A bearing fault detection method, system, device and medium based on adaptive multi-scale enhanced dictionary learning framework

Abstract

A bearing fault detection method, system, device, and medium based on an adaptive multi-scale enhanced dictionary learning framework. The method comprises: step 1, collecting a bearing fault vibration signal; step 2, constructing an adaptive multi-scale enhanced dictionary learning framework to extract a fault pulse signal; and step 3, identifying the fault frequency and its multiples through envelope analysis based on the fault pulse signal extracted in step 2 to determine the fault type. The system, device, and medium are used to implement a bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework. The present invention utilizes multi-scale sub-band screening to weaken the influence of complex and strong interference, designs a threshold adaptive estimation method to avoid manual parameter adjustment, and finally integrates an enhancement strategy of enhance-run-subtract with dictionary learning into a multi-scale transform to construct an adaptive multi-scale enhanced dictionary learning model to improve the performance of multi-scale dictionary learning, thereby enabling adaptive and reliable fault monitoring and diagnosis of bearings.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing fault detection, and in particular relates to a bearing fault detection method, system, equipment and medium based on an adaptive multi-scale enhanced dictionary learning framework. Background Art

[0002] As a key component of rotating systems, rolling bearings are widely used in mechanical equipment such as wind turbines and aircraft engines. Their health determines whether the equipment can operate safely and stably. Vibration-based analysis methods are one of the most commonly used techniques for bearing fault diagnosis. However, modern mechanical equipment often operates in complex environments, causing fault pulses to couple with various interferences (such as noise, harmonics, and random pulses) and be overwhelmed by these complex interferences. Therefore, how to quickly and reliably extract bearing fault characteristics is the foundation and focus of fault detection and health assessment.

[0003] Dictionary learning algorithms, such as K-SVD, can be used for fault feature extraction, and variants based on K-SVD are widely used in bearing fault diagnosis due to their efficiency and flexibility. However, dictionary learning algorithms operating at a single scale are limited in revealing the global characteristics of non-stationary fault signals and have high computational complexity due to the large atomic dimension.

[0004] Patent application number [CN201910079232.3] discloses a planetary bearing fault identification method based on a weighted multi-scale dictionary learning framework. This method uses kurtosis as a weight and integrates it with multi-scale dictionary learning to suppress the harmonic components in the acquired fault signal and compensate for the limitations of single-scale dictionary learning. However, there are still three key issues that deserve attention in practical applications of this method: 1) Fault pulses are easily coupled with harmonic components and are often modulated by the rotation frequency. Unfortunately, dictionary learning algorithms are sensitive to harmonics, so the extracted pulse response is weak and cannot eliminate stubborn interference. 2) The classic fault sensitivity indicator (kurtosis) has been shown to lack robustness to complex and strong interference, which may incorrectly guide the dictionary to learn from the interference. 3) The setting of hyperparameters is an important and open issue. The optimal hyperparameters determined by numerical simulation may fail in actual engineering, and manual parameter adjustment is very laborious.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the invention and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the Invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a bearing fault detection method, system, device and medium based on an adaptive multi-scale enhanced dictionary learning framework, utilize multi-scale sub-band screening to weaken the influence of complex and strong interference, design a threshold adaptive estimation method to avoid manual parameter adjustment, and finally integrate the enhance-run-subtract enhancement strategy with dictionary learning into multi-scale transformation, construct an adaptive multi-scale enhanced dictionary learning model to improve the performance of multi-scale dictionary learning, so as to enable adaptive and reliable fault monitoring and diagnosis of bearings.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] A bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework comprises the following steps:

[0009] Step 1: collecting bearing fault vibration signals;

[0010] Step 2: construct an adaptive multi-scale enhanced dictionary learning framework to extract fault pulse signals;

[0011] Step 3: Based on the fault pulse signal extracted in step 2, the fault frequency and its multiples are identified through envelope analysis to determine the fault type.

[0012] In step 1, the bearing fault vibration signal y∈R n By periodic pulse x∈R n , complex harmonic interference h∈R n and strong background noise e∈R n The structure is expressed as y=x+h+e, and the bearing fault vibration signal y is collected by a vibration acceleration sensor.

[0013] The step 2 specifically includes the following steps:

[0014] Step 2.1, performing Q-modulated wavelet transform on the bearing fault vibration signal collected in step 1 to obtain multi-scale sub-bands;

[0015] Step 2.2, calculating the adaptive periodic modulation intensity APMI of each multi-scale subband in step 2.1;

[0016] Step 2.3, performing multi-scale sub-band screening based on the adaptive periodic modulation intensity (APMI) calculated in step 2.2 to obtain a preferred multi-scale sub-band;

[0017] Step 2.4, adaptively estimating the threshold parameters of the sparse coding stage for the preferred multi-scale sub-band obtained in step 2.3, and setting the wavelet coefficients of the unselected multi-scale sub-bands to 0;

[0018] In step 2.5, the enhancement strategy of enhance-run-subtract and dictionary learning are integrated into the multi-scale transformation, an adaptive multi-scale enhancement dictionary learning model is constructed, and the model is optimized to extract the fault pulse signal.

