Artificial intelligence-based equipment failure prediction system

Abstract

This invention relates to the field of data processing technology, and more particularly to an artificial intelligence-based equipment fault prediction system. The system includes a processor and a memory. The processor executes a computer program in the memory to perform the following steps: obtaining the number of fundamental modes based on the entropy and energy of the vibration acceleration of the target equipment at different preset time scales within the current time period; adaptively adjusting the number of fundamental modes based on the marginal spectral similarity and mutual information between every two adjacent modal components after decomposing the vibration acceleration within the current time period using the fundamental mode numbers, the stability of the center frequency of each modal component at different preset time scales, and the kurtosis and instantaneous frequency of the vibration acceleration within the current time period, to obtain an adaptive number of modes; and decomposing the vibration acceleration within the current time period using the VMD algorithm based on the adaptive number of modes, thereby achieving fault prediction of the target equipment at future moments, improving the accuracy and reliability of fault prediction.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an artificial intelligence-based equipment fault prediction system. Background Technology

[0002] In the fields of industrial production and equipment operation and maintenance, equipment failure prediction is of paramount importance. As industrial equipment becomes increasingly complex and sophisticated, its failure modes are becoming more diverse and subtle. Artificial intelligence, with its powerful data processing and pattern recognition capabilities, can monitor and deeply analyze the massive amounts of data generated during equipment operation in real time, accurately identifying signs of failure in advance and achieving failure prediction. This enables enterprises to plan maintenance in advance, avoid unplanned downtime, ensure continuous and stable production, reduce maintenance costs, extend equipment lifespan, and improve overall production efficiency and safety, playing an irreplaceable and crucial role in the wave of intelligent industrial transformation.

[0003] Traditionally, Variational Mode Decomposition (VMD) algorithms are used to decompose signals into multiple modes with specific center frequencies and bandwidths, enabling accurate analysis of complex signals. These modes are then fed into artificial intelligence models (SVM, CNN, LSTM, etc.) to identify the fault types of industrial equipment. However, in traditional VMD algorithms, the number of modes K usually needs to be manually preset. This fixed setting is difficult to adapt to the dynamic changes in signals under different operating states and fault stages: in a healthy state, the signal is simple and requires fewer modes; in a fault stage, the signal is complex and requires more modes to separate fault features, which a fixed number of modes cannot adequately address. Early fault signals are not readily apparent; if the fixed number of modes is not set correctly, fault features may be mixed with noise, making effective extraction impossible. When multiple faults occur or operating conditions change, the signal composition becomes more complex, and a fixed number of modes can easily lead to mode aliasing or over-decomposition, affecting the accuracy of fault diagnosis. Furthermore, a fixed number of modes K lacks the ability to respond in real time to dynamic changes in signals and cannot adjust the decomposition strategy in a timely manner according to the degradation trend of equipment. This makes it impossible to accurately capture the fault characteristics of industrial equipment, affecting the accuracy and reliability of fault prediction for industrial equipment.

[0004] Therefore, how to adaptively obtain the number of modes in the VMD algorithm, improve the accuracy of signal decomposition, and thus improve the accuracy and reliability of fault prediction for industrial equipment has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide an artificial intelligence-based equipment fault prediction system to solve the problem of how to adaptively acquire the modal number in the VMD algorithm, improve the accuracy of signal decomposition, and thus improve the accuracy and reliability of fault prediction for industrial equipment.

[0006] This invention provides an artificial intelligence-based device fault prediction system, including a memory, a processor, and a computer program stored in the memory and running on the processor. The system is characterized in that the processor executes the computer program to perform the following steps:

[0007] Obtain the vibration acceleration of the target device at each moment in the current time period up to the current moment. Based on the entropy and energy of the vibration acceleration at at least three different preset time scales in the current time period, obtain the number of fundamental modes for decomposing the vibration acceleration in the current time period using the VMD algorithm.

[0008] Based on the kurtosis and instantaneous frequency of the vibration acceleration in the current time period, a first adjustment factor is obtained to adjust the number of basic modes;

[0009] The vibration acceleration in the current time period is decomposed using the number of basic modes to obtain the modal components of the number of basic modes in the current time period. Based on the marginal spectral similarity and mutual information between each pair of adjacent modal components, and the stability of the center frequency of each modal component at different preset time scales in the current time period, a second adjustment factor is obtained to adjust the number of basic modes.

[0010] The basic mode number is adaptively adjusted according to the first adjustment factor and the second adjustment factor to obtain the adaptive mode number. Based on the adaptive mode number, the vibration acceleration in the current time period is decomposed using the VMD algorithm to obtain the decomposition result. Based on the decomposition result, the fault prediction of the target equipment in the future time period is realized.

[0011] Preferably, the step of obtaining the number of fundamental modes for decomposing the vibration acceleration in the current time period using the VMD algorithm based on the entropy and energy of the vibration acceleration at at least three different preset time scales within the current time period includes:

[0012] By downsampling the vibration acceleration in the current time period, the information entropy of the vibration acceleration at each preset time scale is extracted, and the mean of all information entropies is normalized to obtain the signal complexity in the current time period.

[0013] Perform a Hilbert transform on the vibration acceleration in the current time period to obtain the Hilbert envelope of the vibration acceleration in the current time period. Perform an FFT transform on the Hilbert envelope to obtain the envelope spectrum. Sort the peaks in the envelope spectrum in descending order to obtain the peak sequence. Calculate the sum of the energy corresponding to the first preset number of peaks in the peak sequence and its proportion in the total energy of the envelope spectrum. Normalize the proportion to obtain the signal impulse intensity in the current time period.

[0014] The sum of the signal complexity and the signal impact intensity is normalized to obtain the basic coefficient. A preset minimum mode number and a preset maximum mode number are obtained. The preset minimum mode number is subtracted from the preset maximum mode number to obtain the target difference. The product between the basic coefficient and the target difference is calculated to obtain the target result. The sum between the minimum mode number and the target result is calculated to obtain the basic mode number for decomposing the vibration acceleration in the current time period using the VMD algorithm.

