Fault Prediction Method and System for Flattening Equipment Based on Data Analysis

Abstract

The present invention relates to the field of data processing, and particularly to a method and system for predicting the faults of a leveling device based on data analysis. First, sensor arrays at the four corners of the frame are used to synchronously collect signals, and angular domain resampling is performed in combination with the encoder speed to determine a data sequence with spatio-temporal alignment. Subsequently, the energy of specific fault frequencies is extracted through band-pass filtering and envelope demodulation to obtain a normalized observation vector. Then, a theoretical geometric model is constructed based on the physical distance attenuation law, and the direction consistency index between the observation vector and the influence vectors of each support roll is calculated. Finally, the spatial focusing degree is calculated using the kurtosis statistical feature, faults are determined through a dynamic reference line, and the precise positioning of the support roll with a specific fault is achieved based on the maximum similarity index.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically to a method and system for predicting faults in leveling equipment based on data analysis. Background Technology

[0002] As a core piece of equipment in high-precision metal sheet processing, the leveling machine's densely packed support rollers are subjected to alternating heavy loads over long periods, making them highly susceptible to early failures such as surface peeling, which directly affect the leveling quality of the sheet metal. Therefore, accurate early prediction and location of support roller failures are of great significance.

[0003] Existing fault location techniques typically employ vibration analysis based on distributed sensor arrays. The core idea is to establish an energy attenuation model based on geometric distance, based on the physical law that vibration waves attenuate with increasing propagation distance in a solid medium. This type of method usually calculates the theoretical attenuation coefficient from each potential fault point to the sensor and compares it with the actual vibration energy distribution received by the sensor, attempting to infer the specific location of the fault source through changes in the energy gradient.

[0004] However, this conventional method based on the physical distance attenuation model has the following drawbacks. On the one hand, the working load fluctuates frequently during the leveling process, and the sensor may experience sensitivity drift during use. These factors can cause non-fault-related global fluctuations in the absolute energy amplitude, resulting in large errors in model comparisons that rely on absolute energy values. On the other hand, when the support roller is in the early stage of failure, the impact signal it generates is very weak and is often drowned out by mechanical background noise. Existing technologies cannot effectively distinguish specific faulty rollers from numerous support rollers with similar geometric positions through small energy differences, resulting in insufficient spatial resolution and a tendency to cause false alarms or missed alarms. Summary of the Invention

[0005] To address the problem of insufficient spatial resolution in the physical distance attenuation model, which easily leads to false alarms or missed alarms, this invention proposes a data analysis-based fault prediction method for leveling equipment in its first aspect. The method includes: synchronously acquiring multi-channel vibration signals using a distributed sensor array installed on the leveling equipment to obtain vibration data; extracting energy values ​​at a set characteristic frequency from the vibration data of each channel, and normalizing the energy values ​​of all channels to obtain an energy distribution vector; obtaining influence vectors of multiple support rollers to be monitored based on the geometric structure information of the leveling equipment; the influence vectors contain theoretical energy corresponding to each sensor, and the theoretical energy is inversely correlated with the square of the Euclidean distance from the geometric center of the corresponding support roller to the corresponding sensor; calculating the spatial similarity between the energy distribution vector and the influence vectors of each support roller to form a spatial similarity spectrum; calculating the kurtosis of the spatial similarity spectrum to obtain spatial focus; and determining that the equipment has a fault in response to the spatial focus exceeding a set threshold, identifying the support roller with the highest spatial similarity as the fault source.

[0006] Compared to existing technologies that rely solely on a single measurement point or simple time-frequency domain analysis, this approach introduces spatial dimension analysis. By calculating the spatial similarity between the actual collected energy distribution vector and the theoretical influence vector based on a physical distance attenuation model, complex mechanical fault diagnosis is transformed into a spatial pattern matching problem. Using spatial focus as a criterion, environmental background noise interference can be effectively filtered out. In the complex multi-support roller structure of leveling equipment, the specific faulty support roller with the highest spatial similarity can be accurately located, significantly improving the resolution and accuracy of fault source localization.

[0007] Furthermore, the method also includes preprocessing the multi-channel vibration signals. Specifically, the preprocessing involves: acquiring the rotational speed information of the main leveling roller of the leveling equipment; and performing angular domain resampling on the original vibration signals of each channel based on the rotational speed information to obtain time-domain aligned vibration data.

