Dual-band acoustic emission based method and system for diagnosing fault of a draw press gear box

Abstract

The application discloses a double-frequency acoustic emission-based drawing press gear box fault diagnosis method and system, and relates to the technical field of mechanical equipment fault diagnosis. Different frequency acoustic emission sensors are arranged at preset monitoring points of a drawing press gear box, and acoustic emission signals of each point and each frequency are synchronously collected. Time-frequency domain transformation is performed on the signals to obtain a time-frequency point matrix, single-frequency band features based on single sensor signals, double-frequency band comparison features based on comparison relationships of signals of different frequency bands at the same point are extracted, and a multi-dimensional time-frequency statistical feature set is constructed. A health baseline is constructed according to historical normal samples of the same working condition, the Mahalanobis distance of a current feature vector is calculated, the square of the Mahalanobis distance is compared with a chi-square quantile, and it is determined that the gear box has a fault when the square of the Mahalanobis distance is greater than the chi-square quantile. Thus, the problems of signal aliasing under complex working conditions, difficulty in identifying early faults, and fault diagnosis without fault samples are solved, and the accuracy, reliability and adaptability of early fault diagnosis of the drawing press gear box are improved.

Description

Technical Field

[0001] This invention relates to the field of mechanical equipment fault diagnosis technology, and more specifically, to a fault diagnosis method and system for a drawing press gearbox based on dual-band acoustic emission. Background Technology

[0002] Longwall rolling mills are core heavy-duty equipment in automobile manufacturing stamping production lines. Operating under high loads, strong impacts, and low speeds for extended periods, they are prone to progressive failures such as pitting, wear, and cracking in the rolling bearings and meshing gears within the gearbox. These failures can escalate to tooth breakage and shaft breakage, leading to unplanned downtime, significant economic losses, and safety risks. Therefore, achieving early online diagnosis and accurate warning of gearbox faults is crucial for ensuring the stable operation of stamping production lines.

[0003] Acoustic emission technology is suitable for monitoring low-speed, heavy-load gearboxes. It can capture high-frequency elastic waves (20kHz~1MHz) generated by the initiation and propagation of internal cracks in materials. Compared with traditional monitoring methods such as vibration, temperature, and current, it has higher sensitivity to early micro-damage and stronger anti-interference ability. However, applying acoustic emission to online diagnostics of gearboxes in drawing presses still faces several challenges:

[0004] 1) Complex and variable working conditions: The drawing press is subject to stretching pad impact, die contact impact, and sudden stop transient impact, with the timing and amplitude of the impact fluctuating randomly; the production cycle, drawing depth, and load vary greatly with the part formula, resulting in significant working condition drift.

[0005] 2) Complex transmission path: The gearbox has a multi-stage transmission structure. Fault signals need to be rigidly transmitted through gears, shafts, bearings and housing multiple times. High-frequency signals are severely attenuated and low-frequency noise interference is strong.

[0006] 3) Interference from the oil distributor: The reciprocating switching of the oil distributor on the bearing housing and the oil pulsation generate high-amplitude periodic pulses, which significantly change the background noise structure.

[0007] 4) Multi-source signal aliasing: Fault signals, mechanical shocks, emergency stop shocks, and oil circuit pulses intertwine in the time domain and overlap in the frequency domain. Conventional time-domain windowing and frequency-domain filtering are insufficient to separate fault features.

[0008] 5) Scarcity of fault samples: There are a large number of normal samples but very few fault samples, making it difficult to conduct supervised model training; the drift of operating conditions leads to poor generalization ability of fixed thresholds and fixed models, resulting in a high false alarm and false negative rate.

[0009] Existing technologies mostly employ single-band acoustic emission, vibration monitoring, fixed thresholds, or simple statistical methods, without designing dedicated solutions for the operating conditions of drawing presses, oil circuit interference, and signal attenuation characteristics. These technologies fail to effectively address core pain points such as early fault identification, operating condition adaptation, sensor self-diagnosis, and anomaly detection in fault-free samples. Currently, the automotive stamping industry lacks mature and reliable online diagnostic technologies for early faults in drawing press gearboxes. Summary of the Invention

[0010] To overcome the shortcomings of the prior art, the present invention provides a fault diagnosis method and system for a drawing press gearbox based on dual-band acoustic emission, which at least solves the problems in the related art of lacking adaptation to the complex working conditions and signal attenuation characteristics of the drawing press, resulting in the inability to accurately separate fault features, difficulty in identifying early damage, and lack of working condition self-adaptation and sensor self-calibration capabilities, which limit the improvement of online diagnosis reliability and robustness.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] In a first aspect, the present invention provides a fault diagnosis method for a drawing press gearbox based on dual-band acoustic emission, the method comprising:

[0013] At the preset monitoring points of the gearbox of the drawing press, acoustic emission sensors of different frequency bands are arranged to synchronously collect acoustic emission signals of each frequency band at each monitoring point.

[0014] The acquired acoustic emission signals are transformed in the time-frequency domain to obtain a time-frequency point matrix. Single-band sensor features and / or dual-band sensor comparison features are extracted to form a multi-dimensional time-frequency statistical feature set. The single-band sensor features are calculated based on a single sensor signal, and the dual-band sensor comparison features are calculated based on the comparison relationship between sensor signals of different frequency bands at the same monitoring point.