[0015] Preferably, obtaining the first adjustment factor for adjusting the number of fundamental modes based on the kurtosis and instantaneous frequency of the vibration acceleration in the current time period includes:

[0016] Based on the duration of the current time period, obtain the vibration acceleration of the target device in the historical time period up to the previous time, obtain the kurtosis of the vibration acceleration in the historical time period and record it as the historical kurtosis, obtain the kurtosis of the vibration acceleration in the current time period and record it as the current kurtosis, and normalize the absolute value of the difference between the historical kurtosis and the current kurtosis to obtain the degree of kurtosis change at the current time.

[0017] The vibration acceleration at each moment in the current time period is composed into a data sequence, the extreme values ​​in the data sequence are obtained, the absolute values ​​of the extreme values ​​in the data sequence are sorted in descending order, and the extreme values ​​corresponding to the first second preset number of absolute values ​​after sorting are taken as the target extreme values.

[0018] For any target extreme value, if the target extreme value is a maximum value, then obtain the left adjacent minimum value and the right adjacent minimum value of the target extreme value, take the time interval between the time corresponding to the left adjacent minimum value and the time corresponding to the right adjacent minimum value as the denominator, and take the absolute value of the target extreme value as the numerator to obtain the rate of change of the target extreme value.

[0019] If any of the target extreme values ​​is a minimum value, then the left and right adjacent maximum values ​​of the target extreme value are obtained. The time interval between the time corresponding to the left adjacent maximum value and the time corresponding to the right adjacent maximum value is used as the denominator, and the absolute value of the target extreme value is used as the numerator to obtain the rate of change of the target extreme value.

[0020] Obtain the rate of change of each target extreme value, and normalize the average of the rates of change of all target extreme values ​​to obtain the pulse impact degree in the current time period;

[0021] Perform a Hilbert transform on the vibration acceleration in the current time period to obtain the instantaneous frequency at each moment in the current time period. Calculate the coefficient of variation of all instantaneous frequencies and normalize the coefficient of variation to obtain the frequency distribution dispersion in the current time period.

[0022] The sum between the degree of kurtosis variation and the degree of pulse impact is calculated to obtain the sum result. The product between the sum result and the degree of dispersion of the frequency distribution is normalized to obtain the first adjustment factor for adjusting the number of basic modes.

[0023] Preferably, the step of obtaining a second adjustment factor for adjusting the number of basic modes based on the marginal spectral similarity and mutual information between every two adjacent modal components, and the stability of the center frequency of each modal component at different preset time scales within the current time period, includes:

[0024] Based on the marginal spectral similarity and mutual information between each pair of adjacent modal components, and the energy of each modal component in the current time period, the modal aliasing degree after decomposing the vibration acceleration in the current time period using the basic modal number is obtained;

[0025] The stability of the center frequency of each modal component at different preset time scales within the current time period is obtained, and the modal instability is obtained after decomposing the vibration acceleration within the current time period using the basic modal number.

[0026] The product between the modal aliasing degree and the modal instability degree is normalized to obtain a second adjustment factor for adjusting the number of basic modes.

[0027] Preferably, the step of obtaining the modal aliasing degree after decomposing the vibration acceleration in the current time period using the fundamental mode number, based on the marginal spectral similarity and mutual information between every two adjacent modal components and the energy of each modal component in the current time period, includes:

[0028] Obtain the marginal spectrum of each modal component, calculate the cosine similarity of the marginal spectra of every two adjacent modal components, obtain the marginal spectral similarity of every two adjacent modal components, normalize the mean of all marginal spectral similarities, and obtain the frequency domain overlap between different modal components.

[0029] The values ​​of each modal component at each time point in the current time period are combined to form a corresponding data sequence. The mutual information between the data sequences of every two modal components is calculated. The mean of all mutual information is normalized to obtain the information overlap between each modal component.

[0030] The energy of each modal component in the current time period is obtained, and the proportion of the energy corresponding to each modal component in the total energy in the current time period is calculated to obtain the energy ratio of each modal component in the current time period. The information entropy of the energy ratio of all modal components is calculated, and the proportion of the information entropy in the theoretical maximum entropy of all energy ratios is normalized to obtain the energy uniformity.

[0031] The sum of the frequency domain overlap, the information overlap, and the energy uniformity is calculated to obtain the modal aliasing degree after decomposing the vibration acceleration in the current time period using the fundamental mode number.

[0032] Preferably, the stability of the center frequency of each modal component at different preset time scales within the current time period, and the acquisition of the modal instability after decomposing the vibration acceleration within the current time period using the fundamental mode number, include:

[0033] For any modal component, the center frequency of the modal component at different preset time scales within the current time period is extracted, the deviation from the mean of the center frequency of the modal component at different preset time scales is calculated, and the mean of all deviations from the mean is taken as the degree of splitting of the modal component.

[0034] The degree of splitting of all modal components is obtained, and the mean of all splitting degrees is normalized to obtain the modal instability degree after decomposing the vibration acceleration in the current time period using the basic mode number.

[0035] Preferably, the step of adaptively adjusting the number of basic modes based on the first adjustment factor and the second adjustment factor to obtain an adaptive number of modes includes:

[0036] Calculate the sum between constant 1 and the first adjustment factor to obtain the sum result, and calculate the product of the basic mode number and the sum result to obtain the first adjustment result;

[0037] Subtract the second adjustment factor from the constant 1 to obtain the target difference. Calculate the product of the number of basic modes and the target difference to obtain the second adjustment result.

[0038] Calculate the mean between the first adjustment result and the second adjustment result to obtain the number of adjustment modes;

[0039] Obtain a preset minimum number of modes and a preset maximum number of modes. If the number of modes to be adjusted is between the preset minimum number of modes and the preset maximum number of modes, then the number of modes to be adjusted is used as the number of adaptive modes.

