A closed-loop adaptive diagnosis method for steel production line multi-modal working conditions

Abstract

The application discloses a closed-loop adaptive diagnosis method for steel production line multi-mode conditions, and relates to the technical field of equipment state monitoring, comprising: associating mechanical abnormality diagnosis results with speed reduction control instructions triggered thereby, and constructing a speed reduction intervention event data unit comprising a stable section before intervention, a speed reduction transition section and a stable section after intervention; extracting and tracking fault spectrum lines, determining fault orders and fixed diagnosis frequency band abnormal energy changes in combination with actual speeds; determining whether fault spectrum lines migrate out of the fixed diagnosis frequency band and whether the same order impact remains; comparing the residual order with the same order response of normal speed reduction events in the healthy state to determine mechanical abnormality residues, normal speed reduction responses or residual sources that cannot be confirmed; forming a closed-loop feedback state in combination with equipment processing records, limiting sample write-back and model updating of intervened data, and dynamically adjusting the fault spectrum line tracking area. The application can avoid misjudging fault spectrum band escape caused by speed reduction as fault elimination.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring technology, and in particular to a closed-loop adaptive diagnostic method for multimodal operating conditions in steel production lines. Background Technology

[0002] Rolling mills, coilers, straighteners, conveyor rollers, and their associated gearboxes, bearing housings, couplings, and drive motors in steel production lines operate under high loads, strong impacts, continuous speed changes, and multiple specification switching environments for extended periods. Mechanical anomalies such as bearing spalling, tooth surface damage, roller system loosening, coupling misalignment, and increased transmission clearance typically manifest as corresponding changes in vibration signals, acoustic signals, drive current, and load characterization data. Existing diagnostic methods for steel production line equipment usually identify mechanical anomalies by analyzing spectral peak values, envelope peak values, abnormal energy, or multimodal data changes within a fixed diagnostic frequency band. When the anomaly level is high, they trigger equipment speed reduction or load reduction to mitigate the risk of damage from continued operation.

[0003] However, after the equipment speed is reduced, the periodic impact frequency generated by the mechanical fault will shift to a lower frequency direction as the target shaft speed decreases, and may leave the original fixed diagnostic frequency band. At the same time, the vibration amplitude, acoustic intensity, drive current and load data will also decrease due to the speed reduction and load reduction. Existing closed-loop diagnostic methods tend to directly identify the decrease in abnormal energy and the fallback of multimodal signals in the original fixed diagnostic frequency band as the elimination of mechanical abnormalities, and write the data after the speed reduction into the normal state sample or the recovery state sample. However, the order impact corresponding to the mechanical fault may still exist. This causes the data affected by the speed reduction intervention to have a reverse influence on the diagnostic feedback and model update, which masks the uneliminated mechanical abnormalities and thus creates a self-reinforcing problem in the diagnostic results. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as fault spectrum lines migrating out of the fixed diagnostic frequency band after diagnostic triggering deceleration, residual faults being misjudged as eliminated, and interference data being incorrectly involved in the closed-loop model update. Therefore, this invention proposes a closed-loop adaptive diagnostic method for multimodal operating conditions in steel production lines.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution:

[0006] A closed-loop adaptive diagnostic method for multimodal operating conditions in steel production lines includes:

[0007] S1. Associate the mechanical abnormality diagnosis results of the target equipment with the speed reduction control command it triggers, and divide the stable segment before intervention, the speed reduction transition segment and the stable segment after intervention according to the speed change to form a speed reduction intervention event data unit.

[0008] S2. Extract the fault spectrum from the deceleration intervention event data unit, generate a fault order sequence according to the ratio of the fault spectrum frequency to the rotation frequency converted from the actual rotation speed, and determine the abnormal energy change of the fixed diagnostic frequency band before and after the intervention to form the spectrum order tracking result.

[0009] S3. Based on the spectral order tracking results, determine whether the fault spectral line has migrated out of the fixed diagnostic frequency band and whether the corresponding fault order remains in the stable segment after intervention, forming a spectral escape same-order residue determination result containing the residual order.

[0010] S4. Obtain a normal deceleration event comparable to the deceleration intervention event under the health state of the target device, compare the residual order with the response of the same order of the normal deceleration event, and form a residual source determination result.

[0011] S5. Determine the closed-loop feedback status based on the residual source determination results and equipment processing records;

[0012] S6. Determine sample write-back and diagnostic model update based on closed-loop feedback status, and adjust the fault spectrum tracking area according to residual order and actual rotational speed to obtain closed-loop adaptive diagnostic results.

[0013] Preferably, the target equipment is a rolling mill, coiler, straightener, conveyor roller, gearbox, bearing housing, coupling, or drive motor in a steel production line; step S1 also acquires the actual rotational speed, vibration signal, acoustic signal, diagnostic output record, speed control record, equipment processing record, drive current, and load characterization data of the target equipment. The load characterization data includes at least one of rolling force, tension, transmission torque, and motor load rate. The equipment processing record includes maintenance record, component replacement record, fastening record, lubrication record, and retest record.

[0014] Preferably, step S1 further acquires production condition data including product specifications and transmission connection status, and includes: aligning the actual rotational speed, vibration signal, acoustic signal, drive current, load characterization data, diagnostic output record, speed control record, equipment processing record, and production condition data according to a unified time reference; determining the correlation between the mechanical anomaly diagnosis result and the speed reduction control command based on at least one of the event identifier and command source in the control log; using the issuance time of the speed reduction control command as the intervention boundary, determining the pre-intervention stable segment and the post-intervention stable segment based on the rotational speed fluctuation range during the historical normal operation of the target equipment, and determining the speed reduction transition segment based on the actual rotational speed reduction process between the two stable segments.

[0015] Preferably, the fixed diagnostic frequency band is the frequency range corresponding to the mechanical anomaly diagnosis result; step S2 includes: using short-time Fourier transform, envelope spectrum analysis, or order tracking methods to extract periodic impact spectral lines corresponding to the mechanical anomaly diagnosis result from the vibration signal and acoustic signal of the pre-intervention stable segment; continuously tracking the periodic impact spectral lines in the deceleration transition segment and the post-intervention stable segment according to the frequency position change and actual rotational speed change ratio of adjacent analysis windows; converting the actual rotational speed into the rotational frequency of the target shaft system, and dividing the fault spectral line frequency at each analysis time by the rotational frequency at the same analysis time to obtain the fault order sequence; and obtaining the abnormal energy of the pre-intervention stable segment and the post-intervention stable segment by accumulating or integrating the squares of the spectral amplitudes within the fixed diagnostic frequency band.

[0016] Preferably, in step S3, fault spectrum escape and same-order residue are determined when the following conditions are met simultaneously: the abnormal energy of the fixed diagnostic frequency band in the post-intervention stable section is lower than the abnormal energy of the fixed diagnostic frequency band in the pre-intervention stable section; the fault spectrum line continuously migrates to a lower frequency direction as the actual rotational speed decreases in the deceleration transition section; the migrated fault spectrum line is located outside the fixed diagnostic frequency band; the fault spectrum lines before and after migration are in the same order range; the periodic impacts within the order range are still detected in the post-intervention stable section; the order range is determined based on the natural fluctuation of the order in the pre-intervention stable section, the rotational speed measurement error, and the fault spectrum line frequency extraction error.