[0015] Based on a healthy baseline constructed from historical normal acoustic emission samples under the same operating conditions, the Mahalanobis distance of the current multidimensional time-frequency statistical feature vector is calculated, and the square of the Mahalanobis distance is compared with the chi-square quantile.

[0016] When the square of the Mahalanobis distance is greater than the chi-square quantile, it is determined that there is a fault in the gearbox of the drawing press.

[0017] Optionally, the preset monitoring points are determined based on the multi-stage transmission structure of the gearbox, covering the bearing seats and idler shaft bodies corresponding to the high-speed shaft, intermediate shaft, and idler shaft.

[0018] Optionally, the acoustic emission sensors of different frequency bands include a first frequency band sensor and a second frequency band sensor, wherein the operating frequency band of the first frequency band sensor is higher than that of the second frequency band sensor, and there is an overlapping frequency band between the two for sensor status self-diagnosis.

[0019] Optionally, before extracting the single-band sensor features and / or the dual-band sensor contrast features, the method further includes:

[0020] The programmable logic controller status instructions of the drawing press are obtained, and the collected acoustic emission signals are labeled with status tags to distinguish between normal working status data and emergency stop status data.

[0021] Based on the duration of a single production cycle, the labeled acoustic emission data is sliced ​​into fixed time windows so that each data slice contains the same integer number of periodic pulse signals and the mechanical impact signal within a complete cycle.

[0022] The normal working state data includes stable stamping, die contact, drawing and top dead center dwell; the emergency stop data includes emergency braking, dual material alarm and overload protection trigger; the time window length of the fixed time window slice is matched with the single production cycle time, and the slice start point is aligned with the start time of the production cycle.

[0023] Optionally, the single-band sensor characteristics are obtained by statistically calculating the time-frequency point matrix corresponding to a single sensor, including specific frequency band energy ratio, time-frequency kurtosis, time-frequency energy entropy, instantaneous frequency standard deviation, and ridge continuity.

[0024] The dual-band sensor comparison features are obtained by comparing and calculating the time-frequency point matrices corresponding to sensors of different frequency bands at the same monitoring point, including dual-band kurtosis ratio, dual-band pulse index ratio, dual-band energy logarithm ratio, dual-band margin index ratio, and dual-band time-frequency distribution similarity.

[0025] Optionally, before performing fault diagnosis, the validity of the sensor data is also verified, which includes:

[0026] Extract data subsets from sensors of different frequency bands at the same monitoring point within overlapping frequency bands, calculate the correlation coefficient between the data subsets, and compare the current correlation coefficient with the baseline value of the correlation coefficient under historical normal conditions;

[0027] If the comparison result exceeds the preset threshold range, the corresponding sensor is determined to be faulty, and the feature dimension corresponding to the sensor is removed in the subsequent Mahalanobis distance calculation; if all sensors at the same monitoring point are determined to be faulty, a prompt message indicating that the monitoring point is in failure is output.

[0028] Optionally, the health baseline includes the mean vector and covariance matrix calculated from the multidimensional time-frequency statistical feature set of historical normal acoustic emission samples; the health baseline is dynamically updated as the collected samples are updated to adapt to the changes in data distribution caused by the drift of the drawing press operating conditions.

[0029] Alternatively, the Mahalanobis distance can be calculated using the following formula:

[0030] ;

[0031] Where x is the multidimensional time-frequency statistical feature vector extracted from the current monitoring data, μ is the mean vector of the health baseline, and Σ is the covariance matrix of the health baseline.

[0032] A second aspect of the present invention provides a fault diagnosis system for a drawing press gearbox based on dual-band acoustic emission, comprising:

[0033] The synchronous acquisition module is used to deploy acoustic emission sensors of different frequency bands at preset monitoring points in the gearbox of the drawing press, and synchronously acquire acoustic emission signals of each frequency band at each monitoring point.

[0034] The feature extraction module is used to perform time-frequency domain transformation on the acquired acoustic emission signal, extract single-band sensor features and / or dual-band sensor comparison features, and form a multi-dimensional time-frequency statistical feature set; wherein, the single-band sensor features are calculated based on a single sensor signal, and the dual-band sensor comparison features are calculated based on the comparison relationship between high-frequency and low-frequency sensor signals at the same monitoring point;

[0035] The calculation module is used to calculate the Mahalanobis distance of the multidimensional time-frequency statistical feature set based on the health baseline constructed from historical normal acoustic emission samples under the same working conditions, and compare the square of the Mahalanobis distance with the chi-square quantile.

[0036] The diagnostic judgment module is used to determine that there is a fault in the gearbox of the drawing press when the square of the Mahalanobis distance is greater than the chi-square quantile.

[0037] A third aspect of the invention provides that the electronic device includes:

[0038] At least one processor; and

[0039] A memory communicatively connected to the at least one processor; wherein,

[0040] The memory stores a computer program that can be executed by the at least one processor to enable the at least one processor to perform the method described in any one of the first aspects.

