System and method for monitoring state of grinding wheel in full life cycle of centerless grinding machine

Abstract

The invention relates to the technical field of grinding wheel monitoring, in particular to a centerless grinding machine full-life-cycle grinding wheel state monitoring system and method.According to the centerless grinding machine full-life-cycle grinding wheel state monitoring system and method, through combined difference extraction of a main shaft current mean value fluctuation ratio and a power gradient continuous variation value and assisted by transverse comparison of periodic response intervals, effective recognition of time period boundaries of working condition states is achieved; according to the method, operation stage division does not depend on static threshold judgment any more, a track trend is established through linkage judgment of an amplitude kick point and a slope extreme value, a dynamic evolution path of signal response is clearer, the identification resolution of abnormal changes and local deviation trends is improved, response synchronous identification of parameter state changes is achieved, and the identification accuracy is improved. In multi-signal collaborative analysis, directivity comparison between a temperature rise mutation fragment and a current residual error response is carried out, so that the consistency analysis capability of response logic between different monitoring paths is enhanced, and the real-time performance, stability and anti-interference capability of state identification are improved.

Description

Technical Field

[0001] This invention relates to the field of grinding wheel monitoring technology, and in particular to a grinding wheel condition monitoring system and method for the entire life cycle of a centerless grinder. Background Technology

[0002] The field of grinding wheel monitoring technology is specifically applied to the grinding tool status perception and management system of centerless grinding machines. It mainly focuses on the real-time acquisition and analysis of grinding wheel usage status, wear degree, geometric morphology changes and surface integrity to support the assurance of stability, safety and accuracy of the grinding process. This includes grinding wheels used for machining hard and brittle materials and high-precision parts. It is widely used in grinding machines, especially in centerless grinding machines that require high continuous operation and high consistency precision control.

[0003] The centerless grinder's full lifecycle grinding wheel condition monitoring system aims to continuously sense, record data, and predict failures of the grinding wheel throughout its entire lifecycle, from initial installation to retirement and replacement. The system includes multiple monitoring modules and signal processing units, capable of collecting various physical signals generated during grinding wheel operation—including spindle vibration, current load, sound signals, and temperature rise data—without interfering with the grinding process. This enables dynamic assessment of grinding wheel wear, changes in machining performance, and lifespan trends. The system is designed to enhance the intelligence level of the grinding process, ensure transparency of the grinding wheel's operating status, thereby reducing the risk of unexpected downtime, optimizing maintenance cycles, and achieving a dual improvement in equipment utilization and machining consistency.

[0004] Existing technologies mostly employ statistical feature extraction methods using single signal channels, which struggle to accurately depict the stage transition process of grinding wheel behavior under different operating conditions. In the presence of noise interference and complex variable operating conditions, the division of operating stages exhibits significant deviations, leading to accidental maintenance triggers and delayed early warnings. The lack of modeling for the continuous evolution of trend paths makes it difficult to track the state during dynamic changes, particularly lacking transition segment identification logic between the stable processing period and the initial degradation phase. Existing systems have not established a collaborative relationship between multi-condition parameters and behavioral responses, resulting in the inability to effectively identify the impact of control parameter fluctuations. Furthermore, cross-channel data fusion largely relies on simple superposition and logical judgment, lacking an analysis mechanism based on sequence structure and directional response logic. In cases of signal conflict, it is difficult to effectively locate the source of response contradictions, easily leading to misjudgments and omissions, thus affecting the overall reliability and intelligence of the monitoring system. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a grinding wheel condition monitoring system and method for the entire life cycle of a centerless grinder.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a centerless grinder full life cycle grinding wheel condition monitoring system includes:

[0007] Running segment division module: Based on the mean change rate of spindle current fluctuation, the continuous gradient change value of spindle power and the difference between the signal response interval period, extract the peak interval valley region, obtain the slope boundary of the mutation point, determine the stable region of the response period, and generate three time domain boundary block groups;

[0008] Signal evolution tracking module: Based on the three time-domain boundary block groups, it identifies amplitude jumps, extracts the average amplitude segment in the stable region and locates the peak value of the slope point in the attenuation region, calibrates continuous inflection points to construct trajectory segments, and generates segmented dynamic response trajectories.

[0009] State trend fusion module: Based on the segmented dynamic response trajectory, the statistical direction is kept at the length and the range peak segment is selected, the jump dense segment is extracted, the trend change difference matrix is ​​constructed using the residual network, and the behavior pattern sequence is constructed to generate a stage behavior structure identifier set;

[0010] Operating condition consistency identification module: Based on the staged behavior structure identifier set, extract flow offset points and filter speed trend change points, locate abnormal rate fluctuation segments, establish an offset and behavior trend mapping table, and generate parameter behavior deviation trajectory set;

[0011] Collaborative response verification module: Based on the parameter behavior deviation trajectory set, extract temperature change zone and identify current residual difference points, use dynamic time warping to judge conflict segment, extract response direction conflict time segment, and generate cross-parameter signal response consistency set.

[0012] As a further embodiment of the present invention, the three time-domain boundary block group includes a start-up interval segment, a stable interval segment, and a termination interval segment; the segmented dynamic response trajectory includes an amplitude increase trajectory, a uniform amplitude oscillation trajectory, and a slope decay trajectory; the staged behavior structure identifier set includes a direction persistence unit, a jump response unit, and a trend reversal unit; the parameter behavior deviation trajectory set includes a cooling flow offset segment, a spindle speed fluctuation segment, and a feed rate abnormal segment; and the cross-parameter signal response consistency set includes a temperature signal segment, a current residual signal segment, and a response direction conflict segment.

[0013] As a further aspect of the present invention, the running segment division module includes:

[0014] Fluctuation feature extraction submodule: Based on the mean change rate of spindle current fluctuation and the continuous gradient change value of spindle power, construct the mean sliding sequence of spindle current in adjacent time windows and identify the start and end positions of jump segments, extract the gradient recursive sequence formed by continuous points in spindle power and mark the gradient abrupt break boundary point group, combine the extremum point time scales in the superimposed current and power sequences and screen the intersection positions with high overlap density to obtain the set of joint abrupt change feature points of electrical power;

[0015] Periodic stability identification submodule: Based on the set of electrical power joint mutation feature points, the distance between adjacent peaks and valleys between mutation points is statistically analyzed and the valley region of periodic difference amplitude is extracted as a candidate segment. The continuous response amplitude variation trend in the candidate region is extracted and low slope variation segments are marked. The segment is verified to be continuous and complete on the time axis and to form a closed coverage segment. Three time domain boundary block groups are established.