[0017] Preferably, step S4 includes: extracting normal deceleration events from historical operating data confirmed to be in a healthy state after maintenance; screening normal deceleration events comparable to the current deceleration intervention event based on the deceleration start speed, deceleration end speed, speed change process, load change direction, product specifications, and transmission connection status; establishing a natural fluctuation range based on the impact intensity of the screened normal deceleration events at the residual order; comparing the impact intensity of the current deceleration intervention event at the residual order with the natural fluctuation range, and determining whether the periodic impact at the residual order continues from the deceleration transition section to the post-intervention stable section, and whether the periodic impact in the vibration signal and acoustic signal corresponds to the same mechanical rotation period.

[0018] Preferably, step S4 further includes: when the same-order impact intensity in the current deceleration intervention event exceeds the natural fluctuation range, and the same-order impact continues from the deceleration transition segment to the post-intervention stabilization segment, the residual source determination result is determined to be mechanical abnormality residue; when the same-order impact intensity is within the natural fluctuation range, the residual source determination result is determined to be normal deceleration response; when the residual order is not detected in the post-intervention stabilization segment, the residual source determination result is determined to be no residue detected; when no normal deceleration event comparable to the current deceleration intervention event is obtained, or the natural fluctuation range cannot be established based on the normal deceleration event, the residual source determination result is determined to be residual source cannot be confirmed.

[0019] Preferably, step S5 includes: when the residual source determination result is mechanical abnormal residual, and no repair, component replacement, tightening or lubrication treatment has occurred after the intervention boundary, the closed-loop feedback state is determined as an intervention shielding state; when the residual source determination result is no residual detected, the equipment processing record indicates that repair, component replacement, tightening or lubrication treatment was performed after the intervention, and the target equipment is retested under comparable speed and load conditions before the intervention and the residual order is not detected, the closed-loop feedback state is determined as a true recovery state; when the residual source determination result is a normal deceleration response or the residual source cannot be confirmed, or the residual source determination result is no residual detected but there is no equipment processing record and comparable working condition retest result at the same time, the closed-loop feedback state is determined as a state to be verified.

[0020] Preferably, step S6 includes: when the closed-loop feedback state is an intervention masking state, prohibiting the writing of the data of the stable segment after intervention into the normal state sample or the recovery state sample, determining the actual frequency position of the fault spectrum line according to the residual order and the actual rotational speed, and adjusting the fault spectrum line tracking area to the actual frequency position; when the closed-loop feedback state is a true recovery state, determining the data before equipment processing and the data after equipment processing confirmed by comparable operating conditions as fault state data and recovery state data respectively, updating the diagnostic model after order normalization of the fault state data and recovery state data; when the closed-loop feedback state is a pending verification state, maintaining the original fault category label and stopping the use of the data of the stable segment after intervention to update the diagnostic model, re-executing steps S2 to S5 after obtaining new comparable operating condition data, and re-executing step S5 after obtaining new equipment processing records.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] 1. This invention associates the mechanical anomaly diagnosis results with the deceleration control command it triggers, using the issuance time of the deceleration control command as the intervention boundary. It constructs a deceleration intervention event data unit that includes a pre-intervention stable segment, a deceleration transition segment, and a post-intervention stable segment, and continuously tracks the migration trajectory of the fault spectrum line as the actual rotational speed changes. When the abnormal energy decreases within the fixed diagnostic frequency band, this invention further determines whether the impact corresponding to the same mechanical cycle continues to exist after deceleration based on the order relationship between the fault spectrum line frequency and the target shaft rotation frequency. This identifies the false anomaly caused by the fault spectrum line migrating out of the fixed diagnostic frequency band, avoiding misjudging the still-unresolved mechanical anomaly as equipment recovery based solely on vibration amplitude, acoustic intensity, or fixed frequency band energy decrease. This improves the accuracy and interpretability of mechanical anomaly diagnosis under variable speed and multi-load conditions.

[0023] 2. This invention compares the residual order with a comparable normal deceleration event under the healthy state of the target equipment, excluding normal dynamic responses caused by torque adjustment, gear meshing changes, and structural stress release during normal deceleration. It also combines maintenance, component replacement, tightening, lubrication, and comparable operating condition retest records to form an intervention-masked state, a true recovery state, or a state awaiting verification. Based on the closed-loop feedback state, this invention restricts the writing of deceleration intervention data into the normal state sample or recovery state sample, and dynamically adjusts the fault spectrum tracking area according to the residual order and actual speed. It updates the diagnostic model only using data processed by the equipment and verified by comparable operating condition retests, thereby avoiding the formation of an erroneous closed loop where diagnostic results trigger deceleration, deceleration leads to signal degradation, and signal degradation inversely reinforces the original diagnostic results. This improves the long-term stability of the diagnostic model and the reliability of the feedback data. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0025] Figure 1 This is a flowchart illustrating a closed-loop adaptive diagnostic method for multimodal operating conditions in steel production lines, as provided in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0027] Example 1

[0028] This embodiment provides a closed-loop adaptive diagnostic method for multimodal operating conditions in steel production lines. This embodiment uses the drive-side bearing housing of the finishing mill stand in a hot rolling production line as the target equipment, and the main drive shaft passing through the bearing housing as the target shaft system. Vibration sensors are installed in the horizontal and vertical directions of the bearing housing, and acoustic sensors are installed near the bearing housing. The actual rotational speed, speed setpoint, and speed control records of the target shaft system are obtained through the production line control system. The drive current is obtained through the motor drive system. The rolling force and transmission torque are obtained through the rolling control system. Maintenance records, component replacement records, tightening records, lubrication records, and retest records are obtained through the equipment management system. The mechanical anomaly diagnostic results and their output times are obtained through the diagnostic system.

[0029] During implementation, step S1 is executed, recording actual rotational speed, vibration signal, acoustic signal, drive current, load characterization data, diagnostic output record, speed control record, equipment processing record, and production condition data according to a unified time reference. When the diagnostic system outputs a mechanical anomaly diagnosis result corresponding to the bearing anomaly, and the speed control record indicates that a speed reduction control command to reduce the target equipment speed setpoint was issued after the mechanical anomaly diagnosis result was output, and the actual rotational speed decreases according to the speed reduction control command, the mechanical anomaly diagnosis result, the speed reduction control command, and the corresponding operating data are associated as a speed reduction intervention event. The moment the speed reduction control command is issued is used as the intervention boundary. Data on the actual rotational speed remaining at the original operating level before the intervention boundary is extracted as the pre-intervention stable segment. The speed reduction transition segment is extracted from the period from when the actual rotational speed starts to decrease continuously until it enters a new stable operating state. The post-intervention stable segment is extracted from the period when the actual rotational speed completes to decrease and remains at the new operating level. The data in the three data segments are encapsulated to form a speed reduction intervention event data unit.

[0030] Step S2 is executed, using the vibration measurement point directly corresponding to the original mechanical anomaly diagnosis result as the main diagnostic mode. Periodic fault spectrum lines are extracted from the vibration signal of the pre-intervention stable section, and acoustic signals are used as the periodic impact verification mode. In the deceleration transition section and the post-intervention stable section, the fault spectrum lines are continuously tracked according to the direction of change of the actual rotational speed. The actual rotational speed at each analysis time is converted into the rotational frequency of the target shaft system, and the fault order is calculated based on the ratio of the fault spectrum line frequency to the rotational frequency at the same analysis time. At the same time, the vibration anomaly energy of the fixed diagnostic frequency band in the pre-intervention stable section and the post-intervention stable section are calculated respectively. The vibration anomaly energy and acoustic signal energy are calculated separately, and the two types of energy with different physical dimensions are not directly added together. The fault spectrum line migration trajectory, fault order sequence, vibration anomaly energy change of the fixed diagnostic frequency band, and periodic impact detection results of vibration signal and acoustic signal at the corresponding order are combined to form the spectrum order tracking result.

