Direct current motor fault diagnosis method based on improved local mean decomposition and composite multi-scale bubble entropy fusion

Abstract

The application discloses a DC motor fault diagnosis method based on improved local mean decomposition and composite multi-scale bubble entropy fusion and relates to the technical field of motor fault diagnosis. In order to solve the technical defects that the existing motor fault diagnosis technology is insufficient in fault feature extraction, the technical scheme of the application comprises the following steps: collecting a sound signal of a DC motor in a running state as an original signal; decomposing the original signal by using an improved local mean decomposition method to obtain a plurality of product components, and screening out effective feature components with high correlation with the original signal; performing composite multi-scale bubble entropy calculation on the effective feature components to extract a multi-dimensional entropy feature vector representing signal complexity; and inputting the multi-dimensional entropy feature vector into a particle swarm optimized extreme learning machine model for classification to obtain a fault diagnosis result of the DC motor. The application can be widely applied to real-time monitoring and intelligent fault identification of the running state of a DC motor in industrial production.

Description

Technical Field

[0001] This relates to the field of motor fault diagnosis technology, specifically to methods for diagnosing DC motor faults. Background Technology

[0002] DC motors, as important power devices, are widely used in many key fields such as electric vehicles, intelligent manufacturing, and aerospace. With increasingly complex operating environments, motors are prone to mechanical wear, uneven rotor stress, and overload due to long-term operation, leading to typical faults such as rotor shaft bending, blade breakage, and bearing damage. If these faults are not detected and diagnosed in a timely manner, they can cause equipment downtime, production interruptions, and in severe cases, even safety accidents. Therefore, accurate and reliable fault diagnosis of DC motors is of significant practical importance.

[0003] Currently, research in the field of DC motor fault diagnosis mainly focuses on feature extraction and recognition techniques based on vibration, acoustic, and current signals. Traditional diagnostic methods, such as Empirical Mode Decomposition (EMD) and Variational Mode Decomposition (VMD), have made some progress in identifying typical motor faults. For example, Sun Hao et al. studied a method for diagnosing broken bar faults in squirrel-cage motor rotors based on vibration and current signals, pointing out that traditional methods are affected by noise interference in practical applications, impacting diagnostic accuracy. Xia Zhiling et al. proposed a method for diagnosing broken bar faults in asynchronous motor rotors using Variational Mode Decomposition (VMD), demonstrating its advantages in high-noise environments; however, VMD still requires manual adjustment of parameters based on experience, resulting in limited applicability. Furthermore, Wu Liyuan et al. used an improved Nonlinear Mode Decomposition algorithm (NME-EMD) combined with energy entropy analysis of early-stage broken bar fault characteristics in the rotor, which improved diagnostic accuracy but also revealed that mode aliasing is still difficult to completely avoid.

[0004] On the other hand, progress has also been made in the field of blade fault and bearing damage diagnosis. For example, Zhang Junhua et al. used startup signals to diagnose wind turbine blade cracks and fracture states, achieving high accuracy based on time-frequency analysis and pattern recognition technology; Jia Lin et al. used the particle swarm optimization (PSO) algorithm to optimize VMD parameters, thereby improving the distinguishability of bearing inner and outer ring faults. However, these methods are mostly based on signal decomposition and feature extraction at a single scale, lacking comprehensive analysis of signal complexity at different scales, resulting in insufficient generalization ability in practical applications. In addition, traditional signal entropy analysis methods, such as multiscale entropy (MSE) or single-scale bubble entropy (BE), often lose important local feature information due to the coarsening process when analyzing nonlinear and non-stationary signals.

[0005] Therefore, the existing technology has the following problems that urgently need to be solved:

[0006] (1) Traditional signal decomposition methods are susceptible to mode mixing, resulting in insufficient extraction of fault features and reduced diagnostic accuracy;

[0007] (2) The single-scale entropy analysis method is difficult to fully and effectively reflect the complexity characteristics of signals at multiple scales, which limits the effective extraction of deep information of fault signals.

[0008] (3) Existing fault diagnosis models lack automatic optimization of model parameters, resulting in large fluctuations in diagnostic performance and poor generalization ability.

[0009] Therefore, there is an urgent need to propose a DC motor fault diagnosis method that can effectively overcome the above-mentioned shortcomings and improve diagnostic accuracy and robustness. Summary of the Invention

[0010] To address the shortcomings of existing motor fault diagnosis technologies, such as insufficient fault feature extraction, limited effective extraction of deep-level fault signal information, and significant fluctuations in diagnostic performance, the present invention provides the following technical solution:

[0011] A fault diagnosis method for DC motors based on improved local mean decomposition and composite multi-scale bubble entropy fusion includes:

[0012] The steps for collecting the sound signal of a DC motor during operation as the raw signal;

[0013] The process involves decomposing the original signal using an improved local mean decomposition method to obtain several multiplicative integrals, and then selecting effective feature components that are highly correlated with the original signal.

[0014] The steps include performing composite multi-scale bubble entropy calculation on the effective feature components and extracting multidimensional entropy feature vectors that characterize the signal complexity.

[0015] The step of inputting the multidimensional entropy feature vector into the particle swarm optimization extreme learning machine model for classification to obtain the fault diagnosis results of the DC motor.

[0016] Furthermore, a preferred embodiment is provided, in which an acoustic sensor is used to collect sound signals during the operation of the motor in real time. The distance between the acoustic sensor and the motor housing is 10cm, and the two sensors are arranged at 90° orthogonal.