[0016] As a further aspect of the present invention, the signal evolution tracking module includes:

[0017] Amplitude mutation extraction submodule: Based on the three time-domain boundary block groups, calculate the peak amplitude difference segment between multiple adjacent amplitude points in the start segment and mark the location of high-variation areas, filter the amplitude gradient change between continuous wave peaks in the stable segment and extract the weak vibration maintenance segment, generate the sliding slope sequence in the termination segment and locate the peak inflection point of the average direction change, fuse the three key point labels and unify the time reference alignment, and obtain the stage amplitude feature group;

[0018] Trend path construction submodule: Based on the stage amplitude feature group, the distance distribution between the jump point and the slope extreme point is screened and the neighboring point pairs are retained. The direction reversal points are grouped and sorted according to the amplitude extreme value and the main turning segments are identified. The slope change direction of the continuous segment is accumulated and the fluctuating reverse interference segment is removed to generate a segmented dynamic response trajectory.

[0019] As a further aspect of the present invention, the state trend fusion module includes:

[0020] Direction stability identification submodule: Based on the segmented dynamic response trajectory, it determines whether the direction of amplitude change of continuous points in each segment is consistent and performs sequence segmentation processing. It accumulates the switching interval for segments with unchanged direction and extracts continuous peak segments. It counts the amplitude difference between the start and end points in the segment and identifies directional segments with prominent amplitude differences. It uses a residual network to continuously fit the above candidate trend segments and calculates the residual distribution structure index. It performs connectivity verification based on the integrity of the time interval and removes missing segments to obtain a set of stable trend segments.

[0021] Jump Concentration Extraction Submodule: Based on the set of stable trend segments, determine the sudden change in direction of continuous sampling points and mark the extreme value set of slope, calculate the valley interval between two points in the sudden change point group and retain high-density segments, accumulate the total value of amplitude fluctuation in each segment and extract segments with significant deviations, identify the concentrated area of ​​reversal points in the jump segment and record the corresponding time of the boundary, and generate a high-frequency jump trajectory group.

[0022] Behavior sequence construction submodule: Based on the high-frequency jump trajectory group, the number of direction reversal points and the trend duration of adjacent stable segments are combined and judged. The trend progression amplitude is accumulated in the combination relationship and the peak transition area is extracted. The group of points with significant turning amplitude in the behavior trend segment is screened and merged to obtain a continuous behavior template, and a staged behavior structure identifier set is generated.

[0023] As a further aspect of the present invention, the execution process of the residual network includes: constructing a continuous amplitude sequence as the main input layer based on the candidate trend segments of the input and setting a sliding time window step size; using a constant depth structure to superimpose multiple nonlinear transformation units to extract multi-scale variation features within the sequence; connecting the input and output of adjacent layers through an identity mapping path to maintain the dominant trend of the original signal; recording the difference sequence between the output of each layer and the main input to form residual terms; integrating each residual term to generate a residual distribution structure; and outputting a set of structural indicators including residual amplitude, density, slope direction, and change point location.

[0024] As a further aspect of the present invention, the working condition consistency identification module includes:

[0025] Parameter anomaly extraction submodule: Based on the staged behavior structure identifier set, it slides to generate adjacent difference intervals of the cooling flow sequence and extracts the transfer points of the offset anomaly, extracts continuous direction switching segments in the spindle speed change sequence and records the edge position of the change rate, divides the fluctuation intensity increasing segment from the feed rate sequence and calculates the difference between the range and the fluctuation ratio, and obtains parameter anomaly feature segments.

[0026] Offset Trend Mapping Submodule: Based on the parameter anomaly feature fragment, it matches the intersection of the time index of the anomaly point and the time index of the trend change label, compares the co-directionality of the parameter change direction and the trend direction and removes the opposite result group, extracts the time offset between the mapping point pairs and filters out low correlation pairs that exceed half a cycle, and generates a parameter behavior deviation trajectory set.

[0027] As a further aspect of the present invention, the cross-parameter signal response consistency set includes:

[0028] Temperature Segment Extraction Submodule: Based on the parameter behavior deviation trajectory set, extract the temperature rise difference between two consecutive points in the temperature sequence and filter the rising segments with an amplitude greater than twice the historical fluctuation average. Statistically count the frequency of peak occurrences in the rising segments and extract the intervals where the number of peaks per unit time is greater than a specified threshold. Use dynamic time warping to register multiple rising segment sequences with the reference response rhythm sequence and identify the matching structure distribution area. Identify the boundary points within the fluctuation intervals in the continuous rising segments and determine the complete temperature response change segments to obtain a set of abnormal temperature rise segments.

[0029] The residual difference identification submodule: Based on the set of abnormal temperature rise sections, calculate the difference between the fitted value and the actual value of the sliding window in the current sequence on the same time axis and generate a residual time sequence, count the amplitude between the peak and valley values ​​in each residual window and extract points that exceed twice the mean of the residual of the whole sequence, count the change of residual direction in the two periods before and after the abnormal point and remove the group of points with reversed trend distribution to obtain the set of abnormal current residual points.

[0030] The conflict segment determination submodule: Based on the current residual abnormal point set, compare the polarity of the temperature change direction and the current residual direction at the same time point in the temperature rise abnormal segment and filter the group of inconsistent points, extract the continuous inconsistent point sequence and construct the time index sequence group, filter the continuous conflict point segment with a duration of more than one cycle and record the corresponding sampling index range, and generate the cross-parameter signal response consistency set.

[0031] As a further aspect of the present invention, the execution process of the dynamic time warping includes: constructing a distance matrix of two time series based on the temperature change sequence in each heating segment and a preset reference response rhythm sequence; sequentially matching the amplitude difference between each sampling point according to the increasing time step; calculating the cumulative distance on the matching path and recording the cumulative error sum of feasible paths; selecting the alignment path with the cumulative valley distance as a suitable matching trajectory; extracting the position mapping relationship of each corresponding point pair in the alignment path; and outputting the matching structure distribution area containing the time offset alignment relationship, the corresponding position index group, and the cumulative registration error value.

[0032] A method for monitoring the condition of grinding wheels throughout the entire lifecycle of a centerless grinder, wherein the method is based on the aforementioned centerless grinder lifecycle grinding wheel condition monitoring system, includes the following steps:

[0033] S1: Based on the spindle current fluctuation change value, spindle power gradient sequence and periodic signal response interval, the mean difference between adjacent periods is statistically analyzed and the fluctuation distribution change segment is extracted. The power slope change amplitude is calculated and the significant inflection point position is marked. Combined with the periodic boundary difference, the response segment boundary interval is identified and a three-segment time domain boundary block group is established.

