyperbola. This represents the translation term of the independent variable. All the above data have been normalized during the calculation. The second proportional curve parameters are generated by performing hyperbolic fitting on the second mapping sample set. The calculation process is as follows: the pressure representative value is used as the independent variable and the proportion of low energy segments is used as the dependent variable. The fitting residuals are calculated for all sample pairs and their sum of squares is obtained. Then, the hyperbolic parameters are determined under the principle of minimizing the sum of squares to obtain the second proportional curve parameters. The parameters of the first and second proportional curves are summarized to form a hysteresis mapping table containing the "pressure representative value to high energy segment proportional mapping relationship" and the "pressure representative value to low energy segment proportional mapping relationship".

[0026] This scheme segments the acoustic waveform sequence within a reliable window set to form a candidate waveform segment set. Then, it statistically analyzes the candidate waveform segment set by segment amplitude and sorts them in descending order to generate an energy ranking result. Under the same attachment condition, it can bring the energy distribution in different windows to a unified ranking scale, thereby weakening the amplification or compression effect of coupling attenuation on the absolute amplitude. The energy ranking result is calculated by comparing it with the candidate waveform segment set from the beginning and end to generate the proportion of high-energy segments and the proportion of low-energy segments. The proportion can be used to replace the single-point amplitude representation, making the comparison between windows insensitive to occasional strong impacts and improving the distinguishability of weak impacts in the early stage of pitting corrosion. By generating hysteresis mapping tables through hyperbolic fitting of the pressure representative value with the proportion of high-energy segments and the pressure representative value with the proportion of low-energy segments, a reversible mapping can be formed when there are differences in the lifting and lowering paths to compensate for the coupling hysteresis drift caused by thermal cycling and reduce the error caused by over- or under-correction. Furthermore, the pressure representative value is obtained by statistical analysis of the attachment pressure sequence within a set of reliable windows, which can quantify the clamping state of the fixture into a stable parameter and ensure that subsequent threshold lockstep updates have a consistent pressure reference.

[0027] The above describes how to establish hyperbolic mappings of the acoustic waveform sequence from the beginning and end to the corresponding attachment pressure sequence within a trusted window set, generating a hysteresis mapping table. The following describes how to perform offset expansion on the attachment pressure sequence based on the hysteresis mapping table, and then perform boundary discrimination and threshold updates on the offset expansion result and a preset baseline ratio to generate a threshold update set. Please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram of the process for generating a threshold update set provided in an embodiment of this application. Generating the threshold update set specifically includes: The pressure representative values ​​corresponding to the attachment pressure sequence are biased and extended according to the hysteresis mapping table. Perform boundary discrimination on the offset expansion results and the preset baseline ratio, and generate boundary discrimination markers; The preset baseline ratio is obtained by extracting reference waveform segments from the acoustic waveform sequence within a set of trustworthy windows, and performing a ratio mapping from the beginning based on the energy sorting results of the reference waveform segments. The current threshold data is updated based on the boundary discrimination markers to generate a threshold update set. The current threshold data is obtained by combining the reference waveform segment corresponding to the previous confidence window and the preset baseline ratio.

[0028] Among them, the boundary discrimination marker refers to the discrete marker data generated by determining the positional relationship between the preset baseline ratio corresponding to the confidence window and the offset expansion result obtained based on the hysteresis mapping table and the attachment pressure sequence within each confidence window; A reference waveform segment refers to a number of continuous waveform segments selected from the acoustic waveform sequence within a time range defined by a set of reliable windows, which are used to characterize the energy distribution reference of the acoustic emission signal under the current attachment pressure state. The current threshold data refers to the amplitude discrimination benchmark data used to perform trigger discrimination on the acoustic waveform sequence within a certain confidence window.

[0029] The above content will be described in detail below: The pressure representative values ​​corresponding to the attachment pressure sequence are biased and extended according to the hysteresis mapping table: Using the pressure representative value as the lookup input, the proportional mapping value corresponding to the pressure representative value is retrieved from the hysteresis mapping table. The proportional mapping value includes the proportion of high-energy segments and the proportion of low-energy segments. The offset expansion is performed on the scale mapping value to generate the offset expansion result. The specific calculation process is as follows: the difference between the high energy fragment scale and the low energy fragment scale is calculated to obtain the scale span value. The high energy fragment scale and the scale span value are added to obtain the upper bound of the scale. The low energy fragment scale and the scale span value are subtracted to obtain the lower bound of the scale. Thus, the offset expansion result is formed by the lower bound and the upper bound of the scale. Perform boundary discrimination on the offset expansion results and the preset baseline ratio, and generate boundary discrimination markers; The preset baseline ratio is obtained by extracting reference waveform segments from the acoustic waveform sequence within a trusted window set, and performing a ratio mapping from the beginning based on the energy sorting results of the reference waveform segments. Within each reliable time window, a reference waveform segment is extracted from the acoustic waveform sequence that matches the reference position to form a reference waveform segment set. Based on the energy sorting results, a proportional mapping is performed from the beginning to generate the baseline ratio. The calculation process is as follows: In the energy sorting results, the average energy value of all segments is extracted as the energy mean. The set of segments whose energy values ​​are not less than the energy mean in the energy sorting results is determined as the beginning set. The number of segments in the beginning set that belong to the reference waveform segment set is counted as the number of beginning reference segments. At the same time, the total number of segments in the reference waveform segment set is counted as the total number of reference segments. Finally, the ratio of the number of beginning reference segments to the total number of reference segments is used to obtain the baseline ratio corresponding to the reliable time window. The reference position is determined by counting the number of times the position indices of the first and last segments of the energy sorting result appear repeatedly within the reliable time window in both the up and down directions. The segment position index with the highest number of repetitions and consistent position is selected, and the start and end position ranges of the waveform segments corresponding to these segment position indices are summarized as the reference position. Boundary discrimination is performed on the offset expansion result and the baseline ratio to generate boundary discrimination labels. The calculation process is as follows: the difference between the baseline ratio and the lower bound of the offset expansion result ratio is calculated, and the difference between the baseline ratio and the upper bound of the offset expansion result ratio is also calculated. When the difference between the baseline ratio and the lower bound of the offset expansion result ratio is non-negative and the difference between the baseline ratio and the upper bound of the offset expansion result ratio is non-positive, the boundary discrimination label is determined to be within the boundary range. When the difference between the baseline ratio and the lower bound of the offset expansion result ratio is negative, the boundary discrimination label is determined to be below the lower bound. When the difference between the baseline ratio and the upper bound of the offset expansion result ratio is positive, the boundary discrimination label is determined to be above the upper bound, thereby outputting the boundary discrimination label. The current threshold data is updated based on the boundary discrimination markers to generate a threshold update set. The current threshold data is obtained by combining the reference waveform segment corresponding to the previous confidence window with a preset baseline ratio (considering the case of the first confidence window, the starting point). The adjacency relationship between the current trusted window and the previous trusted window on the time axis is determined, and the current threshold data is obtained within the current trusted window as the object of update processing. The process of generating the current threshold data is as follows: when there is a previous trusted window, the reference waveform segment set corresponding to the previous trusted window is extracted from the acoustic waveform sequence within the time range corresponding to the previous trusted window. The amplitude of each sampling point in the reference waveform segment set is squared and averaged to obtain the reference mean square value. Then, the reference mean square value is combined with the corresponding preset baseline ratio to obtain the current threshold data. The combination operation is to multiply the reference mean square value by the corresponding preset baseline ratio. When the current trusted window is the first trusted window and there is no previous trusted window, the set of reference waveform segments corresponding to the trusted window is extracted from the acoustic waveform sequence within the time range corresponding to the first trusted window. The average of the amplitudes of each sampling point in the set of reference waveform segments is obtained by squaring the amplitudes. The reference mean square value is then multiplied by the corresponding preset baseline ratio to obtain the current threshold data of the first trusted window. Based on the boundary discrimination marker, the current threshold data is updated to generate a threshold update set. The update process includes: when the boundary discrimination marker indicates that the current confidence window is within the bias expansion result, the current threshold data is kept as the previous threshold data and the threshold data and its corresponding confidence window index are summarized and written into the threshold update set. When the boundary discrimination marker indicates that the current trusted window has crossed the boundary of the bias expansion result, the current threshold data is replaced with new threshold data obtained by multiplying the current trusted window reference mean square value by the baseline ratio, and the threshold data before replacement, the threshold data after replacement, and the corresponding trusted window index are summarized and written into the threshold update set.

