Complex rolling bearing fault feature extraction method based on adaptive filtering

Abstract

The invention discloses a complex rolling bearing fault feature extraction method based on adaptive filtering, and relates to the technical field of pre-diagnosis and health management, and the method specifically comprises the following steps: obtaining an impact response signal, extracting a wave crest paragraph trend, recognizing an amplitude trend, constructing a trend sequence, extracting a repetitive sequence position and a paragraph number, and obtaining a pre-diagnosis result; and extracting an envelope path and angle change to construct a periodic structure, analyzing a gradient inversion positioning continuous change section, and extracting a connection fragment to participate in filtering to form a continuous response characteristic waveform group. According to the method, repeated rhythm change is identified by extracting wave peak paragraph amplitude trend, an impact signal time sequence contour is continuously tracked, a structure change correlation area is constructed by combining a path trend turning point and a waveform gradient inversion direction, and continuous segments are extracted by using a waveform sequence and a connection position. And the fluctuation characteristics of the impact response in the differentiated time period are reserved, the stability and continuity of the characteristic sequence are enhanced, and the identification capability of the impact change response and the overall continuity of the periodic structure are improved.

Description

Technical Field

[0001] This invention relates to the field of pre-diagnosis and health management technology, and in particular to a method for extracting fault features of complex rolling bearings based on adaptive filtering. Background Technology

[0002] The field of pre-diagnosis and health management technology involves real-time monitoring and fault identification of the operating status of mechanical equipment. Core aspects include acquiring equipment vibration response signals, extracting characteristic indicators representing faults, constructing health assessment models, predicting fault evolution trends, and providing a basis for maintenance decisions. This technology typically employs methods such as vibration signal acquisition, feature quantity construction, time-frequency analysis and processing, and model recognition calculations for fault perception and status assessment. It is particularly suitable for operating environments with complex conditions and drastic speed changes, such as wind turbine gearboxes, aircraft bearings, and locomotive bogies. Its methodological approach mainly involves acquiring characteristic information representing the type and severity of faults through time-domain waveform recognition, frequency-domain frequency recognition, or angle-domain period recognition, based on multi-sensor acquisition, and establishing fault judgment criteria through a fixed computational structure. Traditional methods for extracting fault features of complex rolling bearings involve collecting vibration signals, constructing an objective function based on a kurtosis evaluation function, deriving a fixed-coefficient wavelet filter to amplify the fault impact signal and suppress background noise, identifying fault pulse components through periodic analysis of the signal waveform, or extracting frequency domain feature parameters using spectral envelope demodulation. Some methods further convert the signal to the angle domain to establish a periodic mean model to enhance the ability to identify feature sequences.

[0003] Existing technologies have limitations in resolving complex impact signal structures. In particular, when the vibration waveform changes continuously within the sliding window or the impact rhythm fluctuates violently, relying on fixed filtering parameters to enhance the periodic component can easily lead to the omission of effective response information. During waveform periodic analysis, due to limitations in window selection and insufficient accuracy of peak points, problems such as blurred identification of impact segments or loss of feature sequences often occur. In addition, spectral envelope demodulation is highly sensitive to spurious disturbances and easily introduces non-target frequency interference, resulting in signal structure distortion. In scenarios with large speed fluctuations or frequent fluctuations, it affects the consistency tracking of fault performance and the accurate reconstruction of periodic responses. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a method for extracting fault features of complex rolling bearings based on adaptive filtering. The specific technical solution is as follows: A method for extracting fault features of complex rolling bearings based on adaptive filtering includes the following steps: S1: Obtain the rolling bearing impact response signal sequence, extract the amplitude of the rising segment between peaks, identify peak segments with consistent amplitude trends within the sliding time window, extract the corresponding start and end indices and amplitudes, and obtain the impact response trend identification sequence. S2: Based on the impact response trend identification sequence, extract the peak and trough positions of each segment, determine whether the time interval and amplitude order are repeated, extract the repeated sequence positions and segment numbers, and obtain the impact rhythm repetition calibration group. S3: Based on the index and segment number in the repeated calibration group of the impact rhythm, extract the upper envelope and lower envelope paths, identify the turning areas of the path trend, extract the sequence of continuously changing angle points, and obtain the set of periodic structure paths; S4: Based on the trend segments of the periodic structure path set, extract the gradient trend of the previous and subsequent waveforms, determine whether the gradient direction is reversed, track the position of continuous change of direction, and obtain the gradient symmetric associated window group. S5: Based on the gradient symmetric correlation window group, extract the waveform window within the period, read the amplitude fluctuation sequence, track the connection position of the repeating sequence in the continuous segment, extract the waveform segment to participate in filtering, and obtain the filtering response feature waveform group.

[0005] As a further embodiment of the present invention, the impact response trend identification sequence includes peak segment start and end indexes, corresponding amplitude data within the segment, and identification information indicating that the peak direction remains consistent. The impact rhythm repetition calibration group includes repeated peak position numbers, corresponding segment numbers, and arrangement characteristics of the repetition sequence. The periodic structure path set includes edge paths formed by the upper and lower envelopes, turning points of the edge paths, and changes in waveform angle trends. The gradient symmetry association window group includes waveform dominant trend segments, rising and falling gradient trend pairs, and the time position of the direction reversal point. The filtered response feature waveform group includes extended waveform segments, waveform sequences with consistent repetition counts in the fluctuation sequence, and the connection range between continuous segments.

[0006] As a further aspect of the present invention, the wave peak segment refers to the extraction of the rising segment amplitude sequence between adjacent wave peaks in the rolling bearing impact response signal, the determination of direction by the amplitude change within the sliding time window, and the identification of wave peak segments that rise continuously in the same direction. The path trend turning area refers to the set of boundary points on the time axis that identify changes in direction during continuous movement of the edge path formed by the extracted upper and lower envelopes.

