A spindle defect detection system and method

Abstract

This invention discloses a spindle defect detection system and method, relating to the field of spindle detection technology. It constructs a reliability characterization sequence to depict the time-varying discriminative capability of a first detection method, and identifies a reliability degradation observation segment transitioning from stable to unstable based on this sequence. This segment is used as a unified input to perform adaptive response analysis on multiple candidate detection methods. This transforms the existing method of directly switching based on a single trigger condition into a method that first confirms the actual degradation of discriminative capability and then performs a targeted evaluation. Furthermore, based on the adaptive response results of each candidate detection method in the transition segment, the target detection method is determined, and a smooth switch is completed while ensuring the continuity of the detection process. This avoids erroneous switching caused by local anomalies or short-term fluctuations, and solves the problem of lack of basis for selecting switching targets, thereby improving the stability and accuracy of spindle defect detection.

Description

Technical Field

[0001] This invention relates to the field of spindle inspection technology, and specifically to a spindle defect detection system and method. Background Technology

[0002] As the core moving component in machine tools and various high-speed rotating equipment, the spindle operates under the combined effects of high-speed rotation, alternating loads, thermal coupling, and complex machining disturbances. Its operating condition directly affects machining accuracy, surface quality, equipment stability, and the overall service life of the machine. To promptly detect defects such as bearing wear, rotor imbalance, eccentricity, looseness, rubbing, crack initiation, and abnormal lubrication, existing technologies typically establish defect detection mechanisms based on externally observable information exhibited during spindle operation.

[0003] Specifically, existing spindle defect detection methods generally involve first selecting a single detection method, or a combination of methods with a primary detection method as the core and supplemented by auxiliary methods. Then, corresponding sensors are used to collect relevant physical signals during spindle operation. These signals are then processed, analyzed, and judged to identify the presence and type of defects in the spindle. For example, a common approach is to use vibration sensors to collect vibration signals from the spindle housing, bearing housing, or support components, and then combine this with time-domain, frequency-domain, or time-frequency analysis to extract fault characteristics and identify problems such as imbalance, loosening, and bearing damage. Temperature sensors can also be used to collect temperature rise changes in key parts of the spindle to help determine abnormal friction, poor lubrication, and thermal instability. For electric spindles, changes in the current, voltage, or power of the drive motor can be read to indirectly reflect the mechanical load state. In scenarios more sensitive to early, minor damage, acoustic emission methods are also used to detect localized cracks, minor spalling, or impact events. In other words, existing spindle defect detection methods do not rely on a single technical path but rather offer a variety of selectable detection methods and combinations thereof.

[0004] However, in practical industrial applications, due to considerations of cost control, system complexity, data redundancy, and long-term operational stability, multiple detection methods are typically not deployed in parallel. Instead, a single detection method or a fixed detection structure of "one main detection method plus one auxiliary detection method" is often used, and this detection structure is continuously used to complete spindle defect detection during equipment operation. Building on this, as spindle operating conditions become increasingly complex and the demand for more refined condition monitoring continues to rise, some systems have begun to attempt to pre-set several candidate spindle defect detection methods. During operation, when certain triggering conditions occur, the current spindle defect detection method is adjusted or switched to maintain detection capability under different conditions.

[0005] However, existing dynamic switching detection methods for spindle defects based on trigger conditions typically switch detection methods directly upon the occurrence of an anomaly during the detection process. Their focus is primarily on whether a triggering condition has occurred, rather than assessing whether the original detection method is truly no longer applicable in light of the actual operating state. This can easily lead to erroneous switching, disrupting the continuity and stability of the original detection data, and failing to determine which candidate detection method is more suitable as the target detection method under the current specific conditions. Consequently, the defect detection effect deteriorates, and the detection accuracy decreases. Therefore, how to rationally determine whether a switching detection method is truly necessary during the dynamic switching detection process of spindle defects, and accurately determine the target detection method suitable for the current state when a switching is indeed necessary, remains an unresolved issue in existing technologies. Summary of the Invention

[0006] The purpose of this invention is to solve the problem mentioned in the background art of how to reasonably determine whether it is really necessary to switch the detection mode in combination with the actual situation during the dynamic switching detection of spindle defects, and to accurately determine the target detection mode that is suitable for the current state when it is necessary to switch. Therefore, a spindle defect detection system and detection method are proposed.

[0007] A first aspect of this invention provides a spindle defect detection method. The spindle defect detection is implemented using several detection methods. The detection method corresponding to the spindle defect in the current preset detection cycle is designated as the first detection method. The detection results detected by the first detection method in the current preset detection cycle are analyzed. The method includes: Acquire the independent reference state data formed by the spindle during the reference confirmation stage, and use the independent reference state data as the normal reference reference for the spindle; Within the current preset detection cycle, the current detection data corresponding to the first detection method is obtained, and the current detection data is compared with the normal reference benchmark of the spindle to construct a discrimination reliability characterization sequence that reflects whether the current output of the first detection method is still consistent with the independent benchmark state. Based on the continuous change state of the reliability characterization sequence, the reliability degradation observation section of the first detection method within the current preset detection cycle is determined; Within the reliability degradation observation period, response analysis was performed on multiple pre-defined candidate spindle defect detection methods to obtain the adaptation verification results of each candidate spindle defect detection method; the candidate spindle defect detection method is implemented by at least one sensor in the sensor system; The target detection method is determined based on the adaptation and verification results of each candidate spindle defect detection method, and the judgment power of spindle defect detection is transferred from the first detection method to the target detection method in order to continue spindle defect detection.

[0008] Optionally, the step of constructing a discriminative reliability characterization sequence to reflect the ability of the first detection method to distinguish the spindle state is as follows: The steps of constructing the discriminative reliability characterization sequence include: According to the preset time window length and sliding step size, the detection results output by the first detection method within the current preset detection period are segmented to obtain multiple detection sample windows arranged in chronological order; The signals within each detection sample window are processed by DC removal, amplitude normalization, and time alignment to obtain standardized signal segments; multiple state feature values ​​are extracted from each standardized signal segment and formed into corresponding first feature vectors in a fixed order; Extract second feature vectors of the same type and in the same order as the first feature vector from independent reference state data, and determine the normal reference center based on multiple second feature vectors; Each first feature vector is compared with the normal reference center to obtain the deviation distance of each detection sample window relative to the normal reference center; detection sample windows that meet the preset deviation conditions are selected based on the deviation distance, and the state discrimination vector is determined based on the first feature vector of the selected detection sample window. The offset results of each first feature vector relative to the normal reference center are expanded along the state discrimination vector, and the projection components in the corresponding directions are subtracted to obtain the lateral discretization. The discrimination reliability characterization value of each detection sample window is determined based on the degree of directional expansion and the amount of lateral dispersion. The higher the degree of directional expansion and the lower the amount of lateral dispersion, the higher the corresponding discrimination reliability characterization value. The discrimination reliability characterization values ​​of each detection sample window are arranged in chronological order to obtain the discrimination reliability characterization sequence.

[0009] Optionally, the step of determining the reliability degradation observation section of the first detection method within the current preset detection cycle is as follows: Based on the continuity characteristics, a discrimination capability evolution trajectory is generated to describe the changing trend of the discrimination capability of the first detection method, and based on the discrimination capability evolution trajectory, the observation segment of the reliability decline of the first detection method within the current preset detection cycle is determined.

[0010] Optionally, the step of generating the discrimination capability evolution trajectory to describe the changing trend of the discrimination capability of the first detection method is as follows: According to the chronological order of each discriminant reliability characterization value in the discriminant reliability characterization sequence, the difference between each two adjacent discriminant reliability characterization values ​​is calculated to obtain the adjacent change sequence; For two consecutive adjacent changes in an adjacent change sequence, calculate the degree of consistency in their change directions to obtain the local directional consistency. The first The first discriminant reliability characterization value and the second The arithmetic square root of the product of the local direction consistency is used as the first... Individual discriminant trajectory values; Arrange the discriminative ability trajectory values ​​in chronological order to obtain the discriminative ability evolution trajectory.