[0030] This scheme uses a hysteresis mapping table to bias-extend the representative pressure values ​​corresponding to the attached pressure sequence to obtain the allowable proportional boundary. This allows the threshold adjustment to be separated from the pressure thermal hysteresis path and not miscorrected by a single curve. Boundary discrimination is performed on the bias-extended result and the preset baseline ratio to generate a boundary discrimination mark. This ensures that the threshold update is triggered only when the ratio exceeds the limit, suppressing false alarms caused by strong shocks leading to threshold overshoot. The preset baseline ratio is obtained by extracting reference waveform segments within the set of confidence windows and performing proportional mapping based on their ranking at the beginning of the energy sorting result. This allows the baseline to represent coupling changes with relative ranking, improving cross-window comparability. The current threshold data is updated based on the boundary discrimination mark to generate a threshold update set. The current threshold is obtained by combining the reference waveform segment from the previous confidence window with the preset baseline ratio. This allows the threshold to transition continuously with the historical stable window as the anchor point, reducing false alarms caused by thermal cycling and coupling drift and maintaining triggering stability.

[0031] The above describes the bias expansion of the attachment pressure sequence based on the hysteresis mapping table, and the boundary discrimination and threshold update performed on the bias expansion result and the preset baseline ratio to generate a threshold update set. The following describes the threshold comparison of the acoustic waveform sequence based on the threshold update set, and the grouping, counting, and extreme value selection of the threshold comparison results to generate candidate cycles. Specifically, this includes: Based on the threshold update set, the acoustic waveform sequence is compared with the threshold, and the timing difference is calculated for each pair of adjacent threshold comparison results to generate an interval sequence; The interval sequence is compared and filtered according to the threshold update set to generate a purification interval sequence. The purification interval sequence is then grouped and counted by value, and the purification interval value with the highest occurrence is taken as the candidate cycle for the turnaround.

[0032] Among them, the interval sequence refers to the numerical sequence formed by arranging multiple trigger positions that meet the threshold comparison conditions in chronological order after performing threshold comparison on the acoustic waveform sequence, and calculating the time difference between two adjacent trigger positions one by one. The purification interval sequence refers to the time interval data set that reflects only the true periodic impact characteristics of the gear, obtained by combining the interval sequence with the threshold update set to remove time intervals affected by threshold changes.

[0033] The above content will be described in detail below: Based on the threshold update set, threshold comparisons are performed on the acoustic waveform sequence, and the timing difference is calculated for each pair of adjacent threshold comparison results to generate an interval sequence. The instantaneous amplitude of the acoustic waveform sequence within the reliable time window is compared with the current threshold value point by point. When the instantaneous amplitude of a certain sampling point reaches or exceeds the current threshold value for the first time, the timing position of the sampling point is recorded as the trigger start point. Then, starting from the trigger start point, the instantaneous amplitude is compared with the current threshold value point by point along the timing sequence. The continuous sampling points whose instantaneous amplitude still reaches or exceeds the current threshold value are regarded as the continuous interval of the same trigger process. When a sampling point with an instantaneous amplitude lower than the current threshold value appears, the number of sampling points that are continuously lower than the current threshold value is judged and counted along the timing sequence. When the number of sampling points that are continuously lower than the current threshold value reaches the stability requirement, the timing position of the last sampling point that reaches the stability requirement is recorded as the trigger end point. Thus, the start and end position range of the waveform segment corresponding to the trigger process is determined by the trigger start point and the trigger end point. The stability requirement is to perform a threshold comparison on the acoustic waveform sequence within the reliable time window using the current threshold value and form a "below-threshold indication sequence". The below-threshold indication sequence is marked by whether the sampling point is below the threshold value. All consecutive segments below the threshold are extracted from the below-threshold indication sequence, and the number of consecutive samples in each segment is calculated. The number of consecutive samples is obtained by counting the sampling points of adjacent below-threshold marked points in the same segment, and the number of consecutive samples in each segment is averaged. After determining a start and end position range, the scanning position is moved to the next sampling point after the trigger endpoint and the above comparison and marking process is continued to obtain multiple start and end position ranges corresponding to multiple triggering processes within the reliable time window. Finally, the multiple start and end position ranges are arranged in the order of the triggering start points, and the arranged start and end position ranges are summarized to obtain the event fragment index, which is used as the threshold comparison result. The starting interval value is obtained by subtracting the time positions of the starting points of two adjacent event segments, and all starting interval values ​​are summarized in time sequence according to the event segments to form an interval sequence. Based on the threshold update set, the interval sequence is compared and filtered over time intervals to generate a cleanup interval sequence. The cleanup interval sequence is then grouped and counted by value, and the cleanup interval value with the highest occurrence is selected as the candidate cycle for the turnaround. Extract the corresponding time window index from the threshold update set, and map the time window index to the start and end positions of the update on the time axis through location retrieval, thereby forming multiple threshold update time series intervals; For each interval value in the interval sequence, determine the corresponding interval start event time and interval end event time, and take the time axis range covered by the interval start event time and interval end event time as the time sequence coverage interval of the interval value. For each interval value, the time series coverage interval is compared with each threshold update time series interval. Specifically, the case where the start of the interval value's coverage interval is not earlier than the end of the threshold update time series interval and the end of the coverage interval is not later than the start of the threshold update time series interval is determined to be non-overlapping, and the other cases are determined to be overlapping. When the time series coverage interval of a certain interval value overlaps with any threshold update time series interval, the interval value is removed from the interval sequence to generate a clean interval sequence; The purification interval values ​​in the purification interval sequence are grouped according to their numerical values ​​to form several numerical groups. The number of purification interval values ​​contained in each numerical group is counted to obtain the group count value. Then, the numerical group with the largest group count value is selected from all numerical groups, and the purification interval value corresponding to the numerical group is determined as the purification interval value with the highest frequency. Finally, the purification interval value with the highest frequency is output as the candidate cycle for the cycle. Based on the candidate cycle periods, the threshold comparison results are extrapolated at equal intervals to generate the cycle baseline sequence: Extract the corresponding time position of each comparison exceeding the threshold from the threshold comparison results and sort them in chronological order to form an event starting point sequence. Select the event starting point corresponding to the interval value that is the same as or closest to the value of the candidate cycle in the interval sequence as the benchmark starting point. Then, extrapolate the time axis at equal intervals with the benchmark starting point as the starting point and the candidate cycle as the step size. Specifically, add the benchmark starting point and the candidate cycle successively to obtain a set of benchmark position point sequences arranged in ascending order of time. Calculate the time distance between each event starting point in the event starting point sequence and its two adjacent reference points. Select the reference point with the smallest distance as the assigned reference point of the event starting point. Use the time difference between the event starting point and its assigned reference point as the position offset. This will give you the correspondence between the event starting point and the reference point and the set of position offsets. Use the set of position offsets as the rotation offset set. Finally, summarize the reference point sequence and the correspondence to generate the rotation reference sequence.