[0017] Furthermore, a preferred implementation is provided, in which an integral sliding window strategy is used to calculate the local mean of adjacent extreme point intervals, and a high-smoothness local mean function is obtained by moving average filtering to suppress mode mixing and baseline noise interference.

[0018] Furthermore, a preferred embodiment is provided, wherein the method for screening effective feature components is as follows: calculate the correlation coefficient between each multiplicand and the original signal, and remove components with a correlation coefficient lower than 0.3.

[0019] Furthermore, a preferred embodiment is provided in which the scale factor for the composite multi-scale bubble entropy calculation is 10, the embedding dimension is 2, the time delay is 1, and the similarity tolerance is 0.15 times the standard deviation of the effective feature components.

[0020] Furthermore, a preferred embodiment is provided, wherein in the particle swarm optimization extreme learning machine model, the number of particles in the particle swarm optimization algorithm is 30, the maximum number of iterations is 100, and the number of hidden layer nodes is 100.

[0021] Based on the same inventive concept, this invention also provides a DC motor fault diagnosis device based on improved local mean decomposition and composite multi-scale bubble entropy fusion, comprising:

[0022] A module that collects sound signals from a DC motor during operation as the raw signal;

[0023] The original signal is decomposed using an improved local mean decomposition method to obtain several multiplicative integrals, and modules with effective feature components that are highly correlated with the original signal are selected.

[0024] A module for calculating composite multi-scale bubble entropy on the effective feature components and extracting multi-dimensional entropy feature vectors that characterize signal complexity;

[0025] The module inputs the multidimensional entropy feature vector into the particle swarm optimization extreme learning machine model for classification to obtain the fault diagnosis results of the DC motor.

[0026] Based on the same inventive concept, the present invention also provides a computer storage medium for storing a computing program, wherein when the computer program is read by a computer, the computer executes the method described thereon.

[0027] Based on the same inventive concept, the present invention also provides a computer, including a processor and a storage medium, wherein when the processor reads a computer program stored in the storage medium, the computer executes the method described thereon.

[0028] Based on the same inventive concept, the present invention also provides a computer program product, which, when executed, implements the method described.

[0029] Compared with the prior art, the advantages of the technical solution provided by the present invention are as follows:

[0030] This scheme utilizes an improved Local Means Decomposition (LIMD) method, employing an integral sliding window strategy and adaptive boundary processing. This effectively suppresses endpoint effects and baseline noise, and significantly reduces mode aliasing during signal decomposition, thereby extracting fault-related features more accurately. Compared to the traditional LIMD method, this scheme's LIMD method obtains clearer time-spectrum features, ensuring that effective signals are not interfered with by noise during fault diagnosis, fundamentally improving the accuracy of fault feature extraction.

[0031] This scheme introduces Composite Multiscale Bubble Entropy (CMBE) to analyze the complexity of the decomposed signal. By improving the coarse-grained sequence generation method, it effectively solves the problems of information loss and inaccurate entropy estimation in traditional single-scale or simple multiscale entropy methods under long-scale factors, thus improving the comprehensiveness and stability of signal complexity features. Compared with existing single-scale entropy analysis methods, the CMBE in this scheme can reliably reflect the subtle differences between different fault signals under multi-scale conditions, improving the ability to distinguish fault types.

[0032] This approach applies the Particle Swarm Optimization (PSO) algorithm to the parameter optimization process of Extreme Learning Machine (ELM), achieving adaptive optimization of hidden layer connection weights and bias vectors. This overcomes the performance fluctuation problem caused by random parameter initialization in traditional ELM models. Experimental results show that the PSO-optimized ELM model has higher recognition accuracy and generalization ability, improving accuracy by 2.8125 percentage points compared to the unoptimized ELM model, further enhancing the robustness and stability of fault diagnosis.

[0033] In summary, the LIMD-CMBE-PSO-ELM fusion model proposed in this solution outperforms existing signal processing and diagnostic methods in terms of accuracy, robustness of feature extraction, and generalization performance in the diagnosis of multiple types of faults in DC motors. The overall diagnostic accuracy reaches 99.6875%, providing an effective and advanced solution for intelligent fault diagnosis of DC motors.

[0034] It can be widely used in real-time monitoring and intelligent fault identification of DC motor operating status in industrial production. Attached Figure Description

[0035] Figure 1 A flowchart of a DC motor fault diagnosis method;

[0036] Figure 2 The diagram shows the LIMD decomposition of the normal state signal, where (a) is the time-domain diagram of the LIMD decomposition of the normal state signal, and (b) is the time-spectrum diagram of the LIMD decomposition of the normal state signal.

[0037] Figure 3 The LIMD decomposition diagrams of the rotor shaft bending signal are shown, including (a) the time-domain diagram of the LIMD decomposition of the rotor shaft bending signal and (b) the time-spectrum diagram of the LIMD decomposition of the rotor shaft bending signal.

[0038] Figure 4 The image shows the LIMD decomposition diagram of the blade fracture signal, where (a) is the time-domain diagram of the LIMD decomposition of the blade fracture signal, and (b) is the time-spectrum diagram of the LIMD decomposition of the blade fracture signal.

[0039] Figure 5 The diagram shows the LIMD decomposition of the bearing fault signal, including (a) the time-domain diagram of the LIMD decomposition of the bearing fault signal and (b) the time-spectrum diagram of the LIMD decomposition of the bearing fault signal.

[0040] Figure 6 The bearing fault signal LMD decomposition diagram is shown, where (a) is the time domain diagram of the bearing fault signal LMD decomposition, and (b) is the time spectrum diagram of the bearing fault signal LMD decomposition.