[0011] Optionally, the step of determining the reliability degradation observation section of the first detection method within the current preset detection period based on the discriminant capability evolution trajectory is as follows: Based on the chronological order of the discriminant capability trajectory values ​​in the discriminant capability evolution trajectory, the change between each two adjacent discriminant capability trajectory values ​​is calculated to obtain the trajectory change sequence. For each trajectory change in the trajectory change sequence, the portion less than zero is extracted as the descent component, thus obtaining the trajectory descent sequence. For each trajectory change in the trajectory change sequence, the portion greater than zero is extracted as the recovery component, thus obtaining the trajectory recovery sequence. The cumulative descent is obtained by summing the descent amounts of each trajectory in chronological order. The cumulative descent is obtained by summing the descent amounts of each trajectory in chronological order. Based on the cumulative decrease and cumulative increase, calculate the attenuation dominance ratio at each location; Based on the trajectory values ​​of each discrimination capability and the corresponding attenuation dominance ratio, the discrimination failure pressure value corresponding to each location is calculated; According to the preset sliding section length, the failure pressure value is continuously statistically analyzed, and the proportion of continuous increase of the failure pressure in each sliding section is calculated to obtain the failure penetration rate. From the discriminant capability evolution trajectory corresponding to the stable operation stage of the main shaft, select multiple stable trajectory values ​​and average them to obtain a stable reference value; From the failure penetration rates corresponding to the stable operation phase of the spindle, select multiple stable penetration rates and calculate the average to obtain the penetration reference value; The continuous intervals in the discrimination capability evolution trajectory that satisfy the condition that the current trajectory value is less than the stable reference value and the current failure penetration rate is greater than the penetration reference value are identified as the reliability decline observation segments during the transition from a stable discrimination state to an unstable discrimination state.

[0012] Optionally, the steps for performing response analysis on multiple pre-defined candidate spindle defect detection methods to obtain the adaptation response results of each candidate detection method in the reliability degradation observation section are as follows: Response analysis is performed on multiple pre-defined candidate spindle defect detection methods. Boundary crossing stability value and result path shrinkage rate value are calculated. The boundary crossing stability value and result path shrinkage rate value are added together to obtain the adaptation response value of each candidate detection method.

[0013] Optionally, the steps for calculating the boundary crossing stability value are as follows: Obtain the discrimination result sequence and corresponding discrimination boundary of the candidate spindle defect detection method within the reliability degradation observation section. Subtract the discrimination boundary from each discrimination result and divide it by the sum of the maximum absolute value of all offset values ​​within the reliability degradation observation section and a preset positive number to obtain the signed offset sequence of the boundary. Based on the continuity of the signs of each offset value in the signed offset sequence of the boundary, sampling positions with the same sign and adjacent to each other are divided into multiple continuous segments on the same side and the number of boundary crossings between adjacent segments is recorded. At the same time, the number of sampling points corresponding to each continuous segment on the same side is counted as the segment length. For each continuous segment on the same side, add a preset positive number to the absolute value of each offset value in the segment, multiply them together, and take the square root of the multiplication result according to the length of the corresponding segment to obtain the segment anchorage degree of each continuous segment on the same side. For any two adjacent consecutive segments on the same side, the length symmetry is obtained by dividing the smaller of the two segment lengths by the sum of the larger of the two segment lengths and a preset positive number. The anchorage symmetry is obtained by dividing the smaller of the two segment anchorages by the sum of the larger of the two segment anchorages and a preset positive number. The length symmetry is then multiplied by the anchorage symmetry to obtain the mirror crossing value. The boundary hesitation value is obtained by dividing the mirror crossing value by the sum of the geometric mean of the corresponding two segment anchorages and a preset positive number. The normalized boundary hesitation value is obtained by dividing the boundary hesitation value by the sum of the boundary hesitation value and a constant. The overall boundary hesitation intensity is obtained by subtracting one from each of the normalized boundary hesitation values, adding a preset positive number, multiplying the result, taking the square root of the result according to the number of boundary crossings, and then subtracting one. When the number of boundary crossings is zero, the overall boundary hesitation intensity is recorded as zero. The boundary crossing density is obtained by dividing the number of boundary crossings by the total number of sampling points minus one. The boundary crossing stability value is obtained by multiplying the overall boundary hesitation strength minus one by the boundary crossing density minus one and taking the square root.

[0014] Optionally, the calculation steps for the resulting path shrinkage rate value are as follows: Obtain the discrimination result sequence of candidate detection methods within the reliability degradation observation section, and determine the main direction based on the magnitude relationship between the discrimination result at the end of the sequence and the initial discrimination result; The discrimination results are normalized according to the main direction to obtain the normalized path sequence of the main direction; The leading edge occupancy sequence is obtained by taking the maximum value of each point in the normalized path sequence of the main direction up to the current position in chronological order. Subtract the corresponding value in the normalized path sequence of the main direction from the leading edge occupancy value of each position to obtain the shrinkage gap sequence; In the leading edge occupancy sequence, the position where the current leading edge occupancy value is greater than the previous leading edge occupancy value is determined as the leading edge anchor point, and the interval between adjacent leading edge anchor points is determined as the anchor point segment; For each anchor point segment, add a preset positive number to each retraction gap value inside the segment, multiply them together, and then take the corresponding power according to the number of internal positions to obtain the retraction core value. Divide the number of internal positions of the segment by the total span between adjacent front anchor points to obtain the retraction occupancy ratio. Then multiply the retraction core value and the retraction occupancy ratio and take the square root to obtain the transfer value. For the transfer value of all anchor point segments, subtract each transfer value from the constant and then multiply by the preset positive number. Then, take the corresponding power according to the number of anchor point segments, and finally subtract the processing result from the constant to obtain the overall path transfer strength. The end value is obtained by taking the end value of the normalized path sequence in the main direction and dividing it by the sum of the end value of the leading edge occupant sequence and a preset positive number; Multiply the result obtained by subtracting the overall path yield strength from the constant one by the end retention value and take the square root of the product to obtain the path shrinkage rate.

[0015] A second aspect of the present invention provides a spindle defect detection system, the system comprising: Reference datum module: acquires independent datum state data formed by the spindle during the datum confirmation stage, and uses the independent datum state data as the normal reference datum of the spindle; The discrimination characterization module: within the current preset detection cycle, it acquires the current detection data corresponding to the first detection method, and compares the current detection data with the normal reference benchmark of the spindle to construct a discrimination reliability characterization sequence that reflects whether the current output of the first detection method is still consistent with the independent benchmark state. Observation section module: Based on the continuous change state of the reliability characterization sequence, determine the reliability decline observation section of the first detection method within the current preset detection cycle; Adaptation Response Module: Within the reliability degradation observation period, response analysis is performed on multiple pre-defined candidate spindle defect detection methods to obtain the adaptation verification results of each candidate spindle defect detection method; the candidate spindle defect detection method is implemented by at least one sensor in the sensor system; Spindle defect detection module: Determines the target detection method based on the adaptation verification results of each candidate spindle defect detection method, and transfers the judgment power of spindle defect detection from the first detection method to the target detection method to continue spindle defect detection.