[0041] Compared with the closest prior art, the present invention has the following beneficial effects:

[0042] This invention proposes a fault diagnosis method and system for a drawing press gearbox based on dual-band acoustic emission. This invention can accurately capture high and low frequency acoustic emission signals from the drawing press gearbox, enabling comprehensive analysis of signal characteristics from the current operating status, historical periods, and preceding and following time periods. Furthermore, it can perform sensor self-verification and real-time fault feature identification through dual-band feature comparison. This multi-dimensional signal analysis and status verification capability effectively enhances the dynamic perception of the gearbox's health status, improving the reliability and practicality of the diagnostic process.

[0043] This invention combines the operating data of the drawing press, historical monitoring samples, and signal attenuation characteristics to construct an adaptive health baseline and update it dynamically. It accurately extracts the comparison features of single frequency band and time-frequency domain, enabling personalized and precise identification of early gearbox faults. This avoids the limitations of fixed thresholds and single frequency band monitoring, and significantly improves the accuracy and robustness of fault diagnosis under complex operating conditions.

[0044] The adaptive diagnostic mechanism based on dual-band collaboration and time-frequency statistical features proposed in this invention significantly improves the performance of early fault identification. First, it utilizes the attenuation difference between high and low frequency signals to construct dual-band contrast features, effectively mitigating time-domain and frequency-domain aliasing interference from multi-source signals and improving the accuracy of fault feature separation. Second, it designs a fixed time-window slicing strategy and a Mahalanobis distance anomaly detection model adapted to heavy-load, low-speed operating conditions, eliminating the effects of operating condition drift and random impacts, and fully preserving the early damage characteristics of key components such as gearbox bearings and gears. Attached Figure Description

[0045] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0046] Figure 1 This is a flowchart of a fault diagnosis method for a drawing press gearbox based on dual-band acoustic emission provided by an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram of the structure for verifying the attenuation characteristics of a dual-band acoustic emission sensor provided in an embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of the multi-point arrangement of the dual-band acoustic emission sensor in the gearbox of the drawing press provided in an embodiment of the present invention;

[0049] Figure 4This is a timing characteristic diagram of acoustic emission signals under normal operation and emergency stop states provided in an embodiment of the present invention;

[0050] Figure 5 This is an overall flowchart of online fault diagnosis for a drawing press gearbox provided in an embodiment of the present invention;

[0051] Figure 6 This is a schematic diagram illustrating the principle of time-frequency statistical feature extraction provided in an embodiment of the present invention;

[0052] Figure 7 This is a flowchart of the adaptive anomaly detection logic provided in an embodiment of the present invention;

[0053] Figure 8 This is a schematic diagram of a fault diagnosis device for a drawing press gearbox based on dual-band acoustic emission provided in an embodiment of the present invention;

[0054] Figure 9 This is a schematic diagram of the electronic device structure used to implement the methods and system embodiments of this application;

[0055] 1-Tensile specimen; 2-Low-frequency acoustic emission sensor; 3-Multi-channel signal acquisition unit; 4-Amplifier; 5-High-frequency acoustic emission sensor; 6-Tensile testing machine; 7-Rear bearing of intermediate shaft A; 8-Rear gear of intermediate shaft A; 9-Intermediate shaft A; 10-High-speed shaft; 11-Front gear of intermediate shaft A; 12-Front bearing of intermediate shaft A; 13-Large gear of intermediate shaft A; 14-Intermediate bearing of high-speed shaft; 15-High-speed shaft gear; 16-Front bearing of high-speed shaft; 17-Flywheel 18-Idler wheel, 19-Idler wheel shaft bearing, 20-Idler wheel shaft, 21-Large gear of intermediate shaft B, 22-Front gear of intermediate shaft B, 23-Front bearing of intermediate shaft B, 24-Box and bearing seat, 25-Rear bearing of intermediate shaft B, 26-Rear gear of intermediate shaft B, 27-Intermediate shaft B, 28-High-frequency acoustic emission sensor, 29-Rear bearing of high-speed shaft, 30-Low-frequency acoustic emission sensor; 31-Upper die of drawing press, 32-Sheet metal, 33-Lower die of drawing press. Detailed Implementation

[0056] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of the present invention and are therefore merely examples, and should not be construed as limiting the scope of protection of the present invention.

[0057] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by those skilled in the art to which this invention pertains.

[0058] This invention provides a fault diagnosis method and system for gearboxes of drawing presses based on dual-band acoustic emission, which can be widely used for early damage identification and warning of gearboxes in heavy-duty, low-speed drawing presses in automotive stamping production lines. The embodiments of this invention are described below with reference to the accompanying drawings.

[0059] Example 1: As Figure 1 This embodiment provides a fault diagnosis method for a drawing press gearbox based on dual-band acoustic emission, the method comprising:

[0060] S101 sets up acoustic emission sensors of different frequency bands at preset monitoring points in the gearbox of the drawing press, and synchronously collects acoustic emission signals of each frequency band at each monitoring point.

[0061] S102 performs time-frequency domain transformation on the acquired acoustic emission signal to obtain a time-frequency point matrix, extracts single-band sensor features and / or dual-band sensor comparison features, and constructs a multi-dimensional time-frequency statistical feature set; wherein, the single-band sensor features are calculated based on a single sensor signal, and the dual-band sensor comparison features are calculated based on the comparison relationship of different frequency band sensor signals at the same monitoring point;

[0062] S103 constructs a healthy baseline based on historical normal acoustic emission samples under the same working conditions, calculates the Mahalanobis distance of the current multidimensional time-frequency statistical feature vector, and compares the square of the Mahalanobis distance with the chi-square quantile.