[0041] Figure 7 A schematic diagram of the classification results of the LMD-ELM method;

[0042] Figure 8 A schematic diagram of the classification results of the LMD-PSO-ELM method;

[0043] Figure 9 A schematic diagram of the classification results of the LIMD-ELM method;

[0044] Figure 10 A schematic diagram of the classification results of the LIMD-PSO-ELM method;

[0045] Figure 11 The fitness curve of the LMD-PSO-ELM method is shown.

[0046] Figure 12 The fitness curve of the LIMD-PSO-ELM method is shown. Detailed Implementation

[0047] To make the advantages and benefits of the technical solution provided by the present invention clearer, the technical solution provided by the present invention will now be described in further detail with reference to the accompanying drawings, specifically:

[0048] Implementation Method 1: This implementation method provides a DC motor fault diagnosis method based on improved local mean decomposition and composite multi-scale bubble entropy fusion, including:

[0049] The steps for collecting the sound signal of a DC motor during operation as the raw signal;

[0050] The process involves decomposing the original signal using an improved local mean decomposition method to obtain several multiplicative integrals, and then selecting effective feature components that are highly correlated with the original signal.

[0051] The steps include performing composite multi-scale bubble entropy calculation on the effective feature components and extracting multidimensional entropy feature vectors that characterize the signal complexity.

[0052] The step of inputting the multidimensional entropy feature vector into the particle swarm optimization extreme learning machine model for classification to obtain the fault diagnosis results of the DC motor.

[0053] An acoustic sensor is used to collect sound signals during the operation of the motor in real time. The distance between the acoustic sensor and the motor housing is 10cm, and the two sensors are arranged at 90° orthogonal.

[0054] An integral sliding window strategy is used to calculate the local mean between adjacent extreme points, and a high-smoothness local mean function is obtained by processing with a moving average filter to suppress mode mixing and baseline noise interference.

[0055] The method for screening effective feature components is as follows: calculate the correlation coefficient between each multiplicand and the original signal, and remove components with a correlation coefficient lower than 0.3.

[0056] The scale factor for the composite multi-scale bubble entropy calculation is 10, the embedding dimension is 2, the time delay is 1, and the similarity tolerance is 0.15 times the standard deviation of the effective feature components.

[0057] In the extreme learning machine model with particle swarm optimization, the particle swarm optimization algorithm has 30 particles, a maximum number of iterations of 100, and 100 hidden layer nodes.

[0058] Implementation Method Two: This implementation method further describes the technical solution provided above in detail, specifically:

[0059] The following is the content of the specific implementation method in the patent application document:

[0060] This solution provides a DC motor fault diagnosis method based on the fusion of improved Local Mean Decomposition (LIMD) and Composite Multiscale Bubble Entropy (CMBE). Specifically, the method includes the following steps:

[0061] First, acoustic sensors are used to collect the sound signals generated during the operation of the DC motor in real time. In this embodiment, the Siemens LMS data acquisition system is selected, along with two GPRAS acoustic sensors with a sensitivity of 50mV / Pa for data acquisition. The sensors are arranged 10cm away from the motor housing, orthogonally at 90° to avoid interference. The sampling frequency is set to 12.8kHz to obtain the original sound signals of the motor under normal operating conditions and various fault conditions, including rotor shaft bending, blade breakage, and bearing damage. Each signal group has a length of 50,000 sampling points.

[0062] Secondly, an improved Local Mean Decomposition (LIMD) is performed on the acquired audio signals. Specifically, an integral sliding window method is used to determine the local mean points of the original signal. A highly smooth local mean function is formed by calculating the integral average of the signal within the interval between every two adjacent extreme points. This method effectively reduces the endpoint effects, mode mixing, and baseline noise interference present in the traditional LIMD process, and obtains several multiplicative components (PF components) that are highly correlated with the original signal. Subsequently, the correlation coefficient between each PF component and the original signal is calculated, and redundant components with correlation coefficients lower than a preset threshold (set to 0.3 in this embodiment) are removed, retaining the effective feature components with strong correlation (such as PF1 and PF2 components) for further analysis.

[0063] Next, composite multi-scale bubble entropy (CMBE) feature extraction is performed on the effective feature components obtained in the above steps. Specifically, the effective feature components are first subjected to multi-scale coarsening, dividing the original time series into subsequences with different scale factors. Overlapping subsequences are generated using a sliding window to ensure that the coarsened signal retains sufficient details of the original signal. Then, the corresponding bubble entropy value is calculated at each scale, thereby constructing a multi-dimensional entropy feature vector that reflects the complexity of the signal (in this embodiment, the scale factor is set to 10, forming a 10-dimensional entropy feature vector). The entropy feature vector extracted by this method can effectively quantify the nonlinear complexity information of the signal at different scales, helping to improve the accuracy of fault state classification.

[0064] Finally, the obtained multidimensional entropy feature vector is input into an Extreme Learning Machine (ELM) diagnostic model based on Particle Swarm Optimization (PSO) for classification. Specifically, firstly, the input weights and biases of the hidden layer nodes of the ELM are randomly initialized. The parameters of the ELM model are then globally optimized using the PSO algorithm to obtain the optimal hidden layer parameters that match the distribution of fault feature data. Secondly, the output weights of the ELM model are quickly solved using the least squares method to establish a mapping relationship, thereby achieving accurate identification and diagnosis of the normal state and various fault states (such as rotor shaft bending, blade breakage, and bearing failure) of the DC motor. When using this scheme for diagnosis, a classification accuracy of 99.6875% was achieved on the experimental dataset, significantly higher than the traditional LMD-ELM and the unoptimized LIMD-ELM model, demonstrating obvious superiority and reliability.