[0007] As a further aspect of the present invention, the continuously changing angle point sequence refers to analyzing the local waveform trend angle of each turning point within the path trend turning area, and extracting continuous point segments with angle characteristics such as extensibility, consistent direction, or angle difference less than a preset threshold. The waveform window within a period refers to extracting continuous waveform segments within a period by combining the path structure information provided by the gradient symmetric associated window group, and extending the time range based on the repeatability and connectivity of the amplitude direction sequence.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Obtain the rolling bearing impact response signal sequence, extract the rising segment amplitude group between adjacent peaks, arrange the amplitude points according to time, determine whether the direction of increase or decrease of adjacent amplitudes is continuous and positive, extract the continuous rising segment amplitude, and obtain the continuous increasing segment amplitude sequence. S102: Based on the continuously increasing amplitude sequence, divide the sliding time window, retrieve the peak point sequence within each window, compare the amplitude direction, identify peak segments with the same direction, extract the start and end index positions of the segments, and obtain the segment index group with the same direction. S103: Based on the consistent direction segment index group, extract the corresponding segment amplitude data from the original signal, exclude the direction change segments, extract the start and end indexes and amplitude content, and obtain the impact response trend identification sequence.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the peak start and end index information in the impact response trend identification sequence, analyze the corresponding index of the amplitude value of the peak point and the trough point in each segment in the sequence, and connect the adjacent peak and trough point pairs in chronological order to obtain a set of amplitude position point pairs. S202: Based on the time interval between point pairs and the direction of amplitude change in the set of amplitude position point pairs, extract the sequential information between point pairs, compare the arrangement order in each paragraph, extract the repeating sequence number sequence, and obtain the repeating sorting index set; S203: Based on the peak index information in the repeating sorting index set, extract the corresponding paragraph number, allocate row numbers according to the correspondence between the peak point index and the paragraph number, and obtain the impact rhythm repeating calibration group.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the peak index and segment number in the impact rhythm repetition calibration group, retrieve the upper and lower envelope curves of the corresponding segment, analyze the edge trajectory between the envelopes, trace the continuous direction on the time axis, and obtain the edge change path sequence. S302: Based on the trajectory trend in the edge change path sequence, identify the locations where the waveform contour shifts in direction in adjacent positions, filter the boundary points where the direction changes in a continuous area, and obtain the edge turning position set; S303: Based on the trajectory nodes in the set of edge turning positions, retrieve the edge trajectory within the corresponding time period, define the path extension range according to the trend continuity between nodes, divide the continuous trend into path segments, and obtain a periodic structure path set.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the dominant trend segment of the waveform in the set of periodic structure paths, extract the time series and amplitude series of the corresponding waveforms before and after the segment, and separate the continuous rising segment and the continuous falling segment by the amplitude change trend to obtain the waveform gradient trend pair sequence. S402: Based on the waveform gradient trend pair sequence, by comparing the slope direction of adjacent positions in the rising and falling segments, the time period of the reverse trend is analyzed, the positions of continuous direction change behavior are screened, and the gradient direction turning point group is obtained. S403: Based on the time index in the gradient direction turning point group, track the angular direction continuity between trends, and extract continuous point segments with extension behavior in direction switching to obtain a gradient symmetric association window group.

[0012] As a further aspect of the present invention, the extraction process of the waveform dominant trend segment based on the periodic structure path set is specifically as follows: by continuously judging the amplitude change direction corresponding to each waveform segment in the periodic structure path set, and drawing out continuous segments based on the continuous and consistent behavior of the amplitude change direction between adjacent sampling points, the waveform dominant trend segment is obtained. In the process of decomposing the amplitude change trend, the direction of amplitude change between adjacent sampling points is analyzed from the starting position. The amplitude change segment that continues to rise is classified as an upward trend, the change segment that continues to fall is classified as a downward trend, and the point where there is a change in direction between adjacent trends is marked as the trend turning point. The process of comparing slope direction is as follows: extract two boundary points at the junction of continuous upward and downward trends, determine the continuity of amplitude trend between the previous and subsequent positions, and divide the start and end intervals of turning behavior in the continuous segment of direction switching. In the process of identifying the time period of the reverse trend, the gradient trend is traversed to analyze the trend continuation direction between each trend segment and the adjacent segment in the sequence, and the continuity in time is analyzed according to the switching state of the adjacent trend direction, and the area of ​​trend direction change is circled on the time axis. The process of extracting continuous point segments with extended behavior for direction switching is as follows: the consistency of trend direction is judged for the front and back positions of the identified trend direction change area; the trend state that maintains the same direction within the continuous point segment is extended along the trend direction and defined as the extended direction point segment; and gradient symmetric association window groups are summarized and extracted from the extended point segment.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the path structure index information in the gradient symmetric associated window group, retrieve the waveform window segments within the same period segment, read the amplitude sequence of continuous sampling points in each segment, and extract the change direction between adjacent amplitudes according to the time order to obtain the amplitude direction sequence group; S502: Based on the change direction characteristics in the amplitude direction sequence group, the corresponding recurrence position of the wave direction change sequence in each segment is identified, and the connection range before and after the continuous part of the time index in the sequence is expanded to obtain the sequence extension segment set; S503: Based on the start and end time indices of each group in the sequence extension segment set, extract the periodic waveform data, extract the original sampling point sequence according to the index range corresponding to the segment, and splice each segment of waveform data in order to obtain the filter response feature waveform group.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the amplitude trend of peak segments is extracted to identify repetitive rhythmic changes, the temporal contour of the impact signal is continuously tracked, and the structural change association region is constructed by combining the turning point of the path trend and the direction of waveform gradient reversal. Continuous segments are extracted by using waveform sequence and connection position, the fluctuation characteristics of the impact response in different time periods are preserved, the stability and persistence of the feature sequence are enhanced under the background of non-stationary disturbance, and the ability to identify impact change response and the overall coherence of periodic structure are improved. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a method for extracting fault features of complex rolling bearings based on adaptive filtering, comprising the following steps: S1: Obtain the rolling bearing impact response signal sequence, extract the amplitude of the rising segment between peaks, identify peak segments with consistent amplitude trends within the sliding time window, extract the corresponding start and end indices and amplitudes, and obtain the impact response trend identification sequence.

[0023] Specifically, the rolling bearing impact response signal sequence is obtained, the rising segment amplitude group between adjacent peak points is extracted, the direction of the peak amplitude change sequence within each sliding time window is determined, the fluctuation direction between adjacent amplitudes in the rising segment is compared, the peak segments with consistent directions are identified, the start and end indices and corresponding amplitude data in the segments are extracted, and the impact response trend identification sequence is obtained.