[0040] If the number of adjusted modes is less than the preset minimum number of modes, then the preset minimum number of modes is used as the number of adaptive modes; if the number of adjusted modes is greater than the preset maximum number of modes, then the preset maximum number of modes is used as the number of adaptive modes.

[0041] Preferably, the decomposition result is the value of the several intrinsic mode functions of the adaptive mode at each time point within the current time period.

[0042] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0043] This invention first analyzes the signal complexity of the target device within the current time period based on the entropy and energy of the vibration acceleration at different preset time scales, and obtains the number of basic modes to provide a reasonable starting point for subsequent adaptive adjustments, ensuring that the subsequent adaptive adjustments are based on the basic state of the target device at the current moment. Then, based on the sensitivity of the VMD algorithm to non-stationary signal decomposition and the fault sparsity characteristics, it obtains the first adjustment factor for adjusting the number of basic modes through the kurtosis and instantaneous frequency of the vibration acceleration within the current time period. Simultaneously, it decomposes the vibration acceleration within the current time period using the number of basic modes, and calculates the modal components for each phase after decomposition. The marginal spectral similarity and mutual information between two adjacent modal components, as well as the stability of the center frequency of each modal component at different preset time scales within the current period, are used to analyze the decomposition quality using the basic modal number. This allows for the acquisition of a second adjustment factor to adjust the basic modal number. Finally, the basic modal number is adaptively adjusted by fusing the first and second adjustment factors to obtain an adaptive modal number. This ensures that the modal number is sufficient to capture signal features while avoiding false modes and redundant information caused by over-decomposition. The final adaptive modal number better reflects the actual state of the target device, improving the physical interpretability of VMD decomposition and the accuracy and reliability of subsequent fault prediction. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart of an artificial intelligence-based equipment fault prediction method provided in Embodiment 1 of the present invention. Detailed Implementation

[0046] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0047] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0048] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0049] This invention provides an artificial intelligence-based equipment fault prediction system, including a processor and a memory. The processor executes a computer program stored in the memory to implement an artificial intelligence-based equipment fault prediction method, such as... Figure 1 As shown, the method includes the following steps:

[0050] Step S101: Obtain the vibration acceleration of the target device at each moment in the current time period up to the current moment. Based on the entropy and energy of the vibration acceleration at at least three different preset time scales in the current time period, obtain the number of basic modes for decomposing the vibration acceleration in the current time period using the VMD algorithm.

[0051] Designate any industrial equipment for fault prediction as the target equipment. Depending on the type of fault to be predicted, install an accelerometer (IEPE or charge output type) at key monitoring points. For example, to predict bearing faults, the accelerometer needs to be installed on the side, top, or end face of the bearing housing. At a preset sampling frequency, use the accelerometer to collect the acceleration of the target equipment in real time, and normalize the collected data to obtain the vibration acceleration at each moment. The preset sampling frequency must be set according to the fault type. For example, to predict bearing faults, assuming the characteristic frequency of the bearing fault in the target equipment is at most 100Hz, to satisfy the Nyquist sampling theorem and allow for engineering margins, data collection needs to be performed at a frequency at least 2.5 times the characteristic frequency. That is, the preset sampling frequency should be at least 2.5 times the characteristic frequency, and the sampling rate should be ≥256Hz. This solution predicts faults in the target equipment at the second level. Therefore, it predicts faults in the target equipment based on the vibration acceleration of the target equipment at each moment within 1 second up to the current moment. That is, the current 1 second up to the current moment is used as the current time period for analysis. There are no restrictions here, and the implementer can set it according to the fault type.

[0052] Traditionally, Variational Mode Decomposition (VMD) algorithms are used to decompose vibration acceleration time-series data into multiple modes with specific center frequencies and bandwidths. These modes are then fed into artificial intelligence models (SVM, CNN, LSTM, etc.) to identify the fault types of industrial equipment. The number of modes K in traditional VMD algorithms usually needs to be preset manually. However, due to the complexity of equipment operating environments and the diversity of fault characteristics, the complexity of equipment vibration signals varies significantly with changes in operating conditions and fault characteristics. A single, fixed number of modes cannot guarantee that the decomposed modes can effectively characterize the equipment's fault trends, thus affecting the accuracy and reliability of fault prediction results for industrial equipment. Therefore, in this embodiment of the invention, an adaptive mechanism based on degradation trend entropy is introduced to adaptively adjust the number of modes K in the VMD algorithm. This allows the number of modes to automatically adjust to an appropriate value according to the current actual state of the target equipment, thereby more accurately decomposing the signal, extracting fault features, and providing more reliable data support for artificial intelligence models to predict equipment faults.

[0053] First, the number of basic modes is obtained based on the state of the target device in the current time period. The complexity of the vibration signal varies significantly depending on the device's state, directly reflecting the richness of its components and the breadth of its frequency distribution. In a healthy state or under a single operating condition, the vibration signal is highly regular, with relatively simple components and low complexity. However, as the device degrades, fault trends emerge, or operating conditions change, various impacts and noise components may be mixed into the signal, increasing its complexity. Therefore, in this embodiment of the invention, the entropy and energy of the vibration acceleration of the target device at different preset time scales within the current time period (i.e., within 1 second up to the current moment) are used to characterize the complexity of the vibration signal of the target device in the current time period from different dimensions. This yields the number of basic modes for decomposing the vibration acceleration within the current time period using the VMD algorithm, providing a reasonable starting point for subsequent adaptive adjustments and ensuring that subsequent adaptive adjustments are based on the basic state of the target device in the current time period.