[0065] The specific implementation methods for the above steps are further described in detail below:

[0066] (1) Specific implementation method of data acquisition

[0067] In this embodiment, the experimental subjects were three commonly used miniature DC motors: 555, 775, and 895. The motor operating states included normal operation, rotor shaft bending, blade breakage, and bearing failure. For each state, 400 sets of sound signals were collected, totaling 1600 sets of sample data. During data acquisition, the acoustic sensor sensitivity was set to 50mV / Pa, the sampling frequency was fixed at 12.8kHz, and each signal data set had a length of 50,000 points to ensure the data accurately reflected the motor's operating state.

[0068] (2) Detailed implementation of the improved Local Mean Decomposition (LIMD)

[0069] Using the acquired raw audio signal as input, a local extremum point detection algorithm (with an extremum tolerance threshold of ±5%) is first employed to extract local extrema. Then, the mean of an integral sliding window is calculated within each adjacent extremum point interval to obtain the local statistical features of the signal. Further, a smooth local mean function is obtained through moving average filtering. Based on this, the original signal is progressively decomposed to obtain multiple physically meaningful multiplicative factors (PFs). Subsequently, the correlation coefficient between each PF component and the original signal is calculated, and PF components with a correlation coefficient greater than 0.3 are selected as effective feature components for subsequent feature extraction.

[0070] (3) Specific Implementation Method for Complex Multiscale Bubble Entropy (CMBE) Feature Extraction

[0071] For the selected effective feature components, a scale factor of 10, an embedding dimension of 2, a time delay of 1, and a similarity tolerance of 0.15 times the standard deviation of the original signal were set for coarse-grained sequence processing. A sliding window was used to generate overlapping subsequences to ensure the integrity of signal feature details. At each scale factor, bubble entropy was used to perform complexity analysis on the coarse-grained subsequences, constructing a 10-dimensional entropy feature vector that comprehensively describes the signal complexity features, which served as the input features for the classifier.

[0072] (4) Specific implementation of Particle Swarm Optimization Extreme Learning Machine (PSO-ELM)

[0073] First, the entropy feature vector is input into the ELM model, and the input weight matrix and bias vector of the hidden layer nodes are randomly generated. Using the particle swarm optimization algorithm, with the root mean square error of the diagnostic model on the training set as the fitness function, the ELM hidden layer parameters are globally optimized by iteratively updating the particle positions and velocities. After sufficient iterations, the optimal input weight matrix and bias parameters are obtained. Finally, the output weight matrix is ​​determined based on the least squares method, thereby constructing a high-precision fault diagnosis model, ultimately achieving accurate identification and classification of different states of DC motors.

[0074] In summary, the technical solution provided in this embodiment effectively suppresses mode aliasing and noise interference through the improved Local Means Decomposition (LIMD) method, fully mines signal complexity information through the Composite Multiscale Bubble Entropy (CMBE) method, and achieves accurate optimization of the diagnostic model by combining Particle Swarm Optimized Extreme Learning Machine (PSO-ELM). Ultimately, it greatly improves the accuracy and robustness of DC motor fault diagnosis, fully realizes the technical effects proposed in this solution, and has complete feasibility.

[0075] Implementation Method 3: Combination Figure 1-12 This embodiment describes the technical solution provided above in further detail through specific examples. Specifically:

[0076] Local mean decomposition

[0077] Local Mean Decomposition (LMD) is an adaptive signal decoupling method. Its core objective is to decompose a non-stationary, multi-component complex signal into a series of physically meaningful single-component amplitude-frequency modulated (AM-FM) signals (integral components PF) and a trend term residual. Each PF component is composed of the product of an envelope function and a pure FM signal. The envelope function represents the instantaneous amplitude of the component, while the instantaneous frequency can be obtained by taking the first derivative of the phase function of the pure FM signal. The LMD algorithm extracts PF components step by step through an iterative selection process. The specific decomposition steps are as follows:

[0078] (1) Let the original sound signal be x(n), where n = 0, 1, ..., N, and N is the length of the signal. Find all local maxima and local minima of x(n). The position corresponding to the i-th maximum or minimum point is n. i Calculate n i and adjacent extreme points n i+1 The average value m i :

[0079]

[0080] All averages m i At the extreme point n i and n i+1 Extend the line between the given points using a straight line, and smooth the extended line using the moving average method to obtain the local mean function m. 11 (n), m 11 The first subscript of (n) refers to the first local mean function, the second subscript refers to the first iteration, and so on.

[0081] (2) Calculate n i and adjacent extreme points n i+1 Half the absolute value of the difference yields the local amplitude 'a'. i :

[0082]

[0083] All local amplitudes a i At the extreme point n i and n i+1 Extend the line between the given points using a straight line, and smooth the extended line using the moving average method to obtain the envelope estimation function a. 11 (n), a 11 The first subscript of (n) refers to the first envelope estimation function, the second subscript refers to the first iteration, and so on.