[0024] S2: Based on the shock response trend identification sequence, extract the peak and trough positions of each segment, determine whether the time interval and amplitude order are repeated, extract the position and segment number of the repeated sequence, and obtain the shock rhythm repetition calibration group.

[0025] Specifically, based on the peak start and end index information in the impact response trend identification sequence, the amplitude position point group of the peak and trough in each segment is extracted, the time interval and amplitude change order between adjacent point pairs are compared, it is determined whether the same arrangement order appears repeatedly in the segment, and the peak position number and segment number of the repeated arrangement are filtered to obtain the impact rhythm repetition calibration group.

[0026] S3: Based on the index and segment number in the repeated calibration group of the impact rhythm, extract the upper and lower envelope paths, identify the turning areas of the path trend, extract the sequence of continuously changing angle points, and obtain the set of periodic structure paths.

[0027] Specifically, based on the peak index and segment number in the repeated calibration group of the impact rhythm, the edge change path formed by the upper and lower envelopes in the corresponding segment is extracted, the position segment where the edge path trend turns is identified, it is determined whether the waveform angle trend in the path changes continuously in adjacent positions, the turning point sequence with turning characteristics is extracted, and the periodic structure path set is obtained.

[0028] S4: Based on the trend segments of the periodic structure path set, extract the gradient trend of the previous and subsequent waveforms, determine whether the gradient direction is reversed, track the position of continuous change of direction, and obtain the gradient symmetric association window group.

[0029] Specifically, based on the dominant trend segment of the waveform in the periodic structure path set, the waveform gradient trend pairs of the two segments are called to determine the degree of gradient reversal between the rising and falling segments. The time positions of the abrupt change points in the two directions are compared, and the point sequence with direction reversal and change persistence is extracted to obtain the gradient symmetric association window group.

[0030] S5: Based on gradient symmetric correlation window group, extract waveform windows within the period, read the amplitude fluctuation sequence, track the connection position of the repeating sequence in the continuous segment, extract waveform segments to participate in filtering, and obtain the filtering response feature waveform group.

[0031] Specifically, based on the path structure index information in the gradient symmetric associated window group, waveform window segments within the same period segment are extracted, the amplitude fluctuation sequence within each window segment is read, the position of the waveform sequence with the same number of repetitions in the corresponding fluctuation sequence is determined, the connection range of the sequence between consecutive segments is extended, and the extended waveform segments are extracted to obtain the filter response feature waveform group.

[0032] The impact response trend identification sequence includes the peak segment start and end index, the corresponding amplitude data within the segment, and the identification information that the peak direction remains consistent. The impact rhythm repetition calibration group includes the repeated peak position number, the corresponding segment number, and the arrangement characteristics of the repetition sequence. The periodic structure path set includes the edge path formed by the upper and lower envelopes, the turning position of the edge path, and the changing behavior of the waveform angle trend. The gradient symmetry association window group includes the waveform dominant trend segment, the rising and falling gradient trend pairs, and the time position of the direction reversal point. The filter response feature waveform group includes the extended waveform segment, the waveform sequence with the same number of repetitions in the fluctuation sequence, and the connection range between continuous segments.

[0033] Please see Figure 2 The specific steps of S1 are as follows: S101: Obtain the rolling bearing impact response signal sequence, extract the rising segment amplitude group between adjacent peaks, arrange the amplitude points according to time, determine whether the direction of increase or decrease of adjacent amplitudes is continuous and positive, extract the continuous rising segment amplitude, and obtain the continuous increasing segment amplitude sequence. First, the original vibration signal should be obtained from a vibration sensor installed on the outside of the shaft or bearing housing. This signal, in its physical environment, manifests as non-stationary fluctuations caused by defects, fatigue, or impacts during bearing operation. The signal is read at fixed sampling intervals, with each amplitude corresponding to a sampling point number. In the original signal, the data sequence between adjacent peaks is extracted using the index sequence to define the rising segment range. Within this segment, the amplitude corresponding to all sampling points is retrieved, and amplitude data is extracted point-by-point sequentially to form a rising segment amplitude set. The extraction of each set maintains the same point order as the source signal. For example, if a peak pair corresponds to numbers 126 and 134 in the signal, then a total of 9 sampling points within this interval are extracted. The amplitude values ​​are 1.2, 1.3, 1.5, 1.8, 2.1, 2.3, 2.5, 2.6, and 2.7, respectively. After obtaining the amplitude sets of each rising segment, the continuity of each amplitude point is judged. That is, the numerical difference between the current value and the next value in the amplitude sequence is compared one by one to determine whether they are in a positive growth relationship. If any pair of adjacent amplitude values ​​in a segment shows a non-positive growth trend, it is considered a non-continuous rising segment. Otherwise, the segment is retained as a valid segment. For example, if the comparison result of each pair of values ​​in the aforementioned amplitude set is positive, then the segment is a continuous growth segment. If there is a reverse change pair of 1.5 and 1.3 in another segment, then the segment is not retained. Then, all the amplitude sets of continuous growth segments are included in the output as input data for subsequent analysis to obtain the amplitude sequence of continuous growth segments.

[0034] S102: Based on the amplitude sequence of continuously growing segments, divide the sliding time window, retrieve the peak point sequence within each window, compare the amplitude direction, identify peak segments with the same direction, extract the start and end index positions of the segments, and obtain the index group of segments with the same direction. First, define the time coverage range of each continuously increasing amplitude segment. Based on this, define equal-width sliding time windows, for example, each window covering 30 sampling points, with the window sliding in 10-point increments. Within each time window, number the amplitude sequence within the current coverage range and extract its corresponding peak index, i.e., extract the sampling point position corresponding to the local maximum value from the amplitude points. Then, form a set of peak sequence within each time window. Next, compare the increase / decrease direction of the amplitude in this sequence in chronological order. For every two adjacent peaks, extract their amplitude values ​​and perform a difference calculation, recording their positive and negative directions of change. Finally, count whether there are continuous peaks within the sequence. For sequences with the same direction, such as peak values ​​of 1.5, 1.8, 2.0, and 2.3, the direction is positive. If there are changes in the middle, such as 1.5, 1.8, 1.6, and 2.2, the direction alternates between positive and negative. Only segments with the same direction at the beginning and end need to be retained. Based on the sampling numbers of the start and end peak points in the original signal, the time index range is recorded. If the peak points of a continuous direction are numbered 25, 28, 31, 34, and 37, the time range from 25 to 37 is extracted as the start and end index of the segment. After all sliding window scans are completed, the index set of all segments with the same direction is output to obtain the segment index group with the same direction.

