Feature signal enhancement method for hot die forging press component wear state recognition

Abstract

This invention belongs to the technical field of wear condition identification for hot forging press components, specifically relating to a feature signal enhancement method for wear condition identification of hot forging press components, including: S1, enhancing the feature signal x of inferior vibration signal x poor The repair signal x is obtained through processing. re S2. Construct a closed-open self-complementary top-hat morphological filter FCO-NSTH to enhance the characteristic frequency amplitude of the vibration signal to be enhanced; S3. Adaptively select the optimal structural element parameters of the closed-open self-complementary top-hat morphological filter using a particle swarm optimization algorithm; S4. Use the closed-open self-complementary top-hat morphological filter FCO-NSTH with adaptively selected optimal structural element parameters to enhance the repair signal x. re The characteristic frequency amplitude is enhanced to obtain an enhanced vibration signal; S5. The enhanced vibration signal obtained in S4 is input into the intelligent model to identify and predict the wear state of hot forging press components. This method can accurately identify the wear state of hot forging press components by combining with the intelligent model.

Description

Technical Field

[0001] This invention belongs to the field of wear condition identification technology for hot forging press components, and particularly relates to a feature signal enhancement method for wear condition identification of hot forging press components. Background Technology

[0002] The clutch and brake are two crucial components of a hot forging press. Problems with either the clutch or brake can lead to the following consequences: 1. Continuous Forging: Clutch or brake issues can cause the press to malfunction, resulting in continuous forging. This continuous forging failure is very serious, causing minor damage such as scrapped forgings and mold damage, and potentially leading to equipment damage and personal injury. 2. Overload and Stalling: Clutch or brake problems can also cause the press to overload during operation, resulting in stalling. This overload is difficult to handle; if existing methods fail, it may be necessary to cut the mold or loosen the preload nuts on the press body. 3. Excessive Machine Vibration: Clutch or brake problems can also cause excessive vibration during press operation. Excessive machine vibration can lead to many problems, such as a harsh working environment, reduced press lifespan, and decreased forging precision.

[0003] Wear is a significant factor among the many factors that cause clutch and brake failures. Therefore, identifying wear on components of a hot forging press (i.e., clutches and brakes) is crucial. Currently, a common method for identifying the wear condition of hot forging press components (i.e., clutches and brakes) involves sampling vibration data of the components during operation and inputting it into an intelligent algorithm model for identification, detection, and prediction.

[0004] However, due to the relatively strong vibrations during actual vibration signal acquisition, the acquired signals are easily affected by random factors in the surrounding environment. Furthermore, due to performance degradation or malfunctions of the data acquisition equipment, the acquired vibration signals may contain low-quality data such as data pulses or missing data. Figure 1 The time-domain waveforms of the acquired poor-quality vibration signals are shown, among which, Figure 1 (a) indicates missing data. Figure 1 (b) Data pulses. Poor-quality data cannot accurately reflect the inherent operating patterns of the equipment, destroys the integrity and accuracy of the vibration signal, and weakens the extraction of valuable information. Directly using these poor-quality data will lead to misjudgments of the equipment's operating status. Furthermore, in order to better utilize continuous and complete periodic vibration signals for research on equipment operating status identification methods, these data should not be arbitrarily discarded.

[0005] Although preprocessing the sampled low-quality vibration data can yield higher-quality vibration signals, without sufficient prior knowledge, manually extracting statistical features such as time-domain, frequency-domain, and time-frequency-domain features from the vibration signals cannot guarantee the comprehensiveness and effectiveness of the extracted features. This leads to a decrease in the recognition ability of traditional machine learning methods such as support vector machines and random forests.

[0006] Deep learning methods can adaptively extract deep features from vibration signals without requiring prior knowledge, and have been widely used in equipment fault diagnosis in recent years. However, the features learned by deep learning methods from one-dimensional vibration signals have poor interpretability, making it difficult to provide a physical explanation for the features learned by the deep learning model from one-dimensional vibration signals. This is especially true when the collected vibration signals contain vibration features from other components, resulting in impure vibration signal components. Regardless of the performance of the condition monitoring model built based on deep learning methods in identifying the equipment's operating state, it cannot be guaranteed that the features learned by the model are related to the operating state of the equipment under study. Therefore, if the vibration characteristics of equipment operation can be highlighted or enhanced, while the influence of vibrations from components unrelated to the equipment under study can be weakened or eliminated, the performance and quality of equipment operating state identification based on deep learning will be improved.