[0084] (3) Extract the local mean function m from the original sound signal x(n). 11 (n), thus obtaining the zero-mean function h 11 (n):

[0085] h 11 (n)=x(n)-m 11 (n) (3)

[0086] (4) For h 11 Demodulation is performed using (n), via h 11 (n) divided by the envelope estimation function a 11 (n) is implemented to obtain the demodulated function s. 11 (n):

[0087]

[0088] At this point, it is necessary to determine s. 11 (n) Whether it is a pure frequency modulation function (the amplitude of a pure frequency modulation function is always 1, and -1≤s) 11 (n)≤1. )Calculate s according to the method in step (2). 11 The envelope estimation function a of (n) 12 (n), if a 12 If (n) = 1, then it means s 11 (n) is an ideal pure frequency modulation function; if a 12 (n)≠1, then s needs to be... 11 Using (n) as the initial signal, repeat steps (1) to (4) until a pure frequency modulation function s is obtained. 1o If (n) is reached, and the iteration terminates, then:

[0089]

[0090] in

[0091]

[0092] (5) Multiply all the envelope estimation functions obtained in step (2) during the iteration process to obtain the envelope signal a1(n):

[0093]

[0094] (6) Multiply the envelope signal a1(n) generated in step (5) by the pure frequency modulation function s generated in step (4). 1o The product obtained is the first PF component of the LMD decomposition, denoted as PF1(n):

[0095] PF1(n)=a 1o (n)s 1o (n) (8)

[0096] (7) Finally, subtract the PF1(n) component from the original sound signal x(n) to obtain a new signal u1(n). Use this signal as the original signal and repeat steps (1) to (6) k times until u k Until (n) is a monotonic function:

[0097]

[0098] After multiple iterative decompositions, the original sound signal x(n) is finally decomposed into k PF components and a residual u. k The sum of (n) is in the form of:

[0099]

[0100] Improved Local Mean Decomposition

[0101] Although Local Means Decomposition (LMD) optimizes the symmetry of extreme points and the efficiency of mode separation based on the Empirical Mode Decomposition (EMD) framework, it is still essentially a data-driven adaptive decomposition method. It struggles to completely eliminate mode aliasing, leading to frequency band overlap and energy leakage in the decomposed components, thus affecting the accurate extraction of fault features and the reliability of diagnostic results. To overcome this bottleneck, this paper proposes a local integral mean optimization strategy. By reconstructing the computational flow of the local mean function and envelope estimation, it suppresses the interference of mode aliasing on signal decomposition. Specifically, it abandons the traditional local mean approach and introduces a method based on local integral means. The core steps are as follows:

[0102] (1) Find all the maxima and minima in the original sound signal x(n), and arrange the maxima in ascending order to form (n) k x k (n) k x k ), where n k It is the index of the extreme point, x k It is the signal amplitude corresponding to the extreme point.

[0103] (2) For each pair of adjacent extreme points (n) k x k ) and (n k +1, x k -1), calculate the mean of the local integrals within its interval. in It is the midpoint between adjacent extreme points. It is the integral mean of the signal between adjacent extreme points:

[0104]

[0105] (3) The moving average algorithm is used to calculate the mean of all local integrals. Smoothing is performed to obtain the local mean function m. 11 (n).

[0106] (4) The other steps are the same as steps (2) to (7) of the local mean decomposition.

[0107] The acquired sound signals from four states of a DC motor were decomposed using an improved local mean decomposition method, yielding five product function components. Correlation coefficients were then calculated for these five components to identify feature components strongly correlated with the original signal. Table 1 shows the correlation coefficient calculation results. Analysis indicates that the correlation coefficients of components PF1 and PF2 are both greater than 0.3, and these components contain the main features of the original signal. Components PF3-PF5 mainly contain environmental noise and high-frequency interference, which were removed to reduce feature redundancy. Finally, components PF1 and PF2 were selected as effective feature carriers to replace the original signal input into the Composite Multiscale Bubble Entropy (CMBE) model for feature extraction.

[0108] Table 1. Calculation results of correlation coefficients

[0109]

[0110] Extraction of composite multi-scale bubble entropy feature vector

[0111] Composite Multiscale Bubble Entropy (CMBE) is an improved multiscale entropy analysis method designed to address the problem of decreased entropy estimation accuracy in traditional multiscale bubble entropy (MBE) under long-scale factors due to the shortening of signal coarsening sequences.

[0112] The specific steps of compound multiscale entropy are as follows:

[0113] (1) Select PF1 and PF2 components as effective feature carriers to replace the original signal. The signal at this time is the signal after improved local mean decomposition, denoted as f(n), where n = 0, 1, ..., N, and N is the length of the signal. Perform coarse-grained calculation. For each given scale factor s (representing the time window length), divide the sequence after improved local mean decomposition into non-overlapping windows and calculate the mean of each segment:

[0114]

[0115] In the formula, y (s) (e) is the e-th coarse-grained subsequence with a scale factor of s, f(g) is the g-th sequence after the improved local mean decomposition, and e is from 1 to... It is the coarse-grained sequence number.

[0116] Because the sequence length is shortened to This leads to a decrease in entropy estimation accuracy. A composite processing method using a sliding window generates overlapping subsequences, with a step size Δt = 1 to generate s sets of overlapping subsequences, ensuring the total length of the coarse-grained sequence remains n-s+1. The mean of its subsequences is:

[0117]

[0118] (2) For each coarse-grained subsequence y (s) (e) Calculate the bubble entropy. Given the embedding dimension m and time delay u, construct the phase space vector:

[0119]

[0120] In the formula, F(n) is the phase space vector at point n, and {y(n), y(n+u), ..., y(n+(m-1)u)} are m consecutive samples extracted from the coarse-grained sequence y, n = 1, 2, ..., l-(m-1)u, where l is the subsequence length.