[0054] The specific method for obtaining the number of fundamental modes is as follows:

[0055] Three different preset time scales are set, namely 100Hz, 50Hz and 10Hz. The three preset time scales correspond to high frequency impact components (such as early bearing spalling), mid frequency fault modulation sidebands (such as gear meshing frequency) and low frequency trend changes (such as speed fluctuations or slow-change faults), respectively. This can comprehensively characterize the complexity of vibration signals from a multi-resolution perspective and avoid losing signal characteristics from a single scale. There are no restrictions on the setting of the time scale. Implementers can set it according to the type of target equipment and its operating status.

[0056] By downsampling the vibration acceleration in the current time period, the information entropy of the vibration acceleration at each preset time scale is extracted, and the mean of all information entropies is normalized to obtain the signal complexity in the current time period.

[0057] A Hilbert transform is performed on the vibration acceleration within the current time period (note: this process does not distinguish time scales, but performs the Hilbert transform on the vibration acceleration within the current time period according to a preset sampling frequency) to obtain the Hilbert envelope of the vibration acceleration within the current time period. An FFT transform is then performed on the Hilbert envelope to obtain the envelope spectrum. The acquisition of the envelope spectrum is a prior art technique and will not be elaborated here. The peak values ​​in the envelope spectrum are sorted in descending order to obtain a peak sequence. The sum of the energies corresponding to the first preset number of peaks in the peak sequence and their proportion in the total energy of the envelope spectrum are calculated. The proportion is normalized to obtain the signal impact intensity in the current time period. The first preset quantity is set to 5, which is not limited here. The implementer can set it according to the fluctuation of the envelope spectrum. If there is no obvious peak in the envelope spectrum, it means that the target device is in a relatively healthy state or the operating condition is stable in the current time period. In this case, the first preset quantity can be appropriately reduced. If the peak of the envelope spectrum is relatively significant, it means that the target device is more likely to have a fault trend in the current time period or the operating condition has changed. In this case, the first preset quantity should be appropriately increased to fully reflect the signal characteristics of the target device in the current time period.

[0058] The sum of the signal complexity and the signal impact intensity is normalized to obtain the basic coefficient. A preset minimum mode number and a preset maximum mode number are obtained. The preset minimum mode number is subtracted from the preset maximum mode number to obtain the target difference. The product between the basic coefficient and the target difference is calculated to obtain the target result. The sum between the minimum mode number and the target result is calculated to obtain the basic mode number for decomposing the vibration acceleration in the current time period using the VMD algorithm.

[0059] In one embodiment, the formula for calculating the number of fundamental modes is:

[0060]

[0061] in, Represents the number of fundamental modes. This indicates the preset maximum number of modes. This represents the preset minimum number of modes, and T represents the preset number of time scales. In this embodiment of the invention, T=3. Let represent the information entropy of the vibrational acceleration at the t-th time scale, and EA represent the total energy of the envelope spectrum of the vibrational acceleration within the current time period. This represents the energy corresponding to the p-th peak in the peak sequence corresponding to the envelope spectrum, where P represents a first preset quantity. In this embodiment of the invention, P=5. This represents the normalization function.

[0062] It should be noted that, The larger the value, the more chaotic and complex the vibration signal within the current time period, and the greater the possibility that the target equipment will experience a failure trend or a change in operating conditions within the current time period. In this case, the number of modes should be appropriately increased to prevent under-decomposition. The larger; The larger the value, the stronger the impact on the target equipment during the current time period, and the greater the possibility of the target equipment experiencing a failure or a change in operating conditions during the current time period. In this case, the number of modes should be appropriately increased to prevent under-decomposition. The larger.

[0063] For the preset maximum number of modes and preset minimum number of modes The maximum number of modes is typically set between 2 and 10. The specific setting depends on the signal complexity of the target device during historical faults or operational condition changes. Higher signal complexity allows for a larger maximum number of modes. In this embodiment of the invention, the maximum number of modes is set... , There are no restrictions here; implementers can set it according to the proposed scenario.

[0064] Thus, the number of fundamental modes for decomposing the vibration acceleration in the current time period using the VMD algorithm was obtained.

[0065] Step S102: Based on the kurtosis and instantaneous frequency of the vibration acceleration in the current time period, obtain the first adjustment factor for adjusting the number of basic modes.

[0066] After obtaining the basic modal count, considering the complexity of the actual operating environment and the diversity of fault characteristics of the target equipment, the basic modal count needs to be adjusted. When the target equipment exhibits early fault characteristics or changes in operating conditions, it can lead to non-stationary fluctuations and sparse fault impacts in the vibration signal, making the signal characteristics more complex. In this case, it is necessary to appropriately increase the modal count to ensure that VMD can effectively separate transient components and guarantee the accuracy of the decomposition.

[0067] When the target equipment is in a healthy state, the kurtosis of the vibration signal is relatively stable. However, in the early stages of a fault, when sporadic impacts occur, the kurtosis of the vibration signal fluctuates drastically, requiring more modes to capture transient characteristics at different times. Secondly, when localized damage such as spalling or cracks occurs on the rolling elements, raceways, or tooth surfaces of the target equipment (e.g., bearings, gears, etc.), high-amplitude, narrow-width impacts are generated in the vibration signal. These impacts often occupy a separate frequency band, requiring additional modes for separation. Simultaneously, when the operating conditions of the target equipment change or an early fault occurs, the frequency variation of the vibration signal over time becomes unstable. Based on the above analysis, a first adjustment factor for adjusting the basic mode number can be obtained based on the kurtosis, peak value, and instantaneous frequency of the vibration acceleration of the target equipment in the current time period. Specifically:

[0068] Based on the duration of the current time period (i.e., 1 second), the vibration acceleration of the target device during the historical time period up to the previous time is obtained. The kurtosis of the vibration acceleration during the historical time period is obtained and denoted as the historical kurtosis. The kurtosis of the vibration acceleration during the current time period is obtained and denoted as the current kurtosis. The acquisition of kurtosis is existing technology and will not be elaborated here. The absolute value of the difference between the historical kurtosis and the current kurtosis is normalized to obtain the degree of kurtosis change at the current time, denoted as . ,Right now ,in, Indicates the current kurtosis. Indicates historical kurtosis. Represents the absolute value symbol. Represents the normalization function;

[0069] The vibration acceleration at each moment in the current time period is composed into a data sequence. The extreme values ​​in the data sequence are obtained. The absolute values ​​of the extreme values ​​in the data sequence are sorted in descending order. The extreme values ​​corresponding to the first 3 absolute values ​​after sorting are taken as the target extreme values. There is no limit here. The implementer can set the number of target extreme values ​​according to the specific scenario.