[0007] Therefore, how to accurately identify the wear condition of hot forging press components through intelligent models has become an urgent problem to be solved. Summary of the Invention

[0008] To address the shortcomings of the existing technology, this invention provides a feature signal enhancement method for identifying the wear condition of hot forging press components, which can accurately identify the wear condition of hot forging press components by combining with an intelligent model.

[0009] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0010] A feature signal enhancement method for identifying wear conditions of components in hot forging presses includes the following steps:

[0011] S1. Inferior vibration signal x for hot forging press components poor Sampling, and the sampled poor-quality vibration signal x poor Perform defective data detection and repair processing to obtain the repair signal x. re ;

[0012] S2. Construct a closed-open self-complementary top-hat morphological filter FCO-NSTH to enhance the characteristic frequency amplitude of the vibration signal to be enhanced; the definition of the closed-open self-complementary top-hat morphological filter FCO-NSTH is as follows:

[0013]

[0014] In the formula, x enhance (l) represents the enhanced vibration signal; FCO-NSTH represents the closed-open self-complementary top cap morphological filter. This indicates that an opening operation is performed after a closing operation is performed on the repair signal; This indicates that the opening operation is performed on the repair signal; (x re ·g)(l) represents performing a closing operation on the repair signal; L represents the sampling length of the poor-quality vibration data;

[0015] S3. The optimal structural element parameters of the closed-open self-complementary top-hat morphological filter are adaptively selected through the particle swarm optimization algorithm.

[0016] S4. Use the FCO-NSTH closed-open self-complementary top-hat morphological filter with adaptive selection of optimal structuring element parameters to repair the signal x. re The characteristic frequency amplitude is enhanced to obtain an enhanced vibration signal;

[0017] S5. Input the enhanced vibration signal obtained in S4 into the intelligent model to identify and predict the wear state of hot forging press components.

[0018] Compared with the prior art, the present invention has the following advantages:

[0019] This method utilizes a particle swarm optimization-optimized morphological filter to enhance the amplitude of characteristic frequencies of equipment, highlighting the amplitude of characteristic frequencies related to the equipment's operating state. Morphological filtering is a signal analysis method belonging to nonlinear filtering. This method uses structuring elements to perform translation matching and local correction on the signal from front to back, which can enhance the morphological features of the signal. Furthermore, morphological filtering only involves simple operations such as addition, subtraction, and finding extrema, making it computationally simple and efficient. The closed-open self-complementary top-hat morphological filter FCO-NSTH constructed in this method, compared with conventional morphological filters, not only increases the amplitude of the output signal but also simultaneously extracts positive and negative pulses and enhances detailed features, exhibiting a significant advantage in characteristic amplitude enhancement. It can effectively enhance the amplitude of characteristic frequencies of vibration signals from hot forging press components.

[0020] The restored signal x after enhancement using the closed-open self-complementary top-hat morphological filter FCO-NSTH re This can enhance the vibration characteristics of hot forging press components, eliminate / weaken the impact of vibrations from parts unrelated to the hot forging press components, thereby improving the performance and quality of equipment operation status recognition based on deep learning.

[0021] In summary, this method, combined with an intelligent model, can accurately identify the wear condition of components in a hot forging press.

[0022] Preferably, in S2, the constructed FCO-NSTH uses a cosine-type structuring element for corresponding processing; the expression of the cosine-type structuring element is:

[0023]

[0024] In the formula, l se L is the length of the cosine-type structuring element, used to adjust the period of the cosine function. se Less than the length L of the repair signal se ang se For the angle of the cosine-type structural element, ang se ∈0,...,π, used to adjust the height of the cosine function.

[0025] This setup, where the shape of the cosine-type structural element is similar to the shape of the vibration signal, ensures effective processing of the vibration signal.

[0026] Preferably, in S3, the structural element parameters include length and angle.

[0027] Preferably, in S3, when using the particle swarm optimization algorithm, the number of iterations and the population size are both set to 100, the acceleration factors c1 and c2 are set to 2, the weight coefficient is set to 1, the length optimization range is [2, 50], and the angle optimization range is [0, 180°].