[0063] S104 When the square of the Mahalanobis distance is greater than the chi-square quantile, it is determined that there is a fault in the gearbox of the drawing press.

[0064] It should be noted that, before implementing the above method, an attenuation characteristic testing device was built in this embodiment to verify the attenuation difference between high- and low-frequency acoustic emission sensors. For example... Figure 2 The specific setup method includes: fixing the tensile specimen 1 onto the tensile testing machine 6, and simultaneously installing a low-frequency acoustic emission sensor 2 and a high-frequency acoustic emission sensor 5 near the tensile specimen 1. The output signals of the two sensors are amplified by amplifier 4 and then synchronously acquired by a multi-channel signal acquisition unit 3. Tensile tests verify that the high-frequency acoustic emission sensor 5 exhibits rapid signal attenuation, short propagation distance, and high sensitivity; while the low-frequency acoustic emission sensor 2 exhibits slow signal attenuation, long propagation distance, and wide coverage, providing a basis for subsequent dual-band collaborative monitoring.

[0065] In the above embodiments, multi-point dual-band acoustic emission sensors are arranged at preset monitoring points in the gearbox of the drawing press:

[0066] like Figure 3The gearbox of the drawing press has a multi-stage transmission structure. The core transmission components include the high-speed shaft 10, the intermediate shaft A9, the intermediate shaft B27, and the idler shaft 20, which are installed inside the housing and bearing seat 24.

[0067] High-speed shaft 10: It is equipped with a high-speed shaft gear 15, which is supported by the high-speed shaft front bearing 16, the high-speed shaft intermediate bearing 14, and the high-speed shaft rear bearing 29; the input end is connected to the flywheel 17 for transmitting power.

[0068] Intermediate shaft A9: It is equipped with intermediate shaft A front gear 11, intermediate shaft A rear gear 8, and intermediate shaft A large gear 13, and is supported by intermediate shaft A front bearing 12 and intermediate shaft A rear bearing 7; intermediate shaft A large gear 13 directly meshes with high-speed shaft gear 15 for transmission.

[0069] Intermediate shaft B27: Its structure is symmetrical to that of intermediate shaft A9. It is equipped with intermediate shaft B front gear 22, intermediate shaft B rear gear 26, and intermediate shaft B large gear 21, and is supported by intermediate shaft B front bearing 23 and intermediate shaft B rear bearing 25. Intermediate shaft B large gear 21 meshes with high-speed shaft gear 15 through idler gear 18.

[0070] Idler shaft 20: Idler shaft 18 is supported by idler shaft bearing 19 to realize power reversal and transmission transition.

[0071] To cover all critical bearings and meshing areas, eight monitoring points are set up at the bearing installation positions of the housing and bearing housing 24 corresponding to the high-speed shaft, intermediate shaft A, intermediate shaft B, and idler shaft 20. Each point is simultaneously equipped with a high-frequency acoustic emission sensor 28 and a low-frequency acoustic emission sensor 30, which are used to capture high-frequency damage signals and low-frequency long-distance propagation signals, respectively, to achieve full coverage monitoring of the entire transmission chain.

[0072] During the operation of the drawing press gearbox, the upper die 31 of the drawing press descends, and the sheet metal 32 between it and the lower die 33 undergoes plastic deformation, completing the drawing process. Within a single production cycle, three mechanical impacts occur sequentially: die contact, drawing, and top dead center stop. Simultaneously, the oil circuit distributor periodically operates, generating dense periodic pulses. When emergency braking, dual-material alarm, or overload protection occurs, the machine enters an emergency stop state, and the signal exhibits irregular high-amplitude pulses.

[0073] The dual-band sensors (28, 30) collect acoustic emission signals under the above working conditions in real time, and store them synchronously by the multi-channel signal acquisition unit 3, providing raw data for subsequent preprocessing, feature extraction and anomaly detection.

[0074] The PLC commands differentiate between normal operating conditions and emergency stop conditions. Normal operating conditions include stable stamping, die contact, drawing, and top dead center dwell; emergency stop conditions include emergency braking, dual-material alarm, and overload protection triggering. Based on the duration of a single production cycle, signals are sliced ​​into fixed time windows, ensuring each slice contains the same integer number of hydraulic pulses and one complete mechanical impact, eliminating interference from temporal position uncertainties.

[0075] Data from the high-frequency acoustic emission sensor 28 and the low-frequency acoustic emission sensor 30 in the 60-70kHz crossover frequency band are extracted, the correlation coefficient is calculated and compared with historical benchmarks to determine whether the sensors are faulty. If a single sensor fails, the corresponding feature dimension is removed. If both sensors fail, a point failure prompt is output.