[0034] S2: Based on the three time-domain boundary block groups, obtain the amplitude sequence of each segment and filter the difference mutation region between continuous points, identify the amplitude-maintaining segment in the stable segment and judge the boundary stability, calculate the slope extreme value of continuous points in the decay segment and extract the change concentration area, mark the key points of each segment and construct the cross-segment response trajectory, and generate the segmented dynamic response trajectory.

[0035] S3: Based on the segmented dynamic response trajectory, the duration of segments with consistent directional changes between adjacent points is statistically analyzed and directional sequences with extreme amplitudes are selected. Fluctuation segments with dense jump frequencies are extracted as candidate segments. The extracted segments are continuously fitted using a residual network and the residual feature structure is output. Sequence structure units composed of directional trend combinations are labeled to generate a set of staged behavioral structure identifiers.

[0036] S4: Based on the staged behavior structure identifier set, compare the changing trends of cooling flow, spindle speed and feed rate control parameter sequences during the operation process, extract the interval segments with deviations and match them to the time range of the behavior turning point, construct a mapping relationship between control variables and behavior nodes, and generate a parameter behavior deviation trajectory set.

[0037] S5: Based on the parameter behavior deviation trajectory set, filter continuous segments with sudden increases in the rate of change in the temperature sequence and identify abnormal rising areas; identify the set of points in the current sequence where the fitting residual exceeds the statistical threshold; use dynamic time warping to perform time-series alignment on the multi-source sequences and extract directional variation matching pairs; construct a conflict time period index and mark signal inconsistency blocks; and generate a cross-parameter signal response consistency set.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0039] 1. In this invention, by extracting the joint difference between the mean fluctuation rate of the spindle current and the continuous variation value of the power gradient, and supplementing it with the horizontal comparison of the periodic response interval, the time period boundary of the operating condition is effectively identified, so that the division of the operating phase no longer depends on the static threshold judgment.

[0040] 2. In this invention, the trajectory trend is established by using the linkage judgment of amplitude jump points and slope extreme values, which makes the dynamic evolution path of signal response clearer, improves the identification resolution of abnormal changes and local deviation trends, and realizes synchronous identification of parameter state changes.

[0041] 3. In this invention, by comparing the directionality of temperature rise abrupt segments and current residual responses in multi-signal collaborative analysis, the consistency analysis capability of response logic between different monitoring paths is enhanced, thereby improving the real-time performance, stability and anti-interference capability of state identification. Attached Figure Description

[0042] Figure 1 This is a system flowchart of the present invention;

[0043] Figure 2 This is a schematic diagram of the system framework of the present invention;

[0044] Figure 3 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0046] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0047] Example 1

[0048] Please see Figure 1 This invention provides a technical solution: a centerless grinder full life cycle grinding wheel condition monitoring system, comprising:

[0049] Running segment division module: Based on the mean change rate of spindle current fluctuation, the continuous gradient change value of spindle power and the difference between the signal response interval period, extract the peak interval valley region, obtain the slope boundary of the mutation point, determine the stable region of the response period, and generate three time domain boundary block groups;

[0050] Signal evolution tracking module: Based on three time-domain boundary block groups, it identifies abrupt amplitude jumps, extracts the average amplitude segment in the stable region and locates the peak value of the slope point in the attenuation region, calibrates continuous inflection points to construct trajectory segments, and generates segmented dynamic response trajectories;

[0051] State trend fusion module: Based on segmented dynamic response trajectory, the statistical direction maintains the length and filters out the range peak segment, extracts the jump dense segment, uses residual network to construct trend change difference matrix, constructs behavior pattern sequence, and generates stage behavior structure identifier set;

[0052] Operating condition consistency identification module: Based on the staged behavior structure identifier set, extract flow offset points and filter speed trend change points, locate abnormal speed fluctuation segments, establish an offset and behavior trend mapping table, and generate parameter behavior deviation trajectory set;

[0053] Collaborative response verification module: Based on the parameter behavior deviation trajectory set, it extracts temperature change zones and identifies current residual difference points, uses dynamic time warping to determine conflict segments, extracts conflict time segments in response direction, and generates a set of cross-parameter signal response consistency.

[0054] The three time-domain boundary block groups include the start-up interval, the stable interval, and the termination interval. The segmented dynamic response trajectory includes the amplitude increase trajectory, the average amplitude oscillation trajectory, and the slope decay trajectory. The staged behavior structure identifier set includes the direction continuity unit, the jump response unit, and the trend reversal unit. The parameter behavior deviation trajectory set includes the cooling flow offset segment, the spindle speed fluctuation segment, and the feed rate abnormal segment. The cross-parameter signal response consistency set includes the temperature signal segment, the current residual signal segment, and the response direction conflict segment.

[0055] Please see Figure 2 The runtime segmentation module includes:

[0056] Fluctuation feature extraction submodule: Based on the mean change rate of spindle current fluctuation and the continuous gradient change value of spindle power, construct the mean sliding sequence of spindle current in adjacent time windows and identify the start and end positions of jump segments, extract the gradient recursive sequence formed by continuous points in spindle power and mark the gradient abrupt break boundary point group, combine the extremum point time scales in the superimposed current and power sequences and screen the intersection positions with high overlap density to obtain the set of joint abrupt change feature points of electrical power;

[0057] Periodic stability identification submodule: Based on the electrical power joint mutation feature point set, the distance between adjacent peaks and valleys between mutation points is statistically analyzed and the valley area of ​​periodic difference amplitude is extracted as a candidate segment. The continuous response amplitude variation trend in the candidate area is extracted and low slope variation segments are marked. The segment is verified to be continuous and complete on the time axis and form a closed coverage segment. Three time domain boundary block groups are established.

[0058] The fluctuation feature extraction submodule: Based on the mean change rate of spindle current fluctuation and the continuous gradient change value of spindle power, the sliding window length of the spindle current sequence is set to 30 and the step size is 5. The mean of each window segment is calculated and the sequence is constructed. After first-order difference, the start and end points of the mutation segment with the difference greater than twice the mean of the whole sequence are selected. At the same time, the slope value sequence of the spindle power signal sequence is calculated by recursion at three points. The breakpoints with gradient change amplitude exceeding the threshold of 0.25 are selected as mutation boundaries. Then, the extreme point time indexes are extracted from the current and power sequences respectively. The sets are crossed and the intersection point groups with a time interval of less than 0.1 seconds are counted to generate a set of electrical and power joint mutation feature points.