[0028] This setting ensures the validity of the obtained structural element parameters.

[0029] Preferably, when using the particle swarm optimization algorithm, the envelope entropy of the vibration signal enhanced by FCO-NSTH is used as the fitness function to select the optimal combination of structural element parameters.

[0030] Envelope entropy reflects the periodic transient pulse sequence in a signal. The smaller the envelope entropy, the more pronounced the signal's impulse characteristics are after morphological filtering. This setting ensures the validity of the obtained structural element parameters.

[0031] Preferably, the formula for calculating the envelope entropy is as follows:

[0032]

[0033] In the formula, E evelope Represents the envelope entropy; a evelope (l) represents the characteristic amplitude enhancement signal x enhance(l) (l=1,2,...,L) is the envelope signal sequence obtained after demodulation by Hilbert; H[·] represents the Hilbert transform; κ l It is a evelope The normalized form of (l).

[0034] Preferably, in S1, for inferior vibration signals x poor The process of detecting substandard data includes:

[0035] ① Input poor-quality vibration signal x poor ∈R L Where L represents the sampling length of the poor-quality vibration data;

[0036] ②Let i = 1;

[0037] ③ Calculate the i-th vibration data point p i The k-th distance neighbor NBR k (p i );

[0038] ④ Calculate vibration data points o∈NBR k (p i ) to data point p i The k-th reachable distance RD k (o,p i );

[0039] ⑤ Calculate the i-th vibration data point p i Locally achievable density (LRD) k (p i );

[0040] ⑥ Calculate the i-th vibration data point p i The k-th local outlier LOF k (p i );

[0041] ⑦ If i ≤ L, then let i = i + 1 and repeat steps ③-⑥; otherwise, proceed to step ⑧.

[0042] ⑧ Find the largest n in step ⑦ poor The position index number corresponding to the vibration data point of each LOF value is put into the set PosIndex = {pos_index} i |i=1,2,…,n poor}middle;

[0043] 9. Based on pos_index i ∈PosIndex, the data range [pos_index] i -Th zero pos_index i +Thzero The data within ] is set to zero, where Th zero This indicates the threshold value set to zero.

[0044] ⑩ The signal obtained after the defective data detection is denoted as the defective data detection signal x. LOF .

[0045] This approach considers both local and global attributes. Poor-quality vibration data is determined based on the density of vibration data points relative to their neighborhood. When data points with varying densities exist within the vibration data, the local outlier factor algorithm exhibits excellent detection performance. This ensures that the obtained poor-quality detection signal x is... LOF The effectiveness.

[0046] Preferably, the value of k is 1000, and n poor The value is 30.

[0047] Preferably, in S1, the repair process includes:

[0048] First, using a sparse transformation matrix φ∈R L×L The inferior detection signal x LOF ∈R L Sparse representation is θ∈R L As shown in the following formula: x LOF =φ·θ; where θ is the sparse representation coefficient;

[0049] Then, using the observation matrix The inferior detection signal x LOF Compressed to a low-dimensional space, as shown in the following equation: y observe =ψ·x LOF In the formula, For the observed signal, L observe Indicates the data length of the observed signal;

[0050] Then, x LOF =φ·θ Substitute into y observe =ψ·x LOF , get y observe =ψ·φ·θ=SM·θ, and solve for the sparse representation coefficient θ; where, For the perception matrix;

[0051] Finally, substitute the obtained sparse representation coefficients θ into the following equation: x re =φ·θ; This yields the repair signal x. re ∈R L .

[0052] If a signal possesses compressibility and sparsity, or exhibits sparsity after transformation by a sparse representation matrix, it can be projected onto a low-dimensional space using an observation matrix independent of the sparse representation matrix, thus achieving signal reconstruction. This allows for the rapid acquisition of an effective repair signal x. re .

[0053] Preferably, for inferior detection signals x LOF When performing sparse representation, the discrete cosine transform matrix is ​​used as the sparse representation matrix φ. The element in the i-th row and j-th column of this matrix is ​​calculated by the following formula:

[0054]

[0055] With this setup, the construction of the Discrete Cosine Transform is simple and computationally efficient. Only simple matrix multiplication operations are needed to obtain the sparse representation coefficients of the signal.