[0121] Sort each embedding vector F(n) in ascending order and record the number of element swaps j(n). Calculate the probability distribution of the number of swaps:

[0122]

[0123] In the formula, P(j) is the probability of the number of exchanges, and the time delay u takes the value of 1.

[0124] Calculate the Rényi entropy of order 2 based on the probability distribution:

[0125]

[0126] In the formula, H swaps It is Rényi entropy, p j Let be the probability of the j-th event, and lg be the natural logarithm.

[0127] By adjusting the embedding dimension m, the normalized bubble entropy value is calculated:

[0128]

[0129] In the formula, and represents the Rényi entropy values ​​calculated when the embedding dimensions are m+1 and m, respectively, and BE is the normalized bubble entropy value.

[0130] (3) Calculation of bubble entropy in composite multi-scale systems. For the scale factor s, the mean bubble entropy values ​​of all coarse-grained subsequences are aggregated:

[0131]

[0132] In the formula, CMBE(s) is the composite multiscale bubble entropy value under the scale factor s, n is the total length of the signal after the improved local mean decomposition, and y is the total length of the signal. (s) (e) is the e-th coarse-grained subsequence with a scale factor of s, BE(y (s)(e) is the bubble entropy value of the e-th coarse-grained subsequence, and n-s+1 is the total number of coarse-grained subsequences at scale s.

[0133] The main parameters affecting the accuracy of composite multiscale bubble entropy calculation are the scale factor s, embedding dimension m, time delay u, and similarity tolerance r. When extracting the feature vector of composite multiscale bubble entropy, the scale factor s is set to 10, the embedding dimension m is set to 2, the time delay u is set to 1, and the similarity tolerance r is set to 0.15 × σ, where σ is the standard deviation of the original signal. A multidimensional feature vector T representing the complexity of the acoustic signal is constructed based on the composite multiscale bubble entropy (CMBE). Its dimension is determined by the scale factor s = 10, i.e.: T = [CMBE1, CMBE2, ..., CMBE...]. 10 ].

[0134] PSO-ELM model

[0135] Extreme Learning Machine (ELM), proposed by Huang et al., is an efficient supervised learning framework based on a single-hidden layer feedforward neural network (SLFN). Its core innovation lies in randomly fixing the hidden layer parameters (input weight matrix W and bias vector B) and directly solving for the output weights using the least squares method, thereby significantly reducing the number of training parameters, improving learning speed, and enhancing the model's generalization ability. However, due to the random generation of W and B, they cannot adaptively match the characteristics of data distribution, which may lead to fluctuations in model performance and limited accuracy. To overcome these shortcomings, Particle Swarm Optimization (PSO) is introduced to globally optimize the hidden layer parameters of ELM, constructing a hybrid diagnostic model, PSO-ELM. PSO, based on swarm intelligence, collaboratively searches for individual historical optimal solutions and the global optimal solution of the group, possessing both multi-objective optimization capabilities and efficient convergence characteristics. By dynamically optimizing W and B through PSO, the feature mapping capability and diagnostic accuracy of ELM can be effectively improved, while maintaining its fast training advantage. The specific steps are as follows:

[0136] (1) The core innovation of Extreme Learning Machine (ELM) lies in the random fixation of hidden layer parameters and the interpretation and solution of output weights.

[0137] Hidden layer parameters are randomly fixed: Input weight matrix W∈R d×L With bias vector B∈R L After random initialization, the values ​​are fixed (d is the input feature dimension, and L is the number of hidden layer nodes).

[0138] Analytical solution for output weights: The output weights β∈R are directly calculated using the Moore-Penrose generalized inverse. L×z The mathematical expression is:

[0139]

[0140] In the formula, H∈R v×L Let Z be the hidden layer output matrix, with dimensions v×L, where v is the number of training samples and L is the number of hidden layer nodes; Z∈R v×z Let H be the target matrix with dimensions v×z, where z is the number of output classes. λ is the regularization coefficient, and E is the identity matrix with dimensions the same as H. T H is the same, It is the Moore-Penrose generalized inverse, used to calculate H. T H is the inverse matrix of the regularization term. T It is the transpose of H, with dimensions L×v.

[0141] (2) Particle Swarm Optimization (PSO) Fusion Strategy: To overcome the parameter sensitivity of ELM, the particle swarm optimization algorithm is introduced to construct a hybrid diagnostic model PSO-ELM. The specific steps are as follows:

[0142] First, the input weight matrix W in the hidden layer parameters is concatenated with the bias vector B to form the particle position vector:

[0143] P ii =[vec(W (ii) ),B (ii) ]∈R (d+1)L (20)

[0144] In the formula, P ii It is the position vector of the i-th particle, vec(W (ii) ) indicates that the input weight matrix W (ii) Perform column vectorization operation, B (ii) Let R represent the bias vector of the i-th particle. (d+1)L Let represent the (d+1)L-dimensional space over the real number field.

[0145] Then, the fitness function is designed, with the weighted root mean square error (RMSE) as the target fitness function.

[0146]

[0147] In the formula, v is the number of training samples; z is the number of output categories; tjjll is the ll-th true label of the jj-th sample; β mmll is the connection weight from the mm-th node in the hidden layer to the ll-th node in the output layer; g(·) is the hidden layer activation function (Sigmoid function); It is the weight vector from the input layer to the m-th hidden layer node and the input sample x. jj dot product; b mm It is the bias value of the mm-th hidden layer node.

[0148] (3) Perform particle update and convergence control.