[0034] This scheme performs threshold comparison on the acoustic waveform sequence based on the threshold update set to obtain the threshold comparison result. This ensures that the event extraction maintains a consistent trigger scale as the threshold lockstep changes, reducing false triggers and missed triggers caused by high temperature shocks. The scheme generates an interval sequence by subtracting the execution time sequences of each pair of adjacent threshold comparison results, converting the event time sequence into a statistically representative periodic interval to enhance the identifiability of the cycle. Based on the threshold update set, the interval sequence is compared and filtered to generate a clean interval sequence, eliminating interval distortion intervals caused by threshold updates, thereby improving the stability of cycle estimation. The clean interval sequence is grouped and counted by value, and the clean interval value with the highest occurrence frequency is taken as the candidate cycle. The stable interval with the highest repetition suppresses the influence of occasional noise and outlier intervals, thereby improving the accuracy and repeatability of the candidate cycle.

[0035] The above describes the threshold comparison of the acoustic waveform sequence based on the threshold update set, and the grouping and counting of the threshold comparison results to obtain the extreme values, generating candidate cycles. The following describes the windowing and offset averaging processing of the cycle reference sequence based on the confidence window set to generate a position centroid sequence, and the piecewise fitting of the position centroid sequence and the attachment pressure sequence to generate a phase shift function, specifically including: The rotation reference sequence is windowed and subtracted according to the confidence window set to generate the rotation offset set, and the mean of the rotation offset set is calculated to generate the position centroid sequence. The pressure representative values ​​are grouped according to their numerical similarity within adjacent confidence windows to generate a pressure segment set. Based on the pressure segment set, the position centroid sequence is segmented and statistically analyzed according to pressure segments, and the mean value is calculated to obtain the average offset of the pressure segment. The pressure representative value is then combined with the average offset to generate a mapping pair. Finally, curve fitting is performed on all mapping pairs to generate the phase offset function.

[0036] Among them, the rotation offset set refers to the data set formed by summarizing the relative offsets of each event fragment falling within the confidence window in the rotation reference coordinate system within the time range corresponding to each confidence window; The pressure segment set refers to a set of pressure segments formed by merging representative pressure values ​​extracted from the attached pressure sequence according to their numerical similarity under the confidence window constraint. The average offset refers to a statistical measure that reflects the degree of systematic offset of an event relative to the rotational reference position, calculated based on the rotational reference sequence within multiple confidence windows corresponding to the same pressure segment set. A mapping pair refers to a pair of data structures used in the construction of the phase offset function to characterize the correspondence between "adhesion pressure change" and "overall rotational phase offset change".

[0037] The above content will be described in detail below: The rotational baseline sequence is windowed and subtracted according to a set of confidence windows to generate a rotational offset set. The mean of the rotational offset set is then calculated to generate a centroid sequence. The centroid sequence is generated by averaging the set of rotational offsets. The calculation process is as follows: taking a reliable time window as the unit, the centroid offset value of the reliable time window is obtained by averaging all the position offsets belonging to the same reliable time window. The centroid offset values ​​of each reliable time window are arranged in the time order of the reliable time window and output as the centroid sequence. The pressure representative values ​​are grouped according to their numerical similarity within adjacent confidence windows to generate a pressure segment set: The pressure representative values ​​of adjacent reliable time windows are grouped into adjacent pairs in chronological order, and the adjacent difference is obtained by subtracting the previous pressure representative value from the later pressure representative value in the adjacent pair. The absolute value of adjacent differences is taken to obtain the difference range, and the arithmetic mean of all difference ranges is used as the approximate scale. The pressure representative value of the first reliable time window is determined as the starting representative value of the current segment, and the pressure representative values ​​of subsequent reliable time windows are read in sequence. For each subsequent pressure representative value, the difference between it and the starting representative value of the current segment is calculated. If the difference is less than or equal to the scale, the subsequent pressure representative value is assigned to the current segment, and the arithmetic mean of the pressure representative values ​​in the current segment is updated synchronously as the starting representative value of the current segment update. If the difference is greater than the approximate scale, then the subsequent pressure representative value is used as the starting representative value of the new segment and the next segment is started; After all trusted time windows have been processed, the trusted time window index range contained in each segment and the update start representative value of that segment are summarized to form a set of stress segments and output them. Based on the pressure segment set, the position centroid sequence is segmented by pressure segment, and the mean is calculated to obtain the average offset of the pressure segment: First, determine the range of representative pressure values ​​corresponding to each pressure segment in the pressure segment set as the segment range of that pressure segment, and then locate the time window index set belonging to the segment range of that pressure segment one by one on the time axis. Based on the time window index set, extract the centroid offset values ​​corresponding one-to-one with each time window index in the centroid sequence to form the centroid offset value set for this pressure segment; The average offset of the pressure segment is obtained by averaging the set of centroid offset values. The calculation process is as follows: sum the centroid offset values ​​in the set of centroid offset values ​​of the pressure segment to obtain the total offset, and then divide the total offset by the number of elements in the set of centroid offset values ​​of the pressure segment to obtain the average offset. By combining the representative pressure value with the average offset, mapping pairs are generated. Then, curve fitting is performed on all mapping pairs to generate a phase offset function. The specific calculation formula is as follows: ; In the formula, This represents the candidate set of the fitted function. This indicates the total number of pressure sections. Indicates the first The average offset of each pressure segment Indicates the first The pressure values ​​for each pressure range are represented by the following values. This represents the candidate fitting function. All the data above have been normalized during the calculation.