[0076] The complex operating conditions of a drawing press result in the measured signals from a multi-point dual-band acoustic emission sensor containing various physical information components. For example... Figure 4 As shown, under normal operating conditions, the acoustic emission signal exhibits a development pattern with a single production cycle as its time period. It also contains dense, periodic pulses generated by the near-end oil circuit distributor, and the timing of these pulses is uncertain in the time domain compared to the corresponding production cycle. Within a single production cycle, there are three distinct pulse signals, corresponding to the following physical states: the upper die 31 of the drawing press just contacts the sheet metal 32 on the lower die 33; the upper die 31 continues its downward movement to complete the drawing process; and the upper die 31 is at its top dead center. Different parts have significantly different production cycles, drawing depths, and working loads, resulting in a strong correlation between the acoustic emission characteristics, such as pulse amplitude and pulse period, and the part's structure. Furthermore, during operation, the drawing press inevitably experiences sudden stops (such as emergency braking, dual-material alarms, and overload protection), at which point multiple irregular high-amplitude impact signals will appear in the acoustic emission signal.

[0077] In step S101 above, the preset monitoring points are determined based on the multi-stage transmission structure of the gearbox, covering the bearing seats and idler shaft bodies corresponding to the high-speed shaft, intermediate shaft, and idler shaft.

[0078] The acoustic emission sensors of different frequency bands include a first frequency band sensor and a second frequency band sensor. The operating frequency band of the first frequency band sensor is higher than that of the second frequency band sensor, and there is an overlapping frequency band between the two for sensor status self-diagnosis.

[0079] In step S102 above, before extracting the features of a single-band sensor and / or the comparison features of a dual-band sensor, the method further includes: obtaining the status instructions of the programmable logic controller of the drawing press, labeling the collected acoustic emission signals with status tags, and distinguishing them into normal working status data and emergency stop status data.

[0080] Based on the duration of a single production cycle, the labeled acoustic emission data is sliced ​​into fixed time windows so that each data slice contains the same integer number of periodic pulse signals and the mechanical impact signal within a complete cycle.

[0081] The normal working state data includes stable stamping, die contact, drawing and top dead center dwell; the emergency stop data includes emergency braking, dual material alarm and overload protection trigger; the time window length of the fixed time window slice is matched with the single production cycle time, and the slice start point is aligned with the start time of the production cycle.

[0082] In one embodiment, the single-band sensor characteristics are obtained by statistically calculating the time-frequency point matrix corresponding to a single sensor, including specific frequency band energy ratio, time-spectrum kurtosis, time-spectrum energy entropy, instantaneous frequency standard deviation, and ridge continuity.

[0083] The dual-band sensor comparison features are obtained by comparing and calculating the time-frequency point matrices corresponding to sensors of different frequency bands at the same monitoring point, including dual-band kurtosis ratio, dual-band pulse index ratio, dual-band energy logarithm ratio, dual-band margin index ratio, and dual-band time-frequency distribution similarity.

[0084] Furthermore, prior to fault diagnosis, the validity of the sensor data is verified, including:

[0085] Extract data subsets from sensors of different frequency bands at the same monitoring point within overlapping frequency bands, calculate the correlation coefficient between the data subsets, and compare the current correlation coefficient with the baseline value of the correlation coefficient under historical normal conditions;

[0086] If the comparison result exceeds the preset threshold range, the corresponding sensor is determined to be faulty, and the feature dimension corresponding to the sensor is removed in the subsequent Mahalanobis distance calculation; if all sensors at the same monitoring point are determined to be faulty, a prompt message indicating that the monitoring point is in failure is output.

[0087] In step S102 above, a short-time Fourier transform is performed on the preprocessed signal to generate a time-frequency point matrix. Single-band features (specific frequency band energy ratio, time-frequency kurtosis, time-frequency energy entropy, instantaneous frequency standard deviation, ridge continuity) and dual-band comparative features (kurtosis ratio, impulse index ratio, energy logarithm ratio, margin index ratio, time-frequency distribution similarity) are extracted to form a multi-dimensional time-frequency statistical feature set.

[0088] In one embodiment, a short-time Fourier transform is performed on the preprocessed sliced ​​signal using a Hanning window with a window length of 256 points and an overlap rate of 50%, generating a two-dimensional time-frequency matrix (time × frequency). Single-band features are then calculated based on this matrix.

[0089] Specific frequency band energy ratio: Calculate the proportion of energy in the fault-sensitive frequency band to the total energy;

[0090] Time-frequency kurtosis: reflects the steepness of the time-frequency distribution, and increases significantly during anomalies;

[0091] Time-spectral energy entropy: characterizes the uniformity of energy distribution; the entropy value decreases during a fault.

[0092] Instantaneous frequency standard deviation: reflects the degree of frequency fluctuation; fluctuations intensify during damage.

[0093] Ridge continuity: characterizes the integrity of the time-frequency ridge line of the signal; the continuity deteriorates as the crack propagates.

[0094] Calculate the dual-band contrast features:

[0095] Dual-band kurtosis ratio: high-frequency kurtosis / low-frequency kurtosis, the ratio changes abruptly when damaged;

[0096] Dual-band pulse index ratio: High-frequency pulse index / Low-frequency pulse index;

[0097] Dual-band energy logarithmic ratio: ln(high-frequency energy / low-frequency energy);

[0098] Dual-band margin ratio: High-frequency margin ratio / Low-frequency margin ratio;

[0099] Dual-band time-frequency distribution similarity: Calculate the cosine similarity of the time-frequency matrices of the two bands. The similarity decreases during a fault.