[0019] In step 2.1, the transformation parameters of the Q-wavelet include the quality factor Q, the redundancy r and the decomposition level J. The bearing fault vibration signal y is transformed into the Q-wavelet W T The J+1 layer wavelet coefficients are decomposed as follows:

[0020] W T y→{y1,…,y J ,y J+1}

[0021] Among them, y j (j=1, ..., J+1) is the wavelet coefficient of the j-th multiscale subband;

[0022] In step 2.2, the steps of calculating the adaptive periodic modulation intensity APMI of each multi-scale sub-band are as follows:

[0023] Calculate the possible failure period τ i :

[0024] τ i =round(f s / FCFs)

[0025] Among them, f s represents the sampling frequency, the round(·) function is used to round to the nearest integer, and the fault characteristic frequencies (FCFs) are the outer race (BPFO), inner race (BPFI), cage (FTF), and rolling element (BSF);

[0026] Define the fault cycle search range for each multi-scale sub-band:

[0027]

[0028] Among them, γ j represents a set of four fault cycles and their surrounding coefficients, where L0 and L j denote the original acquired signal y and the length of the j-th multi-scale subband, respectively. C = 10 is a constant that limits the search range.

[0029] Determine when PMI j Multi-scale subbands at maximum and the fault period of the multiscale subband

[0030]

[0031] in,

[0032]

[0033] Among them, R f [·] is the autocorrelation function, Γ(y j )=|y j +i·Hilbert(y j )∣ represents the input signal y j The absolute value of the envelope after Hilbert transformation, where i is an imaginary unit;

[0034] Based on the Multi-scale sub-bands, The fault period T of each multi-scale sub-band can be calculated j :

[0035]

[0036] Among them, T m and L m Respectively represent The period and length of the multi-scale sub-bands;

[0037] Based on T of each multi-scale sub-band j , calculate the adaptive periodic modulation intensity APMI:

[0038]

[0039] in, and They represent the envelope absolute value Γ(y j ) at time shift 0 and T j The autocorrelation function value when ;

[0040] In step 2.3, multi-scale sub-band screening is performed based on the adaptive periodic modulation intensity APMI:

[0041]

[0042] Among them, j opt represents the first preferred multi-scale subband;

[0043] Define the optimal multi-scale subband set S opt :

[0044] S opt ={j opt ,j opt +1,j opt +2};

[0045] The step 2.4 includes the following processes:

[0046] The threshold parameters of the sparse coding stage for adaptive estimation of the preferred multi-scale subband are:

[0047] Calculate the average interference energy e of the signal matrix block in the lth multiscale subband l for:

[0048]

[0049] Among them, N is the dimension of each column in the subsequent dictionary, T l is the failure period calculated in step 2.2, and denote the total energy of the lth multiscale subband and the energy of the periodic pulse, L l is the length of the lth multiscale subband;

[0050] The threshold for the subsequent sparse coding stage of the lth multi-scale subband is estimated to be:

[0051] ε l =z·e l

[0052] Where z = 2 is the threshold gain operator;

[0053] Set the wavelet coefficients of unselected multi-scale subbands to 0;

[0054] In step 2.5, the enhancement strategy of enhance-run-subtract and dictionary learning are integrated into the multi-scale transformation to construct an adaptive multi-scale enhancement dictionary learning model. Only multi-scale subbands are selected for the model. The adaptive multi-scale enhancement dictionary learning model is:

[0055]

[0056] in, is the threshold ε in the lth multi-scale subband l Parameterized multi-scale dictionary learning method, is the denoising wavelet coefficient after the nth iteration, is the denoising wavelet coefficient after the n+1th iteration, y l is the original wavelet coefficient of the lth multi-scale subband, The multi-scale dictionary learning method includes the following optimization process:

[0057] Dictionary learning process:

[0058]

[0059] Among them, D l and A l are the dictionary and representation coefficient of the lth multi-scale subband, and are the dictionary and representation coefficients obtained by solving the l-th multi-scale subband, Indicates that the signal is truncated into small blocks and each small block is moved to the matrix, N represents the dimension of each column in the dictionary, M represents the total number of samples, λ represents the regularization parameter, and P(A) represents the sparse prior of A;

[0060] Signal reconstruction process:

[0061]

[0062] Among them, x l represents the wavelet coefficient of the periodic pulse signal in the lth multi-scale sub-band, Represents the wavelet coefficient of the solved periodic pulse signal in the lth multi-scale subband.

[0063] In step 2.5, using ‖A‖0 as a sparse prior and combining it with the adaptive multi-scale enhanced dictionary learning model, the dictionary learning process can be expressed as:

[0064]

[0065]

[0066] in, and denote the dictionary and representation coefficient of the n+1th iteration in the lth multiscale subband, respectively. and are the dictionary and representation coefficients obtained by the n+1th iteration in the lth multi-scale subband, is the wavelet coefficient of the periodic pulse signal obtained by the n-th iteration in the l-th multi-scale sub-band, and the dictionary learning process is solved by the K-singular value decomposition (K-SVD) algorithm;

[0067] Combined with the adaptive multi-scale enhanced dictionary learning model, the closed-form solution is:

[0068]

[0069] in, Extract the first i blocks, means moving the i-th block back to the reconstructed signal, is the wavelet coefficient of the periodic pulse signal obtained by the n+1th iteration in the lth multiscale subband.

[0070] In step 2.5, after obtaining the wavelet coefficients of the periodic pulse signal obtained by solving each multi-scale sub-band, the fault pulse signal is obtained by inverse Q-modulated wavelet transform W.

[0071] in, represents the wavelet coefficient after the j-th multi-scale sub-band processing, Represents the fault pulse signal obtained by solution.