[0056] Preferably, the process of constructing the observation matrix includes: after detecting poor-quality vibration data, the location numbers of the vibration data to be set to zero are placed into the index set PosIndex = {pos_index}. i |i=1,2,…,n poor In the set, the data length is LL. observe Based on the position index numbers in the index set PosIndex, the identity matrix I∈R is... L×L Delete the corresponding row in the matrix to obtain the observation matrix. Attached Figure Description

[0057] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will now be described in further detail with reference to the accompanying drawings, wherein:

[0058] Figure 1 This is a schematic diagram of the time-domain waveform of the poor-quality vibration signal collected in the background technology;

[0059] Figure 2 This is a flowchart of the method;

[0060] Figure 3 This is a flowchart of the substandard data detection process in Example 1;

[0061] Figure 4 The flowchart is shown in Example 1, illustrating the improved sparsity adaptive matching pursuit algorithm.

[0062] Figure 5 These are schematic diagrams of cosine-type structural elements of different shapes in Example 1;

[0063] Figure 6This is a flowchart illustrating the adaptive selection of structure parameters based on the particle swarm optimization algorithm in Example 1.

[0064] Figure 7 This is a schematic diagram comparing the feature enhancement effects of structural elements of different shapes in Example 1;

[0065] Figure 8 The following are time-domain waveforms and their spectra of some vibration signals collected during the processing in Example 2;

[0066] Figure 9 This is a schematic diagram of the detection of substandard data in Example 2;

[0067] Figure 10 This is a schematic diagram of the repair signal calculated by this method in Example 2;

[0068] Figure 11 A comparison diagram of the envelope spectra of the signal repaired based on the MSAMP method in Example 2, the normal signal, and the inferior signal;

[0069] Figure 12 The envelope spectrum of the signal repaired based on different reconstruction algorithms in Example 2;

[0070] Figure 13 This is a comparison chart of quantitative indicators between different repaired signals and normal signals in Example 2;

[0071] Figure 14 This is a schematic diagram illustrating the Manhattan distance between the signals repaired using different methods and the normal signal, as well as the different repaired signals, in Example 2.

[0072] Figure 15 This is a schematic diagram illustrating the influence of the nearest neighbor number k on the repair result in Example 2;

[0073] Figure 16 n is the number of defective points in Example 2. poor Schematic diagram illustrating the impact on the repair results;

[0074] Figure 17 In Example 2, the threshold Th is set to zero. zero The diagram illustrates the impact of the repair results;

[0075] Figure 18 This is a simulation signal diagram with a signal-to-noise ratio of 5dB in Example 2;

[0076] Figure 19 This is a schematic diagram of the envelope spectrum with a signal-to-noise ratio of 5dB in Example 2;

[0077] Figure 20 This is the simulated signal with a signal-to-noise ratio of 10dB in Example 2;

[0078] Figure 21 This is a schematic diagram of the envelope spectrum with a signal-to-noise ratio of 10dB in Example 2;

[0079] Figure 22 This is a schematic diagram illustrating the repair effects of different repair methods at different signal-to-noise ratios in Example 2;

[0080] Figure 23 This is a schematic diagram of the amplitude enhancement of the measured signal characteristic frequency in Example 2. Detailed Implementation

[0081] The following detailed explanation illustrates the specific implementation methods:

[0082] Example 1

[0083] like Figure 2 As shown in the figure, this embodiment discloses a feature signal enhancement method for identifying the wear condition of components in a hot forging press, including the following steps:

[0084] S1. Inferior vibration signal x for hot forging press components poor Sampling, and the sampled poor-quality vibration signal x poor Perform defective data detection and repair processing to obtain the repair signal x. re .

[0085] In specific implementation, such as Figure 3 As shown, for inferior vibration signal x poor The process of detecting substandard data includes:

[0086] ① Input poor-quality vibration signal x poor ∈R L Where L represents the sampling length of the poor-quality vibration data;

[0087] ②Let i = 1;

[0088] ③ Calculate the i-th vibration data point p i The k-th distance neighbor NBR k (p i );

[0089] ④ Calculate vibration data points o∈NBR k (p i ) to data point p i The k-th reachable distance RD k (o,p i );

[0090] ⑤ Calculate the i-th vibration data point p i Locally achievable density (LRD) k (p i );

[0091] ⑥ Calculate the i-th vibration data point p i The k-th local outlier LOF k (p i );

[0092] ⑦ If i ≤ L, then let i = i + 1 and repeat steps ③-⑥; otherwise, proceed to step ⑧.