[0059] The periodic stability identification submodule: Based on the joint abrupt change feature point set of electrical power, the peak-valley time interval between adjacent abrupt change points is statistically analyzed to form a periodic sequence. The standard deviation of the periodic sequence is calculated by a sliding window with a length of 5, and segments with variance less than half of the total sequence are extracted as candidate segments. Then, a straight line is fitted once in units of 6 points within the candidate segments, and the slope coefficient k is obtained by least squares fitting. Segments with absolute slope values ​​less than the mean of 0.2 are retained. Subsequently, the continuity of the retained segments is verified, and segments with a spacing of no more than 0.05 seconds are connected and segments with a coverage time of less than 0.3 seconds are removed to establish a three-segment time-domain boundary block group.

[0060] Please see Figure 2 The signal evolution tracking module includes:

[0061] Amplitude mutation extraction submodule: Based on three time-domain boundary block groups, calculate the peak amplitude difference segment between multiple adjacent amplitude points in the start segment and mark the location of high-variation areas; filter the amplitude gradient change between continuous wave peaks in the stable segment and extract the weak vibration maintenance segment; generate the sliding slope sequence in the termination segment and locate the peak inflection point of the average direction change; fuse the three key point labels and unify the time reference alignment to obtain the stage amplitude feature group.

[0062] Trend path construction submodule: Based on the stage amplitude feature group, the distance distribution between the jump point and the slope extreme point is screened and the neighboring point pairs are retained. The direction reversal points are grouped and sorted according to the amplitude extreme value and the main turning segments are identified. The slope change direction of continuous segments is accumulated and the fluctuating reverse interference segments are removed to generate segmented dynamic response trajectory.

[0063] Amplitude mutation extraction submodule: Based on three time-domain boundary block groups, the amplitude difference sequence is extracted in the starting segment with a sliding window length of 3 and a step size of 1. The segment with continuous difference greater than twice the mean of the sequence is selected as the amplitude difference peak segment, and the start and end point indices are marked. For the stable segment signal, the amplitude change rate between the two points before and after is calculated and a gradient sequence is constructed using the peak point as the index. The continuous segment with an absolute value less than the threshold of 0.1 is extracted as the weak vibration maintenance segment. The three-point slope of the amplitude sequence of the termination segment is calculated, the average direction slope is obtained, and the position of the slope change sign is marked as the inflection point. The three key point labels are fused, and the labels are indexed and normalized to synchronize the starting point with the zero point position of the unified time axis, so as to obtain the stage amplitude feature group.

[0064] Trend path construction submodule: Based on the stage amplitude feature group, time index sequences of jump points and slope extreme points are constructed respectively. Point pairs with a distance of less than 0.08 seconds are extracted and retained using Euclidean distance calculation. Amplitude sequence groups are divided between each pair of direction reversal points. The segments are sorted by the maximum amplitude and the segment number is recorded. Local slope sequences are extracted using a three-point sliding window with equal interval step size. The signs of slope segments that change continuously in the same direction are accumulated. Interference segments with more than 2 cumulative direction sign changes are filtered out, and segmented dynamic response trajectories are generated.

[0065] Please see Figure 2 The status trend fusion module includes:

[0066] Direction stability identification submodule: Based on segmented dynamic response trajectory, it determines whether the direction of amplitude change of continuous points in each segment is consistent and performs sequence segmentation processing. It accumulates the switching interval for segments with unchanged direction and extracts continuous peak segments. It counts the amplitude difference between the start and end points in the segment and identifies directional segments with prominent amplitude differences. It uses a residual network to continuously fit the above candidate trend segments and calculates the residual distribution structure index. It performs connectivity verification based on the integrity of the time interval and removes missing segments to obtain a set of stable trend segments.

[0067] Jump Concentration Extraction Submodule: Based on the set of stable trend segments, it determines the sudden change in direction of continuous sampling points and marks the extreme value set of slope, calculates the valley interval between two points in the sudden change point group and retains high-density segments, accumulates the total value of amplitude fluctuation in each segment and extracts segments with significant deviations, identifies the concentrated area of ​​reversal points in the jump segment and records the corresponding time of the boundary, and generates a high-frequency jump trajectory group.

[0068] The behavior sequence construction submodule is based on high-frequency jump trajectory groups. It combines the number of direction reversal points with the trend duration of adjacent stable segments to make judgments. It accumulates the trend progression amplitude in the combination relationship and extracts the peak transition zone. It filters the groups of points with significant turning amplitude in the behavior trend segments and merges them to obtain continuous behavior templates, generating a set of staged behavior structure identifiers.

[0069] The directional stability identification submodule: Based on the segmented dynamic response trajectory, it determines whether the amplitude change direction of continuous points in each segment is consistent and performs sequence segmentation. It accumulates the switching interval for segments with unchanged direction and extracts continuous peak segments. It uses a residual network to fit the continuous change of candidate trend segments. It sets the sliding time window length to a fixed 20 points and uses an equidistant sequence to group all amplitude segments into windows. It analyzes the consistency of the amplitude direction of the fitting residuals in each group and obtains the index difference sequence of the fitting residual points. It counts the amplitude difference between the start and end points in the segment and identifies directional segments with prominent amplitude differences. Based on the time series boundary of the trend segment, it performs connectivity verification, filters out discontinuous index segments and removes missing segments, and obtains a set of stable trend segments.

[0070] The jump-point extraction submodule: Based on the set of stable trend segments, it judges the sudden change in direction of continuous sampling points and marks the extreme value set of slope. It performs direction change identification processing on the amplitude change rate in each stable segment, selects the direction reversal of adjacent inflection points as extreme value nodes, calculates the valley interval between two points in the mutation point group and retains high-density segments, uses the equal step size forward accumulation method to calculate the total value of amplitude fluctuation in the complete cycle in each segment and extracts segments with significant deviations, identifies the concentrated area of ​​reversal points in the jump segment and records the corresponding time of the boundary, extracts the mutation cluster area according to the strategy that the mutation frequency is greater than the set threshold, and generates a high-frequency jump trajectory group;

[0071] The behavior sequence construction submodule, based on high-frequency jump trajectory groups, combines and judges the number of direction reversal points with the trend duration of adjacent stable segments. It uses the count of reversal points within amplitude change segments and the length of direction maintenance within adjacent segments to form a two-dimensional combined scalar. In the combined relationship, it accumulates the trend progression amplitude and extracts the peak transition zone. It extracts points with significant changes in amplitude slope in the trend segment and combines them to form local turning regions. It filters groups of points with significant turning amplitude in the behavior trend segment, merges the continuous sequences between turning inflection points, and combines the time labels of stable trend segments to construct continuous behavior templates, generating a set of staged behavior structure identifiers.