[0072] The step 3 specifically includes the following steps:

[0073] Step 3.1, performing square envelope spectrum analysis on the extracted fault pulse signal to obtain a square envelope spectrum curve;

[0074] Step 3.2, searching the square envelope spectrum curve for the maximum peak frequency and its multiples except the rotation frequency;

[0075] In step 3.3, if it is determined that the peak frequency and its multiples are within the set resolution error range with the theoretical fault characteristic frequencies (FCFs) and their multiples, it means that the component corresponding to the theoretical fault characteristic frequencies has failed, thereby completing the detection of the bearing fault.

[0076] The present invention also provides a bearing fault detection system based on an adaptive multi-scale enhanced dictionary learning framework, the detection system comprising:

[0077] Fault vibration signal acquisition module: used to acquire bearing fault vibration signals, which are acquired through a vibration acceleration sensor;

[0078] Multi-scale transformation module: used to implement Q-switched wavelet transform of bearing fault vibration signal; by adjusting the quality factor Q, redundancy r and decomposition level J, the signal output by the fault vibration signal acquisition module is subjected to Q-switched wavelet transform to obtain multi-scale sub-bands;

[0079] Adaptive periodic modulation intensity (APMI) index calculation module: used to calculate the adaptive periodic modulation intensity (APMI) of each multi-scale sub-band; using possible fault cycles to define a fault cycle search range, then estimating the fault cycle of each multi-scale sub-band, and finally calculating the adaptive periodic modulation intensity (APMI) value of the multi-scale sub-band output by the multi-scale transformation module;

[0080] Multi-scale sub-band screening module: used to implement multi-scale sub-band screening based on adaptive periodic modulation intensity (APMI); summing the adaptive periodic modulation intensity (APMI) values ​​of three consecutive multi-scale sub-bands output by the adaptive periodic modulation intensity (APMI) index calculation module, and outputting the three multi-scale sub-bands corresponding to the maximum value as the preferred multi-scale sub-bands;

[0081] A threshold adaptive estimation module is used to implement adaptive estimation of the threshold parameters of the preferred multi-scale sub-band sparse coding stage; calculate the threshold of the sparse coding stage by calculating the average interference energy of the signal matrix block in the preferred multi-scale sub-band output by the multi-scale sub-band screening module and combining it with a given threshold gain operator; and set the wavelet coefficients of the unselected multi-scale sub-bands to 0;

[0082] Fault pulse extraction module: used to extract fault pulse signals; using the constructed adaptive multi-scale enhanced dictionary learning model, the optimal multi-scale sub-band is optimized and solved. After obtaining the processed wavelet coefficients of each multi-scale sub-band, the fault pulse signal is extracted through inverse Q-modulated wavelet transform;

[0083] Fault detection module: used to determine the fault type; based on the fault pulse signal output by the fault pulse extraction module, the fault frequency and its multiples are identified through envelope analysis to determine the fault type.

[0084] The present invention also provides a bearing fault detection device based on an adaptive multi-scale enhanced dictionary learning framework, comprising:

[0085] A memory for storing a computer program for implementing the bearing fault detection method based on the adaptive multi-scale enhanced dictionary learning framework;

[0086] A processor is configured to implement the bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework when executing the computer program.

[0087] The present invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement a bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework.

[0088] Compared with the prior art, the present invention has the following advantages:

[0089] 1. The fault sensitivity index adaptive periodic modulation intensity APMI constructed in the detection method of the present invention is robust to complex and strong interference and can effectively quantify the fault information in each sub-band.

[0090] 2. The sub-band screening method based on adaptive periodic modulation intensity (APMI) in the detection method of the present invention can simplify and accelerate the model learning process and effectively weaken the influence of complex interference.

[0091] 3. The adaptive threshold estimation method based on interference energy in the signal designed in the detection method of the present invention can improve the efficiency and versatility of the model, laying a foundation for online diagnosis.

[0092] 4. The detection method of the present invention combines the enhancement strategy and dictionary learning into multi-scale transformation, which can significantly improve the performance of dictionary learning.

[0093] 5. The framework established by the detection method of the present invention has anti-interference performance and adaptive capabilities, which can improve the real-time and accuracy of fault detection.

[0094] In summary, the various steps of the present invention work together to enhance signal sparsity while weakening the impact of complex strong interference, so that the adaptive multi-scale enhanced dictionary learning framework can more effectively improve the performance of multi-scale dictionary learning. At the same time, the present invention is simple and easy to implement, and its adaptability overcomes manual parameter adjustment, making it suitable for online monitoring and fault diagnosis of bearings.

[0095] The disclosed system, device, and medium are implemented according to the method and also have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 Flowchart of the present invention.

[0097] Figure 2 This is a schematic diagram of the main structure of the bearing fault simulation test bench in this embodiment.

[0098] Figure 3 Schematic diagram of the time domain waveform of the acquisition signal in this embodiment.

[0099] Figure 4 Schematic diagram of the square envelope spectrum of the acquisition signal in this embodiment.

[0100] Figure 5 This is a schematic diagram of extracting the time domain waveform of a fault signal using adaptive multi-scale enhanced dictionary learning in this embodiment.

[0101] Figure 6 This is a schematic diagram of extracting the square envelope spectrum of a fault signal through adaptive multi-scale enhanced dictionary learning in this embodiment.