[0008] Compared to traditional time-domain analysis, this invention eliminates the spectral ambiguity caused by speed fluctuations during the operation of the leveling equipment, ensuring the stability of the vibration signal in the angular domain. This makes the energy extraction at subsequent characteristic frequencies more accurate and improves the detection reliability under non-constant speed conditions.

[0009] Furthermore, obtaining the rotational speed information includes: reading the encoder pulse signal of the main leveling roller drive motor through a counter card; and calculating the instantaneous rotational speed of the main leveling roller based on the pulse signal.

[0010] Furthermore, the energy values ​​at a set characteristic frequency are extracted from the vibration data of each channel, including:

[0011] The vibration data of each channel is bandpass filtered; the filtered vibration data is subjected to Hilbert transform to obtain an analytical signal, and the magnitude of the analytical signal is calculated to obtain an envelope signal; the envelope signal is subjected to fast Fourier transform, and the amplitude at a set characteristic frequency is extracted as the energy value.

[0012] Compared to directly performing spectral analysis on the original signal, this method can effectively demodulate the low-frequency fault impact characteristics modulated by the high-frequency resonant signal. By extracting the amplitude of a set characteristic frequency in the envelope signal as the energy value, it can sensitively capture the impact energy caused by early weak faults from strong background noise, thus improving the signal-to-noise ratio of feature extraction.

[0013] Furthermore, the bandpass filtering process includes: using a Chebyshev Type II bandpass filter to filter the vibration data of each channel; the passband range of the bandpass filter is 2kHz~5kHz.

[0014] This invention specifically preserves the resonant frequency band that is easily triggered by early failures of the support roller. By utilizing the rapid attenuation characteristics of the Chebyshev Type II filter in the stopband, it effectively filters out low-frequency mechanical vibrations and high-frequency interference, thus preserving the integrity of fault characteristic information to the greatest extent.

[0015] Furthermore, the specific method for calculating the theoretical energy is as follows:

[0016] ;

[0017] in Indicates the first When the first support roller fails, the first The theoretical energy corresponding to each sensor; For the first The support roller to the first Euclidean distance between the sensors; This represents the set structural damping constant.

[0018] Furthermore, the method for calculating the spatial focus is as follows:

[0019] ;

[0020] in This indicates the spatial focus. This represents the total number of support rollers; To prevent division by zero constant; and These represent the mean and standard deviation of the spatial similarity spectrum, respectively. The first in the spatial similarity spectrum Spatial similarity of the support rollers.

[0021] Furthermore, the spatial similarity is cosine similarity.

[0022] By defining spatial similarity as cosine similarity, compared to metrics such as Euclidean distance, cosine similarity focuses more on the alignment of vectors in direction rather than their absolute numerical values. This means that fault location results primarily depend on the relative proportions of energy distribution across channels, rather than being affected by fluctuations in the overall vibration amplitude of the equipment, greatly enhancing the algorithm's adaptability to changes in operating conditions.

[0023] Furthermore, the sensing array includes a single-axis piezoelectric accelerometer installed at each of the four corner points of the outer frame of the roller box of the leveling equipment.

[0024] In a second aspect, the present invention provides a data analysis-based fault prediction system for leveling equipment, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the data analysis-based fault prediction method for leveling equipment of the present invention is implemented.

[0025] The technical effects of this invention are as follows:

[0026] The innovation of this invention lies in overcoming the limitations of traditional time-frequency analysis. It utilizes a distributed sensor array and a device geometric model to construct a theoretical energy vector that decays with distance. By calculating the spatial similarity and kurtosis between the actual energy distribution and the theoretical vector, fault diagnosis is transformed into a spatial pattern matching problem. This method effectively solves the challenge of accurately distinguishing and locating specific faulty support rollers in complex multi-roller structures, significantly improving the accuracy and robustness of fault prediction. Attached Figure Description

[0027] Figure 1 This is a schematic flowchart illustrating a data analysis-based fault prediction method for leveling equipment according to an embodiment of the present invention.