[0038] This scheme performs windowed subtraction on the rotational reference sequence according to a set of confidence windows to obtain a rotational offset set, and then averages the results to generate a position centroid sequence. This transforms the rotational position offset from event-level discrete fluctuations to a window-level stable representation, thereby suppressing the interference of threshold jitter and occasional impacts on phase estimation. The attached pressure sequence is averaged within each confidence window to obtain a representative pressure value, and adjacent confidence windows are merged according to their numerical similarity to generate a pressure segment set. This transforms the pressure change from a continuous noise sequence to a segmented steady-state interval, thereby reducing the randomness of the mapping establishment caused by thermal cycling pressure drift. Based on the pressure segment set, the position centroid sequence is segmented and statistically analyzed, and the average offset is obtained. This average offset is then combined with the representative pressure value to form a mapping pair, and curve fitting is performed to generate a phase offset function. This establishes a continuous and calculable relationship between pressure and phase drift, thereby achieving a consistent benchmark for subsequent phase compensation and improving the stability and repeatability of the pitting phase aggregation criterion.

[0039] The above describes the windowing and offset averaging processing performed on the rotation reference sequence based on the confidence window set to generate a position centroid sequence. The position centroid sequence and the attachment pressure sequence are then piecewise fitted to generate a phase shift function. The following describes the compensation processing of the rotation reference sequence based on the phase shift function to generate pitting detection results, specifically including: The rotational reference sequence is compensated and clustered based on the phase shift function to generate a phase band index set. Under the constraint of the phase band index set, the consistency of the acoustic waveform sequence combined with the pressure direction mark is analyzed to generate a pitting evidence score. A weighted synthesis and difference operation are performed on the threshold update set, the pitting erosion evidence score, and the phase band index set to generate a score gradient sequence; Gradient matching is performed on the scoring gradient sequence to generate a fastening suggestion sequence. After executing the fastening suggestion sequence, a verification process is performed to generate pitting detection results.

[0040] Among them, the phase band index set refers to the index set formed in the rotation position space by acoustic emission events after phase compensation under the rotation reference framework, which is used to characterize the concentrated distribution area of ​​pitting features on the gear tooth surface. The pitting evidence score is a comprehensive numerical value used to characterize the strength of the acoustic emission evidence of pitting on the current gear tooth surface that is stable, repeatable, and independent of the direction of the adhesion pressure. Scoring gradient sequence refers to a data sequence used to quantify the trend of the reliability of pitting corrosion diagnosis over time; The recommended tightening sequence refers to a set of minimum tightening adjustments that are recommended to improve the reliability of pitting corrosion detection without adding new sensor data or introducing human experience parameters.