[0102] Figure 7 A schematic diagram of a bearing fault detection system based on an adaptive multi-scale enhanced dictionary learning framework provided by the present invention. DETAILED DESCRIPTION

[0103] The following will refer to Figures 1 to 7 The specific embodiments of the present invention will be described in more detail. Although specific embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Instead, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0104] To facilitate understanding of the embodiments of the present invention, further explanation will be given below using specific embodiments as examples in conjunction with the accompanying drawings, and the accompanying drawings do not constitute a limitation on the embodiments of the present invention.

[0105] like Figure 1 As shown, a bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework comprises the following steps:

[0106] Step 1: collecting bearing fault vibration signals;

[0107] Among them, in step 1, the bearing fault vibration signal y∈R n By periodic pulse x∈R n , complex harmonic interference h∈R n and strong background noise e∈R n The structure is expressed as y=x+h+e, and the bearing fault vibration signal y is collected by a vibration acceleration sensor.

[0108] Step 2: construct an adaptive multi-scale enhanced dictionary learning framework to extract fault pulse signals;

[0109] The specific steps include:

[0110] Step 2.1, performing Q-modulated wavelet transform on the bearing fault vibration signal collected in step 1 to obtain multi-scale sub-bands.

[0111] Among them, in step 2.1, the transformation parameters of the Q-wavelet include the quality factor Q, the redundancy r and the decomposition level J, and the bearing fault vibration signal y is transformed into the Q-wavelet transform W T The J+1 layer wavelet coefficients are decomposed as follows:

[0112] W T y→{y1,…,y J ,y J+1}

[0113] Among them, y j (j=1, ..., J+1) is the wavelet coefficient of the j-th multi-scale subband.

[0114] Step 2.2: Calculate the adaptive periodic modulation intensity (APMI) of each multi-scale subband in step 2.1.

[0115] In step 2.2, the steps for calculating the adaptive periodic modulation intensity APMI of each multi-scale sub-band are as follows:

[0116] Calculate the possible failure period τ i :

[0117] τ i =round(fs / FCFs)

[0118] Among them, f s represents the sampling frequency, the round(·) function is used to round to the nearest integer, and the fault characteristic frequencies (FCFs) are the outer race (BPFO), inner race (BPFI), cage (FTF), and rolling element (BSF);

[0119] Define the fault cycle search range for each multi-scale sub-band:

[0120]

[0121] Among them, γ j represents a set of four fault cycles and their surrounding coefficients, where L0 and L j denote the original acquired signal y and the length of the j-th multi-scale subband, respectively. C = 10 is a constant that limits the search range.

[0122] Determine when PMI j Multi-scale subbands at maximum and the fault period of the multiscale subband

[0123]

[0124] in,

[0125]

[0126] Among them, R f [·] is the autocorrelation function, Γ(y j )=|y j +i·Hilbert(y j )∣ represents the input signal y j The absolute value of the envelope after Hilbert transformation, where i is an imaginary unit;

[0127] Based on the Multi-scale sub-bands, The fault period T of each multi-scale sub-band can be calculated j :

[0128]

[0129] Where T m and L m Respectively represent The period and length of the multi-scale sub-bands;

[0130] Based on T of each multi-scale sub-band j, calculate the adaptive periodic modulation intensity APMI:

[0131]

[0132] in, and They represent the envelope absolute value Γ(y j ) at time shift 0 and T j The autocorrelation function value at .

[0133] Step 2.3: Perform multi-scale sub-band screening based on the adaptive periodic modulation intensity (APMI) calculated in step 2.2 to obtain a preferred multi-scale sub-band.

[0134] In step 2.3, multi-scale sub-band screening is performed based on the adaptive periodic modulation intensity (APMI):

[0135]

[0136] Among them, j opt represents the first preferred multi-scale subband;

[0137] Define the optimal multi-scale subband set S opt :

[0138] S opt ={j opt ,j opt +1,j opt +2}.

[0139] Step 2.4: adaptively estimate the threshold parameters of the sparse coding stage for the preferred multi-scale sub-band obtained in step 2.3, and set the wavelet coefficients of the unselected multi-scale sub-bands to 0.

[0140] Wherein, the step 2.4 includes the following process:

[0141] The threshold parameters of the sparse coding stage for adaptive estimation of the preferred multi-scale subband are:

[0142] Calculate the average interference energy e of the signal matrix block in the lth multiscale subband l for:

[0143]

[0144] Among them, N is the dimension of each column in the subsequent dictionary, T l is the failure period calculated in step 2.2, and denote the total energy of the lth multiscale subband and the energy of the periodic pulse, L l is the length of the lth multiscale subband;

[0145] The threshold for the subsequent sparse coding stage of the lth multi-scale subband is estimated to be:

[0146] ε l =z·e l

[0147] Where z = 2 is the threshold gain operator;

[0148] Set the wavelet coefficients of unselected multi-scale subbands to 0.

[0149] In step 2.5, the enhancement strategy of enhance-run-subtract and dictionary learning are integrated into the multi-scale transformation, an adaptive multi-scale enhancement dictionary learning model is constructed, and the model is optimized to extract the fault pulse signal.