[0031] Execute step S3, based on the spectral order tracking results, determine whether the abnormal vibration energy of the fixed diagnostic frequency band in the stable segment after intervention is lower than the abnormal vibration energy in the stable segment before intervention, determine whether the fault spectrum line continuously migrates to the low frequency direction as the actual rotational speed decreases in the deceleration transition segment, determine whether the migrated fault spectrum line is located outside the fixed diagnostic frequency band, determine whether the fault spectrum lines before and after migration are in the same order range, and determine whether the corresponding periodic impact continues to exist in the stable segment after intervention; when the above conditions are met simultaneously, fault spectrum escape and same-order residue are used as the judgment categories, and the corresponding fault order is used as the residue order, forming the spectrum escape same-order residue judgment result;

[0032] Step S4 involves extracting normal deceleration events from historical operating data of the target equipment, which has been confirmed to be in a healthy state after maintenance. Normal deceleration events comparable to the current deceleration intervention event are selected based on deceleration start speed, deceleration end speed, deceleration duration, load change direction, product specifications, and transmission connection status. Natural fluctuation ranges of vibration signals and effective acoustic signals at the residual order are established. The impact intensity of the current deceleration intervention event at the residual order is compared with the corresponding natural fluctuation range, and it is determined whether the periodic impact continues into the stable phase after intervention. A residual source determination result is formed according to a predetermined mutually exclusive judgment order: mechanical abnormality residue, normal deceleration response, no residue detected, or residue source cannot be confirmed.

[0033] Execute step S5, and retrieve the equipment processing record and retest record from the speed reduction intervention event data unit based on the residual source determination result; when the residual source determination result is mechanical abnormality residue, and no processing that can change the mechanical state of the target equipment occurs after the intervention boundary, the closed-loop feedback state is determined as the intervention shielding state; when the residual source determination result is no residue detected, a processing that can change the mechanical state of the target equipment occurs after the intervention, and no residual order is detected in the retest of the target equipment under comparable operating conditions, the closed-loop feedback state is determined as the true recovery state; other cases where it cannot be confirmed that the mechanical abnormality has been eliminated are determined as the pending verification state;

[0034] In step S6, when the closed-loop feedback state is in the intervention masking state, it is prohibited to write the data of the stable segment after intervention into the normal state sample or the recovery state sample. The actual frequency position of the fault spectrum line under the current operating condition is determined according to the residual order and the actual rotational speed, and the fault spectrum line tracking area is adjusted. When the closed-loop feedback state is in the true recovery state, the data before equipment processing and the data confirmed by comparable operating conditions after equipment processing are respectively determined as fault state data and recovery state data. After order normalization, the model is updated according to the original data format and update process of the existing diagnostic model. When the closed-loop feedback state is in the pending verification state, the original fault category label is maintained, and the diagnostic model is stopped from being updated using the data of the stable segment after intervention, thereby obtaining the closed-loop adaptive diagnostic result.

[0035] Example 2

[0036] This embodiment, based on Embodiment 1, provides a detailed explanation of the acquisition methods for the target equipment, target shaft system, and multimodal data.

[0037] The target equipment refers to equipment in a steel production line that has a rotary transmission relationship and is capable of implementing speed reduction control, including rolling mills, coilers, straighteners, conveyor rollers, gearboxes, bearing housings, couplings, or drive motors. The target shaft system is the shaft system that has a direct transmission relationship with the target mechanical component indicated by the mechanical anomaly diagnosis result and can reflect the actual rotational state of the target mechanical component. When the mechanical anomaly diagnosis result points to the main drive side bearing of the rolling mill, the target shaft system is the main drive shaft passing through the bearing. When the mechanical anomaly diagnosis result points to the output end bearing of the gearbox, the target shaft system is the gearbox output shaft. When the mechanical anomaly diagnosis result points to a rotor anomaly of the drive motor, the target shaft system is the drive motor shaft.

[0038] The actual rotational speed is primarily obtained from a rotary encoder or speed sensor located on the target shaft system. When the target shaft system does not have a separate speed measuring device, the actual rotational speed of the target shaft system is calculated based on the rotational speed fed back by the drive motor encoder and the total transmission ratio from the drive motor to the target shaft system. When there are switchable gear stages, clutches, or different roller system combinations in the transmission path, the total transmission ratio corresponding to the current transmission path is called according to the transmission connection status.

[0039] Vibration signals are collected by acceleration sensors installed in bearing housings, gearbox housings, rolling mill stands, coupling support seats, or drive motor end covers; at least one radial measuring point that can reflect the impact propagation direction of the target mechanical component is set; when the mechanical anomaly diagnosis result is generated by an existing vibration measuring point, that vibration measuring point is determined as the main diagnostic measuring point, and the main diagnostic measuring point, sampling frequency, filtering method, and spectrum normalization method remain unchanged before and after the intervention of the same deceleration intervention event;

[0040] The sampling frequency of the vibration signal is determined based on the highest analysis frequency to be detected, and the sampling condition that the sampling frequency is not less than twice the highest analysis frequency is met; when extracting high-frequency impact using envelope demodulation, the sampling frequency is higher than twice the upper boundary of the resonant frequency band used for envelope demodulation, and anti-aliasing filtering is performed before acquisition;

[0041] Acoustic signals are acquired by sound sensors or acoustic emission sensors located near the target device and avoiding continuous airflow, water flow and direct impact locations; the acoustic signals are used as periodic impact verification modes and are not directly added to the abnormal energy of the vibration signals; when the acoustic sensor fails to meet the data validity conditions due to communication packet loss, range saturation or environmental noise, the corresponding analysis window is marked as invalid acoustic data, and invalid acoustic data is not interpreted as the absence of mechanical abnormalities;

[0042] The drive current is obtained from the frequency converter, motor protection device or motor controller, and includes the effective value of the phase current or the torque current component; the load characterization data includes at least one of rolling force, tension, transmission torque and motor load rate; the drive current and load characterization data are used to determine the load change process and comparability of normal deceleration events, and are not combined with abnormal vibration energy.

[0043] The diagnostic output record includes the mechanical abnormality diagnosis result, fault category, target component, and output time; the speed control record includes the deceleration control command, command issuance time, speed setpoint, command source, and actual execution result; the equipment processing record includes maintenance record, component replacement record, tightening record, lubrication record, and retest record; the retest record includes at least the actual rotational speed, load characterization data, product specifications, transmission connection status, and residual order test results after equipment processing.

[0044] Example 3

[0045] This embodiment, based on Embodiment 2, provides a detailed explanation of the process of multi-source data alignment, correlation between diagnostic results and deceleration control commands, and determination of the three data segments;

[0046] Step S1 also acquires production condition data, which includes product specifications and transmission connection status. Product specifications include the steel type, thickness, width, and production batch of the current steel coil or billet. Transmission connection status includes the currently activated drive motor, gear stage, clutch connection status, roller system combination, and transmission ratio.

[0047] A unified time reference is established in the condition monitoring server, and the system time of the production line control system is used as the main time reference. Vibration acquisition devices, acoustic acquisition devices, drive systems, diagnostic systems and equipment management systems are synchronized with the main time reference through an industrial time synchronization protocol. When the acquisition devices cannot be directly synchronized, a time correction relationship is established based on the correspondence between the timestamps uploaded by the acquisition devices and the main time reference, and the corrected timestamps are used to perform data alignment.