[0072] The execution process of the residual network includes: constructing a continuous amplitude sequence as the main input layer based on the candidate trend segments of the input and setting the sliding time window step size; using a constant depth structure to superimpose multiple nonlinear transformation units to extract multi-scale change features within the sequence; connecting the input and output of adjacent layers through an identity mapping path to maintain the dominant trend of the original signal; recording the difference sequence between the output of each layer and the main input to form residual terms; integrating each residual term to generate a residual distribution structure; and outputting a set of structural indices including residual amplitude, density, slope direction, and change point location.

[0073] Residual networks, according to the formula:

[0074]

[0075] Where: y t The final output value of the residual network at time point t is represented by y, where y represents the output value, t represents the time point index, and W represents the final output value of the residual network at time point t. r Let W represent the residual path weighting coefficient matrix, where W represents the weights and r indicates that a weight belongs to the residual path portion. This represents the summation from index i to n, where n represents the total number of historical nodes used within the window, and i = 1 indicates that the summation starts from the i = 1th historical node. α i f(x) represents the position decay weight coefficient for historical time point index i, α represents the weight coefficient, i represents the relative position of the historical point within the backtracking window, and f(x) represents the position decay weight coefficient for historical time point index i. t-iThe expression represents the output obtained after processing the original value at time point ti in the input sequence through a nonlinear network, where f represents the nonlinear mapping function, x represents the original input amplitude, ti represents the value at the i-th time step forward from time point t, and β represents the weighting coefficient of the original trend branch. t γ represents the raw amplitude value at time point t, x represents the input amplitude signal, t represents the current time, and γ represents the weighting coefficient of the trend acceleration component. This represents the approximation operation of the second derivative of the input sequence x at time point t, where 2 indicates that it is the second derivative, t indicates that the derivative operation is performed with respect to time point t, and dt 2 Represents the square of the time difference;

[0076] Execution process: First, each stable candidate segment is divided into segments using a sliding window of length 20, and each segment constitutes an input sequence x. t For each time point t in the sequence, n historical points (usually 5) are extracted backwards to form x. t-1 ,x t-2 ,...,x t-n Each historical point x t-i The output is fed into a nonlinear mapping function f, and after mapping, it is multiplied by the corresponding historical decay weight α. i , where α i The weights are set according to the principle that the larger the value of i, the smaller the weight. The sum of the weighted results is then multiplied by the overall weight W of the residual path. r This constitutes the residual term, at the current time point x. t Multiplying by the original trend retention coefficient β ensures that a certain proportion of the original signal is retained for trend judgment, which is then combined with the second-order difference result of the input sequence x. Estimate the acceleration characteristics of the trend, multiply by the acceleration coefficient γ to represent the degree of contribution, and sum the above three parts to obtain the final output y. t .

[0077] Please see Figure 2 The operating condition consistency identification module includes:

[0078] The parameter anomaly extraction submodule is based on a set of staged behavior structure identifiers. It slides to generate adjacent difference intervals of the cooling flow sequence and extracts the transfer points of the offset anomaly. It extracts continuous direction switching segments in the spindle speed change sequence and records the edge position of the change rate. It divides the fluctuation intensity increasing segment from the feed rate sequence and calculates the difference between the range and the fluctuation ratio to obtain parameter anomaly feature segments.

[0079] Offset Trend Mapping Submodule: Based on parameter anomaly feature fragments, it matches the intersection of the time index of the anomaly point and the time index of the trend change label, compares the co-directionality of the parameter change direction and the trend direction and removes the opposite result group, extracts the time offset between mapping point pairs and filters out low correlation pairs that exceed half a cycle, and generates a set of parameter behavior deviation trajectories.

[0080] The parameter anomaly extraction submodule, based on a staged behavior structure identifier set, calculates adjacent difference intervals in the cooling flow sequence, sets a window length of 5 and a step size of 1, calculates the absolute value of the difference between the amplitude of the next term and the previous term within each window, constructs a sliding difference sequence, marks the points in the difference sequence that are continuously greater than twice the mean of the cooling flow sequence and extracts the start and end indices as offset anomaly transfer points, extracts the sign change of the spindle speed change sequence, sequentially judges the positive and negative signs of the differences between two adjacent terms and marks the sign flip position index segment, takes the time length between adjacent flip positions as the direction switching cycle, and records the extreme points of the rate of change within each switching cycle as the rate change edge position, divides the feed rate sequence into equal length segments of a fixed segment length of 50, constructs a range sequence for the maximum and minimum differences within each segment, calculates the fluctuation amplitude of the standard deviation of each segment and takes the ratio as the fluctuation intensity index, filters the segments with a ratio greater than a set threshold of 3 as high fluctuation intervals, and obtains parameter anomaly feature segments;

[0081] Offset Trend Mapping Submodule: Based on parameter anomaly feature segments, extract the time index sets of the start and end points of each anomaly segment and the time index set of the behavior trend segment. Perform a set intersection operation on the two sets of indices, retain the segment groups with an intersection number greater than the set minimum matching point number of 10 as the initial candidate point set, compare the consistency between the parameter change direction and the trend change direction in each candidate segment, mark the point groups with the same positive and negative signs and remove the point group groups with opposite direction signs, extract the remaining point segment indices to construct matching point pairs, use the difference calculation method to extract the offset time of the time index between each pair of points, filter out point pairs with offset values ​​greater than half of the set reference period, and generate a parameter behavior deviation trajectory set.

[0082] Please see Figure 2 The set of cross-parameter signal response consistency includes:

[0083] Temperature Segment Extraction Submodule: Based on the parameter behavior deviation trajectory set, extract the temperature rise difference between two consecutive points in the temperature sequence and filter the rising segments with an amplitude greater than twice the historical fluctuation average. Statistically count the frequency of peak occurrences in the temperature rise segment and extract the intervals where the number of peaks per unit time is greater than a specified threshold. Use dynamic time warping to register multiple temperature rise segment sequences with the reference response rhythm sequence and identify the matching structure distribution area. Identify the boundary points within the fluctuation interval in the continuous rising segment and determine the complete temperature response change segment to obtain the set of abnormal temperature rise segments.