[0093] ⑧ Find the largest n in step ⑦ poor The position index number corresponding to the vibration data point of each LOF value is put into the set PosIndex = {pos_index} i |i=1,2,…,n poor}middle;

[0094] 9. Based on pos_index i ∈PosIndex, the data range [pos_index] i -Th zero pos_index i +Th zero The data within ] is set to zero, where Th zero This indicates the threshold value set to zero.

[0095] ⑩ The signal obtained after the defective data detection is denoted as the defective data detection signal x. LOF .

[0096] Where k is 1000, n poor The value is 30.

[0097] This approach considers both local and global attributes. Poor-quality vibration data is determined based on the density of vibration data points relative to their neighborhood. When data points with varying densities exist within the vibration data, the local outlier factor algorithm exhibits excellent detection performance. This ensures that the obtained poor-quality detection signal x is... LOF The effectiveness.

[0098] The aforementioned inferior signal detection algorithm is based on the local outlier factor algorithm. The basic idea of ​​the local outlier factor algorithm is to first calculate the local reachability density of each data point based on the density of the surrounding area, and then use the local reachability density to calculate the local outlier factor value of each data point. The larger the factor value, the greater the degree of outlier the data point is, and the more likely it is to be an inferior point.

[0099] The repair process includes:

[0100] First, using a sparse transformation matrix φ∈R L×L The inferior detection signal x LOF ∈R L Sparse representation is θ∈R LAs shown in the following formula: x LOF =φ·θ; where θ is a sparse representation coefficient, where the value at most positions within this coefficient is zero or close to zero, and the sparsity SPA is used to represent the number of non-zero values ​​within this coefficient. In specific implementation, for the defective detection signal x... LOF When performing sparse representation, the discrete cosine transform matrix is ​​used as the sparse representation matrix φ. The element in the i-th row and j-th column of this matrix is ​​calculated by the following formula:

[0101]

[0102] In actual vibration signals, the signals are not directly sparse; they need to be sparsified using a sparse representation matrix. Based on compressed sensing theory, when signal x... LOF In a sparse representation matrix, only a few values ​​are much greater than zero, while the vast majority of the data are close to zero. Therefore, the signal x can be considered... LOF It possesses compressibility or sparsity. The Discrete Cosine Transform (DCT) is simple to construct and computationally efficient; it only requires simple matrix multiplication to obtain the sparse representation coefficients of the signal. Therefore, this method selects the DCT matrix as the sparse representation matrix.

[0103] Then, using the observation matrix The inferior detection signal x LOF Compressed to a low-dimensional space, as shown in the following equation: y observe =ψ·x LOF In the formula, For the observed signal, L observe Indicates the data length of the observed signal.

[0104] In practice, the construction process of the observation matrix includes: after detecting poor-quality vibration data, the location numbers of the vibration data to be set to zero are placed into the index set PosIndex = {pos_index}. i |i=1,2,…,n poor In the set, the data length is LL. observe Based on the position index numbers in the index set PosIndex, the identity matrix I∈R is... L×L Delete the corresponding row in the matrix to obtain the observation matrix.

[0105] Then, x LOF =φ·θ Substitute into y observe =ψ·x LOF , get y observe =ψ·φ·θ=SM·θ, and solve for the sparse representation coefficient θ; where, This refers to the Sensing Matrix (SM).

[0106] Sensing matrix SM and observed signal y observe All the data has been obtained. Now, it's necessary to solve for the sparse representation coefficients θ. Compressed sensing technology often uses greedy algorithms such as Orthogonal Matching Pursuit (OMP) and Sparsity Adaptive Matching Pursuit (SAMP) to solve for these coefficients. The key to greedy algorithms is finding the best-matching atom (i.e., the sparse representation coefficient θ), then iteratively increasing the number of atoms until a complete support set is found. The atoms in the support set are then linearly combined to solve for the sparse solution of the signal. These algorithms mainly include two basic steps: atom selection and residual update. Residual update first calculates the correlation between the current residual and each column of the sensing matrix, then subtracts this correlation value from the current residual to obtain the updated residual. The correlation is typically represented by the absolute value of the inner product of the sensing matrix and the residual. Atom selection refers to calculating the atom SM in each column of the sensing matrix SM. j The absolute value of the inner product with the residual e is used to select the atoms that best match the residual to form a support set. Then, the atoms in the support set are linearly combined to solve for the sparse representation coefficients.