[0048] Using the vibration signal analysis window as the alignment unit, actual rotational speed, drive current, and load characterization data with timestamps within the same analysis window are mapped to that analysis window. For actual rotational speed and load characterization data with sampling frequencies lower than those of the vibration signal, the median within the analysis window is used as the corresponding data for that analysis window. When data is missing between two adjacent actual rotational speed measurement times, linear interpolation is performed only if the missing duration does not exceed one normal sampling period of actual rotational speed. If the missing duration exceeds one normal sampling period, the corresponding analysis window is marked as invalid rotational speed data and is not included in the fault order calculation.

[0049] For vibration and acoustic signals, if the proportion of missing sampling points in the analysis window exceeds 5%, or the proportion of sampling points reaching the upper or lower limit of the sensor range exceeds 1%, the corresponding mode of the analysis window will be marked as invalid. If the vibration signal of the main diagnostic measurement point is invalid, the fault spectrum tracking of the analysis window will not be performed. If the acoustic signal is invalid, the vibration signal processing results will be retained, but invalid acoustic signals will not be used to form the opposite conclusion.

[0050] Diagnostic output records and deceleration control commands are first associated with event identifiers in the control log; when the two carry the same event identifier, the association is established directly; when they do not carry the same event identifier, the association is established based on the command source field; when the command source field indicates that the deceleration control command is automatically generated by the mechanical abnormality diagnosis system, or generated after confirmation by the operator from a mechanical abnormality alarm, the deceleration control command is associated with the corresponding mechanical abnormality diagnosis result.

[0051] When only the time sequence is available, the speed reduction control command should be within the response time range after the mechanical anomaly diagnosis result, and there should be no product specification switch, planned shutdown, emergency stop, or other independent speed reduction cause between the two; the lower limit of the response time range is zero, and the upper limit is the 95th percentile of the time from the diagnosis output to the issuance of the speed reduction command in the historical effective related events of the target production line; if there are fewer than five historical effective related events, the longest response time specified in the current mechanical anomaly handling procedure of the production line shall be used as the temporary upper limit;

[0052] The moment when the associated deceleration control command is issued is used as the intervention boundary; a ten-second sliding window is used to calculate the coefficient of variation of the actual speed, and the overlap rate of adjacent sliding windows is 50%; the coefficient of variation of the speed is the ratio of the standard deviation of the actual speed within the window to the average value of the actual speed; the stability judgment boundary is the 95th percentile of the coefficient of variation of the speed under the same product specifications and the same transmission connection state during the historical normal operation of the target equipment.

[0053] When the coefficient of variation of rotational speed in six consecutive sliding windows is not higher than the stability judgment boundary, and the average actual rotational speed in the six windows does not show a continuous unidirectional change, the corresponding data is determined as stable operating data; stable operating data continuous with the intervention boundary is extracted from the intervention boundary to form the pre-intervention stable segment; the deceleration transition segment is formed from the moment when the actual rotational speed begins to decrease continuously to the moment when new stable operating data is first formed; continuous stable operating data is extracted from the end of the deceleration transition segment to form the post-intervention stable segment; the pre-intervention stable segment and the post-intervention stable segment each contain at least six effective analysis windows;

[0054] When product specifications or transmission connection status changes during the deceleration process, and the changes alter the target shaft transmission ratio or load propagation path, the current deceleration intervention event is marked as not comparable, and in step S4, the residual source determination result is determined as the residual source cannot be confirmed.

[0055] Example 4

[0056] This embodiment, based on embodiment three, provides a detailed explanation of fixed diagnostic frequency band, fault spectrum extraction, candidate search range, candidate acceptance conditions, fault order calculation, and vibration anomaly energy calculation.

[0057] The fixed diagnostic frequency band is the actual frequency range used when generating mechanical anomaly diagnostic results. The frequency range is read from the diagnostic rule configuration or model configuration corresponding to the diagnostic output record. In this embodiment, the mechanical anomaly diagnostic result points to the periodic impact of the bearing, and the fixed diagnostic frequency band is 145 Hz to 155 Hz.

[0058] The vibration signal from the main diagnostic measurement point is subjected to mean removal and anti-aliasing filtering, and then filtered through a bandpass filter covering the impact resonance zone of the target bearing. The envelope signal is obtained through Hilbert transform, and the envelope spectrum is obtained by fast Fourier transform. The acoustic signal is subjected to mean removal and bandpass filtering using the same analysis window as the vibration signal to obtain the acoustic spectrum. The length of the analysis window is determined according to the condition that each analysis window contains at least five complete impact cycles at the lowest target fault frequency, and adjacent analysis windows overlap by 50%.

[0059] During the stabilization phase before intervention, local spectral peaks that repeat in at least five of the six consecutive effective analysis windows within a fixed diagnostic frequency band are considered as candidate fault spectral lines. The amplitude of the candidate local spectral peak should also be higher than the sum of the median of the background spectral amplitude and three times the absolute median difference in the local frequency range. Local spectral peaks that do not meet this condition are considered as background fluctuations.

[0060] Each candidate fault spectrum line in the stable phase before intervention is converted into a fault order. The candidate fault spectrum line that appears repeatedly in the main diagnostic measurement point, has the smallest degree of fault order dispersion, and matches the frequency corresponding to the original mechanical abnormality diagnosis result is selected as the starting point for fault spectrum line tracking. When there is a periodic spectral peak in the acoustic signal that is in the same order range as the fault spectrum line, it is used as an auxiliary verification without changing the selection of the main diagnostic measurement point.

[0061] No. The rotational frequency of the target axis system in each analysis window is calculated according to the following formula:

[0062]

[0063] In the formula, Indicates the first The rotation frequency of each analysis window, in Hertz; Indicates the first The actual rotational speed of each analysis window, in revolutions per minute;

[0064] Predict the fault spectrum frequency of the current analysis window based on the fault spectrum frequency of the previous analysis window and the actual rotational speed change ratio of adjacent analysis windows:

[0065]

[0066] In the formula, Indicates the first The predicted fault spectrum frequency of each analysis window, in Hertz; Indicates the first The frequency of the fault spectrum line has been determined in each analysis window, in Hertz; and They represent the first The analysis window and the first The actual rotational speed of each analysis window;

[0067] Before starting the tracking, the permissible order deviation is determined based on the natural fluctuation of the fault order during the stable phase before intervention, the actual speed measurement error, and the spectral resolution; in order to predict the fault spectral frequency. Centered on the current rotation frequency, the product of the current rotation frequency and the order tolerance, plus the frequency extraction error, is used as the candidate search half-width, i.e., the candidate search range is:

[0068]

[0069] to

[0070]

[0071] in, Indicates the permissible deviation of the order. This indicates the frequency extraction error; thus, the candidate search range simultaneously covers the range after the natural fluctuations of the order are transformed to the current frequency domain and the frequency extraction error.

[0072] Within the candidate search range, extract candidate spectral peaks that satisfy the local spectral peak amplitude condition, and calculate the candidate distance:

[0073]

[0074] In the formula, Indicates the first The first analysis window Candidate distances for each candidate spectral peak; Indicates the frequency of candidate spectral peaks; Indicates the fault order corresponding to the candidate spectral peak; This represents the median of the fault order during the stable phase before intervention. In this embodiment, the frequency change constraint and the order preservation constraint use the same weight to avoid arbitrarily selecting weights when there are no confirmed historical fault events.

[0075] The candidate acceptance boundary is the 95th percentile of the candidate distance between adjacent effective analysis windows of the determined fault spectral line in the stable segment before intervention. When there are fewer than five effective candidate distances in the stable segment before intervention, the candidate spectral peak with the smallest candidate distance within the candidate search range and which meets the local spectral peak amplitude condition is determined as the current fault spectral line, but at the same time, its fault order is required to be within the order allowable deviation.