[0084] The residual difference identification submodule is based on the set of abnormal temperature rise sections. It calculates the difference between the fitted value and the actual value of the sliding window in the current sequence on the same time axis and generates a residual time series. It counts the amplitude between the peak and valley values ​​in each residual window and extracts the points that exceed twice the mean of the residual of the whole sequence. It counts the change of residual direction in the two periods before and after the abnormal point and removes the group of points with reversed trend distribution to obtain the set of abnormal current residual points.

[0085] The conflict segment determination submodule: Based on the current residual abnormal point set, it compares the polarity of the temperature change direction and the current residual direction at the same time point in the temperature rise abnormal segment and filters the group of inconsistent points. It extracts the continuous inconsistent point sequence and constructs the time index sequence group. It filters the continuous conflict point segments with a duration of more than one cycle and records the corresponding sampling index range, and generates a cross-parameter signal response consistency set.

[0086] Temperature Segment Extraction Submodule: Based on the parameter behavior deviation trajectory set, it performs a first-order difference operation on two adjacent points in the temperature sequence, sets the filtering threshold to twice the historical fluctuation average, retains rising segments with differences greater than the threshold and records the start and end indices of continuous segments, extracts the temperature rise segments and counts the execution peak of each segment sequence, sets the sliding window length to 15 and the step size to 1, counts the number of local maximum occurrences in each window and judges whether the number of peaks in a unit time window is greater than the specified threshold of 5, performs boundary positioning on the window segments that meet the conditions, performs registration processing on multiple temperature rise segment sequences and reference response rhythm sequences using a dynamic time warping method, constructs a distance matrix of two sets of temperature change time sequences, calculates the cumulative distance and obtains the shortest path mapping using a monotonically non-decreasing index path matching rule, performs coordinate merging operation on the matching relationship structure formed in the path, identifies the interval with the most frequent fluctuation changes in the continuous rising segment, extracts local maximum and minimum value indices in the interval, constructs a complete segment index range for the upper and lower boundary points, and obtains a set of abnormal temperature rise segments;

[0087] The residual difference identification submodule: Based on the set of abnormal temperature rise sections, it performs linear regression fitting on the current sequence data that is consistent with the time axis of the temperature rise section with a fixed window length of 20. The step size is set to 1. It constructs a linear least square fitting line for each window segment and calculates the difference between the fitted value and the actual current value to form a residual sequence. It summarizes the peak and valley amplitudes of each residual window and generates a set of difference values. It sets the statistical threshold to twice the mean of the entire residual sequence, filters residual fluctuation points that exceed the threshold, locates the time index segment of the abnormal point, and counts the number of signs of the residual change direction in the two periods before and after the abnormal point. The segment with a difference of more than 80% between positive and negative signs is marked as a direction variation segment. The segment with irregular distribution of sign change is removed to obtain the set of current residual abnormal points.

[0088] The conflict segment determination submodule, based on the current residual abnormal point set, compares the signs of the temperature change direction and the current residual direction at the same time point. It performs synchronous alignment processing on the temperature rise abnormal segment and the current abnormal point concentration segment, extracts the sign values ​​of the temperature sequence and the current residual sequence at each time point in sequence, compares the directional polarity of the two sets of sequences under the same time index, filters out the points with inconsistent signs and records the corresponding time index, and uses an equidistant scanning method to perform cluster window sliding extraction on the inconsistent point sequence to construct an index sequence group of continuous inconsistent point segments. It sets the minimum duration threshold to an integer multiple of the original sampling period, removes point segments with a time span of less than one period, records the start and end index numbers of point segments whose remaining duration meets the conditions, and generates a cross-parameter signal response consistency set.

[0089] The execution process of dynamic time warping includes: constructing a distance matrix between two time series based on the temperature change sequence in each heating segment and a preset reference response rhythm sequence; matching the amplitude difference between each sampling point in sequence according to the increasing time step; calculating the cumulative distance on the matching path and recording the cumulative error of feasible paths; selecting the alignment path with the cumulative valley distance as a suitable matching trajectory; extracting the position mapping relationship of each corresponding point pair in the alignment path; and outputting the matching structure distribution area containing the time offset alignment relationship, the corresponding position index group, and the cumulative registration error value.

[0090] Dynamic time warping, according to the formula:

[0091]

[0092] Where: C(p,q) represents the cumulative registration distance between the p-th temperature sequence point and the q-th reference rhythm point, p represents the index of a sampling point on the time axis in the temperature change sequence, q represents the index of the sampling point on the time axis in the reference response rhythm sequence, κ represents the amplitude scaling factor between the temperature sequence and the rhythm sequence, and u o This represents the amplitude value at the p-th point in the temperature change sequence, where p is the index and v is the value at the p-th point. q This represents the amplitude value at the q-th point in the reference response rhythm sequence, where q is the index, r represents the power coefficient of the distance function (set to 2 to represent the squared difference), and φ... u φ represents the standard deviation of a temperature change series. v The standard deviation of the reference response rhythm sequence, |u p -v q | r This represents the power-law difference measure between the p-th and q-th points of the temperature and rhythm sequences, min{C(p- q)}

[0093] 1,q),C(p,q-1),C(p-1,q-1)} represents the path with the lowest cost among the three cumulative paths that can be chosen at the current position;

[0094] Execution process: First, calculate the standard deviation φ of the two sequences. u With φ v The algorithm calculates a scaling factor κ based on the average amplitude of each signal source to scale the amplitude influence between different source signals. The algorithm iterates through each temperature point u. p With rhythm point v q The absolute amplitude difference between them is used to amplify the height difference point pair by using the exponent r=2, and the difference is expressed through the denominator φ. u +φ v After normalization, the matching cost of the point pair is obtained by multiplying by the scaling factor κ. After the point pair calculation is completed, the entire matching path matrix C(p,q) is filled from the top left corner to the bottom right corner using the minimum cumulative path strategy. Then, the optimal registration path is obtained by backtracking from the endpoint C(P,Q). The segments with structural repetition, point density clustering, and rhythm overlap are found on the path and are used as the matching structure distribution area.

[0095] Please see Figure 3 A method for monitoring the condition of grinding wheels throughout the entire lifecycle of a centerless grinder, which is based on the aforementioned system for monitoring the condition of grinding wheels throughout the entire lifecycle of a centerless grinder, includes the following steps:

[0096] S1: Based on the spindle current fluctuation change value, spindle power gradient sequence and periodic signal response interval, the mean difference between adjacent periods is statistically analyzed and the fluctuation distribution change segment is extracted. The power slope change amplitude is calculated and the significant inflection point position is marked. Combined with the periodic boundary difference, the response segment boundary interval is identified and a three-segment time domain boundary block group is established.