[0016] The beneficial effects of this invention are: This invention proposes a spindle defect detection system and method. By constructing a discrimination reliability characterization sequence reflecting the discrimination capability of a first detection method, the system can characterize the change process of the discrimination capability of the detection method over time. Based on the discrimination reliability characterization sequence, it identifies the reliability degradation observation segment transitioning from a stable discrimination state to an unstable discrimination state. Using this transition segment as a unified input, it performs adaptive response analysis on multiple candidate detection methods. This transforms the existing method of directly switching detection methods based on a single trigger condition into a method that first determines whether the discrimination capability of the first detection method has actually decayed, and then conducts targeted evaluation of candidate detection methods during the actual failure process. Furthermore, based on the adaptive response results of each candidate detection method in the reliability degradation observation segment, it determines the target detection method and completes the switching while maintaining the continuity of the detection process. This effectively avoids the problem of false switching caused by local anomalies or short-term fluctuations, and solves the problem of lack of basis for selecting the switching target in the prior art. Thus, it improves the stability and accuracy of spindle defect detection while ensuring the continuity of detection data. Attached Figure Description

[0017] Figure 1 This is a flowchart of a spindle defect detection method provided in an embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0019] This invention provides a method for detecting spindle defects. See also... Figure 1 Specifically: Spindle defect detection is achieved through several detection methods. The detection method corresponding to the spindle defect in the current preset detection cycle is denoted as the first detection method. The detection results of the first detection method in the current preset detection cycle are then analyzed. The specific steps are as follows: S1: Obtain independent reference state data formed by the spindle during the reference confirmation stage, and use the independent reference state data as the normal reference reference for the spindle. The independent reference state data is collected under at least one stage: spindle factory calibration, no-load test run, standard test piece processing, post-repair acceptance, or manual verification to confirm that there are no abnormalities. S2: Within the current preset detection cycle, obtain the current detection data corresponding to the first detection method, and compare the current detection data with the normal reference benchmark of the spindle to construct a discrimination reliability characterization sequence to reflect whether the current output of the first detection method is still consistent with the independent benchmark state; S3: Based on the continuous change state of the reliability characterization sequence, determine the reliability decline observation section of the first detection method within the current preset detection cycle; the reliability decline observation section is used to characterize the observation section where the output result of the first detection method deviates continuously from the normal reference benchmark of the spindle, but the complete failure of the first detection method has not yet been directly confirmed; the candidate spindle defect detection method is implemented by at least one sensor in the sensor system; S4: Within the reliability degradation observation period, response analysis is performed on multiple pre-set candidate spindle defect detection methods to obtain the adaptation verification results of each candidate spindle defect detection method. S5: Determine the target detection method based on the adaptation verification results of each candidate spindle defect detection method, and transfer the judgment power of spindle defect detection from the first detection method to the target detection method to continue spindle defect detection.

[0020] In one embodiment, S1: Obtain independent reference state data formed by the spindle during the reference confirmation stage, and use the independent reference state data as the normal reference reference for the spindle. The independent reference state data is collected under at least one stage: spindle factory calibration, no-load test run, standard test piece processing, post-repair acceptance, or manual verification confirming no abnormalities. Specifically, during data collection, a "benchmark confirmation process" can be set up before the spindle enters formal defect detection: First, the spindle is placed in any stable stage among factory calibration, no-load test run, standard test piece machining, post-repair acceptance, or manual verification confirming no abnormalities. The CNC system records the speed command, load, current, feed status, and machining program segment information for that stage. Simultaneously, vibration sensors, temperature sensors, and acoustic emission sensors installed on the spindle bearing housing, spindle box, or electric spindle housing synchronously collect vibration signals, temperature rise signals, impact signals, or acoustic emission signals. During data collection, unstable segments such as spindle start-up, shutdown, speed increase / decrease, tool change, and sudden feed changes are eliminated. Only continuous time windows with stable speed, load changes within allowable range, no alarm records, and spindle status confirmed as normal by manual verification or acceptance rules are retained. The sensor signals within these time windows are then time-aligned, denoised, and standardized to form independent benchmark status data. For example, during the post-maintenance acceptance phase, the spindle can be run unloaded at 3000 r / min, 6000 r / min, and 9000 r / min for several minutes respectively. After each speed stabilizes, the corresponding root mean square value of vibration, characteristic frequency band energy, bearing temperature rise, current fluctuation, and number of acoustic emission pulses can be collected. The collected results that are confirmed to be without abnormalities can be used as a reference benchmark for subsequent judgment of the normal state of the spindle.

[0021] S2: Within the current preset detection cycle, the steps of acquiring the current detection data corresponding to the first detection method, comparing the current detection data with the normal reference benchmark of the spindle, and constructing a reliability characterization sequence to reflect whether the current output of the first detection method is still consistent with the independent benchmark state are as follows: According to the preset time window length and sliding step size, the detection results output by the first detection method within the current preset detection period are segmented to obtain multiple detection sample windows arranged in chronological order. Among them, when the first detection method is a single detection method, each detection sample window contains only the signal data collected by the sensor corresponding to the single detection method within the corresponding time period. When the first detection method is a detection combination consisting of a main detection method and an auxiliary detection method, each detection sample window contains the signal data collected by the sensor corresponding to the main detection method and the sensor corresponding to the auxiliary detection method within the same time period. The signals within each detection sample window are processed by DC removal, amplitude normalization, and time alignment to obtain corresponding standardized signal segments, which are then saved in a preset order. For each detection sample window, multiple state feature values ​​are extracted from the standardized signal segment, and arranged or spliced ​​in a fixed order to form a corresponding feature vector, which is denoted as the first feature vector; the second feature vector is denoted as the third feature vector. The first feature vector corresponding to each detection sample window Represented as: ;in, Indicates the first The first detection sample window corresponds to the first Each state feature value, This represents the dimension of the eigenvector. Indicates transpose; Feature vectors corresponding to multiple detection sample windows are selected from independent baseline state data. Each feature vector corresponding to a detection sample window is denoted as a second feature vector. The mean of all second feature vectors is used as the normal reference center, i.e., the average value at each position in the second feature vector is calculated to obtain the normal reference center. Its expression is: In the formula, n represents the number of test sample windows used to construct the normal reference center. Indicates the first The feature vector corresponding to each historical detection sample window; Subtract the value at the corresponding position of the normal reference center from the value at each position in each first feature vector, and then square, sum, and square root the differences to obtain the deviation distance corresponding to each detection sample window. ; In the formula, Indicates the first The deviation distance of each detection sample window Indicates the first The first detection sample window corresponds to the first Each state feature value, Indicates the normal reference center number The value at each position, Indicates the number of feature dimensions; Sort the samples by deviation distance from largest to smallest, select a predetermined number of detection sample windows with the largest deviation distance (the predetermined number is at least [number]), and average the values ​​at each position in their feature vectors to obtain the deviation from the reference center. The calculation formula is: In the formula, Indicates the number of selected test sample windows. Indicates the number after sorting by deviation distance. The feature vector corresponding to each detection sample window; Values ​​at positions deviating from the reference center Subtract the values ​​corresponding to the normal reference center from each position. This yields the state distinction vector; Subtract the normal reference center from the feature vector of each detection sample window to obtain the offset result. Multiply this offset result positionally by the state discrimination vector and sum them to obtain the directional expansion amount, and normalize it according to the overall magnitude of the state discrimination vector; the calculation formula is: In the formula, Indicates the first The directional expansion amount of each detection sample window The state discrimination vector is represented by the first... The value at each position, This represents a preset positive number to prevent the denominator from being zero; it is generally set to a value of... ; The projection component is obtained by multiplying the directional expansion amount by the state discrimination vector position by position. This projection component is then subtracted from the offset result to obtain the remaining offset result. The remaining offset result is squared, summed, and squared, and then normalized to obtain the lateral discretization amount. The calculation formula is as follows: In the formula, Indicates the first The lateral dispersion of each detection sample window; This represents a preset positive number to prevent the denominator from being zero; it is generally set to a value of [value missing]. ; The discriminative reliability characterization value for each detection sample window is obtained by jointly calculating the directional expansion and lateral discretization, where a larger directional expansion and a smaller lateral discretization result correspond to a larger value. The corresponding formula is: In the formula, Indicates the first The discrimination reliability characterization value of each detection sample window; The discrimination reliability characterization values ​​of each detection sample window are arranged in chronological order to obtain the discrimination reliability characterization sequence. , This represents the reliability characterization sequence. Indicates the number of sample windows to be detected.