[0070] For any target extreme value, if the target extreme value is a maximum value, then obtain the left adjacent minimum value and the right adjacent minimum value of the target extreme value, take the time interval between the time corresponding to the left adjacent minimum value and the time corresponding to the right adjacent minimum value as the denominator, and take the absolute value of the target extreme value as the numerator to obtain the rate of change of the target extreme value.

[0071] If any of the target extreme values ​​is a minimum value, then the left and right adjacent maximum values ​​of the target extreme value are obtained. The time interval between the time corresponding to the left adjacent maximum value and the time corresponding to the right adjacent maximum value is used as the denominator, and the absolute value of the target extreme value is used as the numerator to obtain the rate of change of the target extreme value.

[0072] Obtain the rate of change of each target extreme value, and normalize the average of the rates of change of all target extreme values ​​to obtain the pulse impact degree in the current time period;

[0073] A Hilbert transform is applied to the vibration acceleration within the current time period to obtain the instantaneous frequency at each moment within the current time period. The acquisition of instantaneous frequencies is a prior art technique and will not be elaborated upon here. The coefficient of variation (CV) of all instantaneous frequencies is calculated and normalized to obtain the frequency distribution dispersion within the current time period. ,in, The standard deviation of all instantaneous frequencies. This represents the mean of all instantaneous frequencies. Represents the normalization function;

[0074] The sum between the degree of kurtosis change and the degree of pulse impact is calculated to obtain the sum result. The product between the sum result and the degree of frequency distribution dispersion is normalized to obtain the first adjustment factor for adjusting the number of basic modes.

[0075] In one embodiment, the formula for calculating the first adjustment factor is:

[0076]

[0077] Where B represents the first adjustment factor, This indicates the degree of kurtosis change at the current moment, and A represents the number of target extrema. This represents the absolute value of the a-th objective extremum. This represents the time interval between the maximum or minimum values ​​adjacent to the left and right of the a-th target extreme value. It indicates the degree of dispersion in the frequency distribution.

[0078] It should be noted that, The larger the value of B, the more severe the kurtosis fluctuation of the vibration signal, and the greater the possibility that the target equipment will fail in the current period. At this time, more modes are needed to capture the transient characteristics at different times, and thus the larger B is. The larger B is, the greater the pulse amplitude and the greater the impact. The greater the possibility of fault impact in the vibration signal of the target equipment in the current time period. Such fault impact often occupies a separate frequency band, and it is necessary to increase the number of modes for separation in order to capture the fault characteristics of the target equipment. Therefore, the larger B is. The larger B is, the more chaotic the vibration frequency distribution of the target equipment is in the current time period. The greater the possibility of early failure or change of operating conditions of the target equipment in the current time period, the more modal number decomposition components are needed to more accurately capture the fault characteristics of the target equipment. Therefore, the larger B is.

[0079] Thus, the first adjustment factor for adjusting the number of basic modes was obtained.

[0080] Step S103: Decompose the vibration acceleration in the current time period using the number of basic modes to obtain the modal components of the number of basic modes in the current time period. Based on the marginal spectral similarity and mutual information between each pair of adjacent modal components, and the stability of the center frequency of each modal component at different preset time scales in the current time period, obtain a second adjustment factor to adjust the number of basic modes.

[0081] The first adjustment factor obtained in the above steps is mainly based on the non-stationary characteristics of vibration signals such as kurtosis fluctuations and pulse impacts, tending to focus on the intrinsic characteristics of the vibration signal without considering the intrinsic quality and stability of the decomposed modes. Therefore, in this step, based on the number of basic modes obtained above, the VMD algorithm is used to decompose the vibration acceleration of the target device in the current time period, obtaining the modal components of the basic modes in the current time period. Then, the intrinsic quality of these modal components is analyzed to obtain a second adjustment factor for adjusting the number of basic modes. This ensures that the number of modes is sufficient to capture signal characteristics while avoiding false modes and redundant information caused by over-decomposition, making the final adaptive number of modes more consistent with the actual state of the target device, and improving the physical interpretability of VMD decomposition and the accuracy of subsequent diagnosis.

[0082] The second adjustment factor is obtained as follows:

[0083] Considering that VMD over-decomposition will produce two modal components with similar frequencies and overlapping physical meanings, the number of modes should be appropriately reduced to prevent over-decomposition. Therefore, the marginal spectrum of each modal component can be obtained, and the cosine similarity of the marginal spectrum of each two adjacent modal components (adjacent modal components refer to the two modal components whose center frequencies are closest to each other on the frequency axis) can be calculated to obtain the marginal spectrum similarity of each two adjacent modal components. The acquisition of marginal spectrum and cosine similarity is an existing technology and will not be elaborated here. The mean of all marginal spectrum similarities is normalized using the norm() function to obtain the frequency domain overlap between different modal components, denoted as By.