[0097] S2: Based on three time-domain boundary block groups, obtain the amplitude sequence of each segment and filter the difference mutation region between continuous points, identify the amplitude-maintaining segment in the stable segment and judge the degree of boundary stability, calculate the slope extreme value of continuous points in the decay segment and extract the change concentration area, mark the key points of each segment and construct the cross-segment response trajectory, and generate segmented dynamic response trajectory.

[0098] S3: Based on the segmented dynamic response trajectory, the duration of the segment with consistent directional changes between adjacent points is statistically analyzed and the directional sequences with extreme amplitudes are selected. The fluctuation segments with dense jump frequencies are extracted as candidate segments. The residual network is used to continuously fit the selected segments and output the residual feature structure. The sequence structure units composed of directional trend combinations are labeled to generate a set of staged behavior structure identifiers.

[0099] S4: Based on the staged behavior structure identifier set, compare the changing trends of cooling flow, spindle speed and feed rate control parameter sequences during the operation process, extract the interval segments with deviations and match them to the time range of the behavior inflection point, construct the mapping relationship between control variables and behavior nodes, and generate parameter behavior deviation trajectory set.

[0100] S5: Based on the set of deviation trajectories of parameter behavior, filter continuous segments with sudden increases in the rate of change in the temperature sequence and identify abnormal rising areas; identify the set of points in the current sequence where the fitting residual exceeds the statistical threshold; use dynamic time warping to perform time-series alignment of multi-source sequences and extract directional variation matching pairs; construct a conflict time period index and mark signal inconsistency blocks; and generate a set of cross-parameter signal response consistency.

[0101] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A centerless grinder's full lifecycle wheel condition monitoring system, characterized in that, The system includes: Running segment division module: Based on the mean change rate of spindle current fluctuation, the continuous gradient change value of spindle power and the difference between the signal response interval period, extract the peak interval valley region, obtain the slope boundary of the mutation point, determine the stable region of the response period, and generate three time domain boundary block groups; Signal evolution tracking module: Based on the three time-domain boundary block groups, it identifies amplitude jumps, extracts the average amplitude segment in the stable region and locates the peak value of the slope point in the attenuation region, calibrates continuous inflection points to construct trajectory segments, and generates segmented dynamic response trajectories. State trend fusion module: Based on the segmented dynamic response trajectory, the statistical direction is kept at the length and the range peak segment is selected, the jump dense segment is extracted, the trend change difference matrix is ​​constructed using the residual network, and the behavior pattern sequence is constructed to generate a stage behavior structure identifier set; Operating condition consistency identification module: Based on the staged behavior structure identifier set, extract flow offset points and filter speed trend change points, locate abnormal rate fluctuation segments, establish an offset and behavior trend mapping table, and generate parameter behavior deviation trajectory set; Collaborative response verification module: Based on the parameter behavior deviation trajectory set, extract temperature change zone and identify current residual difference points, use dynamic time warping to judge conflict segment, extract response direction conflict time segment, and generate cross-parameter signal response consistency set.

2. The centerless grinding machine full life cycle grinding wheel condition monitoring system according to claim 1, characterized in that, The three time-domain boundary block groups include a start-up interval, a stable interval, and a termination interval. The segmented dynamic response trajectory includes an amplitude increase trajectory, a uniform amplitude oscillation trajectory, and a slope decay trajectory. The staged behavior structure identifier set includes a direction persistence unit, a jump response unit, and a trend reversal unit. The parameter behavior deviation trajectory set includes a cooling flow offset segment, a spindle speed fluctuation segment, and a feed rate abnormal segment. The cross-parameter signal response consistency set includes a temperature signal segment, a current residual signal segment, and a response direction conflict segment.

3. The centerless grinding machine full life cycle grinding wheel condition monitoring system according to claim 1, characterized in that, The runtime segment division module includes: Fluctuation feature extraction submodule: Based on the mean change rate of spindle current fluctuation and the continuous gradient change value of spindle power, construct the mean sliding sequence of spindle current in adjacent time windows and identify the start and end positions of jump segments, extract the gradient recursive sequence formed by continuous points in spindle power and mark the gradient abrupt break boundary point group, combine the extremum point time scales in the superimposed current and power sequences and screen the intersection positions with high overlap density to obtain the set of joint abrupt change feature points of electrical power; Periodic stability identification submodule: Based on the set of electrical power joint mutation feature points, the distance between adjacent peaks and valleys between mutation points is statistically analyzed and the valley region of periodic difference amplitude is extracted as a candidate segment. The continuous response amplitude variation trend in the candidate region is extracted and low slope variation segments are marked. The segment is verified to be continuous and complete on the time axis and to form a closed coverage segment. Three time domain boundary block groups are established.

4. The centerless grinding machine full life cycle grinding wheel condition monitoring system according to claim 1, characterized in that, The signal evolution tracking module includes: Amplitude mutation extraction submodule: Based on the three time-domain boundary block groups, calculate the peak amplitude difference segment between multiple adjacent amplitude points in the start segment and mark the location of high-variation areas, filter the amplitude gradient change between continuous wave peaks in the stable segment and extract the weak vibration maintenance segment, generate the sliding slope sequence in the termination segment and locate the peak inflection point of the average direction change, fuse the three key point labels and unify the time reference alignment, and obtain the stage amplitude feature group; Trend path construction submodule: Based on the stage amplitude feature group, the distance distribution between the jump point and the slope extreme point is screened and the neighboring point pairs are retained. The direction reversal points are grouped and sorted according to the amplitude extreme value and the main turning segments are identified. The slope change direction of the continuous segment is accumulated and the fluctuating reverse interference segment is removed to generate a segmented dynamic response trajectory.

5. The centerless grinding machine full life cycle grinding wheel condition monitoring system according to claim 1, characterized in that, The state trend fusion module includes: Direction stability identification submodule: Based on the segmented dynamic response trajectory, it determines whether the direction of amplitude change of continuous points in each segment is consistent and performs sequence segmentation processing. It accumulates the switching interval for segments with unchanged direction and extracts continuous peak segments. It counts the amplitude difference between the start and end points in the segment and identifies directional segments with prominent amplitude differences. It uses a residual network to continuously fit the above candidate trend segments and calculates the residual distribution structure index. It performs connectivity verification based on the integrity of the time interval and removes missing segments to obtain a set of stable trend segments. Jump Concentration Extraction Submodule: Based on the set of stable trend segments, determine the sudden change in direction of continuous sampling points and mark the extreme value set of slope, calculate the valley interval between two points in the sudden change point group and retain high-density segments, accumulate the total value of amplitude fluctuation in each segment and extract segments with significant deviations, identify the concentrated area of ​​reversal points in the jump segment and record the corresponding time of the boundary, and generate a high-frequency jump trajectory group. Behavior sequence construction submodule: Based on the high-frequency jump trajectory group, the number of direction reversal points and the trend duration of adjacent stable segments are combined and judged. The trend progression amplitude is accumulated in the combination relationship and the peak transition area is extracted. The group of points with significant turning amplitude in the behavior trend segment is screened and merged to obtain a continuous behavior template, and a staged behavior structure identifier set is generated.