[0076] If there are no candidate peaks in an analysis window that meet the search range, peak amplitude, and order tolerance conditions, the analysis window is marked as missing spectral lines, and the peaks are not forcibly selected by expanding the candidate search range.

[0077] The fault order for each analysis window is calculated using the following formula:

[0078]

[0079] In the formula, Indicates the first The fault order of each analysis window is a dimensionless value. This indicates the frequency of the fault spectrum lines in the analysis window; Indicates the rotation frequency of the same analysis window;

[0080] The abnormal energy of the fixed diagnostic frequency band is calculated using only the vibration signal from the main diagnostic measurement point; the acoustic signal is not combined with the abnormal vibration energy. The abnormal vibration energy of the fixed diagnostic frequency band within an analysis window is:

[0081]

[0082] In the formula, This represents the abnormal vibration energy within a fixed diagnostic frequency band; Indicating the vibration envelope spectrum of the main diagnostic measurement point, the first... spectral amplitude at each frequency point; and These represent the frequency point numbers corresponding to the lower and upper boundaries of the fixed diagnostic frequency band, respectively.

[0083] The vibration anomaly energy of each effective analysis window in the stable segment before and after intervention is calculated separately, and the median of the vibration anomaly energy of each segment is used as the representative value of the vibration anomaly energy of the corresponding data segment. The representative value of vibration anomaly energy, the fault spectral migration trajectory, the fault order sequence, and the periodic impact detection results of vibration and acoustic signals are combined to form the spectral band order tracking result.

[0084] Example 5

[0085] This embodiment, based on embodiment four, provides a detailed explanation of the judgment process for order tolerance, continuous migration of spectral lines, fault band escape, and residue of the same order.

[0086] Calculate the absolute deviation of the fault order relative to the median fault order for each effective analysis window during the pre-intervention stable phase, and take the 95th percentile of the absolute deviation as the natural fluctuation of the order; the permissible deviation of the order is determined according to the following formula:

[0087]

[0088] In the formula, Indicates the permissible deviation of the order; It represents the 95th percentile of the absolute deviation of the fault order from the median fault order during the stable phase before intervention. This represents the median of the fault order during the stable phase before intervention; This indicates the maximum relative error of the actual rotational speed measurement; The error in extracting the frequency of the fault spectral line is represented by the larger of the spectral resolution and the peak localization error obtained by calibration. Indicates the minimum effective rotational frequency in the current deceleration intervention event; to As a range of the same order;

[0089] Compare the representative values ​​of abnormal vibration energy in the fixed diagnostic frequency band during the stable segment after intervention with those before intervention; when the representative value of abnormal vibration energy in the stable segment after intervention is lower than that in the stable segment before intervention, it is determined that the abnormal vibration energy in the fixed diagnostic frequency band has decreased.

[0090] The fault spectrum migration trajectory in the deceleration transition section is checked in chronological order. When the number of effective analysis windows in which the frequency of the fault spectrum decreases with the actual speed decreases is not less than 80% of the total number of effective analysis windows in the deceleration transition section, and there are at least three consecutive effective analysis windows whose frequency change direction is consistent with the actual speed change direction, the fault spectrum is determined to continuously migrate to the low frequency direction.

[0091] An isolated spectral line missing window is allowed to be located between two valid tracking windows; when the fault spectral lines before and after the missing window are both within the candidate search range formed according to the actual rotational speed ratio, the continuous migration path is confirmed using the valid windows before and after, but the fault spectral line frequency is not artificially generated for the missing window; when two consecutive spectral line missing windows appear, or when the number of spectral line missing windows in the deceleration transition section exceeds 20% of the total number of valid analysis windows, the continuous migration of the fault spectral line is not confirmed, and the residual judgment result of the same order of spectral band escape is determined as an uncertain residual state;

[0092] The fault spectrum frequency in the stable segment after intervention is compared with the fixed diagnostic frequency band boundary; when the fault spectrum frequency is lower than the lower boundary of the fixed diagnostic frequency band or higher than the upper boundary of the fixed diagnostic frequency band, it is determined that the fault spectrum is outside the fixed diagnostic frequency band.

[0093] When the fault order corresponding to the fault spectrum line in the stable segment after intervention is within the same order range, it is determined that the fault spectrum line before and after migration is within the same order range; a narrow-band extraction region determined by the candidate search range described in Example 4 is set at the frequency position corresponding to the residual order, and bandpass filtering and envelope processing are performed on the vibration signal of the main diagnostic measurement point; when no less than five of the six effective analysis windows in the stable segment after intervention detect periodic impacts within the same order range, it is determined that the periodic impacts of the same order continue to exist;

[0094] When the acoustic signal is valid, the same order detection is performed on the acoustic signal; when the difference between the fault order detected by the acoustic signal and the fault order of the main diagnostic measurement point is not greater than the order allowable deviation, it is determined that the acoustic signal provides consistent auxiliary evidence; when the acoustic signal is valid but the detection result is inconsistent with the vibration signal, the vibration result is not directly rejected, but the conflict is passed to step S4 for residual source determination.

[0095] When the abnormal vibration energy in the fixed diagnostic frequency band decreases, the fault spectrum line continuously migrates to the lower frequency direction, the migrated fault spectrum line is located outside the fixed diagnostic frequency band, the fault spectrum lines before and after migration are in the same order range, and the periodic impact continues to exist in the stable segment after intervention, it is determined that fault spectrum escape and same-order residue have occurred, and the median of the fault order in the stable segment after intervention is determined as the residual order.

[0096] Example 6

[0097] This embodiment, based on embodiment five, provides a detailed explanation of normal deceleration event extraction, comparability screening, establishment of natural fluctuation range, and multimodal comparison.

[0098] Extract the time range from the equipment processing records where the target mechanical component has been confirmed healthy after maintenance, passed acceptance after component replacement, or has not shown any mechanical abnormality diagnosis results during continuous and stable operation; within the time range, search for actual speed reduction processes caused by normal production scheduling, product rhythm adjustment, or planned shutdown to form candidate normal speed reduction events; if there is only a single no-alarm record but no maintenance confirmation, acceptance record, or continuous and stable operation record, the corresponding event will not be identified as a candidate normal speed reduction event in a healthy state;

[0099] The relative differences in starting speed, ending speed, duration, and average load between candidate normal deceleration events and the current deceleration intervention event are calculated separately. In this embodiment, comparable normal deceleration events meet the following conditions: the relative difference in starting speed is no higher than 10%; the relative difference in ending speed is no higher than 10%; the relative difference in duration is no higher than 20%; the relative difference in average load is no higher than 15%; the direction of speed change and the direction of load change are the same; the steel type belongs to the same process category, and the thickness and width are within the same specification range in the existing product process table of the production line; the transmission connection status is the same.

[0100] Candidate normal speed reduction events that meet the above conditions are determined to be comparable normal speed reduction events; when there are no fewer than five comparable normal speed reduction events, the normal speed reduction events are used to establish a natural fluctuation range; when there are fewer than five, a formal natural fluctuation range is not directly established, and the residual source determination result is determined to be that the residual source cannot be confirmed, or a planned speed reduction is implemented after the target equipment is confirmed to be healthy through maintenance, to supplement to no fewer than five normal speed reduction events;

[0101] For the residual order, in the stable phase after intervention in each normal deceleration event, the actual frequency position corresponding to the residual order is determined according to the actual rotational speed, and the envelope spectrum energy or narrowband impact root mean square value of the main diagnostic measurement point is extracted near the position. In this embodiment, the narrowband impact root mean square value is uniformly used as the vibration impact intensity in the same deceleration intervention event and its normal deceleration event, and the impact intensity calculation method is not changed between events. When the acoustic signal is valid, the same analysis window is used to extract the acoustic narrowband impact root mean square value as the acoustic impact intensity.