[0100] In one embodiment, such as Figure 5As shown, based on the acoustic emission characteristics of the gearbox of the drawing press, the designed online diagnosis process for early faults of the drawing gearbox mainly includes three key links: building a historical monitoring database, self-diagnosis of sensor status at monitoring points, and online diagnosis of gearbox faults. (1) Building a historical monitoring database: ① First, the PLC instructions distinguish between normal operation and emergency stop. For different part structures, high-frequency acoustic emission monitoring data and low-frequency acoustic emission monitoring data of each monitoring point are collected sequentially within a set sampling time. ② Second, based on a single production cycle, fixed time window samples are divided for these monitoring data so that each time window sample contains the same integer number of oil circuit distributor pulses and mechanical impacts within a complete cycle, effectively avoiding the influence of the uncertainty of the time domain position of oil circuit distributor pulses and mechanical impacts. ③ Then, signal features that are more sensitive to fault information are extracted from each fixed time window sample. ④ Finally, the interaction law between features and equipment status is deeply explored through mathematical models to form an abnormality detection standard. (2) Sensor status self-diagnosis at monitoring points: ① First, before the monitoring task begins, the part structure and the status of the drawing press are identified sequentially through PLC instructions, and the corresponding historical monitoring database is called; ② Second, for the current monitoring data of the high-frequency acoustic emission sensor and the current monitoring data of the low-frequency acoustic emission sensor, the cross-frequency band data (60-70 kHz) is extracted using the frequency band filtering method, and the cross-frequency band data is processed by combining correlation analysis and historical data comparison to realize the self-diagnosis of the current sensor status. (3) Gearbox fault online diagnosis: ① First, after completing the sensor status self-diagnosis, the same fixed time window sample division mechanism and feature extraction method are used to construct features; ② Second, the features are compared with the abnormal detection standard. If they do not meet the judgment standard, the fault in the gearbox is indicated by the on-site alarm.

[0101] The complex operating conditions and random noise interference of the press gearbox environment cause the acquired acoustic emission signals to exhibit nonlinear and non-stationary characteristics. Furthermore, common forms of impact fatigue can cause potential fault signals to appear simultaneously with signals from mechanical impacts, sudden stops, and oil circuit pulses. The resulting "time-domain overlap and frequency-domain aliasing" characteristics make it difficult for conventional time-domain and frequency-domain analyses to accurately extract fault-sensitive features. For example... Figure 6As shown, feature extraction is performed on samples within a fixed time window within a single production cycle. A short-time Fourier transform is used to convert the signal analysis domain to the time-frequency domain. Statistical feature calculations are performed based on the time-frequency point matrix. Potential fault signals are stripped away by identifying differences in the distribution of features in the time-frequency space. Depending on the sensor object, statistical features can be divided into single-band sensor features (such as specific frequency band energy ratio, time-frequency kurtosis, time-frequency energy entropy, instantaneous frequency standard deviation, ridge continuity, etc.) and dual-band sensor contrast features (dual-band kurtosis ratio, dual-band impulse index ratio, dual-band energy logarithmic ratio, dual-band margin index ratio, dual-band time-frequency distribution similarity, etc.). The inconsistent attenuation characteristics between high-frequency acoustic emission sensor signals and high-frequency acoustic emission sensor signals inherently possess anti-time-domain and frequency-domain aliasing capabilities for dual-band sensor contrast features, providing a reliable guarantee for high-quality, high-sensitivity, and robust feature extraction.

[0102] In step S103 above, based on the healthy baseline constructed from historical normal acoustic emission samples under the same operating conditions, the Mahalanobis distance of the current multidimensional time-frequency statistical feature vector is calculated, and the square of the Mahalanobis distance is compared with the chi-square quantile. The healthy baseline includes the mean vector and covariance matrix calculated from the multidimensional time-frequency statistical feature set of historical normal acoustic emission samples. The healthy baseline is dynamically updated as the collected samples are updated to adapt to the changes in data distribution caused by the drift of the drawing press operating conditions.

[0103] The Mahalanobis distance is calculated using the following formula:

[0104] ;

[0105] Where x is the multidimensional time-frequency statistical feature vector extracted from the current monitoring data, μ is the mean vector of the health baseline, and Σ is the covariance matrix of the health baseline.

[0106] In one embodiment, during normal equipment operation, 1000 to 5000 sets of normal samples under the same operating conditions are collected, feature vectors are extracted to construct a health baseline, and the mean vector μ and covariance matrix Σ are calculated. The baseline is dynamically updated with the addition of new normal samples each day to adapt to operating condition drift.

[0107] During online monitoring, the current signal feature vector x is extracted in real time and substituted into the Mahalanobis distance formula to calculate DM: ;

[0108] Mahalanobis distance squared DM 2 obey 2 (N) distribution, where N is the feature dimension. With a confidence level of 99.7%, the corresponding quantiles are obtained from the chi-square table. If the current DM... 2If the value is greater than the quantile, an early fault in the gearbox is determined, triggering an audible and visual alarm and indicating the fault location; otherwise, it is determined to be normal operation.

[0109] In step S104 above, the chi-square quantile is determined by looking up the chi-square distribution table based on the effective dimension of the multidimensional time-frequency statistical feature vector and the preset confidence level.