6. The centerless grinding machine full life cycle grinding wheel condition monitoring system according to claim 1, characterized in that, The execution process of the residual network includes: constructing a continuous amplitude sequence as the main input layer based on the candidate trend segments of the input and setting a sliding time window step size; using a constant depth structure to superimpose multiple nonlinear transformation units to extract multi-scale variation features within the sequence; connecting the input and output of adjacent layers through an identity mapping path to maintain the dominant trend of the original signal; recording the difference sequence between the output of each layer and the main input to form residual terms; integrating each residual term to generate a residual distribution structure; and outputting a set of structural indicators including residual amplitude, density, slope direction, and change point location.

7. The centerless grinding machine full life cycle grinding wheel condition monitoring system according to claim 1, characterized in that, The operating condition consistency identification module includes: Parameter anomaly extraction submodule: Based on the staged behavior structure identifier set, it slides to generate adjacent difference intervals of the cooling flow sequence and extracts the transfer points of the offset anomaly, extracts continuous direction switching segments in the spindle speed change sequence and records the edge position of the change rate, divides the fluctuation intensity increasing segment from the feed rate sequence and calculates the difference between the range and the fluctuation ratio, and obtains parameter anomaly feature segments. Offset Trend Mapping Submodule: Based on the parameter anomaly feature fragment, it matches the intersection of the time index of the anomaly point and the time index of the trend change label, compares the co-directionality of the parameter change direction and the trend direction and removes the opposite result group, extracts the time offset between the mapping point pairs and filters out low correlation pairs that exceed half a cycle, and generates a parameter behavior deviation trajectory set.

8. The centerless grinding machine full life cycle grinding wheel condition monitoring system according to claim 1, characterized in that, The cross-parameter signal response consistency set includes: Temperature Segment Extraction Submodule: Based on the parameter behavior deviation trajectory set, extract the temperature rise difference between two consecutive points in the temperature sequence and filter the rising segments with an amplitude greater than twice the historical fluctuation average. Statistically count the frequency of peak occurrences in the rising segments and extract the intervals where the number of peaks per unit time is greater than a specified threshold. Use dynamic time warping to register multiple rising segment sequences with the reference response rhythm sequence and identify the matching structure distribution area. Identify the boundary points within the fluctuation intervals in the continuous rising segments and determine the complete temperature response change segments to obtain a set of abnormal temperature rise segments. The residual difference identification submodule: Based on the set of abnormal temperature rise sections, calculate the difference between the fitted value and the actual value of the sliding window in the current sequence on the same time axis and generate a residual time sequence, count the amplitude between the peak and valley values ​​in each residual window and extract points that exceed twice the mean of the residual of the whole sequence, count the change of residual direction in the two periods before and after the abnormal point and remove the group of points with reversed trend distribution to obtain the set of abnormal current residual points. The conflict segment determination submodule: Based on the current residual abnormal point set, compare the polarity of the temperature change direction and the current residual direction at the same time point in the temperature rise abnormal segment and filter the group of inconsistent points, extract the continuous inconsistent point sequence and construct the time index sequence group, filter the continuous conflict point segment with a duration of more than one cycle and record the corresponding sampling index range, and generate the cross-parameter signal response consistency set.

9. The centerless grinding machine full life cycle grinding wheel condition monitoring system according to claim 1, characterized in that, The execution process of the dynamic time warping includes: constructing a distance matrix of two time series based on the temperature change sequence in each heating segment and a preset reference response rhythm sequence; matching the amplitude difference between each sampling point in an increasing time step; calculating the cumulative distance on the matching path and recording the cumulative error sum of feasible paths; selecting the alignment path with the cumulative valley distance as a suitable matching trajectory; extracting the position mapping relationship of each corresponding point pair in the alignment path; and outputting the matching structure distribution area containing the time offset alignment relationship, the corresponding position index group, and the cumulative registration error value.

10. A method for monitoring the condition of grinding wheels throughout their entire lifecycle in a centerless grinder, characterized in that, The centerless grinding machine full life cycle grinding wheel condition monitoring system according to any one of claims 1-9 includes the following steps: S1: Based on the spindle current fluctuation change value, spindle power gradient sequence and periodic signal response interval, the mean difference between adjacent periods is statistically analyzed and the fluctuation distribution change segment is extracted. The power slope change amplitude is calculated and the significant inflection point position is marked. Combined with the periodic boundary difference, the response segment boundary interval is identified and a three-segment time domain boundary block group is established. S2: Based on the three time-domain boundary block groups, obtain the amplitude sequence of each segment and filter the difference mutation region between continuous points, identify the amplitude-maintaining segment in the stable segment and judge the boundary stability, calculate the slope extreme value of continuous points in the decay segment and extract the change concentration area, mark the key points of each segment and construct the cross-segment response trajectory, and generate the segmented dynamic response trajectory. S3: Based on the segmented dynamic response trajectory, the duration of segments with consistent directional changes between adjacent points is statistically analyzed and directional sequences with extreme amplitudes are selected. Fluctuation segments with dense jump frequencies are extracted as candidate segments. The extracted segments are continuously fitted using a residual network and the residual feature structure is output. Sequence structure units composed of directional trend combinations are labeled to generate a set of staged behavioral structure identifiers. S4: Based on the staged behavior structure identifier set, compare the changing trends of cooling flow, spindle speed and feed rate control parameter sequences during the operation process, extract the interval segments with deviations and match them to the time range of the behavior turning point, construct a mapping relationship between control variables and behavior nodes, and generate a parameter behavior deviation trajectory set. S5: Based on the parameter behavior deviation trajectory set, filter continuous segments with sudden increases in the rate of change in the temperature sequence and identify abnormal rising areas; identify the set of points in the current sequence where the fitting residual exceeds the statistical threshold; use dynamic time warping to perform time-series alignment on the multi-source sequences and extract directional variation matching pairs; construct a conflict time period index and mark signal inconsistency blocks; and generate a cross-parameter signal response consistency set.

Images