[0041] The above content will be described in detail below: The threshold comparison result is time-series subtracted from the cycle reference sequence and the absolute value is taken to generate the event cycle offset sequence. The event rotation offset sequence is compensated and clustered based on the phase offset function to generate a phase band index set: Input the pressure representative value into the phase offset function to obtain the offset compensation amount corresponding to the confidence window, where the offset compensation amount is the offset value output by the phase offset function corresponding to the pressure representative value; For each event rotation offset falling within the trusted window, compensation calculation is performed one by one. Specifically, the offset compensation amount is subtracted from the rotation offset of each event to obtain the compensated offset. Then, the compensated offsets in all time windows are summarized in the order of events to form the compensated position distribution. Cluster analysis is performed on the compensated offsets to generate a phase band index set. The specific process is as follows: First, the compensated offsets are divided into multiple adjacent position intervals according to the rotation position coordinates, and the number of compensated offsets in each position interval is counted to obtain the interval count value. Then, the interval count values ​​are arranged in position order to form a phase density sequence. In the phase density sequence, the interval with the largest count value is selected as the initial dense interval. The difference between the count values ​​of adjacent intervals is calculated with the initial dense interval as the center to obtain the density gradient value. The sign and magnitude of the density gradient value are compared to determine the density decay direction from the center outward. Then, adjacent intervals are included in the cluster range one by one along the density decay direction. Each time an interval is included, the count value of the included interval is recalculated and compared with the average count value of the included intervals. When the difference between the count value of the included interval and the average count value of the included intervals is in the same direction, the included interval is confirmed as part of the same cluster bandwidth, thus obtaining the phase band range composed of multiple adjacent position intervals. Extract and summarize the event segment index entries corresponding to all events whose compensated offsets fall within the phase band range to form a phase band event set, and output the start and end position ranges of the event segments corresponding to the phase band event set in chronological order as a phase band index set; The attachment pressure sequence is normalized based on the hysteresis mapping table to generate proportional weights. The specific calculation formula is as follows: ; In the formula, Indicates time The proportional mapping value, This indicates the lower bound of the proportion used for normalization. This represents the upper bound of the proportion used for normalization. Indicates time Pressure value, This represents the uplink mapping relationship in the hysteresis mapping table. This represents the downlink mapping relationship in the hysteresis map table. Indicates time Pressure direction markings Indicates the upward direction. Indicates the downward direction; all the above data have been normalized during the calculation. Under the constraints of the phase band index set and threshold comparison results, the energy comparison between inside and outside the band of the acoustic waveform sequence is calculated and weighted according to the proportional weight, and then combined with the pressure direction mark to generate an evidence consistency matrix: The set of start and end positions of event segments corresponding to the phase band range is determined based on the phase band index set, and the waveform segments corresponding to this set are defined as the in-band waveform segment set. At the same time, the waveform segments corresponding to the set of start and end positions of event segments that do not fall within the phase band range are defined as the out-of-band waveform segment set. The sampling point indices in the in-band waveform segment set are mapped and aligned with the threshold comparison results to obtain the threshold-constrained set of valid in-band sampling points within the in-band waveform segment set. Similarly, the threshold-constrained set of valid out-of-band sampling points is obtained within the out-of-band waveform segment set. The in-band energy and out-of-band energy are calculated separately. The in-band energy is calculated as follows: take the amplitude of each sampling point in the set of effective sampling points in the band and square it to obtain the square amplitude. Then sum all the square amplitudes to obtain the in-band energy sum. At the same time, count the number of sampling points in the set of effective sampling points in the band to obtain the number of sampling points in the band. Finally, divide the in-band energy sum by the number of sampling points in the band to obtain the in-band energy mean. The calculation process of out-of-band energy is as follows: take the amplitude of each sampling point in the set of effective out-of-band sampling points and perform a square operation to obtain the square amplitude. Then, sum all the square amplitudes to obtain the out-of-band energy sum. At the same time, count the number of sampling points in the set of effective out-of-band sampling points to obtain the number of out-of-band sampling points. Finally, divide the out-of-band energy sum by the number of out-of-band sampling points to obtain the out-of-band energy mean. The energy contrast between inside and outside the band is calculated to form an energy contrast value. The calculation process of the energy contrast value is to compare the average energy inside the band with the average energy outside the band to obtain the energy contrast value. Thus, the energy contrast value characterizes the degree of energy enhancement inside the band relative to outside the band when both phase band constraints and threshold constraints are simultaneously met. The energy comparison values ​​are weighted according to the proportional weights to form a weighted comparison value. The calculation process of the weighted comparison value is to multiply the energy comparison value by the proportional weights to obtain the weighted comparison value. The weighted comparison values ​​are then grouped by direction using pressure direction markers and summarized to generate an evidence consistency matrix. Consistency scores are calculated on the evidence consistency matrix to generate pitting erosion evidence scores. The specific calculation formula is as follows: ; In the formula, This indicates the row number of the consistency matrix. This indicates the number of columns in the consistency matrix. The first element of the evidence consistency matrix represents the... Line number The weighting coefficients of the matrix cell values ​​in the column. The first element of the evidence consistency matrix represents the... Line number The matrix cell values ​​of the columns, all of which have been normalized during the calculation; Weighted synthesis and differencing operations are performed on the threshold update set, pitting evidence score, and phase band index set to generate a score gradient sequence: The threshold update intensity sequence is generated from the threshold update set. The calculation process is as follows: In each time window, the threshold update records contained in the time window are summarized. The difference between the updated threshold value and the original threshold value of each threshold update record is taken as the absolute value to obtain the single update amplitude value. Then, all single update amplitude values ​​in the time window are summed to obtain the total threshold update amplitude of the time window. At the same time, the number of threshold update records in the time window is counted to obtain the threshold update frequency value of the time window. Finally, the total threshold update amplitude value and the threshold update frequency value are multiplied to obtain the threshold update intensity value of the time window, thus forming a threshold update intensity sequence in the order of the time windows. The phase band width sequence and phase band drift sequence are generated from the phase band index set. The calculation process is as follows: In each time window, the start position and end position of the phase band corresponding to the phase band index set are read. The phase band width value of the time window is obtained by subtracting the start position from the end position. The phase band width sequence is formed by summing the phase band width values ​​in the order of the time windows. The start position difference value is obtained by subtracting the phase band start position of the two adjacent time windows respectively. The end position difference value is obtained by subtracting the phase band end position of the two adjacent time windows respectively. The absolute values ​​of the start position difference value and the end position difference value are taken and summed to obtain the phase band drift amount of the two adjacent time windows. Thus, the phase band drift sequence is formed in the order of the time windows. The process of generating an evidence difference sequence for pitting corrosion evidence scoring is as follows: the difference between the pitting corrosion evidence scores of two adjacent time windows is calculated to obtain the change in evidence, and the results are summarized in the order of the time windows to form an evidence difference sequence. Weighted synthesis is performed to form a comprehensive change sequence. The calculation process is as follows: First, the threshold update intensity sequence is summed to obtain the total threshold intensity; the phase band drift sequence is summed to obtain the total drift; and the absolute values ​​of the evidence difference sequence are summed to obtain the total evidence change. Then, the total threshold intensity, the total drift, and the total evidence change are added to obtain the total sum. The threshold weight is obtained by the ratio of the total threshold intensity to the total sum; the evidence weight is obtained by the ratio of the total evidence change to the total sum; and the drift weight is obtained by the ratio of the total drift to the total sum. Subsequently, difference operations and weighted synthesis operations are performed on each adjacent time window pair. The calculation process for generating the scoring gradient sequence is as follows: the threshold update intensity value corresponding to the adjacent time window is subtracted to obtain the threshold intensity difference value; the phase bandwidth value corresponding to the adjacent time window is subtracted and the absolute value is taken to obtain the bandwidth change value; the evidence difference value corresponding to the adjacent time window is taken as the evidence change value; finally, the threshold weight is multiplied by the threshold intensity difference value, the evidence weight is multiplied by the evidence change value, and the drift weight is multiplied by the bandwidth change value, and the three are normalized and summed to obtain the gradient score value of the adjacent time window. All gradient score values ​​are summarized in the order of the time windows to obtain the scoring gradient sequence. Discrete action matching is performed between the target pressure range and the scoring gradient sequence to generate a tightening suggestion sequence; The target pressure range is generated by filtering and sampling the attached pressure sequence according to a set of reliable windows and taking extreme values. The attached pressure sequence is filtered and sampled using a set of reliable windows. The pressure sample values ​​that fall within the time window of each set of reliable windows are summarized in chronological order to form a pressure sample set. Extreme value calculation is performed on the pressure sample set to generate the target pressure range. The extreme value calculation includes comparing each pressure sample set and determining the minimum pressure value as the lower limit of the target pressure and the maximum pressure value as the upper limit of the target pressure. The target pressure range is thus formed by the lower limit of the target pressure and the upper limit of the target pressure. The sampled values ​​of the attachment pressure sequence corresponding to the end time window in the reliable window set are numerically summarized and the average value is calculated to obtain the current pressure representative value. The pressure center value of the interval is obtained by summing the lower limit of the target pressure and the upper limit of the target pressure and dividing by two. At the same time, the interval width value is obtained by subtracting the upper limit of the target pressure and the lower limit of the target pressure. Then, the pressure offset is obtained by subtracting the current pressure representative value from the interval center pressure value. The sign of the pressure offset is encoded as the action direction code. The offset ratio value is obtained by comparing the absolute value of the pressure offset with the interval width value to represent the degree of deviation. In the attachment pressure sequence covered by the confidence window set, the difference between adjacent sampled values ​​is calculated and the absolute value is taken to obtain the pressure difference set. The pressure difference set is sorted and the value at the middle position is taken as the small step size value. At the same time, the extreme value of the pressure difference set is compared and the maximum value is taken as the large step size value. Then, the small step size value and the large step size value are summed and divided by two to obtain the medium step size value. Thus, a discrete step size set composed of small step size value, medium step size value and large step size value is formed, which correspond to the three discrete action intensities of small-amplitude tightening, medium-amplitude tightening and large-amplitude tightening, respectively. The scoring gradient sequence is normalized to form a matching quantity that can be compared with the discrete step size set. Specifically, the absolute value of the scoring gradient sequence is taken to obtain the gradient magnitude sequence. The maximum gradient magnitude value is obtained by comparing the extreme values ​​of the gradient magnitude sequence. The gradient magnitude value that is aligned with the end time window of the confidence window set in the gradient magnitude sequence is taken as the current gradient magnitude value. Then, the gradient ratio value is obtained by comparing the current gradient magnitude value with the maximum gradient magnitude value. The gradient scale value is used to calculate the recommended step size value, which is obtained by a linear combination of the small step size value, the gradient scale value, and the large step size value. That is, the recommended step size value is obtained by adding the results of multiplying the small step size value and the gradient scale value by the difference between the large step size value and the small step size value. Next, the recommended step size value is subtracted from the small step size value, medium step size value, and large step size value in the discrete step size set, and the absolute value is taken. The discrete step size value corresponding to the smallest absolute value is selected as the matching step size value, and the discrete action intensity corresponding to the matching step size value is encoded as the action intensity code. The action direction code, action intensity code, matching step size value, target pressure range, current pressure representative value, and offset ratio value are combined in a fixed field order to form a tightening suggestion sequence that can be directly parsed by the execution end. After executing the recommended fastening sequence, the attachment pressure sequence and acoustic waveform sequence are reviewed and recalculated to generate a review pitting corrosion evidence score and a review evidence consistency matrix. The review pitting corrosion evidence score and the review evidence consistency matrix are then compared with the pitting corrosion evidence score and the evidence consistency matrix respectively using differential synthesis to generate the pitting corrosion detection results. After executing the recommended fastening sequence, the attachment pressure sequence and acoustic waveform sequence are re-acquired, and the same detection link before and after fastening is reviewed and recalculated to generate a review pitting corrosion evidence score and a review evidence consistency matrix. The evidence score difference is calculated by subtracting the pitting erosion evidence score from the review pitting erosion evidence score. The consistency matrix difference is calculated by comparing the review evidence consistency matrix and the evidence consistency matrix item by item in the corresponding matrix cells, counting the number of cells with different values ​​in the corresponding matrix cells to obtain the number of difference cells, and then comparing the number of difference cells with the total number of matrix cells to obtain the consistency matrix difference. The pitting detection results are generated by combining the evidence score difference value and the consistency matrix difference value. When the evidence score difference value is non-negative and the consistency matrix difference value does not increase, the detection conclusion that pitting is established is output, and the verification pitting evidence score is added as risk information. When the evidence score difference value is negative or the consistency matrix difference value increases, the detection conclusion dominated by coupling artifacts is output, and a verification failure flag is given.