[0028] Figure 2 This is a schematic diagram illustrating the instantaneous frequency change curve after bandpass filtering of the fault impact resonance frequency band in an embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram illustrating the comparison of the effects of Hilbert transform and envelope extraction on the filtered signal in an embodiment of the present invention;

[0030] Figure 4 This is a schematic planar diagram illustrating the relationship between the sensor array layout and the geometric location of the support roller fault source in an embodiment of the present invention;

[0031] Figure 5 This is a schematic diagram illustrating the Euclidean distance matrix heat map from each support roller to each sensor node in an embodiment of the present invention.

[0032] Figure 6 This is a schematic bar chart illustrating the comparison of the spatial orientation consistency index distribution patterns under normal and fault states in an embodiment of the present invention.

[0033] Figure 7 This is a schematic statistical analysis diagram illustrating the automatic locking and identification of faulty support rollers based on dynamic baselines in an embodiment of the present invention.

[0034] Figure 8 This is a schematic diagram illustrating the structural block diagram of a data analysis-based fault prediction system for leveling equipment according to an embodiment of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0037] Example of a data analysis-based fault prediction method for leveling equipment:

[0038] like Figure 1 As shown, the data analysis-based fault prediction method for leveling equipment of the present invention includes:

[0039] S101. Multi-channel vibration signals are synchronously acquired based on a distributed array to construct a spatiotemporally aligned basic data sequence.

[0040] In this embodiment, to capture high-frequency impact signals and ensure spatial distinguishability, uniaxial piezoelectric accelerometers are installed at the four corners of the outer frame of the leveling machine roller box, constructing a system containing... A sensor array with multiple measuring points. The sensitivity of the accelerometer is exemplarily set to... Frequency response range coverage to This is to meet the requirement of recording early, subtle fault characteristics.

[0041] During the data acquisition process, the acquisition card uses The system simultaneously acquires the raw vibration signals of all channels at a high sampling rate. At the same time, it reads the photoelectric encoder pulse signals of the servo motor via a high-speed counter card, calculating the instantaneous rotational speed of the main leveling roller in real time. To eliminate frequency ambiguity caused by speed fluctuations, the system uses the calculated speed information to perform angular domain resampling of the time-domain signal, converting the non-stationary time-domain signal into an angularly stationary signal, ensuring that the data from all subsequent channels are strictly aligned on the time and angular axes.

[0042] S102. Use bandpass filtering and envelope demodulation to extract fault feature energy and obtain a normalized energy distribution vector.

[0043] After acquiring the synchronization data, the processing unit first processes each channel. (in Signal conditioning is performed. Considering that support roller peeling faults typically generate high-frequency resonant waves, this embodiment sets a passband range of [missing information]. to The Chebyshev Type II bandpass filter filters the original signal to suppress low-frequency mechanical noise and high-frequency electromagnetic interference.

[0044] like Figure 2 As shown, to accurately capture subtle early spalling characteristics, the bandpass filter was configured to track high-frequency resonant components. The figure illustrates the instantaneous frequency fluctuations of the signal, with the dashed line indicating the resonant center frequency of the support roller bearing assembly (approximately 3500 Hz). Through the action of the Chebyshev Type II filter, the system locked onto this frequency band, filtering out irrelevant low-frequency rotational components.

[0045] Subsequently, a Hilbert Transform is performed on the filtered signal to obtain an analytic signal, and its magnitude is calculated to obtain the envelope signal. A Fast Fourier Transform (FFT) is then performed on this envelope signal to extract the characteristic frequencies corresponding to support roller faults. The amplitude energy at that point. Since the kinematic parameters of the support rollers are consistent, this characteristic frequency... All support rollers in the same row are the same.

[0046] like Figure 3 As shown, the envelope clearly depicts the energy fluctuation profile of the impact signal, and the subsequent FFT analysis is based on this envelope signal to extract feature frequencies. The amplitude energy at that point provides the raw intensity data for the subsequent vector construction.

[0047] To eliminate the influence of absolute amplitude caused by sensitivity drift of different sensors and fluctuations in operating load, the system does not directly use the absolute value of energy, but instead calculates a normalized energy distribution vector. Specifically, the energy value extracted from each channel is divided by the sum of the energies of all channels, such that... It only characterizes the relative intensity distribution of the fault impact energy received at the four corner points in space.

[0048] S103. Construct a theoretical geometric model based on the physical distance decay law to calculate spatial similarity.

[0049] Since the leveling machine frame is constructed of high-rigidity alloy steel, the propagation of vibration waves within it follows the attenuation law of sound waves in a solid medium. Therefore, this embodiment defines a geometric feature space based on physical dimensions.