[0084] The mutual information between different modal components reflects the independence of the modal components. If the mutual information is high, it indicates that the information in the modal components is relatively redundant and the independence is poor. In this case, the number of modes should be appropriately reduced to prevent over-decomposition. Therefore, in this embodiment of the invention, the value of each modal component at each time point in the current time period is used to form a corresponding data sequence. The mutual information between the data sequences of every two modal components is calculated. The acquisition of mutual information is a prior art and will not be described in detail here. The mean of all mutual information is normalized to obtain the information overlap between each modal component, denoted as HI. In the formula, This represents the number of fundamental modes, which is the number of modal components. This represents the number of pairwise combinations of all modal components. This represents the mutual information between the data sequences of the two modal components in the u-th group. Represents the normalization function;

[0085] In high-quality decompositions, the energy of each modal component typically exhibits significant differences and highly uneven distribution. Therefore, the decomposition quality of the vibration signal using the fundamental mode number can be further evaluated based on the energy differences of each modal component. This involves obtaining the energy of each modal component in the current time period, calculating the proportion of each modal component's energy in the total energy of the current time period, obtaining the energy ratio of each modal component in the current time period, calculating the information entropy of the energy ratios of all modal components, and then taking the theoretical maximum entropy of the information entropy across all energy ratios. , The proportion of the fundamental mode number (i.e., the number of modal components) is normalized to obtain the energy uniformity, denoted as HE. ,in, Information entropy representing all energy proportions Represents the theoretical maximum entropy. Represents the normalization function. The larger the value, the more uniform the energy distribution of each mode. At this point, the possibility of over-decomposition is greater, and the number of modes needs to be appropriately reduced.

[0086] The sum of the frequency domain overlap, the information overlap, and the energy uniformity is calculated to obtain the modal aliasing degree after decomposing the vibration acceleration in the current time period using the fundamental mode number;

[0087] The change in the center frequency of a modal component at different time scales can reflect modal stability. Therefore, for any modal component, the center frequency of the modal component at different preset time scales (in this embodiment, the preset time scales are 100Hz, 50Hz, and 10Hz) within the current time period is extracted. The general calculation process is as follows: First, the modal value of the modal component at each moment within the current time period is obtained. Then, the modal value of the modal component at each moment within the current time period is resampled according to different preset time scales and a corresponding array is formed to obtain the array of the modal component at different preset time scales. FFT transformation is performed on the array at each preset time scale to convert the time domain signal into a frequency domain signal, thereby obtaining the spectrum of the modal component at each preset time scale within the current time period. Then, the center frequency is obtained from the spectrum. The acquisition of the center frequency is an existing technology and will not be described in detail here. The deviation from the mean of the center frequency of the modal component at different preset time scales is calculated, and the mean of all deviations from the mean is taken as the degree of splitting of the modal component.

[0088] Obtain the degree of splitting of all modal components, normalize the mean of all splitting degrees, and obtain the modal instability degree after decomposing the vibration acceleration in the current time period using the basic modal number;

[0089] The product between the modal aliasing degree and the modal instability degree is normalized to obtain a second adjustment factor for adjusting the number of basic modes.

[0090] In one embodiment, the formula for calculating the second adjustment factor is:

[0091]

[0092] Where C represents the second adjustment factor, By represents the frequency domain overlap between different modal components, HI represents the information overlap between each modal component, and HE represents the energy uniformity. This represents the number of fundamental modes, or the number of modal components. T represents the number of preset time scales. In this embodiment of the invention, T=3. This represents the center frequency of the y-th modal component at the t-th preset time scale. This represents the mean center frequency of the y-th modal component at different preset time scales. This represents the normalization function.

[0093] It should be noted that the larger By is, the more similar the frequencies of the various modal components are, and the greater the possibility of over-decomposition. Therefore, the number of fundamental modes needs to be appropriately reduced. Consequently, the larger C is, the greater the reduction in the number of fundamental modes. The larger the value of C, the higher the mutual information between different modal components. The information in the modal components is more redundant and the independence is poor. In this case, the number of basic modes should be reduced appropriately. Therefore, the larger C is, the greater the reduction in the number of basic modes. The larger C is, the closer the energies of the modal components are to each other, meaning the energy distribution among the modal components is more uniform. At this point, the possibility of over-decomposition is greater, and it is necessary to appropriately reduce the number of fundamental modes. Therefore, the larger C is, the greater the reduction in the number of fundamental modes. The larger the value of C, the greater the difference in the center frequencies of the modal components at different time scales. This indicates that the modal components are more unstable and of lower quality, and the possibility of over-decomposition is greater. Therefore, it is necessary to appropriately reduce the number of fundamental modes. Consequently, the larger C is, the greater the reduction in the number of fundamental modes.

[0094] Thus, the second adjustment factor for adjusting the number of basic modes was obtained.

[0095] Step S104: Adaptively adjust the basic mode number according to the first adjustment factor and the second adjustment factor to obtain the adaptive mode number. Based on the adaptive mode number, decompose the vibration acceleration in the current time period using the VMD algorithm to obtain the decomposition result. Based on the decomposition result, realize the fault prediction of the target equipment in the future time period.

[0096] After obtaining the first and second adjustment factors for adjusting the basic modal number through the above steps, the basic modal number is adaptively adjusted using the first and second adjustment factors to obtain the adaptive modal number. This allows the modal components obtained from the adaptive modal number decomposition to reflect the complexity of the vibration signal of the target equipment, adapt to the non-stationarity of the signal and the sparsity of the fault impact, and ensure the quality of the decomposed modal components. This enables dynamic and accurate tracking of equipment fault trends and provides more reliable signal decomposition results for equipment fault prediction.

[0097] The specific implementation method for obtaining the adaptive modal number is as follows:

[0098] Calculate the sum between constant 1 and the first adjustment factor to obtain the sum result, and calculate the product of the basic mode number and the sum result to obtain the first adjustment result;

[0099] Subtract the second adjustment factor from the constant 1 to obtain the target difference. Calculate the product of the number of basic modes and the target difference to obtain the second adjustment result.

[0100] The mean between the first adjustment result and the second adjustment result is calculated to obtain the number of adjustment modes that decompose the vibration acceleration in the current time period using the VMD algorithm.