[0102] The median impact intensity of each normal deceleration event in the stable phase after intervention is taken as the representative impact intensity of that normal deceleration event; natural fluctuation ranges are established for vibration impact intensity and acoustic impact intensity respectively, and the impact intensities of the two modes are not directly added together;

[0103] The range of natural fluctuations is determined according to the following formula:

[0104]

[0105]

[0106] In the formula, Indicates the lower boundary of the natural fluctuation range; Indicates the upper boundary of the natural fluctuation range; This represents the median impact intensity across multiple normal deceleration events; This represents the difference between the third quartile and the first quartile, which represent the impact intensity.

[0107] When the impact intensity of the current deceleration intervention event is equal to the upper boundary of the natural fluctuation range, it is considered to be within the natural fluctuation range; only when the impact intensity is greater than the upper boundary of the natural fluctuation range is it determined that the impact intensity exceeds the natural fluctuation range.

[0108] The vibration signal from the main diagnostic measurement point is a necessary input for determining the residual source. If the main diagnostic measurement point is invalid, the residual source determination result will be determined as the residual source cannot be confirmed. The acoustic signal is an auxiliary input. If the acoustic signal is invalid, it will not be interpreted as the absence of mechanical abnormality. If the acoustic signal is valid, the acoustic impact intensity and fault order will be used as the basis for consistency verification.

[0109] Example 7

[0110] This embodiment, based on Embodiment Six, provides a detailed explanation of the mutually exclusive classification order of the residual source determination results;

[0111] To ensure that any deceleration intervention event results in only one residual source determination, the determination shall be performed in the following priority order;

[0112] First, check whether at least five comparable normal deceleration events have been obtained, whether the natural fluctuation range of the main diagnostic measurement point can be established, whether the current deceleration intervention events are comparable, and whether the main diagnostic measurement point is effective; if any condition is not met, the residual source determination result will be determined as the residual source cannot be confirmed, and subsequent classification will not be performed.

[0113] Second, under the condition that the reference conditions are met, check whether residual order is detected in the stable segment after intervention; when no periodic impact within the same order range is detected in five of the at least six effective analysis windows in the stable segment after intervention, the residual source determination result is determined to be no residual detected; the absence of residual detection only means that no residual order is detected in the current operating data, and does not mean that the target device has been truly restored;

[0114] Third, when residual order is detected in the stable phase after intervention, compare the vibration impact intensity of the main diagnostic measurement point with the natural vibration fluctuation range; when the vibration impact intensity is greater than the upper boundary of the natural vibration fluctuation range, and the periodic impact continues from the deceleration transition phase to the stable phase after intervention, further examine the acoustic signal.

[0115] When the acoustic signal is invalid, since the main diagnostic measurement point has formed valid evidence that it continuously exceeds the range of natural fluctuations, the result of the residual source determination is determined to be mechanical abnormal residual.

[0116] If the acoustic signal is valid and periodic impacts are detected within the same order range of the acoustic signal, or if the intensity of the acoustic impact is greater than the upper boundary of the natural acoustic fluctuation range, the result of the residual source determination will be determined as mechanical abnormal residual.

[0117] If the acoustic signal is valid, but the periodic impact detected by the acoustic signal is not in the same order range as the main diagnostic measurement point, and the acoustic impact intensity is within the natural fluctuation range, the residual source determination result will be determined as the residual source cannot be confirmed, so as to avoid forcibly outputting mechanical abnormality residue when there is a conflict in the multimodal conclusions.

[0118] Fourth, when the vibration and impact intensity of the main diagnostic measurement point is within the natural vibration fluctuation range, and the acoustic impact intensity is also within the natural acoustic fluctuation range when the acoustic signal is effective, and the periodic impact does not continue from the deceleration transition section to the stable section after intervention, the residual source determination result is determined as a normal deceleration response.

[0119] Fifth, the following situations, where the conditions for both mechanical abnormality residue and normal deceleration response integrity cannot be met simultaneously, are uniformly determined as cases where the source of residue cannot be confirmed: the vibration and impact intensity at the main diagnostic measurement point exceeds the natural fluctuation range, but the periodic impact only appears briefly in the deceleration transition section; the vibration and impact intensity at the main diagnostic measurement point is within the natural fluctuation range, but the periodic impact continues until the stable section after intervention; the fault order of the vibration signal and the effective acoustic signal are not in the same order range; the vibration signal and the effective acoustic signal point to opposite intensity judgment results; the current deceleration intervention event simultaneously involves significant changes in product specifications, transmission connection status, or load regime.

[0120] The residue source determination results only include the residue source category and the residue level corresponding to the mechanical abnormality residue; the impact intensity, natural fluctuation range, and normal deceleration event number are saved as process records in the deceleration intervention event data unit.

[0121] Example 8

[0122] This embodiment, based on embodiment seven, provides a detailed explanation of the process for determining the intervention masking state, the actual recovery state, and the state pending verification;

[0123] Step S5 receives the residual source determination result output in step S4, and reads the equipment processing record and retest record after the intervention boundary from the deceleration intervention event data unit formed in step S1;

[0124] When the residual source is determined to be mechanical abnormality residue, check whether there are maintenance records, component replacement records, tightening records or lubrication records that can change the state of the target mechanical component after the intervention boundary; if there are no such equipment processing records, the closed-loop feedback state is determined to be the intervention shielding state; the intervention shielding state indicates that the decrease in vibration abnormal energy in the fixed diagnostic frequency band is caused by deceleration and fault spectrum migration, and cannot be used as evidence that the mechanical abnormality has been eliminated.

[0125] When the source of residue is determined to be no residue detected, the equipment processing record and retest record are checked simultaneously. The closed-loop feedback state is determined to be the true recovery state only when both of the following conditions are met: after intervention, maintenance, component replacement, tightening or lubrication treatment corresponding to the original mechanical abnormality and capable of changing the state of the target mechanical component is carried out; after the equipment is processed, the target equipment is running at a speed, load, product specifications and transmission connection state comparable to that before intervention, and the original residual order is not detected in six consecutive effective analysis windows.

[0126] If only equipment processing is performed but comparable operating conditions are not retested, or if no residual order is detected only under low-speed and low-load conditions after speed reduction, it is uncertain whether it is a true recovery state.

[0127] When the residual source determination result is a normal deceleration response or the residual source cannot be confirmed, the closed-loop feedback status is determined to be a pending verification status; when the residual source determination result is no residual detected, but there are no equipment processing records and comparable operating condition retest results at the same time, the closed-loop feedback status is also determined to be a pending verification status.

[0128] The closed-loop feedback status includes the status category and the residual order corresponding to the intervention masking status; the fault status data range and recovery status data range corresponding to the actual recovery status are determined according to the equipment processing completion time and the comparable operating condition retest period; the supplementary data conditions required for the status to be reviewed are written into the speed reduction intervention event data unit, so that step S6 can determine the subsequent review path.

[0129] Example 9

[0130] This embodiment describes the specific execution process of closed-loop adaptive diagnosis based on embodiment eight;

[0131] When the closed-loop feedback state is in the intervention masking state, the post-intervention stable segment data in the deceleration intervention event data unit is locked, and the data is prohibited from being written into the normal state sample or recovery state sample. The deceleration intervention-affected identifier, residual order, and intervention boundary are written into the data. The center frequency of the fault spectrum is determined based on the residual order and the current actual rotational speed.

[0132]

[0133] In the formula, The center frequency of the fault spectrum line at the current actual rotational speed is expressed in Hertz. Indicates the residual order; This indicates the current actual rotational speed, expressed in revolutions per minute.