[0022] It should be noted that the above steps are illustrated with an example. For instance, if the spindle is stably machining at 6000 r / min within the current preset detection cycle, the vibration signal collected by the first detection method is divided into windows every 0.5 seconds and slid every 0.1 seconds to obtain 10 detection sample windows. From each sample window, four state features are extracted in a fixed order: root mean square value of vibration, peak value, characteristic frequency band energy, and temperature rise rate, forming 10 first feature vectors. At the same time, 20 reference sample windows are extracted from the no-load test run reference data that has been confirmed to be without abnormalities in S1. The same four types of features are also extracted and averaged to obtain the normal reference center, for example, the normal reference center is "2.0, 5.0, 1.5, 0.2". The system compares the current 10 first feature vectors with the normal reference center and calculates the deviation distance. Assuming the 6th, 7th, and 8th windows have the largest deviation distances, the system averages the feature vectors of these three sample windows to obtain the deviation from the reference center, for example, "3.2, 7.1, 2.4, 0.6". Then, it subtracts the normal reference center from the deviation to obtain the state discrimination vector "1.2, 2.1, 0.9, 0.4", representing the most significant deviation direction of the current first detection method relative to the normal state. Subsequently, for each current detection sample window, the system first calculates its offset relative to the normal reference center, then determines the extent of expansion of this offset in the direction of the state discrimination vector and the lateral dispersion after deviating from that direction. For example, the feature vector of the 7th window is "3.1, 6.9, 2.3, 0.55", and its offset is basically in the same direction as the state discrimination vector, with a relatively large lateral dispersion. The large opening and small lateral dispersion of the window indicate that although it deviates from the normal state, the deviation direction is concentrated, thus a high discrimination reliability characterization value of 0.82 can be obtained. The feature vector of the 9th window is "2.8, 5.2, 3.6, 0.15". Although it also deviates from the normal reference center, it is mainly characterized by a sudden increase in one feature and the failure of other features to follow suit. After deducting the state-distinguishing direction, there is still a large lateral residue, thus a lower discrimination reliability characterization value of 0.43 is obtained. Finally, the characterization values ​​of the 10 sample windows are arranged in chronological order as "0.88, 0.84, 0.81, 0.79, 0.76, 0.83, 0.82, 0.70, 0.43, 0.46". This sequence is used as the discrimination reliability characterization sequence to reflect whether the current output of the first detection method and the independent normal reference benchmark still maintain a consistent direction and a concentrated deviation relationship.

[0023] It's important to note that the core advantage of constructing the reliability characterization sequence using the above method is that, unlike moving averages and standard deviations which only consider the magnitude of signal value changes or fluctuations over time, this method first compares the current detection sample window with the normal reference center formed by the independent baseline state. It then determines whether the deviation unfolds along an interpretable state-distinguishing direction and whether there are chaotic lateral residuals in the deviation direction. This allows it to distinguish between ordered deviations under real-state changes and disordered deviations caused by output drift, noise disturbances, or discrimination distortion in the first detection method. In other words, moving averages and standard deviations can only indicate whether the signal has increased or fluctuated drastically, but they cannot indicate whether this change still maintains structural consistency with the normal baseline state. This step, through the directional expansion, lateral dispersion, and the projection-dissociation relationship between the directional expansion and lateral dispersion, can determine whether the current output of the first detection method is still changing in a reasonable direction around the independent baseline state. Therefore, it is more suitable for characterizing whether the detection method itself is still reliable, avoiding mistaking normal operating condition fluctuations for detection failure, and also avoiding mistaking abnormal outputs with seemingly stable values ​​but deviating from the baseline for reliability.

[0024] In one embodiment, S3: Based on the continuous change state of the reliability characterization sequence, determine the reliability degradation observation segment of the first detection method within the current preset detection cycle; the step of using the reliability degradation observation segment to characterize the observation segment in which the output result of the first detection method deviates continuously from the normal reference benchmark of the spindle, but the complete failure of the first detection method has not yet been directly confirmed, is as follows: Based on the continuity characteristics, a discrimination capability evolution trajectory is generated to describe the changing trend of the discrimination capability of the first detection method, and based on the discrimination capability evolution trajectory, the observation segment of the reliability decline of the first detection method within the current preset detection cycle is determined.

[0025] In one implementation, the step of generating a discrimination capability evolution trajectory based on continuous features to describe the changing trend of the discrimination capability of the first detection method is as follows: Based on the chronological order of the discriminant reliability characterization values ​​in the discriminant reliability characterization sequence, the difference between every two adjacent discriminant reliability characterization values ​​is calculated to obtain the adjacent change sequence; the corresponding calculation formula is: ; Indicates the first Each adjacent change; For two consecutive adjacent changes in an adjacent change sequence, the degree of consistency in their change directions is calculated to obtain the local directional consistency degree; the formula for calculation is: In the formula, Indicates the first Local directional consistency This represents a preset positive number to prevent the denominator from being zero; it is generally set to a value of [value missing]. ; The first The first discriminant reliability characterization value and the second The arithmetic square root of the product of the local direction consistency is used as the first... Individual discriminant trajectory value The calculation formula is as follows: In the formula, Indicates the first Individual discriminant trajectory values; Arrange the discriminant ability trajectory values ​​in chronological order to obtain the discriminant ability evolution trajectory. ; In the formula, This represents the evolutionary trajectory of discriminative ability.

[0026] It should be noted that, to illustrate the above steps with an example, when determining the reliability characterization sequence as "0.86, 0.82, 0.79, 0.73, 0.68, 0.66", the adjacent changes in the first step can be calculated as "-0.04, -0.03, -0.06, -0.05, -0.02", indicating whether the characterization value decreases or increases between adjacent time windows and the magnitude of the decrease. In the second step, when taking two consecutive adjacent changes "-0.04" and "-0.03", since both are decreasing and their values ​​are close, the corresponding local direction consistency is close to 1. However, when taking "-0.06" and "-0.02", the results are different. When the difference in the magnitude of the decrease is greater, the corresponding local directional consistency is smaller. In the third step, if the representation value corresponding to a certain position is 0.79 and the local directional consistency is 0.91, the discriminative ability trajectory value obtained after combining the two remains at a high level, indicating that although the discriminative ability of this detection method is changing, the overall change is smooth and locally stable. However, if the representation value corresponding to another position is 0.73 and the local directional consistency is 0.55, the obtained trajectory value will decrease significantly, indicating that the discriminative ability of this segment begins to become unstable. In the fifth step, after arranging the trajectory values ​​obtained from each position in order, a discriminative ability evolution trajectory can be formed.

[0027] The core advantage of constructing the discriminant capability evolution trajectory using this method is that it doesn't simply look at whether the discriminant reliability characterization value itself is high or low, nor does it only consider whether a single decrease is significant. Instead, it incorporates the "current level of the characterization value," "whether adjacent changes are continuous and consistent," and "whether local changes have a smooth transition relationship" into the same trajectory value. This allows the discriminant capability evolution trajectory to reflect the continuous change process of the first detection method's reliability from stability to decline. This avoids two types of misjudgments: first, when a characterization value decreases occasionally but the direction is inconsistent or the local changes are discontinuous, it will not be directly identified as a reliability decline; second, although some characterization values ​​do not drop sharply, if they show a slow decline in the same direction over multiple consecutive windows, this will gradually be reflected in the trajectory. Compared to directly using moving averages, single-point thresholds, or adjacent difference judgments, this method is better able to identify the reliability decay trend formed by "persistence, directional consistency, and local continuity," making it more suitable for determining subsequent reliability decline observation segments, rather than misjudging instantaneous noise, local rebounds, or single-time operating condition fluctuations as failures of the first detection method.