[0110] In one embodiment, as mentioned above, the acoustic emission signal characteristics of the gearbox of a drawing press are influenced by numerous factors, and the mode of using a fixed model and fixed threshold for anomaly detection is difficult to meet actual needs; moreover, the press is in normal working condition in most cases, and the failure of important components is not known. Therefore, it is necessary to construct an adaptive anomaly detection standard using normal samples. Figure 7 As shown, after sensor self-diagnosis, the sensor status can result in three outcomes: "both frequency band sensors are normal," "one sensor is faulty," and "both frequency band sensors are faulty." When both frequency band sensors are faulty, the monitoring point will directly lose its status identification capability and requires immediate sensor repair. When both frequency band sensors are normal, N-dimensional time-frequency statistical features are extracted according to the construction process of the historical data monitoring database, and then the historical monitoring database is called to obtain the health baseline (mean and covariance) of N-dimensional normal samples. When one sensor is faulty, N-dimensional time-frequency statistical features are extracted according to the construction process of the historical data monitoring database, and K-dimensional features related to the sensor are removed, and then the historical monitoring database is called to obtain the health baseline (mean and covariance) of (NK)-dimensional normal samples. The Mahalanobis distance transform, which has excellent adaptability to high dimensions, is used to process multidimensional time-frequency statistical features. The press is in normal working condition in most cases, and the randomness of the state influencing factors makes the correlation features of the acoustic emission signal approximately follow a normal distribution; under the premise that the N-dimensional or (NK)-dimensional time-frequency statistical features follow a multivariate normal distribution, the square of the Mahalanobis distance follows a chi-square distribution. Based on the corresponding dimension and confidence level, the chi-square quantile is determined. When the Mahalanobis distance of the multidimensional time-frequency statistical features obtained from certain monitoring data exceeds this chi-square quantile, it is determined that a critical component in the gearbox has malfunctioned during operation. The Mahalanobis distance is only related to the healthy baseline of normal samples and the monitoring data samples, while the chi-square quantile is determined based on the dimension of the time-frequency statistical features. Both are independent of the formulation parameters of the production process, which makes this anomaly detection method highly adaptive.

[0111] Example 2: Based on the same technical concept described above, Example 2 of this invention also provides a fault diagnosis system for a drawing press gearbox based on dual-band acoustic emission, such as... Figure 8 As shown, it includes: a synchronous acquisition module 210, a feature extraction module 220, a calculation module 230, and a diagnostic judgment module 240. Among them:

[0112] The synchronous acquisition module 210 is used to arrange acoustic emission sensors of different frequency bands at preset monitoring points in the gearbox of the drawing press, and synchronously acquire acoustic emission signals of each frequency band at each monitoring point.

[0113] The feature extraction module 220 is used to perform time-frequency domain transformation on the acquired acoustic emission signal, extract single-band sensor features and / or dual-band sensor comparison features, and form a multi-dimensional time-frequency statistical feature set; wherein, the single-band sensor features are calculated based on a single sensor signal, and the dual-band sensor comparison features are calculated based on the comparison relationship between high-frequency and low-frequency sensor signals at the same monitoring point;

[0114] The calculation module 230 is used to calculate the Mahalanobis distance of the multidimensional time-frequency statistical feature set based on the health baseline constructed from historical normal acoustic emission samples under the same working conditions, and compare the square of the Mahalanobis distance with the chi-square quantile.

[0115] The diagnostic judgment module 240 is used to determine that there is a fault in the gearbox of the drawing press when the square of the Mahalanobis distance is greater than the chi-square quantile.

[0116] Example 3: This embodiment of the invention also provides an electronic device corresponding to Example 1. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it causes the processor to perform the steps of any one of the methods described in S101-S104.

[0117] like Figure 9 As shown, the electronic device may include: at least one processor 31, at least one network interface 35, user interface 34, memory 36, and at least one communication bus 32.

[0118] The communication bus 32 is used to enable communication between these components.

[0119] The user interface 34 may include a display screen and a camera. Optionally, the user interface 34 may also include a standard wired interface and a wireless interface.

[0120] The network interface 35 may optionally include a standard wired interface or a wireless interface (such as a WIFI interface).

[0121] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 36, and by calling data stored in the memory 36. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0122] The memory 36 may include random access memory (RAM) or read-only memory. Optionally, the memory 36 may include a non-transitory computer-readable storage medium. The memory 36 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 36 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 36 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 9 As shown, the memory 36, which serves as a computer storage medium, may include an operating system, a network communication module, and a user interface module.

[0123] exist Figure 9In the electronic device shown, the user interface 34 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call the application program stored in the memory 36, which is a fault diagnosis method for a drawing press gearbox based on dual-band acoustic emission. When executed by one or more processors 31, the electronic device executes one or more of the fault diagnosis methods for a drawing press gearbox based on dual-band acoustic emission as described in steps S101-S104 of the above embodiment.

[0124] Those skilled in the art will clearly understand that the technical solutions of this application can be implemented using software and / or hardware. In this specification, "module" refers to software and / or hardware capable of independently performing or cooperating with other components to perform a specific function. Hardware may include, for example, a field-programmable gate array (FPGA), an integrated circuit (IC), etc.