[0134] The frequency half-width of the fault spectral tracking region is:

[0135]

[0136] In the formula, The frequency half-width represents the tracking region; Indicates the current rotation frequency; Indicates the permissible deviation of the order; Indicates frequency extraction error; to The new fault spectral line tracking region is identified, and periodic impacts within the region are continuously extracted from subsequent vibration and acoustic signals.

[0137] Once the target equipment returns to comparable speed and load conditions, steps S2 to S5 are executed again. If it is still determined to be a mechanical abnormality residue or intervention shielding state, the original mechanical abnormality diagnosis result is maintained, and the fault state is not reduced due to the decrease in vibration abnormality energy in the fixed diagnostic frequency band during low speed.

[0138] When the closed-loop feedback state is the true recovery state, the data corresponding to the original mechanical abnormality before the equipment processing is determined as fault state data, and the data confirmed by comparable working conditions that no residual order was detected after the equipment processing is determined as recovery state data. The cumulative rotation angle of the target shaft system is calculated based on the actual rotation speed at each sampling time, and the vibration signal and acoustic signal are resampled at equal rotation angle intervals so that the signals under different actual rotation speeds are represented by the same number of sampling points per revolution, thus completing the order normalization.

[0139] This invention does not change the model structure, input data format, loss function, or training algorithm of the existing diagnostic model. It only determines whether data is allowed to enter the original update process of the existing diagnostic model based on the closed-loop feedback state. The fault state data retains the fault category label corresponding to the original mechanical abnormality diagnosis result, and the recovery state data adopts the normal state label already defined in the existing diagnostic model. The data after order normalization is organized according to the original input format of the existing diagnostic model.

[0140] When the existing diagnostic model is a threshold rule or statistical diagnostic model, the impact intensity reference range and alarm boundary at the residual order are recalculated using fault state data and recovery state data; when the existing diagnostic model is a machine learning model or neural network model, the two types of validated data are input into the original training process of the existing model, and incremental training or periodic retraining is performed according to the existing training configuration.

[0141] Independent validation data uses data that was fixed before the model update and was not involved in this update; the updated diagnostic model is only released when the recall rate of the updated model for the original mechanical anomaly category is not lower than that of the unupdated model and the false alarm rate of the normal state is not higher than that of the unupdated model; otherwise, the unupdated model is retained; thus, the model update content of this invention is limited to the admission of validated samples, order normalization and release control, and does not require the redesign of the diagnostic model.

[0142] When the closed-loop feedback status is pending verification, the original fault category label is maintained, and the fault regression pattern is not updated using the data from the stable segment after intervention; when new comparable operating condition data is obtained, the data is associated with the original speed reduction intervention event data unit, and steps S2 to S5 are re-executed; when only new equipment processing records are obtained, the new equipment processing records are written into the original speed reduction intervention event data unit, and step S5 is re-executed; when the new equipment processing records are accompanied by comparable operating condition retest data, it is re-determined whether the true recovery status is met.

[0143] The final result is a closed-loop adaptive diagnostic result that includes closed-loop feedback status, sample write-back permission, diagnostic model update permission, fault spectrum tracking region, and subsequent verification status.

[0144] Example 10

[0145] This embodiment provides a set of exemplary calculation data throughout steps S1 to S6 to illustrate the specific calculation process of the present invention;

[0146] The target equipment is the bearing housing on the drive side of the finishing mill stand, and the target shaft system is the main drive shaft. The diagnostic system detects periodic impacts around 150 Hz in the vibration envelope spectrum of the main diagnostic measurement point and outputs the mechanical abnormality diagnosis results corresponding to the bearing abnormality. The fixed diagnostic frequency band is 145 Hz to 155 Hz.

[0147] After the mechanical anomaly diagnosis results are output, the control system issues a speed reduction control command; the actual rotational speed in the pre-intervention stable phase is 900 rpm, corresponding to a rotational frequency of 15 Hz; the actual rotational speed in the post-intervention stable phase is 600 rpm, corresponding to a rotational frequency of 10 Hz; the pre-intervention stable phase and the post-intervention stable phase each contain six effective analysis windows;

[0148] The vibration anomaly energy of the fixed diagnostic frequency band was calculated using the same main diagnostic measurement point, the same preprocessing method, and the same spectrum normalization method. The representative value of the vibration anomaly energy in the stable segment before intervention was 8.2, and the representative value of the vibration anomaly energy in the stable segment after intervention was 1.1, which determined that the vibration anomaly energy of the fixed diagnostic frequency band decreased.

[0149] Before intervention, the fault spectrum frequency in the stable phase was 150 Hz, corresponding to a fault order of 10. During the deceleration transition phase, the fault spectrum migrated to lower frequencies as the actual speed decreased. After intervention, the fault spectrum frequency in the stable phase was 100 Hz, still corresponding to a fault order of 10. The fault spectrum had already fallen below the lower boundary of the fixed diagnostic frequency band, confirming that the fault spectrum had migrated out of the fixed diagnostic frequency band.

[0150] The 95th percentile of the absolute deviation of the fault order in the stable phase before intervention is 0.08, the maximum relative error of the actual speed measurement is 0.2%, the fault spectrum frequency extraction error is 0.5 Hz, and the minimum effective rotational frequency of the current deceleration intervention event is 10 Hz. Therefore, the permissible order deviation is:

[0151]

[0152] Therefore, the same order range is 9.85 to 10.15; the fault order of the stable segment after intervention is within the range, and five of the six effective analysis windows detect the corresponding periodic impact, determining that fault spectrum escape and same order residue occurred, with the residue order being ten.

[0153] Six comparable normal deceleration events were extracted from the target equipment under healthy conditions. The vibration-representing impact intensities at the residual order of these normal deceleration events were 0.72, 0.76, 0.79, 0.83, 0.88, and 0.91, respectively. The median of the data was 0.81, the first quartile was 0.76, the third quartile was 0.88, and the interquartile range was 0.12. Therefore, the upper boundary of the natural vibration fluctuation range was:

[0154]

[0155] The vibration at the residual order of the current deceleration intervention event represents an impact intensity of 2.6, which is greater than the upper boundary of the natural fluctuation range of 0.99. Furthermore, the periodic impact extends from the deceleration transition phase to the stable phase after intervention. The effective acoustic signal also detected periodic impacts within the same order range. Therefore, the residual source is determined to be mechanical abnormality residual.

[0156] The equipment processing records show that only speed reduction control was performed after the intervention boundary, and no bearing replacement, tightening, lubrication or other treatments that could change the mechanical state of the target bearing were implemented. Therefore, the closed-loop feedback state was determined to be the intervention masking state.

[0157] The current actual rotational speed is 600 revolutions per minute, and the residual order is ten. Therefore, the center frequency of the fault spectrum is:

[0158]

[0159] With an order tolerance of 0.15, a current rotation frequency of 10 Hz, and a frequency extraction error of 0.5 Hz, the half-width of the tracking region is:

[0160]

[0161] Therefore, the 98 Hz to 102 Hz range is defined as the new fault spectrum tracking region. Data from the stable segment after intervention is not written into the normal state sample or the recovery state sample. The diagnostic system continues to monitor periodic impacts in the tracking region, thereby avoiding the erroneous identification of the abnormal vibration energy decrease caused by the fault spectrum moving out of the original fixed diagnostic frequency band as the elimination of mechanical abnormalities.