[0035] S103: Based on the consistent direction segment index group, extract the corresponding segment amplitude data from the original signal, exclude the direction change segments, extract the start and end indexes and amplitude content, and obtain the impact response trend identification sequence. First, for the start and end positions of each segment in the group, retrieve the amplitude value groups within the corresponding range in the original signal sequence. Extract all data point values ​​within the coverage area of ​​each segment by traversing the sequence. During this process, compare the monotonicity trend of the amplitude sequence within each segment. If there is amplitude reversal in the sequence, for example, if the previous data point is 1.6 and the next data point is 1.3, indicating an inconsistency in direction, then the segment is determined to contain a direction change portion. During execution, such segments need to be removed from subsequent processing. In practice, the sign of consecutive differences in the sequence can be judged. Consecutive identical signs can be used to determine whether the direction remains consistent within a segment. For example, in... Within the segment index range of 120 to 150, the amplitude sequence is 1.2, 1.4, 1.5, 1.3, and 1.6. If the direction consistency is interrupted due to a downward trend from 1.5 to 1.3, the current segment is excluded. The segment index range where the direction is always positive or negative is retained. After confirming the direction consistency, the complete amplitude data of the segment in the original signal is extracted again to form a continuous data segment without direction reversal. At the same time, its start and end index numbers and corresponding amplitude sequences are recorded. The entire consistent direction segment index group is processed in this way, and finally the start and end sampling point numbers and corresponding data content of each segment are output to obtain the impact response trend identification sequence.

[0036] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the peak start and end index information in the impact response trend identification sequence, analyze the corresponding index of the amplitude value of the peak point and the trough point in each segment in the sequence, and connect the adjacent peak and trough point pairs in chronological order to obtain the set of amplitude position point pairs. First, retrieve the start and end positions of the peaks for each data segment from the sequence. For example, extract the data segment with start index 85 and end index 97. Then, based on these start and end indexes, determine the amplitude sequence of all points within this data segment. Identify the peaks and troughs within this amplitude sequence. A peak can be defined as the point with the largest value among adjacent points, and a trough as the point with the smallest value among adjacent points. After identification, read the index numbers of the peaks and adjacent troughs one by one. For example, if the peak is at index 88 and the trough is at index 91, it indicates a decreasing trend in amplitude within this data segment. Then continue... The next pair of peaks and troughs is searched in the subsequent index of this data segment. During the process, it is necessary to ensure that there are no other larger amplitude interference values ​​between the two points. For the process of judging the direction of amplitude change, the direction change can be confirmed by comparing the amplitude values ​​before and after, and the point pairs whose direction changes do not conform to the continuous up and down fluctuation relationship are eliminated. After the effective connection of peaks and troughs is completed, they are connected according to the time sequence of each pair of peak and trough points in the original sequence to form a set of corresponding relationships. For example, the amplitude point pairs with indices 88 to 91 are connected, followed by the point pairs with indices 94 to 97. No sorting or arrangement transformation is performed to ensure the consistency of their time sequence, and the set of amplitude position point pairs is obtained.

[0037] S202: Based on the time interval between point pairs and the direction of amplitude change in the set of amplitude position point pairs, extract the sequential information between point pairs, compare the arrangement order in each paragraph, extract the repeating sequence number sequence, and obtain the repeating sorting index set; First, each pair of peaks and troughs in the set is decomposed into two basic parameters: the positional interval between the two points in the sequence, and the sign and direction of the amplitude difference from peak to trough or from trough to peak. For example, if the first point in a pair is a peak at index 88 with an amplitude of 2.1, and the second point is a trough at index 91 with an amplitude of 1.2, then the interval is 3 and the direction is downward. Then, all the point pairs are encoded sequentially according to their order of appearance in the original sequence. The encoding format can be set by first concatenating the time interval and direction into a set of feature values, and then assigning sequential numbers to each set of feature values. For example, the first set is 3D (representing an interval of 3 and a downward direction), and the second set is 5U (representing an interval of 5 and an upward direction). The obtained encoding sequence is 3D, 5U, 4D, etc. Then, according to the preset window division method, the encoding sequence is read in segments. The encoding order in each segment is compared item by item to identify the situation where the same encoding order appears in consecutive segments. For example, if one window segment is 3D, 5U, 4D, and another window segment is also 3D, 5U, 4D, it is determined to be a repeating sequence. By comparing the correspondence between the starting position index and the number sequence index, the starting position index values ​​of the first and second occurrences of the repeating sequence are extracted and merged into a unified sequence. The process continues to traverse all possible repetitions in subsequent segments to avoid missing any repeating segments. Finally, the index numbers corresponding to the positions of all repeating sequences are extracted to obtain the repeating sorting index set.

[0038] S203: Based on the peak index information in the repeating sorted index set, extract the corresponding paragraph number, allocate row number according to the correspondence between peak point index and paragraph number, and obtain the impact rhythm repeat calibration group; First, the repeating sorted index set is traversed. Each index corresponds to a peak point identified as repeating in the original amplitude sequence. Using the start and end segment range information established in the previous steps, the complete segment in which the peak point is located is retrieved. A one-to-one mapping relationship is established between each peak index and its corresponding segment number. For example, if the peak point with index 112 is identified as part of the repeating pattern and is in the periodic segment with segment number 6, it is mapped to "112-6". All "index-segment number" pairs are then collected in a numbered table. Next, all segment numbers in the mapping relationship are called to generate a segment partitioning index sequence, according to the order of the peak points in the original signal. The data is assigned row numbers sequentially. For example, if peak indices 112, 134, and 156 appear sequentially in the original sequence and correspond to segment numbers 6, 7, and 8 respectively, then row numbers 1, 2, and 3 are assigned to the three sets of data. These are then nested into the annotation result set of the "index-segment number-row number" correspondence to form a visual index table. If multiple peak indices fall into the same segment, their numbers are updated incrementally according to their order of appearance in the original sequence. For example, if indices 112 and 117 both fall into segment number 6, then they are assigned numbers 1 and 2 to distinguish multiple repetitive feature points within the segment. Finally, all segment numbers, indices, and row numbers are integrated to obtain the impact rhythm repetition calibration group.