[0101] In one embodiment, the formula for calculating the number of adjusted modes by decomposing the vibration acceleration in the current time period using the VMD algorithm is as follows:

[0102]

[0103] in, This represents the number of adjusted modes used to decompose the vibration acceleration within the current time period using the VMD algorithm. B represents the number of fundamental modes, C represents the first adjustment factor, and C represents the second adjustment factor.

[0104] It should be noted that a larger B indicates more severe kurtosis fluctuations in the vibration signal, larger pulse amplitudes, and more chaotic vibration frequency distributions. This increases the likelihood of early failures or changes in operating conditions of the target equipment within the current time period. Therefore, more modal decomposition time components are needed to more accurately capture the fault characteristics of the target equipment. The larger C is, the more similar the frequencies and the closer the energies are among the modal components obtained from the fundamental mode number decomposition, or the greater the difference in the center frequencies of the modal components at different time scales. This results in poorer quality of the modal components obtained from the fundamental mode number decomposition, a higher probability of over-decomposition, and a need for fewer modes. The smaller.

[0105] To ensure the reasonableness of the adaptive mode number, the adaptive mode number should be between a preset minimum mode number and a preset maximum mode number. In this embodiment of the invention, the preset minimum mode number is 3, and the preset maximum mode number is 8. If the mode number is adjusted... If the number is between 3 and 8, the modal number will be adjusted. As an adaptive mode number;

[0106] If the number of modes is adjusted If the number of modes is less than the preset minimum number of modes (3), then the preset minimum number of modes (3) will be used as the adaptive number of modes. If the number of modes is adjusted... If the number of modes is greater than the preset maximum number of modes (8), then the preset maximum number of modes (8) will be used as the number of adaptive modes.

[0107] Thus, the adaptive mode number for decomposing the vibration acceleration in the current time period using the VMD algorithm is obtained, denoted as . Furthermore, based on the adaptive mode number The VMD algorithm is used to decompose the vibration acceleration of the target device in the current time period to obtain... Each modal component, that is The value of each intrinsic mode function (IMF) at each time point within the current time period.

[0108] Furthermore, according to Based on the values ​​of each intrinsic mode function (IMF) at each time point within the current time period, the failure of the target equipment at future time points is predicted. The general process is as follows: Based on The system extracts the values ​​of each intrinsic mode function (EMF) at each time point within the current time period to identify fault-sensitive features of the target device. These features include amplitudes at fault characteristic frequencies in the envelope spectrum, modal energy proportions, instantaneous frequency fluctuations, and temporal, frequency, and entropy characteristics of modal components. Based on these fault-sensitive features, a regression model from the feature sequence to the remaining useful life (RUL) is established using recurrent neural networks (such as LSTM), temporal convolutional networks, or traditional similarity matching methods (such as dynamic time warping). Alternatively, a classifier is trained to identify the current stage (healthy, early degradation, or severe failure) and predict the target device's health status within the next second. By dynamically updating the number of modes and continuously decomposing them, the system can continuously track the drift of fault frequencies and the transfer of modal energy in the target device, thereby achieving accurate prediction of device faults.

[0109] It is worth noting that the focus of this invention is on how to obtain the adaptive modal number in the VMD algorithm, thereby more accurately decomposing the signal, extracting fault features, and providing more reliable data support for artificial intelligence models to predict equipment faults. Analyzing the decomposed modal components using an artificial intelligence model to predict equipment faults is existing technology and will not be elaborated upon here.

[0110] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An artificial intelligence-based equipment fault prediction system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it performs the following steps: Obtain the vibration acceleration of the target device at each moment in the current time period up to the current moment. Based on the entropy and energy of the vibration acceleration at at least three different preset time scales in the current time period, obtain the number of fundamental modes for decomposing the vibration acceleration in the current time period using the VMD algorithm. Based on the kurtosis and instantaneous frequency of the vibration acceleration in the current time period, a first adjustment factor is obtained to adjust the number of basic modes; The vibration acceleration in the current time period is decomposed using the number of basic modes to obtain the modal components of the number of basic modes in the current time period. Based on the marginal spectral similarity and mutual information between each pair of adjacent modal components, and the stability of the center frequency of each modal component at different preset time scales in the current time period, a second adjustment factor is obtained to adjust the number of basic modes. The basic mode number is adaptively adjusted according to the first adjustment factor and the second adjustment factor to obtain the adaptive mode number. Based on the adaptive mode number, the vibration acceleration in the current time period is decomposed using the VMD algorithm to obtain the decomposition result. Based on the decomposition result, the fault prediction of the target equipment in the future time period is realized. The step of obtaining a first adjustment factor for adjusting the number of fundamental modes based on the kurtosis and instantaneous frequency of the vibration acceleration in the current time period includes: Based on the duration of the current time period, obtain the vibration acceleration of the target device in the historical time period up to the previous time, obtain the kurtosis of the vibration acceleration in the historical time period and record it as the historical kurtosis, obtain the kurtosis of the vibration acceleration in the current time period and record it as the current kurtosis, and normalize the absolute value of the difference between the historical kurtosis and the current kurtosis to obtain the degree of kurtosis change at the current time. The vibration acceleration at each moment in the current time period is composed into a data sequence, the extreme values ​​in the data sequence are obtained, the absolute values ​​of the extreme values ​​in the data sequence are sorted in descending order, and the extreme values ​​corresponding to the first second preset number of absolute values ​​after sorting are taken as the target extreme values. For any target extreme value, if the target extreme value is a maximum value, then obtain the left adjacent minimum value and the right adjacent minimum value of the target extreme value, take the time interval between the time corresponding to the left adjacent minimum value and the time corresponding to the right adjacent minimum value as the denominator, and take the absolute value of the target extreme value as the numerator to obtain the rate of change of the target extreme value. If any of the target extreme values ​​is a minimum value, then the left and right adjacent maximum values ​​of the target extreme value are obtained. The time interval between the time corresponding to the left adjacent maximum value and the time corresponding to the right adjacent maximum value is used as the denominator, and the absolute value of the target extreme value is used as the numerator to obtain the rate of change of the target extreme value. Obtain the rate of change of each target extreme value, and normalize the average of the rates of change of all target extreme values ​​to obtain the pulse impact degree in the current time period; Perform a Hilbert transform on the vibration acceleration in the current time period to obtain the instantaneous frequency at each moment in the current time period. Calculate the coefficient of variation of all instantaneous frequencies and normalize the coefficient of variation to obtain the frequency distribution dispersion in the current time period. Calculate the sum between the degree of kurtosis change and the degree of pulse impact to obtain the sum result. Normalize the product between the sum result and the degree of frequency distribution dispersion to obtain the first adjustment factor for adjusting the number of basic modes. The step of obtaining a second adjustment factor for adjusting the number of basic modes based on the marginal spectral similarity and mutual information between every two adjacent modal components, and the stability of the center frequency of each modal component at different preset time scales within the current time period, includes: Based on the marginal spectral similarity and mutual information between each pair of adjacent modal components, and the energy of each modal component in the current time period, the modal aliasing degree after decomposing the vibration acceleration in the current time period using the basic modal number is obtained; The stability of the center frequency of each modal component at different preset time scales within the current time period is obtained, and the modal instability is obtained after decomposing the vibration acceleration within the current time period using the basic modal number. The product between the modal aliasing degree and the modal instability degree is normalized to obtain a second adjustment factor for adjusting the number of basic modes.