[0042] This scheme compensates for the rotational baseline sequence by performing clustering based on the phase shift function and generates a phase band index set. This transforms the overall phase drift caused by the change in attachment pressure into a correctable shift, stabilizing the event cluster position and reducing phase band false expansion. Under the constraint of the phase band index set, the acoustic waveform sequence is combined with pressure direction markers for consistency analysis and pitting evidence score is generated. This suppresses one-sided overestimation of evidence caused by coupling scaling and local distortion, ensuring that pitting evidence remains comparable in both uplink and downlink paths and reducing false alarms. The threshold update set, pitting evidence score, and phase band index set are weighted, synthesized, and differentially processed to generate a scoring gradient sequence. This unifies the threshold lockstep stability, evidence strength, and phase band convergence to the same scale of change, improving the discriminative reliability of the diagnosis. The scoring gradient sequence is gradient matched to generate a tightening suggestion sequence, which is then verified to generate pitting detection results. This allows for rapid recovery of the high-confidence pressure window with minimal tightening actions and verification within the same link, thereby improving the consistency and repeatability of the conclusions.

[0043] The above describes the compensation processing of the rotational reference sequence based on the phase shift function to generate pitting corrosion detection results. The following describes the compensation and clustering of the rotational reference sequence based on the phase shift function to generate a phase band index set. Under the constraints of the phase band index set, a consistency analysis is performed on the acoustic waveform sequence combined with pressure direction markings to generate a pitting corrosion evidence score. Specifically, this includes: The threshold comparison results are subtracted from the cycle reference sequence to generate an event cycle offset sequence. The event cycle offset sequence is then compensated and clustered according to the phase offset function to generate a phase band index set. The attached pressure sequence is normalized according to the hysteresis mapping table to generate proportional weights. Under the constraints of the phase band index set and the threshold comparison results, the energy comparison inside and outside the band of the acoustic waveform sequence is calculated and weighted according to the proportional weights. Then, the evidence consistency matrix is ​​generated by combining the pressure direction mark. The consistency degree of the evidence consistency matrix is ​​calculated to generate a pitting erosion evidence score.

[0044] Among them, the event rotation offset sequence refers to the numerical sequence formed by calculating the relative offset position of the event starting point in the rotation coordinate for each sound emission event determined by the threshold comparison result, using the rotation reference sequence as the reference coordinate system, and arranging them in the order of event occurrence. Proportional weighting refers to normalized weighted data used to measure the comparability and credibility of acoustic emission evidence under different attachment pressure conditions. The evidence consistency matrix is ​​a two-dimensional data structure generated by structuring the evidence representation of acoustic waveform sequences in different pressure change directions.

[0045] This part has already been described in detail above, so I will not repeat it here.

[0046] This scheme generates an event cycle offset sequence by subtracting the threshold comparison result from the cycle reference sequence. Then, it uses a phase offset function for compensation and clustering to generate a phase band index set. This eliminates the overall drift in cycle position caused by changes in attachment pressure, allowing events to cluster under a unified phase coordinate system and stably delineate suspected pitting phase bands. Based on a hysteresis mapping table, the attachment pressure sequence is normalized to generate proportional weights. Under the constraints of the phase band index set and the threshold comparison result, the energy comparison between the acoustic waveform sequence and the in-band and out-of-band regions is calculated and weighted proportionally. Combined with pressure direction markings, an evidence consistency matrix is ​​generated, making the in-band enhancement evidence comparable under uplink and downlink conditions and suppressing non-phase band noise distortion. The consistency of the evidence consistency matrix is ​​calculated to generate a pitting evidence score, compressing the cross-directional consistent phase band energy enhancement into a stable quantitative index, thereby reducing false alarms and improving the reliability of pitting detection.

[0047] The above describes the compensation clustering of the rotational reference sequence based on the phase shift function to generate a phase band index set. Under the constraints of the phase band index set, a consistency analysis is performed on the acoustic waveform sequence combined with the pressure direction marker to generate a pitting corrosion evidence score. The following describes gradient matching of the score gradient sequence to generate a tightening suggestion sequence. After executing the tightening suggestion sequence, a verification process is performed to generate the pitting corrosion detection results, specifically including: Discrete action matching is performed between the target pressure range and the scoring gradient sequence to generate a tightening suggestion sequence; The target pressure range is generated by filtering and sampling the attached pressure sequence according to a set of reliable windows and taking extreme values. After executing the recommended fastening sequence, the attachment pressure sequence and acoustic waveform sequence are reviewed and recalculated to generate a review pitting corrosion evidence score and a review evidence consistency matrix. The review pitting corrosion evidence score and the review evidence consistency matrix are then compared with the pitting corrosion evidence score and the evidence consistency matrix respectively to perform differential synthesis and discrimination, and pitting corrosion detection results are generated.