[0150] In step 2.5, the enhancement strategy of enhance-run-subtract and dictionary learning are integrated into the multi-scale transformation to construct an adaptive multi-scale enhancement dictionary learning model. Only multi-scale subbands are selected for the model. The adaptive multi-scale enhancement dictionary learning model is:

[0151]

[0152] in, is the threshold ε in the lth multi-scale subband l Parameterized multi-scale dictionary learning method, is the denoising wavelet coefficient after the nth iteration, is the denoising wavelet coefficient after the n+1th iteration, y l is the original wavelet coefficient of the lth multi-scale subband, The multi-scale dictionary learning method includes the following optimization process:

[0153] Dictionary learning process:

[0154]

[0155] Among them, D l and A l are the dictionary and representation coefficient of the lth multi-scale subband, and are the dictionary and representation coefficients obtained by solving the l-th multi-scale subband, Indicates that the signal is truncated into small blocks and each small block is moved to the matrix, N represents the dimension of each column in the dictionary, M represents the total number of samples, λ represents the regularization parameter, and P(A) represents the sparse prior of A;

[0156] Signal reconstruction process:

[0157]

[0158] Among them, x l represents the wavelet coefficient of the periodic pulse signal in the lth multi-scale sub-band, Represents the wavelet coefficient of the solved periodic pulse signal in the lth multi-scale subband.

[0159] In a preferred embodiment of the method, ‖A‖0 is used as a sparse prior and combined with an adaptive multi-scale enhanced dictionary learning model. The dictionary learning process can be expressed as:

[0160]

[0161]

[0162] in, and denote the dictionary and representation coefficient of the n+1th iteration in the lth multiscale subband, respectively. and are the dictionary and representation coefficients obtained by the n+1th iteration in the lth multi-scale subband, is the wavelet coefficient of the periodic pulse signal obtained by the n-th iteration in the l-th multi-scale sub-band, and the dictionary learning process is solved by the K-singular value decomposition (K-SVD) algorithm.

[0163] In a preferred embodiment of the method, the signal reconstruction process is a convex optimization problem with a closed-form solution. Combined with the adaptive multi-scale enhanced dictionary learning model, the closed-form solution is:

[0164]

[0165] in, Indicates extracting the i-th block from the original signal, means moving the i-th block back to the reconstructed signal, is the wavelet coefficient of the periodic pulse signal obtained by the n+1th iteration in the lth multiscale subband.

[0166] In a preferred embodiment of the method, in step 2.5, after obtaining the wavelet coefficients of the periodic pulse signal obtained by solving each multi-scale sub-band, that is, the wavelet coefficients of each multi-scale sub-band, the fault pulse signal is obtained by inverse Q-modulated wavelet transform W.

[0167]

[0168] in, represents the wavelet coefficient after the j-th multi-scale sub-band processing, Represents the fault pulse signal obtained by solution.

[0169] Step 3: Based on the fault pulse signal extracted in step 2, the fault frequency and its multiples are identified through envelope analysis to determine the fault type.

[0170] The step 3 specifically includes the following steps:

[0171] Step 3.1, performing square envelope spectrum analysis on the extracted fault pulse signal to obtain a square envelope spectrum curve;

[0172] Step 3.2, searching the square envelope spectrum curve for the maximum peak frequency and its multiples except the rotation frequency;

[0173] In step 3.3, if it is determined that the peak frequency and its multiples are within the set resolution error range with the theoretical fault characteristic frequencies (FCFs) and their multiples, it means that the component corresponding to the theoretical fault characteristic frequencies has failed, thereby completing the detection of the bearing fault.

[0174] In order to further understand the present invention, in one embodiment, Figure 1 Flowchart of the present invention; Figure 1 As shown, a bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework includes the following steps:

[0175] Step 1: collecting bearing fault vibration signal y;

[0176] Step 2: construct an adaptive multi-scale enhanced dictionary learning framework to extract fault pulse signals;

[0177] The specific steps include:

[0178] Step 2.1, performing Q-modulated wavelet transform on the bearing fault vibration signal collected in step 1 to obtain multi-scale sub-bands;

[0179] Step 2.2, calculating the adaptive periodic modulation intensity APMI of each multi-scale sub-band in step 2.1;

[0180] Step 2.3: Perform multi-scale sub-band screening based on the APMI calculated in step 2.2 to obtain the optimal multi-scale sub-band;

[0181] Step 2.4, adaptively estimating the threshold parameters of the sparse coding stage for the preferred multi-scale sub-band obtained in step 2.3, and setting the wavelet coefficients of the unselected multi-scale sub-bands to 0;

[0182] In step 2.5, the enhancement strategy of enhance-run-subtract and dictionary learning are integrated into the multi-scale transformation, an adaptive multi-scale enhancement dictionary learning model is constructed, and the model is optimized to extract the fault pulse signal.

[0183] Step 3: Based on the fault pulse signal extracted in step 2, the fault frequency and its multiples are identified through envelope analysis to determine the fault type.

[0184] The step 3 specifically includes the following steps:

[0185] Step 3.1, performing square envelope spectrum analysis on the extracted fault pulse signal to obtain a square envelope spectrum curve;

[0186] Step 3.2, searching the square envelope spectrum curve for the maximum peak frequency and its multiples except the rotation frequency;

[0187] In step 3.3, if it is determined that the peak frequency and its multiples are within the set resolution error range with the theoretical fault characteristic frequencies (FCFs) and their multiples, it means that the component corresponding to the theoretical fault characteristic frequencies has failed, thereby completing the detection of the bearing fault.

[0188] The above embodiments constitute the complete technical solution of this invention. Unlike existing technologies, these embodiments enhance signal sparsity while simultaneously mitigating the effects of complex and strong interference. Consequently, this framework can more effectively improve the performance of multi-scale dictionary learning. Furthermore, the framework's adaptability enables flexible and reliable bearing fault monitoring and diagnosis in practical applications.