[0149] First, perform a speed update:

[0150]

[0151] In the formula, v (t+1)ii is the velocity of the i-th particle at time t+1, w is the inertia weight (w = 0.7), c1 and c2 are learning factors (c1 and c2 are both 2), and random numbers r1 and r2 are randomly generated in the range [0,1]. p best,ii It is the optimal position of the i-th particle, g best It is the globally optimal position, p (t)ii It is the position of the i-th particle at time t.

[0152] Then, perform a location update:

[0153]

[0154] In the formula, p (t+1)ii p is the position of the i-th particle at time t+1. (t)ii It is the position of the i-th particle at time t.

[0155] Finally, the termination condition is met when RMSE < 10. -3 Or reach the maximum number of iterations T max When the value is 100, the global optimal solution P* is output, which is equivalent to outputting the global optimal solution g. best , * indicates the optimal solution.

[0156] (4) Substitute the optimized {W*,B*} into the ELM framework, and calculate the output weight β* to form the PSO-ELM classifier.

[0157] The flowchart of the PSO-ELM model is as follows: Figure 1 As shown.

[0158] Experimental Results and Analysis

[0159] The acoustic signals of normal state, rotor shaft bending, blade fracture, and bearing failure were decomposed using an improved Local Means Decomposition (LIMD) method. Five physically meaningful product function (PF) components were extracted for each fault state. The decomposition results (including the time-domain PF components and the time spectrum) are shown below. Figures 2 to 5 As shown.

[0160] To verify the optimization effect of the improved Local Integral Mean Decomposition (LIMD), a bearing fault signal is used as an example to compare and analyze the decomposition performance of LIMD and the classic LMD. Figure 6 The time-domain plot and time spectrum of bearing fault signal decomposition based on LMD are shown, combined with Figure 5 Based on the LIMD decomposition results, the following conclusions can be drawn: Comparing the time-domain and time-spectrum diagrams of the bearing fault signal decomposition by LIMD and LMD, it can be found that LIMD decomposition is more uniform in the extraction of high-frequency components of PF1 and PF2, reducing mode aliasing problems. At the same time, LIMD decomposition is less sensitive to noise and the decomposition results are more stable. In contrast, although LMD decomposition can decompose the signal step-by-step from high-frequency to low-frequency components, it suffers from mode aliasing problems in the extraction of high-frequency components and is more sensitive to noise. Therefore, LIMD is superior to LMD in signal decomposition performance.

[0161] The experiment collected acoustic signals from four types of DC motors: normal state, rotor shaft bending, blade breakage, and bearing failure. Each state contained 400 samples (sampling frequency of 12800Hz, duration of 2000s, and data length of 50000 bytes per sample), totaling 1600 sets of data. These were divided into a training set (1280 sets) and a test set (320 sets) at a 4:1 ratio, with labels coded as c = [1,2,3,4]. The Extreme Learning Machine (ELM) configuration was as follows: hidden layer node count L = 100, activation function sigmoid; input layer dimension d = 10, output layer dimension z = 4. Particle Swarm Optimization (PSO) parameter settings were: particle count 30, maximum number of iterations T. max The inertia weight w is 0.7, the learning factors c1 and c2 are both 2, and the random numbers r1 and r2 are randomly generated in the range [0,1]. The weight coefficient α = 0.7. First, the signal is decomposed using both the traditional LMD decomposition method and the improved LIMD decomposition method. Then, the Pearson correlation coefficient between each product function (PF) component and the original signal is calculated, and components with correlation coefficients below 0.3 are removed. After reconstructing the retained PF components, the composite multi-scale bubble entropy value of the reconstructed signal is calculated, and the recognition of four types of acoustic signal data is achieved using an Extreme Learning Machine (ELM) classifier and an Extreme Learning Machine (ELM) optimized by Particle Swarm Optimization (PSO). The classification results based on the LMD-ELM method, LMD-PSO-ELM method, LIMD-ELM method, and LIMD-PSO-ELM method are as follows: Figure 7 , Figure 8 , Figure 9 , Figure 10As shown in the figures, by analyzing these four figures, we can see that the LMD-ELM method has a test set accuracy of 50.625%. Feature noise causes the acoustic signals of a DC motor in normal condition, a DC motor with blade breakage, and a DC motor with bearing failure to be confused with each other. The LMD-PSO-ELM method improves the accuracy to 79.0625%, which sufficiently demonstrates that PSO optimization can significantly suppress parameter sensitivity and improve the accuracy of fault diagnosis. The LIMD-ELM method has a test set accuracy of 96.875%, indicating that LIMD decomposition enhances feature separability and has a stronger ability to improve fault diagnosis accuracy than PSO optimization. The LIMD-PSO-ELM method has the highest accuracy at 99.6875%, which verifies that the PSO algorithm and the LIMD decomposition algorithm can be synergistically optimized to improve the accuracy of fault diagnosis.