[0039] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the peak index and segment number in the repeated calibration group of the impact rhythm, retrieve the upper and lower envelope curves of the corresponding segment, analyze the edge trajectory between the envelopes, trace the continuous direction on the time axis, and obtain the edge change path sequence. First, signal data within the corresponding segment interval is retrieved, and boundary data of upper and lower envelope curves are extracted. Equal-length waveform segments are then cut from each envelope using the peak point as a positioning reference, forming corresponding upper and lower amplitude trajectories. Subsequently, the data is horizontally expanded along the time axis, and the amplitude difference between the upper and lower envelopes is calculated for each time point, with the direction of boundary change marked. The continuous trend of the edge trajectory is extracted, and all points with the same direction of change are linked sequentially in time to form a series of trajectory segments with consistent directions between adjacent points. Then, the endpoints of each trajectory segment are direction-determined, and the points where the direction changes are identified as edge trend change nodes. In actual operation... During processing, if a segment of the envelope contains continuous peaks and the change amplitude of the upper envelope is always higher than that of the lower envelope, it is determined to be a positive trajectory segment; otherwise, it is a negative trajectory segment. By synchronously processing the envelope curves of multiple segments, comparing their trajectory segment change trends, extracting trajectory segments that change synchronously in multiple periodic segments, and sequentially splicing them along the time dimension to obtain a complete change path chain. For example, if the continuously tracked edge trajectory maintains a similar upward trend in each segment between the 6th and 8th segments, then this continuous trend is written as a path segment into the edge trajectory sequence. Finally, the continuous tracking operation of multiple trajectory segments is completed to obtain the edge change path sequence.

[0040] S302: Based on the trajectory trend in the edge change path sequence, identify the location points where the waveform contour shifts in direction in adjacent positions, filter the boundary points where the direction changes in continuous areas, and obtain the edge turning position set; First, the time sequence of each trajectory is traversed, and the amplitude offset trend between the two consecutive points and the center point in each group is determined using a three-point sliding method. If the trends on both sides are different, the center point is marked as a direction reversal point. In this process, a minimum amplitude threshold for judging direction changes needs to be set to avoid misjudgments due to noise. For example, in a certain trajectory segment, if the amplitude increments of the two consecutive points are +0.04 and -0.05 respectively, and the amplitude change of the center point is 0, then this point is considered a direction reversal point. Subsequently, such change points are selected from the entire trajectory sequence as preliminary direction reversal markers. Further identification is made as to whether these points are within a continuous interval on the time axis. If a group of reversal points all appear within a certain time period and their interval is less than the set maximum continuity interval, then it is determined that there is a stable region. A defined direction switching trend is established, and the corresponding point group is included in the continuous change region. At the same time, in order to eliminate false direction reversals caused by short-term oscillations, point series with frequent fluctuations in amplitude direction within five consecutive points are directly eliminated. Only segments with consistent amplitude direction and minimum continuous length requirements are retained. In actual processing, for example, if an amplitude trend of "increase-decrease-increase-decrease-increase" is observed between points 18 and 24 in the edge trajectory, the segment is eliminated due to repeated direction changes. However, if points 30 to 36 continuously show an "increase-increase-decrease-decrease-decrease" trend, then point 32 in this segment is retained as the core node of direction offset. Finally, the direction trend evaluation of all trajectories is completed in the above manner, and key points with significant boundary turning within each segment are screened out to obtain the edge turning position set.

[0041] S303: Based on the trajectory nodes in the edge turning position set, retrieve the edge trajectory within the corresponding time period, define the path extension range according to the trend continuity between nodes, divide the continuous trend into path segments, and obtain the periodic structure path set; First, retrieve the complete curve data for the corresponding time period from the edge trajectory sequence. Using each pair of adjacent turning nodes as the beginning and end endpoints of the reference interval, extract the edge amplitude sequence contained within it along the time series direction. Within the sequence, sequentially read the amplitude change direction of each continuous point. By comparing the increasing and decreasing trends between adjacent points, identify segments with the same continuous direction. Waveform segments with the same continuous direction of change are considered as a trend segment. In a real-world scenario, if the edge trajectory from point 45 to point 53 shows a continuously increasing state, this segment is recorded as a complete trend segment. Then, continue detecting several trend segments. Multiple continuous trajectory segments with the same direction can usually be formed within a turning node sequence, completing the above... After the above operation, for each trajectory node sequence, its extensibility is determined by whether the connection between the preceding and following trend segments is continuous. When the time interval between the starting point of the following trend segment and the end of the preceding trend segment is lower than the preset interval threshold, it is considered that its trend has continuity, and thus it is classified into a unified segment for processing. In actual operation, for example, when the interval between points 70 and 77 is a decreasing segment, while the interval between points 78 and 85 is still decreasing, and the time interval between the two segments is less than 2 sampling points, then the two segments are spliced ​​into a complete trajectory segment. Finally, in this way, the edge path segments with continuity and directional consistency between each trajectory node are sorted out, and all path segment numbers are merged into a set to obtain a periodic structure path set.

[0042] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the dominant trend segment of the waveform in the periodic structure path set, extract the time series and amplitude series of the corresponding waveforms before and after the segment, and separate the continuous rising segment and the continuous falling segment by the amplitude change trend to obtain the waveform gradient trend pair sequence. First, locate the start and end points of each dominant trend segment from the path set using the index. Then, retrieve 5 to 10 points before and after each segment to form a complete reference segment. Within this segment, sequentially read the time series and amplitude series of the corresponding waveforms. Set the traversal step size to 1, and read adjacent amplitude values ​​in the time series order. Compare the value of the latter with the former; if the latter is greater than the former, it is considered an upward trend; otherwise, it is a downward trend. Traverse the entire sequence in this way, marking the direction of amplitude change in each segment. Then, segment the continuous sequence of points with consistent directions based on the breakpoints of direction changes, forming continuous upward and downward segments. Record the start and end indexes at each segment boundary. For example, the amplitude between points 20 and 28... If the amplitude continues to increase, the segment is classified as a continuous upward segment. If the amplitude continuously decreases between points 29 and 35, it is classified as a continuous downward segment. In practical applications, for example, if the dominant trend segment is indexed from points 45 to 65, and after expansion, the complete reference sequence is points 40 to 70, with amplitude data of 1.2, 1.5, 1.9, 2.3, 2.6, 2.4, 2.1, 1.9, 1.6, 1.2, and 1.0 respectively, then points 40 to 44 are the upward segment, points 45 to 50 are the downward segment, points 51 to 54 are the new upward segment, and points 55 to 58 are the downward segment again. Through this trend breakpoint division method based on amplitude changes, multiple upward or downward sequence segments are formed, ultimately resulting in a waveform gradient trend pair sequence.