[0050] First, for each support roller to be monitored Assuming the total number is Based on the known geometric center coordinates of the CAD drawing, calculate its distance to the _____th _____. Euclidean distance of each sensor .

[0051] like Figure 4 As shown, in the physical space of this embodiment, four accelerometers (S1-S4) are located at the four corners of the rectangular frame, while support rollers (R1-R8) are distributed in the central area. The figure illustrates the significant differences in the physical path length from the fourth support roller (R4) to the different sensors when it is set as the fault source. For example, R4 is only 721 mm away from sensor S2, while it is 1000 mm away from sensor S1, and the furthest sensor S4 is 1844 mm away.

[0052] Subsequently, the following calculation method is used to obtain the first... Each support roller pair The influence vector of each sensor :

[0053] ;

[0054] in Indicates the assumption that the first When the first support roller fails, the first The relative energy theoretically received by each sensor; For the first The support roller to the first Euclidean distance between the sensors; The structural damping constant is taken as a value in this embodiment. It is used to correct nonlinear singularities under near-field effects.

[0055] As the distance between the fault point and the sensor As the value increases, the squared term in the denominator causes a rapid decrease in the value, thus reflecting the diffusion law of spherical wave energy decaying inversely with the square of the distance in isotropic solid media. Furthermore, a constant is introduced... This prevents numerical divergence when the distance approaches zero, ensuring the numerical stability of the model.

[0056] like Figure 5 As shown, the system calculated the Euclidean distance matrix from all support rollers (R1-R8) to the four sensors (S1-S4). It can be seen that each support roller has a set of non-repeating distance combinations. For example, R1 is closest to S1 (632mm), while the distance from R8 to S4 (1442mm) is symmetrically distributed with the distance from S1 (1442mm).

[0057] Next, the system calculates the observed energy distribution vector. The influence vector calculated for each support roller The directional consistency index between them, also known as spatial similarity:

[0058] ;

[0059] in For the first The directional consistency index of each support roller; Represents the dot product of vectors; The L2 norm of a vector is represented by its magnitude.

[0060] The above calculation is actually finding the cosine of the angle between two vectors. This is relevant when the actual observed energy distribution pattern... Direction and the first The theoretical influence vector corresponding to each support roller When the directions of the vectors highly overlap, the index approaches 1. Since cosine similarity is only related to the vector direction and not to the magnitude, this algorithm can effectively filter out global amplitude fluctuations caused by changes in operating load, thereby extracting spatial features that are purely determined by the location of the fault source.

[0061] S104. Statistical spatial similarity spectrum discrete features and calculation of spatial focus to achieve adaptive fault locking.

[0062] This embodiment utilizes the statistical concept of kurtosis to measure the spatial focus of fault characteristics. First, all the values ​​calculated in step S103 are... The directional consistency index of each support roller constitutes a spatial similarity spectrum vector. Then, the mean of the vector is calculated. and standard deviation Based on the above statistics, spatial focus is calculated using the following formula. :

[0063] ;

[0064] in Indicates the dimensionless spatial focus; This represents the total number of support rollers; To prevent division by zero constant, this embodiment takes... .

[0065] The fourth power operation in the above formula acts as a non-linear amplification of the differences. When the system is in a healthy state, the orientation is consistent at all locations. The differences are small, and the distribution is close to Gaussian white noise. The value is low (close to 3). And once a certain support roller... Early flaking occurs, and its corresponding It will be slightly higher than other rollers. After the fourth power calculation, the tiny difference is significantly amplified, leading to... A spike appears.

[0066] like Figure 6 As shown, it illustrates the distribution of directional consistency index under two typical conditions. Under normal conditions, the consistency coefficient of each support roller is relatively uniformly distributed with no obvious protrusions, corresponding to a low kurtosis value; while under fault conditions, the index at position R4 is significantly higher than that at other positions, exhibiting a distinct peak shape.

[0067] Finally, the system uses a dynamic baseline for decision-making. The system is maintained in real time. Historical moving average and historical volatility variance For example, retrieving a data window from the past hour. The decision logic is as follows: If... Then it is determined to be That is, to discover the fault, among which This represents the baseline adjustment coefficient, which is preferably 1 in this embodiment. At this time, the system automatically extracts the vector. The index with the largest value in the middle This index is the specific faulty support roller number that is locked, thus enabling precise location of the support roller fault.