[0162] Figure 11 , Figure 12 These are the fitness curves for particle swarm optimization in the LMD-PSO-ELM method and the LIMD-PSO-ELM method, respectively. By analyzing these two graphs, we can see that... Figure 11 In the first 20 iterations, the accuracy rapidly increased from 0.5 to 0.7, indicating that the PSO algorithm quickly explored a relatively optimal solution region in the early stages. From 20 to 70 iterations, the accuracy gradually increased to around 0.75, reflecting the dynamic balance between global search and local optimization in the particle swarm optimization, possibly due to feature interference caused by mode aliasing. From 70 to 100 iterations, initial diagnostics were performed. Experimental results show that the accuracy of this method increased by 4% in a period of 320 test samples, after which the accuracy remained constant, indicating that the PSO algorithm gradually converged and eventually stabilized at the optimal solution. The low accuracy may be due to insufficient particle diversity or excessively high inertial weights, affecting the fine-grained search capability in later stages. Figure 12 In the initial iterations, the accuracy started at around 97%, validating that the improved LIMD method directly provides highly discriminative features through noise suppression and mode separation. In the early iterations, the accuracy improved relatively quickly, from 97% to 99%. In the middle iterations, the accuracy remained around 99.5%, indicating that the PSO algorithm was approaching the global optimum. At the 90th iteration, the accuracy improved by approximately 0.3%, indicating that the PSO algorithm had found the global optimum, reaching its highest accuracy of 96.875%. The LIMD-PSO-ELM method only requires 10 iterations to approach peak performance, demonstrating its high efficiency. Figure 11 The initial accuracy of 96.6875% was significantly higher than Figure 10The 50.625% result indicates that the LIMD decomposition algorithm provides a higher signal-to-noise ratio feature input through integral sliding window and noise suppression.

[0163] The above description of several specific embodiments further details the technical solution provided by the present invention in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the above-described specific embodiments are not intended to limit the present invention. Any reasonable modifications and improvements to the present invention, combinations of embodiments, and equivalent substitutions based on the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A DC motor fault diagnosis method based on improved local mean decomposition and composite multi-scale bubble entropy fusion, characterized in that, include: The steps for collecting the sound signal of a DC motor during operation as the raw signal; The process involves decomposing the original signal using an improved local mean decomposition method to obtain several multiplicative integrals, and then selecting effective feature components that are highly correlated with the original signal. The steps include performing composite multi-scale bubble entropy calculation on the effective feature components and extracting multi-dimensional entropy feature vectors that characterize signal complexity. The step of inputting the multidimensional entropy feature vector into the particle swarm optimization extreme learning machine model for classification to obtain the fault diagnosis result of the DC motor. Specifically: An integral sliding window strategy is used to calculate the local mean between adjacent extreme points, and a high-smoothness local mean function is obtained through moving average filtering to suppress mode mixing and baseline noise interference; Find the original sound signal x ( n All the maximum and minimum points in the extrema are arranged in ascending order to form a... ,in n k It is the index of the extreme point. x k It is the signal amplitude corresponding to the extreme point; For each pair of adjacent extreme points and Calculate the mean of the local integrals within its interval. ,in It is the midpoint between adjacent extreme points. It is the integral mean of the signal between adjacent extreme points: All local integral means are smoothed using a sliding average algorithm to obtain a local mean function m 11 ( n ).

2. The DC motor fault diagnosis method based on improved local mean decomposition and composite multi-scale bubble entropy fusion according to claim 1, characterized in that, An acoustic sensor is used to collect sound signals during the operation of the motor in real time. The distance between the acoustic sensor and the motor housing is 10cm, and the two sensors are arranged at 90° orthogonal.

3. The DC motor fault diagnosis method based on improved local mean decomposition and composite multi-scale bubble entropy fusion according to claim 1, characterized in that, The method for screening effective feature components is as follows: calculate the correlation coefficient between each multiplicand and the original signal, and remove components with a correlation coefficient lower than 0.

3.

4. The DC motor fault diagnosis method based on improved local mean decomposition and composite multi-scale bubble entropy fusion as described in claim 1, characterized in that, The scale factor for the composite multi-scale bubble entropy calculation is 10, the embedding dimension is 2, the time delay is 1, and the similarity tolerance is 0.15 times the standard deviation of the effective feature components.

5. The DC motor fault diagnosis method based on improved local mean decomposition and composite multi-scale bubble entropy fusion according to claim 1, characterized in that, In the extreme learning machine model with particle swarm optimization, the particle swarm optimization algorithm has 30 particles, a maximum number of iterations of 100, and 100 hidden layer nodes.

6. A DC motor fault diagnosis device based on improved local mean decomposition and composite multi-scale bubble entropy fusion, characterized in that, include: A module that collects sound signals from a DC motor during operation as the raw signal; The original signal is decomposed using an improved local mean decomposition method to obtain several multiplicative integrals, and modules with effective feature components that are highly correlated with the original signal are selected. A module for calculating composite multi-scale bubble entropy on the effective feature components and extracting multi-dimensional entropy feature vectors that characterize signal complexity; The module inputs the multidimensional entropy feature vector into the particle swarm optimization extreme learning machine model for classification to obtain the fault diagnosis results of the DC motor. Specifically: An integral sliding window strategy is used to calculate the local mean between adjacent extreme points, and a high-smoothness local mean function is obtained through moving average filtering to suppress mode mixing and baseline noise interference; Find the original sound signal x ( n All the maximum and minimum points in the extrema are arranged in ascending order to form a... ,in n k It is the index of the extreme point. x k It is the signal amplitude corresponding to the extreme point; For each pair of adjacent extrema points and the local integral mean value within their interval is calculated where is the midpoint between the adjacent extrema points, is the integral mean value of the signal between the adjacent extrema points: All local integral means are smoothed using a sliding average algorithm to obtain a local mean function m 11 ( n ).

7. A computer storage medium for storing computing programs, characterized in that, When the computer program is read by the computer, the computer executes the method of claim 1.

8. A computer comprising a processor and a storage medium, characterized in that When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 1.

9. Computer program product as computer program, characterized in that When the computer program is executed, it implements the method of claim 1.

Images