[0043] S402: Based on the waveform gradient trend pair sequence, by comparing the slope direction of adjacent positions in the rising and falling segments, the time period of the reverse trend is analyzed, the position of continuous direction change behavior is screened, and the gradient direction turning point group is obtained. First, read the amplitude value and time index between every two adjacent points in the rising segment, calculate the amplitude difference and index difference between adjacent points, and then use this amplitude difference and index difference as a slope index to determine the direction. After processing all point pairs in the rising segment, a complete rising segment slope direction sequence is obtained. Similarly, extract the amplitude difference and index difference between adjacent points in the falling segment as the corresponding slope direction sequence. Then, connect the end point of the rising segment with the start point of the falling segment and determine whether the slope direction at the junction of the two segments has switched from positive to negative. When the slope of the end point of the rising segment is positive and the slope of the starting point of the immediately following falling segment is negative, it is marked as a direction reversal point. This process is repeated in the continuous waveform. The judgment operation screens all connection positions that satisfy the reverse change, thereby capturing the specific location where the direction change behavior occurs in the continuous trend. For example, in the waveform gradient trend, if a certain upward segment from index 21 to index 30 has a continuously positive slope direction, while the immediately following downward segment from index 31 to index 38 has a continuously negative slope direction, then the point between index 30 and index 31 is a direction switching point. Continuing to process the subsequent upward and downward pairs in the same way, multiple such direction reversal connection points can be identified. Finally, all points where the direction has changed significantly are output according to their index values ​​in the original waveform sequence, resulting in a gradient direction turning point group.

[0044] S403: Based on the time index in the gradient direction turning point group, track the angle and direction continuation between trends, and extract continuous point segments with extension behavior of direction switching to obtain gradient symmetric association window group. First, select the waveform position point corresponding to each time index. Take a fixed number of data points before and after each point to form a local waveform segment. Then, calculate the amplitude difference and lateral spacing between adjacent point pairs within the local segment. Determine the trend direction based on the amplitude and spacing changes of each group. Convert the slope of point pairs in the same direction into angle values ​​and continue tracking to identify whether the angle direction is continuously extended. When the continuous angle value changes are small and the direction is consistent, mark the corresponding point segment as the trend continuation area. Further select the turning point as the starting point to trace forward to the end point of the previous trend segment and backward to the starting point of the next trend segment, forming a set of bidirectional connected trend windows. Then, calculate the value within this window as a unit. The overall trend direction and average angle within a waveform segment are used to form a mirror trend group with two windows that have similar trends but opposite directions and similar absolute angle values. For example, for a point with a turning point index of 124, the preceding segment from index 116 to 123 shows a continuous upward trend with an average angle of 15°, while the following segment from index 125 to 132 shows a continuous downward trend with an average angle of -16°. The angle difference is 31°, and the opposite directions satisfy the symmetrical trend characteristics. The current window combination is then screened out as a group of candidate point segments. By processing all turning points in this way, the final output is a combination of all point segments that exhibit this kind of reverse trend extension behavior, resulting in a gradient symmetrical associated window group.

[0045] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the path structure index information in the gradient symmetric associated window group, retrieve the waveform window segment within the same period segment, read the amplitude sequence of continuous sampling points in each segment, and extract the change direction between adjacent amplitudes according to the time order to obtain the amplitude direction sequence group; First, the path sequence segments indicated in each index are called. The sampling point range corresponding to each path segment is sequentially detected in the continuously sampled waveform data. Then, based on the start and end positions of the path segments, the waveform data within that time interval is extracted into a continuous sequence of amplitude sampling values. Starting from the first amplitude point, the magnitude comparison between the current amplitude and its subsequent amplitude is calculated sequentially. When the subsequent amplitude is greater than the current amplitude, the direction of that pair of sampling points is defined as positive; otherwise, it is defined as negative. If they are equal, it is defined as zero. Subsequently, the same direction identification process is performed on each pair of adjacent points in the entire sampling sequence according to the original timeline order. Finally, these direction information are concatenated to form the amplitude direction sequence of the current segment. For example, if the consecutive amplitude sampling values ​​in a certain segment are 0.2, 0.5, 0.4, 0.6, 0.6, and 0.3, then their amplitude directions are positive, negative, positive, zero, and negative, respectively. Each segment's direction information corresponds one-to-one with the original amplitude sequence, facilitating subsequent overall trend identification of waveform variation patterns. To ensure the integrity of sampling coverage, it is necessary to simultaneously confirm whether the sampling frequency is consistent with the periodic characteristics of the waveform events. If the sampling frequency is higher than the waveform characteristic change rate, the extracted direction information has detailed continuity. If the sampling frequency is too low, appropriate interpolation smoothing processing is required for the changes between amplitude points to repair missing trends. Finally, the amplitude direction sequence extracted from each periodic segment is saved independently, resulting in an amplitude direction sequence group. S502: Based on the change direction characteristics in the amplitude direction sequence group, the corresponding recurrence position of the wave direction change sequence in each segment is identified, and the connection range before and after the continuous part of the time index in the sequence is expanded to obtain the sequence extension segment set; First, the system iterates through the continuous direction data segments in each amplitude direction sequence, scanning the arrangement of positive, negative, and zero values ​​within each segment to identify recurring patterns. For example, segments of the type "positive, positive, negative" or "negative, positive, positive" that appear repeatedly in a direction sequence, appearing more than twice in different segments, are marked with their first and last indexes. After recording the start and end indices within the time series range, the system continues to expand the boundaries forward and backward. By determining whether the direction sequence within the expanded area still maintains the changing direction characteristics of the current segment, if the directional change at a certain boundary continues the previous trend, the expansion continues; if the continuous directional change is interrupted, the expansion operation stops. Assuming that the direction at the starting time point of a repeating segment in the original sequence is positive, and the subsequent two consecutive points are positive and negative, forming a local trend of "positive, positive, negative," the system determines this to be a repeatable pattern and records the first occurrence of this pattern. The three time point indices are set as the start point, midpoint, and end point, respectively. Then, it is checked whether there is a continuous group of points in the same positive direction before the start point. If so, the start point is extended forward, and the process is repeated to determine whether it can be traced back further until the directions are no longer consistent. The direction segment after the end point is processed in the same way. When the expansion range on both sides of the segment is established, the time point indices in the entire range are uniformly organized into a continuous index interval. Then, the remaining sequence is scanned and expanded. Finally, after all the regions where the repeating pattern appears are extracted, multiple sets of time series segments with start and end index information are formed. For example, when the "positive, negative, positive" pattern repeats 3 times in different period segments, and each repetition extends 2 to 3 points before and after, the final possible index segments are [5, 10], [22, 27], and [40, 46]. Each set of index segments retains the time axis order for subsequent waveform processing, and finally, the sequence extension segment set is obtained.