[0048] The target pressure range refers to the range of pressure values ​​extracted from the attachment pressure sequence under the current testing conditions that can simultaneously meet the requirements of "pressure morphology stability" and "diagnostic reliability improvement". The review pitting evidence score refers to the comprehensive quantitative result obtained by recalculating pitting-related evidence based on the reviewed attachment pressure sequence and acoustic waveform sequence during the review period formed after the execution of the fastening recommendation sequence, following a data processing link that is completely consistent with the initial pitting evidence score. The verification evidence consistency matrix refers to the comprehensive quantitative result obtained by recalculating the evidence consistency matrix based on the verified attachment pressure sequence and acoustic waveform sequence during the verification period formed after the execution of the tightening recommendation sequence, following a data processing link that is completely consistent with the initial evidence consistency matrix.

[0049] This part has already been described in detail above, so I will not repeat it here.

[0050] This scheme generates a tightening suggestion sequence by performing gradient matching on the scoring gradient sequence, so that the selection of tightening action is constrained by the rate of change of confidence, reducing coupling fluctuations and false triggers caused by blindly tightening multiple times. Discrete action matching is performed on the target pressure range and the scoring gradient sequence to generate a tightening suggestion sequence, so that the action amplitude and number of times are consistent with the pressure deviation and confidence improvement requirements, thereby increasing the probability of entering a stable detection state after tightening. The target pressure range is generated by sampling and taking the extreme values ​​of the attachment pressure sequence according to the set of confidence windows. This ensures that the target range is derived from the stable window pressure statistics, reducing the impact of temperature cycling and transient jumps on the target setting. After executing the fastening recommendation sequence, the attachment pressure sequence and acoustic waveform sequence are reviewed and recalculated to generate a review pitting corrosion evidence score and a review evidence consistency matrix. This ensures that the conclusion is based on the evidence from the recalculation of the same link after fastening, suppressing false trends caused by coupling drift. The review pitting corrosion evidence score and the pitting corrosion evidence score, as well as the review evidence consistency matrix and the evidence consistency matrix, are differentiated and synthesized to generate pitting corrosion detection results. This allows pitting corrosion and coupling artifacts to be distinguished by the difference direction and consistency changes, improving the stability and verifiability of the results.

[0051] Example 2: Please see Figure 5 A gear tooth surface pitting damage acoustic emission detection system, including: The data acquisition module is used to acquire the attachment pressure sequence and acoustic waveform sequence of the target object; The acoustic signal processing module is used to perform morphological consistency judgment on the attachment pressure sequence according to the sliding window, generate a set of reliable windows, and establish hyperbolic mappings of the acoustic waveform sequence from the beginning and end and the attachment pressure sequence of the corresponding reliable window within the set of reliable windows, and generate a hysteresis mapping table. The threshold update module is used to perform bias expansion on the attachment pressure sequence according to the hysteresis mapping table, and to perform boundary discrimination and threshold update on the bias expansion result and the preset baseline ratio to generate a threshold update set. The benchmark generation module is used to perform threshold comparison on the acoustic waveform sequence according to the threshold update set, and to group and count the threshold comparison results and take the extreme number to generate a cycle candidate period. Then, based on the cycle candidate period, the threshold comparison results are extrapolated at equal intervals to generate a cycle benchmark sequence. The function fitting module is used to perform windowing and offset averaging on the rotation reference sequence based on the set of reliable windows, generate the position centroid sequence, and perform piecewise fitting on the position centroid sequence and the attachment pressure sequence to generate the phase offset function. The detection result generation module is used to compensate the rotation reference sequence according to the phase offset function and generate pitting detection results.

[0052] This embodiment has the same technical effects as Embodiment 1.

[0053] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. The data mentioned in this application, when used for calculations, have undergone normalization and other preprocessing to achieve dimensional uniformity.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An acoustic emission detection method for pitting damage on gear tooth surfaces, applied to the acoustic emission detection of pitting damage on gear tooth surfaces inside an external gearbox in a high-temperature kiln, characterized in that... Includes the following steps: Obtain the attachment pressure sequence and acoustic waveform sequence of the target object; The attachment pressure sequence is subjected to morphological consistency discrimination by sliding window to generate a set of reliable windows. Within the set of reliable windows, a hyperbola mapping is established for the acoustic waveform sequence from the beginning and end and the attachment pressure sequence of the corresponding reliable window, and a hysteresis mapping table is generated. The attachment pressure sequence is biased and expanded according to the hysteresis mapping table, and boundary discrimination and threshold update are performed on the bias expansion result and the preset baseline ratio to generate a threshold update set. The acoustic waveform sequence is compared with the threshold update set, and the threshold comparison results are grouped, counted, and the extreme number is taken to generate a cycle candidate period. Then, the threshold comparison results are extrapolated at equal intervals according to the cycle candidate period to generate a cycle reference sequence. The rotation reference sequence is subjected to windowing and offset averaging processing according to the set of reliable windows to generate a position centroid sequence. The position centroid sequence and the attachment pressure sequence are then piecewise fitted to generate a phase offset function. The rotation reference sequence is compensated according to the phase offset function to generate pitting detection results.

2. The acoustic emission detection method for pitting damage on gear tooth surfaces according to claim 1, characterized in that: After generating the trusted window set, the process also includes generating pressure direction identifiers, specifically including: Within the set of trusted windows, the difference between each pair of adjacent attachment pressure sequences is calculated to generate a pressure difference sequence, and the pressure difference sequence is then counted by sign. The sign counting includes counting the number of elements greater than zero in the pressure difference sequence to obtain a positive count and counting the number of elements less than zero to obtain a negative count. The proportions of the positive and negative counts in the non-zero elements of the pressure difference sequence are calculated respectively, and the magnitude relationship of the proportions is determined to generate a pressure direction identifier.

3. The acoustic emission detection method for pitting damage on gear tooth surfaces according to claim 2, characterized in that: Within the set of trusted windows, a hyperbolic mapping is established for the acoustic waveform sequence from the beginning and end of the waveform and the attachment pressure sequence of the corresponding trusted window, respectively, to generate a hysteresis mapping table. Specifically, this includes: Within the set of trustworthy windows, the acoustic waveform sequence is segmented to generate a set of candidate waveform segments. The set of candidate waveform segments is then sorted by amplitude and descending order to generate an energy ranking result. The energy sorting result is compared with the candidate waveform segment set from the beginning and end, respectively, to generate the proportion of high-energy segments and the proportion of low-energy segments. Hyperbolic fitting is performed between the representative pressure value and the proportion of the high-energy segment, and between the representative pressure value and the proportion of the low-energy segment, respectively, to generate a hysteresis mapping table; The pressure representative value is obtained by statistical analysis of the attachment pressure sequence within the set of trusted windows.