[0068] like Figure 7 As shown, the final decision based on the dynamic baseline is presented. During the inspection, the directional consistency index of support roller R4 was not only the maximum value in the entire field, but also significantly greater than the background noise level, meeting the trigger condition for spatial focusing. Based on this, the system automatically identified R4 as the fault source.

[0069] Example of a data analysis-based fault prediction system for leveling equipment:

[0070] On the other hand, the present invention also provides a fault prediction system for leveling equipment based on data analysis. For example... Figure 8 As shown, the fault prediction system for leveling equipment based on data analysis includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the fault prediction method for leveling equipment based on data analysis according to the first aspect of the present invention.

[0071] The fault prediction system for leveling equipment based on data analysis also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0072] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained by such a computer-readable medium.

Claims

1. A fault prediction method for leveling equipment based on data analysis, characterized in that, The method includes: synchronously acquiring multi-channel vibration signals using a distributed sensor array installed on a leveling device to obtain vibration data; extracting the energy value at a set characteristic frequency from the vibration data of each channel, and normalizing the energy values ​​of all channels to obtain an energy distribution vector; Based on the geometric structure information of the leveling equipment, the influence vectors of multiple support rollers to be monitored are obtained; the influence vectors contain the theoretical energy corresponding to each sensor, and the theoretical energy is inversely correlated with the square of the Euclidean distance from the geometric center of the corresponding support roller to the corresponding sensor; The theoretical energy is calculated as follows: , Indicates the first When the first support roller fails, the first The theoretical energy corresponding to each sensor For the first The support roller to the first The Euclidean distance between the sensors This represents the set structural damping constant; Calculate the spatial similarity between the energy distribution vector and the influence vectors of each support roller to form a spatial similarity spectrum; calculate the kurtosis of the spatial similarity spectrum to obtain the spatial focus, the spatial focus being calculated as follows: , Indicates the spatial focus, The total number of support rollers, To prevent division by zero constant, and Let represent the mean and standard deviation of the spatial similarity spectrum, respectively. The first in the spatial similarity spectrum Spatial similarity of the support rollers; In response to the spatial focus exceeding a set threshold, a fault is determined in the equipment, and the support roller with the greatest spatial similarity is identified as the source of the fault.

2. The fault prediction method for leveling equipment based on data analysis according to claim 1, characterized in that, It also includes preprocessing the multi-channel vibration signal, the preprocessing specifically being: Obtain the rotational speed information of the main leveling roller of the leveling equipment; Based on the rotational speed information, the original vibration signals of each channel are resampled in the angular domain to obtain time-domain aligned vibration data.

3. The fault prediction method for leveling equipment based on data analysis according to claim 2, characterized in that, Obtaining the rotational speed information includes: The encoder pulse signal of the main leveling roller drive motor is read through the counter card; The instantaneous rotational speed of the main straightening roller is calculated based on the pulse signal.

4. The fault prediction method for leveling equipment based on data analysis according to claim 1, characterized in that, Extract the energy value at a set characteristic frequency from the vibration data of each channel, including: Bandpass filtering is applied to the vibration data of each channel; The filtered vibration data are subjected to Hilbert transform to obtain an analytical signal, and the magnitude of the analytical signal is calculated to obtain an envelope signal. Perform a Fast Fourier Transform on the envelope signal and extract the amplitude at a set characteristic frequency as the energy value.

5. The fault prediction method for leveling equipment based on data analysis according to claim 4, characterized in that, The bandpass filtering process includes: The vibration data of each channel were filtered using a Chebyshev Type II bandpass filter; The passband range of the bandpass filter is 2kHz to 5kHz.

6. The fault prediction method for leveling equipment based on data analysis according to claim 1, characterized in that, The spatial similarity is cosine similarity.

7. The fault prediction method for leveling equipment based on data analysis according to claim 1, characterized in that, The sensing array includes: A single-axis piezoelectric accelerometer is installed at each of the four corners of the outer frame of the roller box of the leveling equipment.

8. A fault prediction system for leveling equipment based on data analysis, characterized in that, It includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the fault prediction method for leveling equipment based on data analysis as described in any one of claims 1 to 7 is implemented.

Images