[0046] S503: Based on the start and end time indices of each group in the sequence extension segment set, extract the periodic waveform data, extract the original sampling point sequence according to the index range corresponding to the segment, and splice each segment of waveform data in order to obtain the filter response feature waveform group. First, based on the acquired multiple time index segments, the start and end positions of each segment are traversed and confirmed. The original periodic waveform data is retrieved, and the sampling point range covered by each segment is located. Within each time index range, the amplitude value at the corresponding position in the original data is read point by point. During the reading process, each time index point is used as a pointer to advance sequentially, and the amplitude values ​​of the corresponding sampling points are stored in the original order of appearance. Simultaneously, the start and end numbers and the total number of sampling points for each data segment are recorded for subsequent processing. Assuming a set of index segments is [120, 145], then starting from the 120th point in the original waveform array, data is read sequentially up to the 145th point, extracting a total of 26 data points. For example, if the original waveform is [...1.2, 1.5, 1.7, 1.6, ...], then the sub-waveforms read in this segment are [1.2, 1.5, 1.7, 1.6, ...] up to the 145th position. After completion, the process moves to the next set. For each index segment, the above data reading process is repeated to ensure that the waveform data of each segment is independently retained according to the sampling segment corresponding to its own index. After all index segments have been read, the waveforms of each segment are numbered and assembled in sequence to maintain their relative temporal position in the periodic response. The splicing operation is performed sequentially using a pointer positioning method. No interpolation is performed between each data segment during the splicing process. A direct sequential connection method is used to connect the end of the first segment to the beginning of the second segment without interruption, preserving the amplitude differences and continuous trends shown by the data itself in the waveform. If there are differences in the length of each segment, the number of data points in the overall waveform formed after splicing will vary from segment to segment. For example, if the first segment has 30 points, the second segment has 25 points, and the third segment has 28 points, the splicing result will be a continuous waveform composed of 83 points. Finally, the waveform data corresponding to all index segments are spliced ​​to form a set of periodic response data with representative characteristics, resulting in a set of filtered response characteristic waveforms.

[0047] In this embodiment of the invention, the repeating rhythm changes are identified by extracting the amplitude trend of the peak segment, the temporal contour of the impact signal is continuously tracked, and the structural change association region is constructed by combining the turning point of the path trend and the direction of waveform gradient reversal. Continuous segments are extracted by using waveform sequence and connection position, the fluctuation characteristics of the impact response in different time periods are preserved, the stability and persistence of the feature sequence are enhanced under the background of non-stationary disturbance, and the ability to identify impact change response and the overall coherence of periodic structure are improved.

[0048] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for extracting fault features of complex rolling bearings based on adaptive filtering, characterized in that, Includes the following steps: S1: Obtain the rolling bearing impact response signal sequence, extract the amplitude of the rising segment between peaks, identify peak segments with consistent amplitude trends within the sliding time window, extract the corresponding start and end indices and amplitudes, and obtain the impact response trend identification sequence. S2: Based on the impact response trend identification sequence, extract the peak and trough positions of each segment, determine whether the time interval and amplitude order are repeated, extract the repeated sequence positions and segment numbers, and obtain the impact rhythm repetition calibration group. S3: Based on the index and segment number in the repeated calibration group of the impact rhythm, extract the upper envelope and lower envelope paths, identify the turning areas of the path trend, extract the sequence of continuously changing angle points, and obtain the set of periodic structure paths; S4: Based on the trend segments of the periodic structure path set, extract the gradient trend of the previous and subsequent waveforms, determine whether the gradient direction is reversed, track the position of continuous change of direction, and obtain the gradient symmetric associated window group. S5: Based on the gradient symmetric correlation window group, extract the waveform window within the period, read the amplitude fluctuation sequence, track the connection position of the repeating sequence in the continuous segment, extract the waveform segment to participate in filtering, and obtain the filtering response feature waveform group.

2. The method for extracting complex rolling bearing fault features based on adaptive filtering according to claim 1, characterized in that, The impact response trend identification sequence includes the peak segment start and end index, the corresponding amplitude data within the segment, and the identification information that the peak direction remains consistent. The impact rhythm repetition calibration group includes the repeated peak position number, the corresponding segment number, and the arrangement characteristics of the repetition sequence. The periodic structure path set includes the edge path formed by the upper and lower envelopes, the turning position of the edge path, and the changing behavior of the waveform angle trend. The gradient symmetry association window group includes the waveform dominant trend segment, the rising and falling gradient trend pairs, and the time position of the direction reversal point. The filtered response feature waveform group includes the extended waveform segment, the waveform sequence with the same number of repetitions in the fluctuation sequence, and the connection range between continuous segments.