[0189] like Figure 2 Figure 2 shows a simplified diagram of the main structure of a bearing fault simulation test bench according to an embodiment of the present invention. The main body is driven by a motor, and the electric spindle is connected to the main spindle via a flexible coupling. Support bearings 1 and 2 are two support bearings. Axial and radial loads are applied to the test bearing (H7015C). Two acceleration sensors are mounted on the bearing seat. The speed is 2000 r / min and the sampling frequency is 20,000 Hz. Based on the bearing rotational frequency (RF) and the bearing's geometric parameters, the four FCFs (BPFO, BPFI, BSF, and FTF) are calculated to be 275.9 Hz, 357.4 Hz, 123.1 Hz, and 14.5 Hz, respectively. When the test bearing was disassembled, localized spalling of the outer ring was found.

[0190] like Figure 3 As shown, the collected time domain vibration signal does not show any obvious periodic pulses.

[0191] like Figure 4 As shown in Figure 1, it can be found that there are a large number of unknown low-frequency interference components in the square envelope spectrum of the original acquired signal, and the second-order BPFO is completely submerged. Therefore, it is necessary to perform more reliable fault diagnosis.

[0192] The following is a bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework to process the collected signals of the above test bench, in which the quality factor Q = 2, the redundancy r = 5, the number of decomposition levels J = 12, the dimension of each column in the dictionary N = 4, the number of dictionaries L = 2N = 8, the number of enhancements is 3, and the original collected signal is divided into blocks with maximum overlap to determine the total number of samples M.

[0193] like Figure 5 、 Figure 6 As shown in FIG, a schematic diagram of the extracted signal and its square envelope spectrum based on the adaptive multi-scale enhanced dictionary learning framework of this embodiment; Figure 5 As shown in Figure 2, an obvious periodic pulse signal can be observed. The bearing characteristic frequencies (BPFO and BPFO*2) and the bearing rotation frequency (RF) are shown in Figure 2. Figure 6 The low-frequency interference component is clearly suppressed, indicating a faulty outer ring of the bearing in this embodiment. Actual disassembly confirms the accuracy of the diagnostic results. Therefore, the present invention can extract effective periodic pulse features for bearing fault detection and diagnosis under complex interference.

[0194] like Figure 7 As shown, the present invention also provides a bearing fault detection system based on an adaptive multi-scale enhanced dictionary learning framework, the detection system comprising:

[0195] The fault vibration signal acquisition module is configured to: acquire the bearing fault vibration signal in step 1, wherein the bearing fault vibration signal is acquired by a vibration acceleration sensor;

[0196] The multi-scale transformation module is configured to: implement the Q-modulated wavelet transform of the bearing fault vibration signal in step 2.1; perform Q-modulated wavelet transform on the signal output by the fault vibration signal acquisition module to obtain multi-scale sub-bands by adjusting the quality factor Q, redundancy r and decomposition level J;

[0197] The adaptive periodic modulation intensity (APMI) index calculation module is configured to: implement the calculation of the adaptive periodic modulation intensity (APMI) of each multi-scale sub-band in step 2.2; define a fault cycle search range using possible fault cycles, then estimate the fault cycle of each multi-scale sub-band, and finally calculate the adaptive periodic modulation intensity (APMI) value of the multi-scale sub-band output by the multi-scale transformation module;

[0198] The multi-scale sub-band screening module is configured to: implement the multi-scale sub-band screening based on the adaptive periodic modulation intensity (APMI) in step 2.3; sum the adaptive periodic modulation intensity (APMI) values ​​of three consecutive multi-scale sub-bands output by the adaptive periodic modulation intensity (APMI) index calculation module, and output the three multi-scale sub-bands corresponding to the maximum value as the preferred multi-scale sub-bands;

[0199] The threshold adaptive estimation module is configured to: implement adaptive estimation of the threshold parameters of the preferred multi-scale sub-band sparse coding stage in step 2.4; calculate the threshold of the sparse coding stage by calculating the average interference energy of the signal matrix block in the preferred multi-scale sub-band output by the multi-scale sub-band screening module and combining it with a given threshold gain operator; and set the wavelet coefficients of the unselected multi-scale sub-bands to 0;

[0200] The fault pulse extraction module is configured to: extract the fault pulse signal in step 2.5; optimize and solve the preferred multi-scale sub-bands using the constructed adaptive multi-scale enhanced dictionary learning model, obtain the processed wavelet coefficients of each multi-scale sub-band, and extract the fault pulse signal through inverse Q-modulated wavelet transform;

[0201] The fault detection module is configured to: determine the fault type in step 3; and identify the fault frequency and its multiples through envelope analysis based on the fault pulse signal output by the fault pulse extraction module to determine the fault type.

[0202] The present invention also provides a bearing fault detection device based on an adaptive multi-scale enhanced dictionary learning framework, comprising:

[0203] A memory for storing a computer program for implementing the bearing fault detection method based on the adaptive multi-scale enhanced dictionary learning framework;

[0204] A processor is configured to implement the bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework when executing the computer program.

[0205] The present invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it is capable of detecting a bearing fault based on an adaptive multi-scale enhanced dictionary learning framework.

[0206] It should be noted that certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different nouns to refer to the same component. This specification and claims do not use the difference in nouns as a way to distinguish components, but use the difference in the functions of the components as the criterion for distinction. As mentioned throughout the specification and claims, "including" or "comprising" is an open term, so it should be interpreted as "including but not limited to". The subsequent description of the specification is a preferred embodiment of the present invention, but the description is based on the general principles of the specification and is not intended to limit the scope of the invention. The scope of protection of the present invention shall be as defined in the attached claims.