2. The equipment fault prediction system based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the number of fundamental modes for decomposing the vibration acceleration in the current time period using the VMD algorithm, based on the entropy and energy of the vibration acceleration at at least three different preset time scales within the current time period, includes: By downsampling the vibration acceleration in the current time period, the information entropy of the vibration acceleration at each preset time scale is extracted, and the mean of all information entropies is normalized to obtain the signal complexity in the current time period. Perform a Hilbert transform on the vibration acceleration in the current time period to obtain the Hilbert envelope of the vibration acceleration in the current time period. Perform an FFT transform on the Hilbert envelope to obtain the envelope spectrum. Sort the peaks in the envelope spectrum in descending order to obtain the peak sequence. Calculate the sum of the energy corresponding to the first preset number of peaks in the peak sequence and its proportion in the total energy of the envelope spectrum. Normalize the proportion to obtain the signal impulse intensity in the current time period. The sum of the signal complexity and the signal impact intensity is normalized to obtain the basic coefficient. A preset minimum mode number and a preset maximum mode number are obtained. The preset minimum mode number is subtracted from the preset maximum mode number to obtain the target difference. The product between the basic coefficient and the target difference is calculated to obtain the target result. The sum between the minimum mode number and the target result is calculated to obtain the basic mode number for decomposing the vibration acceleration in the current time period using the VMD algorithm.

3. The equipment fault prediction system based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the modal aliasing degree after decomposing the vibration acceleration in the current time period using the basic mode number, based on the marginal spectral similarity and mutual information between each pair of adjacent modal components and the energy of each modal component in the current time period, includes: Obtain the marginal spectrum of each modal component, calculate the cosine similarity of the marginal spectra of every two adjacent modal components, obtain the marginal spectral similarity of every two adjacent modal components, normalize the mean of all marginal spectral similarities, and obtain the frequency domain overlap between different modal components. The values ​​of each modal component at each time point in the current time period are combined to form a corresponding data sequence. The mutual information between the data sequences of every two modal components is calculated. The mean of all mutual information is normalized to obtain the information overlap between each modal component. The energy of each modal component in the current time period is obtained, and the proportion of the energy corresponding to each modal component in the total energy in the current time period is calculated to obtain the energy ratio of each modal component in the current time period. The information entropy of the energy ratio of all modal components is calculated, and the proportion of the information entropy in the theoretical maximum entropy of all energy ratios is normalized to obtain the energy uniformity. The sum of the frequency domain overlap, the information overlap, and the energy uniformity is calculated to obtain the modal aliasing degree after decomposing the vibration acceleration in the current time period using the fundamental mode number.

4. The equipment fault prediction system based on artificial intelligence according to claim 1, characterized in that, The stability of the center frequency of each modal component at different preset time scales within the current time period, and the modal instability after decomposing the vibration acceleration within the current time period using the fundamental mode number, include: For any modal component, the center frequency of the modal component at different preset time scales within the current time period is extracted, the deviation from the mean of the center frequency of the modal component at different preset time scales is calculated, and the mean of all deviations from the mean is taken as the degree of splitting of the modal component. The degree of splitting of all modal components is obtained, and the mean of all splitting degrees is normalized to obtain the modal instability degree after decomposing the vibration acceleration in the current time period using the basic mode number.

5. The equipment fault prediction system based on artificial intelligence according to claim 1, characterized in that, The step of adaptively adjusting the number of basic modes based on the first adjustment factor and the second adjustment factor to obtain an adaptive number of modes includes: Calculate the sum between constant 1 and the first adjustment factor to obtain the sum result, and calculate the product of the basic mode number and the sum result to obtain the first adjustment result; Subtract the second adjustment factor from the constant 1 to obtain the target difference. Calculate the product of the number of basic modes and the target difference to obtain the second adjustment result. Calculate the mean between the first adjustment result and the second adjustment result to obtain the number of adjustment modes; Obtain a preset minimum number of modes and a preset maximum number of modes. If the number of modes to be adjusted is between the preset minimum number of modes and the preset maximum number of modes, then the number of modes to be adjusted is used as the number of adaptive modes. If the number of adjusted modes is less than the preset minimum number of modes, then the preset minimum number of modes is used as the number of adaptive modes; if the number of adjusted modes is greater than the preset maximum number of modes, then the preset maximum number of modes is used as the number of adaptive modes.

6. The equipment fault prediction system based on artificial intelligence according to claim 1, characterized in that, The decomposition result is the value of the several intrinsic mode functions of the adaptive mode at each time point within the current time period.

Images