3. The method for extracting fault features of complex rolling bearings based on adaptive filtering according to claim 1, characterized in that, The term "peak segment" refers to the extraction of the rising amplitude sequence between adjacent peak points in the rolling bearing impact response signal, the determination of direction by amplitude changes within a sliding time window, and the identification of peak segments that rise continuously in the same direction. The path trend turning area refers to the set of boundary points on the time axis that identify changes in direction during continuous movement of the edge path formed by the extracted upper and lower envelopes.

4. The method for extracting fault features of complex rolling bearings based on adaptive filtering according to claim 1, characterized in that, The continuously changing angle point sequence refers to the process of analyzing the local waveform trend angle of each turning point within the path trend turning area and extracting continuous point segments with angle characteristics such as extensibility, consistent direction, or angle difference less than a preset threshold. The waveform window within a period refers to extracting continuous waveform segments within a period by combining the path structure information provided by the gradient symmetric associated window group, and extending the time range based on the repeatability and connectivity of the amplitude direction sequence.

5. The method for extracting fault features of complex rolling bearings based on adaptive filtering according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain the rolling bearing impact response signal sequence, extract the rising segment amplitude group between adjacent peaks, arrange the amplitude points according to time, determine whether the direction of increase or decrease of adjacent amplitudes is continuous and positive, extract the continuous rising segment amplitude, and obtain the continuous increasing segment amplitude sequence. S102: Based on the continuously increasing amplitude sequence, divide the sliding time window, retrieve the peak point sequence within each window, compare the amplitude direction, identify peak segments with the same direction, extract the start and end index positions of the segments, and obtain the segment index group with the same direction. S103: Based on the consistent direction segment index group, extract the corresponding segment amplitude data from the original signal, exclude the direction change segments, extract the start and end indexes and amplitude content, and obtain the impact response trend identification sequence.

6. The method for extracting fault features of complex rolling bearings based on adaptive filtering according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the peak start and end index information in the impact response trend identification sequence, analyze the corresponding index of the amplitude value of the peak point and the trough point in each segment in the sequence, and connect the adjacent peak and trough point pairs in chronological order to obtain a set of amplitude position point pairs. S202: Based on the time interval between point pairs and the direction of amplitude change in the set of amplitude position point pairs, extract the sequential information between point pairs, compare the arrangement order in each paragraph, extract the repeating sequence number sequence, and obtain the repeating sorting index set; S203: Based on the peak index information in the repeating sorting index set, extract the corresponding paragraph number, allocate row numbers according to the correspondence between the peak point index and the paragraph number, and obtain the impact rhythm repeating calibration group.

7. The method for extracting fault features of complex rolling bearings based on adaptive filtering according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the peak index and segment number in the impact rhythm repetition calibration group, retrieve the upper and lower envelope curves of the corresponding segment, analyze the edge trajectory between the envelopes, trace the continuous direction on the time axis, and obtain the edge change path sequence. S302: Based on the trajectory trend in the edge change path sequence, identify the locations where the waveform contour shifts in direction in adjacent positions, filter the boundary points where the direction changes in a continuous area, and obtain the edge turning position set; S303: Based on the trajectory nodes in the set of edge turning positions, retrieve the edge trajectory within the corresponding time period, define the path extension range according to the trend continuity between nodes, divide the continuous trend into path segments, and obtain a periodic structure path set.

8. The method for extracting fault features of complex rolling bearings based on adaptive filtering according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the dominant trend segment of the waveform in the set of periodic structure paths, extract the time series and amplitude series of the corresponding waveforms before and after the segment, and separate the continuous rising segment and the continuous falling segment by the amplitude change trend to obtain the waveform gradient trend pair sequence. S402: Based on the waveform gradient trend pair sequence, by comparing the slope direction of adjacent positions in the rising and falling segments, the time period of the reverse trend is analyzed, the positions of continuous direction change behavior are screened, and the gradient direction turning point group is obtained. S403: Based on the time index in the gradient direction turning point group, track the angular direction continuity between trends, and extract continuous point segments with extension behavior in direction switching to obtain a gradient symmetric association window group.

9. The method for extracting fault features of complex rolling bearings based on adaptive filtering according to claim 8, characterized in that, The extraction process of the dominant trend segment of the waveform based on the periodic structure path set is as follows: by continuously judging the amplitude change direction corresponding to each waveform in the periodic structure path set, and drawing out continuous segments based on the continuous and consistent behavior of the amplitude change direction between adjacent sampling points, the dominant trend segment of the waveform is obtained. In the process of decomposing the amplitude change trend, the direction of amplitude change between adjacent sampling points is analyzed from the starting position. The amplitude change segment that continues to rise is classified as an upward trend, the change segment that continues to fall is classified as a downward trend, and the point where there is a change in direction between adjacent trends is marked as the trend turning point. The process of comparing slope direction is as follows: extract two boundary points at the junction of continuous upward and downward trends, determine the continuity of amplitude trend between the previous and subsequent positions, and divide the start and end intervals of turning behavior in the continuous segment of direction switching. In the process of identifying the time period of the reverse trend, the gradient trend is traversed to analyze the trend continuation direction between each trend segment and the adjacent segment in the sequence, and the continuity in time is analyzed according to the switching state of the adjacent trend direction, and the area of ​​trend direction change is circled on the time axis. The process of extracting continuous point segments with extended behavior for direction switching is as follows: the consistency of trend direction is judged for the front and back positions of the identified trend direction change area; the trend state that maintains the same direction within the continuous point segment is extended along the trend direction and defined as the extended direction point segment; and gradient symmetric association window groups are summarized and extracted from the extended point segment.

10. The method for extracting fault features of complex rolling bearings based on adaptive filtering according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on the path structure index information in the gradient symmetric associated window group, retrieve the waveform window segments within the same period segment, read the amplitude sequence of continuous sampling points in each segment, and extract the change direction between adjacent amplitudes according to the time order to obtain the amplitude direction sequence group; S502: Based on the change direction characteristics in the amplitude direction sequence group, the corresponding recurrence position of the wave direction change sequence in each segment is identified, and the connection range before and after the continuous part of the time index in the sequence is expanded to obtain the sequence extension segment set; S503: Based on the start and end time indices of each group in the sequence extension segment set, extract the periodic waveform data, extract the original sampling point sequence according to the index range corresponding to the segment, and splice each segment of waveform data in order to obtain the filter response feature waveform group.

Images