[0028] In one implementation, the step of determining the reliability degradation observation segment of the first detection method within the current preset detection period based on the discriminative capability evolution trajectory is as follows: Based on the chronological order of the discriminant capability trajectory values ​​in the discriminant capability evolution trajectory, the change between each two adjacent discriminant capability trajectory values ​​is calculated to obtain the trajectory change sequence. For each trajectory change in the trajectory change sequence, the portion less than zero is extracted as the descent component, thus obtaining the trajectory descent sequence. For each trajectory change in the trajectory change sequence, the portion greater than zero is extracted as the recovery component, thus obtaining the trajectory recovery sequence. The cumulative descent is obtained by summing the descent amounts of each trajectory in chronological order. The cumulative descent is obtained by summing the descent amounts of each trajectory in chronological order. Based on the cumulative decrease and cumulative increase, calculate the attenuation dominance ratio at each location; the formula is: In the formula, Indicates the first The dominant attenuation ratio, This represents a preset positive number to prevent the denominator from being zero; it is generally set to a value of [value missing]. ; and They represent the first The cumulative decrease and the cumulative decrease; Based on the discrimination capability trajectory values ​​and the corresponding attenuation dominance ratios, the discrimination failure pressure values ​​at each location are calculated; the calculation formula is: In the formula, Indicates the first One failure pressure value; Based on the preset sliding section length, the failure pressure value is continuously statistically analyzed, and the proportion of continuous increase in failure pressure within each sliding section is calculated to obtain the failure penetration rate; the calculation formula is: In the formula, Indicates the first Failure penetration rate with each location as the endpoint Indicates the preset sliding segment length. This represents an indicator function, which takes the value 1 when the condition inside the parentheses is true, and takes the value 0 otherwise. From the discriminant capability evolution trajectory corresponding to the stable operation phase of the spindle, multiple stable trajectory values ​​are selected and averaged to obtain a stable reference value. The formula for calculating the stable reference value is: In the formula, This indicates the number of trajectory values ​​during the stable phase. Indicates the stable phase One trajectory value; From the failure penetration rates corresponding to the stable operation phase of the spindle, multiple stable penetration rates are selected and averaged to obtain the penetration reference value; the formula for calculating the penetration reference value is: , Indicates the reference value throughout. This indicates the penetration rate during the stable phase. Indicates the stable phase A penetration rate; The continuous interval in the discrimination capability evolution trajectory that satisfies "the current discrimination capability trajectory value is less than the stable reference value and the current failure penetration rate is greater than the penetration reference value" is determined as the reliability degradation observation segment of the first detection method within the current preset detection period; the reliability degradation observation segment is represented as: In the formula, Ψ represents the observation section of reliability degradation. Indicates the starting position of the reliability degradation observation segment. This indicates the end point of the reliability degradation observation segment.

[0029] The core advantage of this approach is that it doesn't simply judge the reliability decline of the first detection method based on whether the discrimination capability trajectory value at a certain moment is below a threshold. Instead, it simultaneously examines the cumulative decline and recovery of the trajectory value over time, as well as whether the failure pressure continuously increases within the continuous sliding segment. This allows it to distinguish between "accidental decline," "short-term jitter," "local rebound," and "persistent reliability decay." Specifically, the cumulative decline and cumulative recovery can reflect whether the current change is dominated by decline or recovery. The decay dominance ratio can avoid misjudging a one-off fluctuation as a reliability decline, further clarifying the failure pressure value. The method combines the current trajectory level with the downward trend, and the failure penetration rate is used to confirm whether this pressure has been continuously increasing over a period of time. Finally, it is compared with the stable reference value and penetration reference value formed during the stable operation phase of the spindle, so that the identified reliability decline observation section has the dual basis of "below the normal stable level" and "failure pressure continues to penetrate". Therefore, this method is more suitable for identifying the continuous process of the first detection method gradually transitioning from a stable and usable state to a reliability decline state than simple threshold judgment, moving average judgment or single change judgment, and can reduce misjudgments caused by industrial signal noise, short-term load changes or local rebound.

[0030] In one embodiment, S3: The step of taking the reliability degradation observation section as input, performing response analysis on multiple pre-defined candidate spindle defect detection methods respectively, and obtaining the adaptation response results of each candidate detection method in the reliability degradation observation section is as follows: Response analysis is performed on multiple pre-defined candidate spindle defect detection methods. Boundary crossing stability value and result path shrinkage rate value are calculated. The boundary crossing stability value and result path shrinkage rate value are added together to obtain the adaptation response value of each candidate detection method.

[0031] In one implementation, the steps for calculating the boundary crossing stability value are as follows: Obtain the discrimination result sequence of a candidate spindle defect detection method within the reliability degradation observation period, and determine the discrimination boundary corresponding to the candidate spindle defect detection method; Subtract the discrimination boundary from each discrimination result in the discrimination result sequence to obtain the boundary offset value corresponding to each sampling position; then take the maximum value among the absolute values ​​of all boundary offset values ​​in the reliability decline observation section, and divide each boundary offset value by the sum of the maximum value and the preset positive number to obtain the signed boundary offset sequence, so that the discrimination results of different candidate detection methods are uniformly mapped to the same boundary offset scale; Based on the sign continuity of the offset values ​​of each sampling position in the signed offset sequence of the boundary, sampling positions with the same sign and adjacent to each other are divided into the same continuous segment on the same side, resulting in multiple continuous segments on the same side arranged in time order, and the connection position between two adjacent continuous segments on the same side is recorded as a boundary crossing. Count the number of sampling locations contained in each continuous segment on the same side, and determine the segment length of the corresponding continuous segment on the same side based on the number of sampling locations. For each continuous segment on the same side, the absolute value of the signed offset value of the boundary corresponding to each sampling position in the segment is extracted. A preset positive number is added to each absolute value and then multiplied together. The result of the multiplication is then processed by taking the corresponding power according to the segment length of the continuous segment on the same side to obtain the segment anchorage degree of the continuous segment on the same side, which characterizes the stability of the continuous segment on the same side as a whole away from the discrimination boundary. For each pair of adjacent continuous segments on the same side, first, the smaller of the two segment lengths is divided by the sum of the larger of the two segment lengths and a preset positive number to obtain the length symmetry; then, the smaller of the two segment anchorage degrees is divided by the sum of the larger of the two segment anchorage degrees and a preset positive number to obtain the anchorage symmetry; finally, the length symmetry and the anchorage symmetry are multiplied to obtain the mirror crossing value of the pair of adjacent continuous segments on the same side, which characterizes the degree of symmetry of the dwelling states on both sides before and after a boundary crossing; For each pair of adjacent continuous segments on the same side, the corresponding mirror crossing value is first divided by the sum of the geometric mean of the anchorage of the pair of adjacent continuous segments on the same side and a preset positive number to obtain the boundary hesitation value. Then, the boundary hesitation value is divided by the sum of the boundary hesitation value and a constant one to obtain the normalized boundary hesitation value, so that the degree of hesitation corresponding to each boundary crossing is compressed to between zero and one. For the normalized boundary hesitation value corresponding to all boundary crossings, subtract each normalized boundary hesitation value from the constant and add a preset positive number. Multiply the results together and take the corresponding power root of the multiplication result according to the number of boundary crossings. Then subtract the processed result from the constant to obtain the overall boundary hesitation intensity. When the number of boundary crossings is zero, the overall boundary hesitation intensity is directly recorded as zero. Divide the number of boundary crossings by the total number of sampling locations within the observation segment where reliability decreases by one to obtain the boundary crossing density, which characterizes the frequency of boundary crossings within a unit segment length. Subtract the overall boundary hesitation intensity and boundary crossing density from the constant, multiply the two, and take the square root of the product to obtain the boundary crossing stability value corresponding to the candidate spindle defect detection method. The greater the boundary crossing stability value, the less boundary crossing, the more stable the stay on both sides of the boundary, and the weaker the hesitation near the boundary.