[0207] Although the embodiments of the present invention have been described above with reference to the accompanying drawings, the present invention is not limited to the above-mentioned specific embodiments and application fields. The above-mentioned specific embodiments are merely illustrative and instructive, and are not restrictive. A person skilled in the art, guided by this specification and without departing from the scope of protection of the claims of the present invention, may also devise various forms, all of which fall within the scope of protection of the present invention.

Claims

1. A bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework, characterized by: The following steps are involved: Step 1: collecting bearing fault vibration signals; Step 2: construct an adaptive multi-scale enhanced dictionary learning framework to extract fault pulse signals; The step 2 specifically includes the following steps: Step 2.1, performing Q-modulated wavelet transform on the bearing fault vibration signal collected in step 1 to obtain multi-scale sub-bands; Step 2.2, calculating the adaptive periodic modulation intensity APMI of each multi-scale sub-band in step 2.1; In step 2.2, the steps of calculating the adaptive periodic modulation intensity APMI of each multi-scale sub-band are as follows: Calculate the possible failure period τ i : τ i =round(f s / FCFs) Among them, f s represents the sampling frequency, the round(·) function is used to round to the nearest integer, and the fault characteristic frequencies FCFs are outer ring BPFO, inner ring BPFI, cage FTF, and rolling element BSF respectively; Define the fault cycle search range for each multi-scale sub-band: Among them, γ j represents a set of four fault cycles and their surrounding coefficients, where L0 and L j denote the original acquired signal y and the length of the j-th multi-scale subband, respectively. C = 10 is a constant that limits the search range. Determine when PMI j Multi-scale subbands at maximum and the fault period of the multiscale subband in, in, is the autocorrelation function, Γ(y j )=|y j +i·Hilbert(y j )∣ represents the input signal y j The absolute value of the envelope after Hilbert transformation, where i is an imaginary unit; Based on the Multi-scale sub-bands, The fault period T of each multi-scale sub-band can be calculated j : Among them, T m and L m Respectively represent The period and length of the multi-scale sub-bands; Based on T of each multi-scale sub-band j , calculate the adaptive periodic modulation intensity APMI: in, and They represent the envelope absolute value Γ(y j ) at time shift 0 and T j The autocorrelation function value when ; Step 2.3, performing multi-scale sub-band screening based on the adaptive periodic modulation intensity (APMI) calculated in step 2.2 to obtain a preferred multi-scale sub-band; Step 2.4, adaptively estimating the threshold parameters of the sparse coding stage for the preferred multi-scale sub-band obtained in step 2.3, and setting the wavelet coefficients of the unselected multi-scale sub-bands to 0; In step 2.5, the enhancement strategy of enhance-run-subtract and dictionary learning are integrated into the multi-scale transformation to construct an adaptive multi-scale enhancement dictionary learning model, which is then optimized to extract the fault pulse signal. In step 2.5, the enhancement strategy of enhance-run-subtract and dictionary learning are integrated into the multi-scale transformation to construct an adaptive multi-scale enhancement dictionary learning model. Only multi-scale subbands are selected for the model. The adaptive multi-scale enhancement dictionary learning model is: in, is the threshold ε in the lth multi-scale subband l Parameterized multi-scale dictionary learning method, is the denoising wavelet coefficient after the nth iteration, is the denoising wavelet coefficient after the n+1th iteration, y l is the original wavelet coefficient of the lth multi-scale subband, Step 3: Based on the fault pulse signal extracted in step 2, the fault frequency and its multiples are identified through envelope analysis to determine the fault type.

2. The bearing fault detection method based on the adaptive multi-scale enhanced dictionary learning framework according to claim 1 is characterized by: In step 1, the bearing fault vibration signal y∈R n By periodic pulse x∈R n , complex harmonic interference h∈R n and strong background noise e∈R n The structure is expressed as y=x+h+e, and the bearing fault vibration signal y is collected by a vibration acceleration sensor.

3. The bearing fault detection method based on the adaptive multi-scale enhanced dictionary learning framework according to claim 1 is characterized in that: In step 2.1, the transformation parameters of the Q-wavelet include the quality factor Q, the redundancy r and the decomposition level J. The bearing fault vibration signal y is transformed into the Q-wavelet W T The J+1 layer wavelet coefficients are decomposed as follows: W T y→{y1,…,y J ,y J+1 } Among them, y j (j=1, ..., J+1) is the wavelet coefficient of the j-th multiscale subband; In step 2.3, multi-scale sub-band screening is performed based on the adaptive periodic modulation intensity APMI: Among them, j opt represents the first preferred multi-scale subband; Define the optimal multi-scale subband set S opt : S opt ={j opt ,j opt +1,j opt +2}; The step 2.4 includes the following processes: The threshold parameters of the sparse coding stage for adaptive estimation of the preferred multi-scale subband are: Calculate the average interference energy e of the signal matrix block in the lth multiscale subband l for: Among them, N is the dimension of each column in the subsequent dictionary, T l is the failure period calculated in step 2.2, and denote the total energy of the lth multiscale subband and the energy of the periodic pulse, L l is the length of the lth multiscale subband; The threshold for the subsequent sparse coding stage of the lth multi-scale subband is estimated to be: ε l =z·e l Where z = 2 is the threshold gain operator; Set the wavelet coefficients of unselected multi-scale subbands to 0; In step 2.5, the multi-scale dictionary learning method includes the following optimization process: Dictionary learning process: Among them, D l and A l are the dictionary and representation coefficient of the lth multi-scale subband, and are the dictionary and representation coefficients obtained by solving the l-th multi-scale subband, Indicates that the signal is truncated into small blocks and each small block is moved to the matrix, N represents the dimension of each column in the dictionary, M represents the total number of samples, λ represents the regularization parameter, and P(A) represents the sparse prior of A; Signal reconstruction process: Among them, x l represents the wavelet coefficient of the periodic pulse signal in the lth multi-scale sub-band, Represents the wavelet coefficient of the solved periodic pulse signal in the lth multi-scale subband.