[0162] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A closed-loop adaptive diagnostic method for multimodal operating conditions in steel production lines, characterized in that, Includes the following steps: S1. Associate the mechanical abnormality diagnosis results of the target equipment with the speed reduction control command it triggers, and divide the stable segment before intervention, the speed reduction transition segment and the stable segment after intervention according to the speed change to form a speed reduction intervention event data unit. S2. Extract the fault spectrum from the deceleration intervention event data unit, generate a fault order sequence according to the ratio of the fault spectrum frequency to the rotation frequency converted from the actual rotation speed, and determine the abnormal energy change of the fixed diagnostic frequency band before and after the intervention to form the spectrum order tracking result. S3. Based on the spectral order tracking results, determine whether the fault spectral line has migrated out of the fixed diagnostic frequency band and whether the corresponding fault order remains in the stable segment after intervention, forming a spectral escape same-order residue determination result containing the residual order. S4. Obtain a normal deceleration event comparable to the deceleration intervention event under the health state of the target device, compare the residual order with the response of the same order of the normal deceleration event, and form a residual source determination result. S5. Determine the closed-loop feedback status based on the residual source determination results and equipment processing records; S6. Determine sample write-back and diagnostic model update based on closed-loop feedback status, and adjust the fault spectrum tracking area according to residual order and actual rotational speed to obtain closed-loop adaptive diagnostic results.

2. The closed-loop adaptive diagnostic method for multi-modal operating conditions in steel production lines according to claim 1, characterized in that, The target equipment is a rolling mill, coiler, straightener, conveyor roller, gearbox, bearing housing, coupling, or drive motor in a steel production line; step S1 also acquires the actual rotational speed, vibration signal, acoustic signal, diagnostic output record, speed control record, equipment processing record, drive current, and load characterization data of the target equipment. The load characterization data includes at least one of rolling force, tension, transmission torque, and motor load rate. The equipment processing record includes maintenance record, component replacement record, fastening record, lubrication record, and retest record.

3. The closed-loop adaptive diagnostic method for multi-modal operating conditions in steel production lines according to claim 2, characterized in that, Step S1 also acquires production condition data including product specifications and transmission connection status, and includes: aligning the actual rotational speed, vibration signal, acoustic signal, drive current, load characterization data, diagnostic output record, speed control record, equipment processing record, and production condition data according to a unified time reference; determining the correlation between the mechanical anomaly diagnosis result and the speed reduction control command based on at least one of the event identifier and command source in the control log; using the issuance time of the speed reduction control command as the intervention boundary, determining the pre-intervention stable segment and the post-intervention stable segment based on the rotational speed fluctuation range during the historical normal operation of the target equipment, and determining the speed reduction transition segment based on the actual rotational speed reduction process between the two stable segments.

4. The closed-loop adaptive diagnostic method for multi-modal operating conditions in steel production lines according to claim 3, characterized in that, The fixed diagnostic frequency band is the frequency range corresponding to the mechanical anomaly diagnosis result; step S2 includes: using short-time Fourier transform, envelope spectrum analysis, or order tracking methods to extract periodic impact spectral lines corresponding to the mechanical anomaly diagnosis result from the vibration and acoustic signals of the pre-intervention stable segment; continuously tracking the periodic impact spectral lines in the deceleration transition segment and the post-intervention stable segment according to the frequency position change and actual rotational speed change ratio of adjacent analysis windows; converting the actual rotational speed into the rotational frequency of the target shaft system, and dividing the fault spectral line frequency at each analysis time by the rotational frequency at the same analysis time to obtain the fault order sequence; and obtaining the abnormal energy of the pre-intervention stable segment and the post-intervention stable segment by accumulating or integrating the squares of the spectral amplitudes within the fixed diagnostic frequency band.

5. The closed-loop adaptive diagnostic method for multi-modal operating conditions in steel production lines according to claim 4, characterized in that, In step S3, fault spectrum escape and same-order residue are determined when the following conditions are met simultaneously: the abnormal energy of the fixed diagnostic frequency band in the stable segment after intervention is lower than the abnormal energy of the fixed diagnostic frequency band in the stable segment before intervention; the fault spectrum line continuously migrates to the low-frequency direction as the actual speed decreases in the deceleration transition segment; and the migrated fault spectrum line is located outside the fixed diagnostic frequency band. The fault spectrum before and after migration are in the same order range; the periodic impacts within the order range are still detected in the stable segment after the intervention; the order range is determined based on the natural fluctuations of the order in the stable segment before the intervention, the speed measurement error, and the fault spectrum frequency extraction error.

6. The closed-loop adaptive diagnostic method for multi-modal operating conditions in steel production lines according to claim 5, characterized in that, Step S4 includes: extracting normal deceleration events from historical operating data confirmed to be in a healthy state after maintenance; screening normal deceleration events comparable to the current deceleration intervention event based on the deceleration start speed, deceleration end speed, speed change process, load change direction, product specifications, and transmission connection status; establishing a natural fluctuation range based on the impact intensity of the screened normal deceleration events at the residual order; comparing the impact intensity of the current deceleration intervention event at the residual order with the natural fluctuation range, and determining whether the periodic impact at the residual order continues from the deceleration transition section to the post-intervention stable section, and whether the periodic impacts in the vibration signal and acoustic signal correspond to the same mechanical rotation period.

7. The closed-loop adaptive diagnostic method for multi-modal operating conditions in steel production lines according to claim 6, characterized in that, Step S4 further includes: when the same-order impact intensity in the current deceleration intervention event exceeds the natural fluctuation range, and the same-order impact continues from the deceleration transition segment to the post-intervention stabilization segment, the residual source determination result is determined to be mechanical abnormality residue; when the same-order impact intensity is within the natural fluctuation range, the residual source determination result is determined to be normal deceleration response; when the residual order is not detected in the post-intervention stabilization segment, the residual source determination result is determined to be no residue detected; when no normal deceleration event comparable to the current deceleration intervention event is obtained, or the natural fluctuation range cannot be established based on the normal deceleration event, the residual source determination result is determined to be residual source cannot be confirmed.

8. The closed-loop adaptive diagnostic method for multi-modal operating conditions in steel production lines according to claim 7, characterized in that, Step S5 includes: when the residual source determination result is mechanical abnormal residual, and no repair, component replacement, tightening or lubrication treatment has occurred after the intervention boundary, the closed-loop feedback state is determined as the intervention shielding state; when the residual source determination result is no residual detected, the equipment processing record indicates that repair, component replacement, tightening or lubrication treatment was performed after the intervention, and the target equipment is retested under comparable speed and load conditions before the intervention and the residual order is not detected, the closed-loop feedback state is determined as the true recovery state; when the residual source determination result is normal deceleration response or the residual source cannot be confirmed, or the residual source determination result is no residual detected but there is no equipment processing record and comparable working condition retest result at the same time, the closed-loop feedback state is determined as the pending verification state.

9. The closed-loop adaptive diagnostic method for multi-modal operating conditions in steel production lines according to claim 8, characterized in that, Step S6 includes: when the closed-loop feedback state is in the intervention masking state, prohibiting the writing of the data of the stable segment after intervention into the normal state sample or the recovery state sample, determining the actual frequency position of the fault spectrum line according to the residual order and the actual rotational speed, and adjusting the fault spectrum line tracking area to the actual frequency position; when the closed-loop feedback state is in the true recovery state, determining the data before equipment processing and the data after equipment processing confirmed by comparable operating conditions as fault state data and recovery state data respectively, updating the diagnostic model after order normalization of the fault state data and recovery state data; when the closed-loop feedback state is in the pending verification state, maintaining the original fault category label and stopping the use of the data of the stable segment after intervention to update the diagnostic model, re-executing steps S2 to S5 after obtaining new comparable operating condition data, and re-executing step S5 after obtaining new equipment processing records.

Images