[0032] It should be noted that the data involved in the above boundary crossing stability value calculation process all come from the actual detection output results of each candidate spindle defect detection method within the reliability degradation observation section. The discrimination result sequence is obtained directly from the sensor acquisition signal of the corresponding detection method through its existing discrimination model or feature judgment logic. For example, the vibration detection method acquires vibration signals through a vibration sensor and obtains the discrimination value through time-frequency analysis; the temperature detection method acquires temperature rise data through a temperature sensor and obtains the discrimination value through threshold or rate of change judgment; and the electrical signal detection method obtains the load change discrimination result through current or power signal analysis. The reliability degradation observation section is identified in the previous steps based on the discrimination reliability characterization sequence and discrimination capability evolution trajectory, and the detection data within the corresponding time range is extracted accordingly. The discrimination boundary is the discrimination threshold that each candidate detection method pre-sets in its existing detection model or is calibrated through historical normal operation data. Therefore, the above value calculation can be completed without introducing additional data or external information.

[0033] It should be noted that the boundary crossing stability value is essentially used to measure whether a candidate detection method can still maintain stable, consistent and unwavering discrimination behavior when making judgments around its own discrimination boundary within the observation section of reliability degradation; it does not simply look at whether the detection method has crossed the boundary, but rather whether it exhibits unstable phenomena such as frequent back-and-forth jumps, brief stops on both sides, and lingering close to the boundary when crossing near the boundary. A higher boundary crossing stability value indicates fewer boundary crossings by the detection method within the transition zone. Even if crossings occur, the dwell time on both sides of the boundary is more stable, the dwelling area is more asymmetrical, and the overall distance from the boundary is greater. This suggests that the detection method has a clearer judgment of the principal axis state at this stage and is less prone to fluctuating between "normal" and "abnormal." Therefore, it indicates that the detection method still has strong boundary discrimination stability in the current reliability decline observation zone and is more suitable as the target detection method after switching. Conversely, a smaller value indicates that the detection method's judgment result is close to the boundary for a long time and frequently crosses back and forth. For example, the judgment result repeatedly shows "slightly above the threshold - slightly below the threshold - above the threshold again" within a certain period. This usually means that the detection method's representation of the current principal axis state has become hesitant and unreliable. If this detection method is still used at this time, it is easy to cause misjudgment or unstable detection after switching.

[0034] In one implementation, the steps for calculating the result path shrinkage rate are as follows: Obtain the discrimination result sequence of candidate spindle defect detection methods within the reliability degradation observation section, and compare the magnitude pattern of the discrimination result at the end of the discrimination result sequence with that at the beginning of the discrimination result sequence; when the discrimination result at the end is not less than the discrimination result at the beginning of the discrimination result, determine the direction of increasing the discrimination result as the main direction; when the discrimination result at the end is less than the discrimination result at the beginning of the discrimination result, determine the direction of decreasing the discrimination result as the main direction. According to the main direction, the discrimination results within the reliability degradation observation section are normalized. Specifically, when the increasing direction is the main direction, the minimum value in the discrimination result sequence is subtracted from each discrimination result, and then divided by the sum of the difference between the maximum and minimum values ​​in the discrimination result sequence and a preset positive number. When the decreasing direction is the main direction, the maximum value in the discrimination result sequence is subtracted from each discrimination result, and then divided by the sum of the difference between the maximum and minimum values ​​in the discrimination result sequence and a preset positive number, thus obtaining the main direction normalized path sequence. According to the time sequence, the maximum value up to the current position is extracted point by point in the normalized path sequence of the main direction to obtain the front edge occupancy sequence, so that the front edge occupancy value corresponding to each position represents the foremost position that the result path has advanced to before that position; Subtract the corresponding value in the normalized path sequence from the leading edge position value at each position in the leading edge position sequence to obtain the shrinkage gap sequence, which represents the degree of retreat at each position relative to the leading edge position that has been reached. In the leading edge occupancy sequence, the position where the current leading edge occupancy value is greater than the previous leading edge occupancy value is determined as the leading edge anchor point, and all leading edge anchor points are arranged in chronological order; the interval between any two adjacent leading edge anchor points is determined as an anchor point segment, so as to decompose the entire result path into multiple continuous structural units of "refresh leading edge - shrinkage occurs - refresh leading edge again"; For each anchor point segment, extract the retraction gap value corresponding to each position inside the anchor point segment, add a preset positive number to each retraction gap value and multiply them together, and then take the corresponding power root of the multiplication result according to the number of positions inside the anchor point segment to obtain the retraction core value of the anchor point segment; when there are no internal positions in an anchor point segment, the retraction core value of the anchor point segment is directly recorded as zero. For each anchor point segment, the number of internal positions within that anchor point segment is counted, and the number of internal positions is divided by the total span between the two adjacent leading-edge anchor points to obtain the retraction occupancy ratio of that anchor point segment, which represents the proportion of the retraction portion in the entire leading-edge refresh span of that anchor point segment. For each anchor point segment, the corresponding retraction core value is multiplied by the retraction occupancy ratio, and the square root of the product is taken to obtain the transfer value of that anchor point segment, so as to simultaneously characterize the depth of retraction and the length of retraction within that anchor point segment. For the transfer value of all anchor point segments, subtract each transfer value from the constant and add a preset positive number. Multiply the results together and take the corresponding power according to the number of anchor point segments. Then subtract the processed result from the constant to obtain the overall path transfer strength. When the number of anchor point segments is less than or equal to zero, the overall path transfer strength is directly recorded as zero. Take the end value of the normalized path sequence in the main direction, and divide the end value by the sum of the end value of the leading edge occupancy sequence and a preset positive number to obtain the end retention value, which characterizes the degree of retention of the leading edge position by the resulting path at the end of the reliability decline observation section. Subtract the overall path yield strength from the constant, multiply the result by the end retention value, and then take the square root of the product to obtain the result path shrinkage rate. The larger the result path shrinkage rate is, the weaker the shrinkage, the less the yield, and the more fully the end retains the front result during the advancement process.

[0035] It should be noted that the data involved in the calculation of the above-mentioned path shrinkage rate value all come from the continuous discrimination output results of the candidate detection methods within the reliability degradation observation section. Specifically, the acquisition method is as follows: First, during the spindle operation, the raw operating signal is collected in real time through corresponding sensors (such as vibration sensors, displacement sensors, current sensors, or temperature sensors), and the raw operating signal is preprocessed, including noise reduction, drift reduction, and resampling processing with a unified time reference. Then, the preprocessed signal is input into the corresponding candidate detection method, which then uses its predetermined discrimination model (such as a threshold discrimination model, a state scoring model, or anomaly detection model) to make the judgment. The model outputs the discrimination result values ​​time-by-time, thus forming a discrimination result sequence arranged in chronological order. Based on this, according to the preset discrimination failure identification rules (e.g., the discrimination result is close to the discrimination boundary or frequently crosses the boundary in multiple consecutive sampling points), the corresponding time segment is automatically extracted as the reliability decline observation segment, and all discrimination results in the segment are extracted to form the discrimination result sequence. The discrimination result can be a normalized state score value or anomaly degree value, and its numerical range has been uniformly limited in the preprocessing or model output stage, so that the calculation of the subsequent result path shrinkage rate value can be completed without introducing additional external data or unknown factors.