4. The bearing fault detection method based on the adaptive multi-scale enhanced dictionary learning framework according to claim 3 is characterized by: In step 2.5, using ‖A‖0 as a sparse prior and combining it with the adaptive multi-scale enhanced dictionary learning model, the dictionary learning process can be expressed as: in, and denote the dictionary and representation coefficient of the n+1th iteration in the lth multiscale subband, respectively. and are the dictionary and representation coefficients obtained by the n+1th iteration in the lth multi-scale subband, is the wavelet coefficient of the periodic pulse signal obtained by the n-th iteration in the l-th multi-scale sub-band, and the dictionary learning process is solved by the K-singular value decomposition K-SVD algorithm; Combined with the adaptive multi-scale enhanced dictionary learning model, the closed-form solution is: in, Indicates extracting the i-th block from the original signal, means moving the i-th block back to the reconstructed signal, is the wavelet coefficient of the periodic pulse signal obtained by the n+1th iteration in the lth multiscale subband.

5. The bearing fault detection method based on the adaptive multi-scale enhanced dictionary learning framework according to claim 4 is characterized in that: In step 2.5, after obtaining the wavelet coefficients of the periodic pulse signal obtained by solving each multi-scale sub-band, the fault pulse signal is obtained by inverse Q-modulated wavelet transform W. in, represents the wavelet coefficient after the j-th multi-scale sub-band processing, Represents the fault pulse signal obtained by solution.

6. The bearing fault detection method based on the adaptive multi-scale enhanced dictionary learning framework according to claim 1 is characterized in that: The step 3 specifically includes the following steps: Step 3.1, performing square envelope spectrum analysis on the extracted fault pulse signal to obtain a square envelope spectrum curve; Step 3.2, searching the square envelope spectrum curve for the maximum peak frequency and its multiples except the rotation frequency; In step 3.3, if it is determined that the peak frequency and its multiples are within the set resolution error range with the theoretical fault characteristic frequency FCFs and its multiples, it means that the component corresponding to the theoretical fault characteristic frequency has failed, thereby completing the detection of the bearing fault.

7. A bearing fault detection system based on an adaptive multi-scale enhanced dictionary learning framework according to the method of claim 1, characterized in that: include: Fault vibration signal acquisition module: used to acquire bearing fault vibration signals, which are acquired through a vibration acceleration sensor; Multi-scale transformation module: used to implement Q-switched wavelet transform of bearing fault vibration signal; by adjusting the quality factor Q, redundancy r and decomposition level J, the signal output by the fault vibration signal acquisition module is subjected to Q-switched wavelet transform to obtain multi-scale sub-bands; Adaptive periodic modulation intensity (APMI) index calculation module: used to calculate the adaptive periodic modulation intensity (APMI) of each multi-scale sub-band; using possible fault cycles to define a fault cycle search range, then estimating the fault cycle of each multi-scale sub-band, and finally calculating the adaptive periodic modulation intensity (APMI) value of the multi-scale sub-band output by the multi-scale transformation module; Multi-scale sub-band screening module: used to implement multi-scale sub-band screening based on adaptive periodic modulation intensity (APMI); summing the adaptive periodic modulation intensity (APMI) values ​​of three consecutive multi-scale sub-bands output by the adaptive periodic modulation intensity (APMI) index calculation module, and outputting the three multi-scale sub-bands corresponding to the maximum value as the preferred multi-scale sub-bands; A threshold adaptive estimation module is used to implement adaptive estimation of the threshold parameters of the preferred multi-scale sub-band sparse coding stage; the threshold of the sparse coding stage is calculated by calculating the average interference energy of the signal matrix block in the preferred multi-scale sub-band output by the multi-scale sub-band screening module and combining it with a given threshold gain operator; Set the wavelet coefficients of unselected multi-scale subbands to 0; Fault pulse extraction module: used to extract fault pulse signals; using the constructed adaptive multi-scale enhanced dictionary learning model, the optimal multi-scale sub-band is optimized and solved. After obtaining the processed wavelet coefficients of each multi-scale sub-band, the fault pulse signal is extracted through inverse Q-modulated wavelet transform; Fault detection module: used to determine the fault type; based on the fault pulse signal output by the fault pulse extraction module, the fault frequency and its multiples are identified through envelope analysis to determine the fault type.

8. A bearing fault detection device based on an adaptive multi-scale enhanced dictionary learning framework, characterized by: include: A memory for storing a computer program for implementing a bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework according to any one of claims 1 to 6; A processor is configured to implement the bearing fault detection method based on an adaptive multi-scale enhanced dictionary learning framework as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it can implement the bearing fault detection method based on the adaptive multi-scale enhanced dictionary learning framework described in any one of claims 1 to 6.

Images