[0125] Specifically, the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0126] The code of the computer program can be in the form of source code, object code, executable file, or some intermediate form.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A fault diagnosis method for a drawing press gearbox based on dual-band acoustic emission, characterized in that, The method includes: At the preset monitoring points of the gearbox of the drawing press, acoustic emission sensors of different frequency bands are arranged to collect acoustic emission signals of each frequency band at each monitoring point simultaneously. The acquired acoustic emission signals are transformed in the time-frequency domain to obtain a time-frequency point matrix. Single-band sensor features and / or dual-band sensor comparison features are extracted to form a multi-dimensional time-frequency statistical feature set. The single-band sensor features are calculated based on a single sensor signal, and the dual-band sensor comparison features are calculated based on the comparison relationship between sensor signals of different frequency bands at the same monitoring point. Based on a healthy baseline constructed from historical normal acoustic emission samples under the same operating conditions, the Mahalanobis distance of the current multidimensional time-frequency statistical feature vector is calculated, and the square of the Mahalanobis distance is compared with the chi-square quantile. When the square of the Mahalanobis distance is greater than the chi-square quantile, the gearbox of the drawing press is determined to be faulty.

2. The method according to claim 1, characterized in that, The preset monitoring points are determined based on the multi-stage transmission structure of the gearbox, covering the bearing seats and idler shaft bodies corresponding to the high-speed shaft, intermediate shaft, and idler shaft.

3. The method according to claim 1, characterized in that, The acoustic emission sensors of different frequency bands include a first frequency band sensor and a second frequency band sensor. The operating frequency band of the first frequency band sensor is higher than that of the second frequency band sensor, and there is an overlapping frequency band between the two for sensor status self-diagnosis.

4. The method according to claim 1, characterized in that, Before extracting the single-band sensor features and / or the dual-band sensor comparison features, the method further includes: The programmable logic controller status instructions of the drawing press are obtained, and the collected acoustic emission signals are labeled with status tags to distinguish between normal working status data and emergency stop status data. Based on the duration of a single production cycle, the labeled acoustic emission data is sliced ​​into fixed time windows so that each data slice contains the same integer number of periodic pulse signals and the mechanical impact signal within a complete cycle. The normal working state data includes stable stamping, die contact, drawing and top dead center dwell; the emergency stop data includes emergency braking, dual material alarm and overload protection trigger; the time window length of the fixed time window slice is matched with the single production cycle time, and the slice start point is aligned with the start time of the production cycle.

5. The method according to claim 1, characterized in that, The single-band sensor characteristics are obtained by statistically calculating the time-frequency point matrix corresponding to a single sensor, including specific frequency band energy ratio, time-spectrum kurtosis, time-spectrum energy entropy, instantaneous frequency standard deviation, and ridge continuity. The dual-band sensor comparison features are obtained by comparing and calculating the time-frequency point matrices corresponding to sensors of different frequency bands at the same monitoring point, including dual-band kurtosis ratio, dual-band pulse index ratio, dual-band energy logarithm ratio, dual-band margin index ratio, and dual-band time-frequency distribution similarity.

6. The method according to claim 3, characterized in that, Before troubleshooting, the validity of the sensor data is verified, which includes: Extract data subsets from sensors of different frequency bands at the same monitoring point within overlapping frequency bands, calculate the correlation coefficient between the data subsets, and compare the current correlation coefficient with the baseline value of the correlation coefficient under historical normal conditions; If the comparison result exceeds the preset threshold range, the corresponding sensor is determined to be faulty, and the feature dimension corresponding to the sensor is removed in the subsequent Mahalanobis distance calculation; if all sensors at the same monitoring point are determined to be faulty, a prompt message indicating that the monitoring point is in failure is output.

7. The method according to claim 1, characterized in that, The health baseline includes the mean vector and covariance matrix calculated from the multidimensional time-frequency statistical feature set of historical normal acoustic emission samples; the health baseline is dynamically updated as the collected samples are updated to adapt to the changes in data distribution caused by the drift of the drawing press operating conditions.

8. The method according to claim 1, characterized in that, The Mahalanobis distance is calculated using the following formula: ； Where x is the multidimensional time-frequency statistical feature vector extracted from the current monitoring data, μ is the mean vector of the health baseline, and Σ is the covariance matrix of the health baseline.

9. A fault diagnosis system for a drawing press gearbox based on dual-band acoustic emission, characterized in that, include: The synchronous acquisition module is used to deploy acoustic emission sensors of different frequency bands at preset monitoring points in the gearbox of the drawing press, and synchronously acquire acoustic emission signals of each frequency band at each monitoring point. The feature extraction module is used to perform time-frequency domain transformation on the acquired acoustic emission signal, extract single-band sensor features and / or dual-band sensor comparison features, and form a multi-dimensional time-frequency statistical feature set; wherein, the single-band sensor features are calculated based on a single sensor signal, and the dual-band sensor comparison features are calculated based on the comparison relationship between high-frequency and low-frequency sensor signals at the same monitoring point; The calculation module is used to calculate the Mahalanobis distance of the multidimensional time-frequency statistical feature set based on the health baseline constructed from historical normal acoustic emission samples under the same working conditions, and compare the square of the Mahalanobis distance with the chi-square quantile. The diagnostic judgment module is used to determine that there is a fault in the gearbox of the drawing press when the square of the Mahalanobis distance is greater than the chi-square quantile.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-8.

Images