[0036] It should be noted that the result path retraction rate essentially measures whether a candidate detection method, within the reliability decline observation period, can maintain its established discrimination trend after its judgment results advance along the main direction, without frequently exhibiting the phenomenon of "advancing a certain distance in the abnormal direction and then retreating again." It does not reflect the level of the judgment result at a single moment, but rather the stable advancement capability of the entire judgment result path over time. The larger the result path retraction rate, the smaller the overall retraction gap, the shorter the retraction duration, and the lower the degree of path concession in the transition section of the candidate detection method. On the other hand, the final judgment result retains the previously reached leading position more fully. Therefore, it can better indicate that the detection method has strong continuity and self-consistency in its response to changes in the main axis state in the current transition section, and is less likely to repeatedly overturn its own judgment after forming an abnormal trend. The reason is that if a detection method is reliable, then when the main axis state gradually transitions from stable to abnormal, its judgment result will usually continue to evolve along a certain direction. Even if there are local fluctuations, it will not deviate from the established overall trend for a long time. For example, if the discrimination result of a candidate detection method gradually increases from 0.30 to 0.75 in a transitional zone, and then only slightly drops back to 0.71, it indicates that it maintains the abnormal trend well, and the corresponding result path shrinkage rate is relatively large. On the other hand, if the discrimination result of another detection method increases from 0.30 to 0.75 and then drops back to 0.42, then increases to 0.68 and then drops back again, it indicates that its discrimination path shrinks frequently and the transition is obvious, and the representation of the current main axis state is unstable, and the corresponding result path shrinkage rate will be smaller.

[0037] In one embodiment, S5: The step of determining the target detection method based on the adaptation verification results of each candidate spindle defect detection method, and transferring the judgment weight of spindle defect detection from the first detection method to the target detection method to continue spindle defect detection is as follows: The adaptation response results of each candidate detection method are compared, and the candidate detection method with the largest adaptation response result is determined as the target detection method. Switch the current primary detection method to target detection method to continue spindle defect detection.

[0038] It should be noted that, for example, if the current first detection method is vibration detection (relying on a vibration sensor), candidate methods include temperature detection (relying on a temperature sensor), acoustic emission detection (relying on an acoustic emission sensor), and vibration-current combined detection. If the adaptation response results of the three are 0.62, 0.81, and 0.74 respectively, it indicates that the acoustic emission detection method has the most stable response and the best tolerance to abnormal conditions within the current reliability decline observation period. The system then determines the acoustic emission detection method as the target detection method and continues to output a judgment on whether there are defects such as crack initiation, impact abnormalities, or local spalling on the spindle, thereby avoiding the first detection method from continuing to dominate the detection results after its reliability declines.

[0039] Based on the same inventive concept, embodiments of the present invention also provide a spindle defect detection system. It includes: Reference datum module: acquires independent datum state data formed by the spindle during the datum confirmation stage, and uses the independent datum state data as the normal reference datum of the spindle. The independent datum state data is collected under at least one stage: spindle factory calibration, no-load test run, standard test piece machining, post-repair acceptance, or manual verification to confirm that there are no abnormalities. The discrimination characterization module: within the current preset detection cycle, it acquires the current detection data corresponding to the first detection method, and compares the current detection data with the normal reference benchmark of the spindle to construct a discrimination reliability characterization sequence that reflects whether the current output of the first detection method is still consistent with the independent benchmark state. Observation section module: Based on the continuous change state of the reliability characterization sequence, determine the reliability decline observation section of the first detection method within the current preset detection cycle; the reliability decline observation section is used to characterize the observation section where the output result of the first detection method deviates continuously from the normal reference benchmark of the spindle, but the complete failure of the first detection method has not yet been directly confirmed; Adaptation Response Module: Within the reliability degradation observation period, response analysis is performed on multiple pre-defined candidate spindle defect detection methods to obtain the adaptation verification results of each candidate spindle defect detection method; the candidate spindle defect detection method is implemented by at least one sensor in the sensor system; Spindle defect detection module: Determines the target detection method based on the adaptation verification results of each candidate spindle defect detection method, and transfers the judgment power of spindle defect detection from the first detection method to the target detection method to continue spindle defect detection.

[0040] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should still fall within the scope of the claims of the present invention.

Claims

1. A method for detecting defects in a spindle, characterized in that, Spindle defect detection is achieved through several detection methods. The detection method corresponding to the spindle defect in the current preset detection cycle is denoted as the first detection method. The detection results of the first detection method in the current preset detection cycle are then analyzed. The specific steps are as follows: Acquire the independent reference state data formed by the spindle during the reference confirmation stage, and use the independent reference state data as the normal reference reference for the spindle; Within the current preset detection cycle, the current detection data corresponding to the first detection method is obtained, and the current detection data is compared with the normal reference benchmark of the spindle to construct a discrimination reliability characterization sequence that reflects whether the current output of the first detection method is still consistent with the independent benchmark state. Based on the continuous change state of the reliability characterization sequence, the reliability degradation observation section of the first detection method within the current preset detection cycle is determined; Within the reliability degradation observation period, response analysis was performed on multiple pre-defined candidate spindle defect detection methods to obtain the adaptation verification results of each candidate spindle defect detection method. The defect detection method for candidate spindles is implemented by at least one sensor in the sensor system; The target detection method is determined based on the adaptation and verification results of each candidate spindle defect detection method, and the judgment power of spindle defect detection is transferred from the first detection method to the target detection method in order to continue spindle defect detection.

2. The spindle defect detection method according to claim 1, characterized in that, The steps for constructing a discriminative reliability characterization sequence that reflects the ability of the first detection method to distinguish spindle states are as follows: The steps for constructing the discriminative reliability characterization sequence include: According to the preset time window length and sliding step size, the detection results output by the first detection method within the current preset detection period are segmented to obtain multiple detection sample windows arranged in chronological order; The signals within each detection sample window are processed by DC removal, amplitude normalization, and time alignment to obtain standardized signal segments; multiple state feature values ​​are extracted from each standardized signal segment and formed into corresponding first feature vectors in a fixed order; Extract second feature vectors of the same type and in the same order as the first feature vector from independent reference state data, and determine the normal reference center based on multiple second feature vectors; Each first feature vector is compared with the normal reference center to obtain the deviation distance of each detection sample window relative to the normal reference center; detection sample windows that meet the preset deviation conditions are selected based on the deviation distance, and the state discrimination vector is determined based on the first feature vector of the selected detection sample window. The offset results of each first feature vector relative to the normal reference center are expanded along the state discrimination vector, and the projection components in the corresponding directions are subtracted to obtain the lateral discretization. The discrimination reliability characterization value of each detection sample window is determined based on the degree of directional expansion and the amount of lateral dispersion. The higher the degree of directional expansion and the lower the amount of lateral dispersion, the higher the corresponding discrimination reliability characterization value. The discrimination reliability characterization values ​​of each detection sample window are arranged in chronological order to obtain the discrimination reliability characterization sequence.

3. The spindle defect detection method according to claim 1, characterized in that, The steps for determining the reliability degradation observation section of the first detection method within the current preset detection cycle are as follows: Based on the continuity characteristics, a discrimination capability evolution trajectory is generated to describe the changing trend of the discrimination capability of the first detection method, and based on the discrimination capability evolution trajectory, the observation segment of the reliability decline of the first detection method within the current preset detection cycle is determined.