4. The acoustic emission detection method for pitting damage on gear tooth surfaces according to claim 3, characterized in that: The attachment pressure sequence is bias-extended according to the hysteresis mapping table, and boundary discrimination and threshold update are performed on the bias-extended result and the preset baseline ratio to generate a threshold update set, specifically including: The pressure representative value corresponding to the attachment pressure sequence is biased and expanded according to the hysteresis mapping table; Perform boundary discrimination on the offset expansion results and the preset baseline ratio, and generate boundary discrimination markers; The preset baseline ratio is obtained by extracting a reference waveform segment from the acoustic waveform sequence within the set of trusted windows, and performing a ratio mapping from the beginning of the energy sorting result based on the reference waveform segment. The current threshold data is updated based on the boundary discrimination marker to generate a threshold update set; wherein the current threshold data is obtained by combining the reference waveform segment corresponding to the previous trusted window and the preset baseline ratio.

5. The acoustic emission detection method for pitting damage on gear tooth surfaces according to claim 1, characterized in that: The process of performing threshold comparisons on the acoustic waveform sequence based on the threshold update set, and then grouping and counting the threshold comparison results and taking the extreme values ​​to generate candidate cycles specifically includes: The acoustic waveform sequence is compared with the threshold update set, and the timing difference is calculated for each pair of adjacent threshold comparison results to generate an interval sequence. The interval sequence is compared and filtered according to the threshold update set to generate a purification interval sequence. The purification interval sequence is then grouped and counted by numerical value, and the purification interval value with the highest occurrence frequency is taken as the candidate cycle for the cycle change.

6. The acoustic emission detection method for pitting damage on gear tooth surfaces according to claim 3, characterized in that: Based on the set of reliable windows, the rotation reference sequence is subjected to windowing and offset averaging processing to generate a position centroid sequence. The position centroid sequence and the attachment pressure sequence are then piecewise fitted to generate a phase offset function, specifically including: The rotation reference sequence is subjected to window subtraction processing according to the confidence window set to generate a rotation offset set, and the mean of the rotation offset set is calculated to generate a position centroid sequence. The pressure representative values ​​are grouped according to their numerical similarity within adjacent confidence windows to generate a pressure segment set. Based on the pressure segment set, the position centroid sequence is segmented and statistically analyzed according to pressure segments, and the mean value is calculated to obtain the average offset of the pressure segment. The pressure representative value and the average offset are then combined to generate a mapping pair. Finally, curve fitting is performed on all mapping pairs to generate a phase offset function.

7. The acoustic emission detection method for pitting damage on gear tooth surfaces according to claim 6, characterized in that: The pitting detection results are generated by compensating the rotation reference sequence according to the phase offset function, specifically including: The rotational reference sequence is compensated and clustered according to the phase offset function to generate a phase band index set. Under the constraint of the phase band index set, the consistency analysis of the acoustic waveform sequence combined with the pressure direction mark is performed to generate a pitting evidence score. The threshold update set, the pitting evidence score, and the phase band index set are weighted, synthesized, and differentially processed to generate a score gradient sequence. Gradient matching is performed on the scoring gradient sequence to generate a fastening suggestion sequence. After executing the fastening suggestion sequence, a verification process is performed to generate pitting detection results.

8. The acoustic emission detection method for pitting damage on gear tooth surfaces according to claim 7, characterized in that: The rotational reference sequence is compensated and clustered according to the phase offset function to generate a phase band index set. Under the constraints of the phase band index set, the acoustic waveform sequence is combined with the pressure direction mark for consistency analysis to generate a pitting evidence score, specifically including: The threshold comparison result is subtracted from the cycle reference sequence to generate an event cycle offset sequence. The event cycle offset sequence is then compensated and clustered according to the phase offset function to generate a phase band index set. The attachment pressure sequence is normalized according to the hysteresis mapping table to generate proportional weights. Under the constraint of the phase band index set and threshold comparison results, the energy comparison between the inside and outside bands of the acoustic waveform sequence is calculated and weighted according to the proportional weights. Then, the evidence consistency matrix is ​​generated by combining the pressure direction mark. The consistency degree of the evidence consistency matrix is ​​calculated to generate a pitting erosion evidence score.

9. The acoustic emission detection method for pitting damage on gear tooth surfaces according to claim 8, characterized in that: Gradient matching is performed on the scoring gradient sequence to generate a fastening suggestion sequence. After executing the fastening suggestion sequence, a verification process is performed to generate pitting detection results, specifically including: Discrete action matching is performed between the target pressure range and the scoring gradient sequence to generate a tightening suggestion sequence; The target pressure range is generated by filtering and sampling the attachment pressure sequence according to the set of confidence windows and taking extreme values. After executing the fastening recommendation sequence, the attachment pressure sequence and the acoustic waveform sequence are reviewed and recalculated to generate a review pitting corrosion evidence score and a review evidence consistency matrix. The review pitting corrosion evidence score and the review evidence consistency matrix are then compared with the pitting corrosion evidence score and the evidence consistency matrix respectively to perform differential synthesis discrimination and generate pitting corrosion detection results.

10. An acoustic emission detection system for pitting damage on gear tooth surfaces, characterized in that, include: The data acquisition module is used to acquire the attachment pressure sequence and acoustic waveform sequence of the target object; The acoustic signal processing module is used to perform morphological consistency judgment on the attachment pressure sequence according to the sliding window, generate a set of reliable windows, and establish a hyperbola mapping for the acoustic waveform sequence from the beginning and end and the attachment pressure sequence of the corresponding reliable window within the set of reliable windows, and generate a hysteresis mapping table. The threshold update module is used to perform bias expansion on the attachment pressure sequence according to the hysteresis mapping table, and perform boundary discrimination and threshold update on the bias expansion result and the preset baseline ratio to generate a threshold update set. The benchmark generation module is used to perform threshold comparison on the acoustic waveform sequence according to the threshold update set, and to group and count the threshold comparison results and take the extreme number to generate a cycle candidate period. Then, based on the cycle candidate period, the threshold comparison results are extrapolated at equal intervals to generate a cycle benchmark sequence. The function fitting module is used to perform windowing and offset averaging processing on the rotation reference sequence according to the set of reliable windows to generate a position centroid sequence, and to perform piecewise fitting on the position centroid sequence and the attachment pressure sequence to generate a phase offset function. The detection result generation module is used to compensate the rotation reference sequence according to the phase offset function to generate pitting detection results.

Images