4. The spindle defect detection method according to claim 3, characterized in that, The steps for generating the discrimination capability evolution trajectory to describe the changing trend of the discrimination capability of the first detection method are as follows: According to the chronological order of each discriminant reliability characterization value in the discriminant reliability characterization sequence, the difference between each two adjacent discriminant reliability characterization values ​​is calculated to obtain the adjacent change sequence; For two consecutive adjacent changes in an adjacent change sequence, calculate the degree of consistency in their change directions to obtain the local directional consistency. The first The first discriminant reliability characterization value and the second The arithmetic square root of the product of the local direction consistency is used as the first... Individual discriminative trajectory values; Arrange the discriminative ability trajectory values ​​in chronological order to obtain the discriminative ability evolution trajectory.

5. The spindle defect detection method according to claim 3, characterized in that, The steps for determining the reliability degradation observation section of the first detection method within the current preset detection period based on the discriminant capability evolution trajectory are as follows: Based on the chronological order of the discriminant capability trajectory values ​​in the discriminant capability evolution trajectory, the change between each two adjacent discriminant capability trajectory values ​​is calculated to obtain the trajectory change sequence. For each trajectory change in the trajectory change sequence, the portion less than zero is extracted as the descent component, thus obtaining the trajectory descent sequence. For each trajectory change in the trajectory change sequence, the portion greater than zero is extracted as the recovery component, thus obtaining the trajectory recovery sequence. The cumulative descent is obtained by summing the descent amounts of each trajectory in chronological order. The cumulative descent is obtained by summing the descent amounts of each trajectory in chronological order. Based on the cumulative decrease and cumulative increase, calculate the attenuation dominance ratio at each location; Based on the trajectory values ​​of each discrimination capability and the corresponding attenuation dominance ratio, the discrimination failure pressure value corresponding to each location is calculated; According to the preset sliding section length, the failure pressure value is continuously statistically analyzed, and the proportion of continuous increase of the failure pressure in each sliding section is calculated to obtain the failure penetration rate. From the discriminant capability evolution trajectory corresponding to the stable operation stage of the main shaft, select multiple stable trajectory values ​​and average them to obtain a stable reference value; From the failure penetration rates corresponding to the stable operation phase of the spindle, select multiple stable penetration rates and calculate the average to obtain the penetration reference value; The continuous intervals in the discrimination capability evolution trajectory that satisfy the condition that the current discrimination capability trajectory value is less than the stable reference value and the current failure penetration rate is greater than the penetration reference value are identified as the reliability decline observation segments in the transition from the stable discrimination state to the unstable discrimination state.

6. The spindle defect detection method according to claim 1, characterized in that, The steps for performing response analysis on multiple pre-defined candidate spindle defect detection methods to obtain the adaptation response results of each candidate detection method in the reliability degradation observation section are as follows: Response analysis is performed on multiple pre-defined candidate spindle defect detection methods. Boundary crossing stability value and result path shrinkage rate value are calculated. The boundary crossing stability value and result path shrinkage rate value are added together to obtain the adaptation response value of each candidate detection method.

7. The spindle defect detection method according to claim 6, characterized in that, The steps for calculating the stability value of boundary crossing are as follows: Obtain the discrimination result sequence and corresponding discrimination boundary of the candidate spindle defect detection method within the reliability degradation observation section. Subtract the discrimination boundary from each discrimination result and divide it by the sum of the maximum absolute value of all offset values ​​within the reliability degradation observation section and a preset positive number to obtain the signed offset sequence of the boundary. Based on the continuity of the signs of each offset value in the signed offset sequence of the boundary, sampling positions with the same sign and adjacent to each other are divided into multiple continuous segments on the same side and the number of boundary crossings between adjacent segments is recorded. At the same time, the number of sampling points corresponding to each continuous segment on the same side is counted as the segment length. For each continuous segment on the same side, the absolute value of each offset value within the segment is added to a preset positive number, and then multiplied together. The result of the multiplication is then squared according to the length of the corresponding segment to obtain the segment anchorage of each continuous segment on the same side. For any two adjacent consecutive segments on the same side, the length symmetry is obtained by dividing the smaller of the two segment lengths by the sum of the larger of the two segment lengths and a preset positive number. The anchorage symmetry is obtained by dividing the smaller of the two segment anchorages by the sum of the larger of the two segment anchorages and a preset positive number. The length symmetry is then multiplied by the anchorage symmetry to obtain the mirror crossing value. The boundary hesitation value is obtained by dividing the mirror crossing value by the sum of the geometric mean of the corresponding two segment anchorages and a preset positive number. The normalized boundary hesitation value is obtained by dividing the boundary hesitation value by the sum of the boundary hesitation value and a constant. The overall boundary hesitation intensity is obtained by subtracting one from each of the normalized boundary hesitation values, adding a preset positive number, multiplying the result, taking the square root of the result according to the number of boundary crossings, and then subtracting one. When the number of boundary crossings is zero, the overall boundary hesitation intensity is recorded as zero. The boundary crossing density is obtained by dividing the number of boundary crossings by the total number of sampling points minus one. The boundary crossing stability value is obtained by multiplying the total boundary hesitation strength by one minus the boundary crossing density and taking the square root.

8. The spindle defect detection method according to claim 1, characterized in that, The steps for calculating the path shrinkage rate are as follows: Obtain the discrimination result sequence of candidate detection methods within the reliability degradation observation section, and determine the main direction based on the magnitude relationship between the discrimination result at the end of the sequence and the initial discrimination result; The discrimination results are normalized according to the main direction to obtain the normalized path sequence of the main direction; The leading edge occupancy sequence is obtained by taking the maximum value of each point in the normalized path sequence of the main direction up to the current position in chronological order. Subtract the corresponding value in the normalized path sequence of the main direction from the leading edge occupancy value of each position to obtain the shrinkage gap sequence; In the leading edge occupancy sequence, the position where the current leading edge occupancy value is greater than the previous leading edge occupancy value is determined as the leading edge anchor point, and the interval between adjacent leading edge anchor points is determined as the anchor point segment; For each anchor point segment, add a preset positive number to each retraction gap value inside the segment, multiply them together, and then take the corresponding power according to the number of internal positions to obtain the retraction core value. Divide the number of internal positions of the segment by the total span between adjacent front anchor points to obtain the retraction occupancy ratio. Then multiply the retraction core value and the retraction occupancy ratio and take the square root to obtain the transfer value. For the transfer value of all anchor point segments, subtract each transfer value from the constant and then multiply by the preset positive number. Then, take the corresponding power according to the number of anchor point segments, and finally subtract the processing result from the constant to obtain the overall path transfer strength. The end value is obtained by taking the end value of the normalized path sequence in the main direction and dividing it by the sum of the end value of the leading edge occupant sequence and a preset positive number; Multiply the result obtained by subtracting the overall path yield strength from the constant one by the end retention value and take the square root of the product to obtain the path shrinkage rate.

9. A spindle defect detection system, used to implement the spindle defect detection method according to any one of claims 1-8, characterized in that, The system includes: Reference datum module: acquires independent datum state data formed by the spindle during the datum confirmation stage, and uses the independent datum state data as the normal reference datum of the spindle; The discrimination characterization module: within the current preset detection cycle, it acquires the current detection data corresponding to the first detection method, and compares the current detection data with the normal reference benchmark of the spindle to construct a discrimination reliability characterization sequence that reflects whether the current output of the first detection method is still consistent with the independent benchmark state. Observation section module: Based on the continuous change state of the reliability characterization sequence, determine the reliability decline observation section of the first detection method within the current preset detection cycle; Adaptation Response Module: Within the reliability degradation observation period, response analysis is performed on multiple pre-defined candidate spindle defect detection methods to obtain the adaptation verification results of each candidate spindle defect detection method; the candidate spindle defect detection method is implemented by at least one sensor in the sensor system; Spindle defect detection module: Determines the target detection method based on the adaptation verification results of each candidate spindle defect detection method, and transfers the judgment power of spindle defect detection from the first detection method to the target detection method to continue spindle defect detection.

Images