[0107] The SAMP algorithm, when the true sparsity SPA is unknown, incorporates a backtracking approach, adjusting the size of the support set by varying the value of the iterative residual. When the residual satisfies a condition, the iteration ends, and the final iteration step size is considered the estimated true sparsity SPAE (also representing the support set size), thus achieving reconstruction. However, this algorithm suffers from a fixed step size problem. If the step size is much smaller than the true sparsity SPA, a large number of iterations are required; if the step size is large, the estimated sparsity SPAE will differ significantly from the true SPA, leading to substantial signal reconstruction errors. To address these issues, this paper proposes an improved Sparsity Adaptive Matching Pursuit (MSAMP) algorithm. First, it calculates an initial sparsity estimate SPAE0 using a sparsity estimation method, serving as the initial step size, i.e., the initial support set size. Then, combining regularization and variable step size, it selects the optimal atoms through two thresholding steps to obtain a relatively accurate sparsity estimate SPAE. Finally, it achieves accurate signal reconstruction based on the obtained optimal support set.

[0108] like Figure 4 As shown, the specific steps of the algorithm implementation are as follows:

[0109] ① Initialization. Initial residual e0 = y observeIteration count iter=1, index set Pre-selection Candidate set Support set length L SupS =SPAE0.

[0110] ② Threshold selection. Calculate the threshold SM of each atom in the sensing matrix SM. j (SM j (representing the j-th column of the perception matrix) and the residual e iter-1 The inner product of the sets u = {u j |u j =| <SM j ,e iter-1 The expression `>|,j=1,2,…,L}` represents selecting values ​​greater than ρ1ε1 ​​from u and placing them into the set u. InVS ={u invs |u invs In the set ∈u, invs∈InVS, where the index set InVS represents the position index number corresponding to the value that satisfies the condition; ||u|| represents the 2-norm of set u, and ρ1 is an empirical value, which is taken as 3.9 here based on the vibration signal of this method.

[0111] ③ Regularization. From the inner product set u InVS Select the condition that satisfies {|u c |≤2|u f |,u c ,u f ∈u InVS The inner product of} is given, and its position index is placed in the preselection set PreSS. If the inner product set u InVS No condition is met {|u c |≤2|u f |,u c ,u f ∈u InVS If the inner product of} is given, then select from set u a set satisfying the condition {|u}. c |≤2|u f |,u c ,u f The inner product of ∈u} and its position index number are placed in the preselection set PreSS.

[0112] ④ Select the corresponding column SM in the perception matrix SM according to the index number in the preselection set PreSS. j The matrix consisting of columns j∈PreSS is defined as SM PreSS .

[0113] ⑤ Update the candidate set CaS. Solve the following equation using the least squares method:

[0114] ;

[0115] From θ PreSS Select values ​​that satisfy ρ²ε² and add their corresponding position indices to the candidate set CaS, where ||θ PreSS ||1 represents θ PreSS The 1-norm of ρ2 is an empirical value, which is taken as 0.21 here.

[0116] ⑥ Update the support set SupS;

[0117] If the number of elements in the candidate set CaS is greater than the length L of the support set SupS Then the top L in the candidate set will be... SupS The items form the support set SupS;

[0118] If the number of elements in the candidate set CaS is less than the length L of the support set SupS If the support set SupS is equal to the candidate set CaS, then the support set SupS is equal to the candidate set CaS.

[0119] ⑦ Select the corresponding column in the perception matrix SM based on the position index number in the support set SupS, and define the matrix composed of these columns as SM. SupS Solve the following equation using the least squares method:

[0120]

[0121] And update the residuals

[0122] ⑧ If condition |e is met iter |≤10e-6 or iter=L observe If the set of index numbers of the non-zero terms in the sparse representation coefficients θ equals the support set SupS, then stop the iteration and proceed to step ⑨; if the condition |e iter |≥β·|e iter-1 (β is called the error ratio coefficient, which is taken as 1.1 here), then proceed to the next iteration iter = iter + 1 and update the support set length L. SupS =L SupS +iter, return to step ②; otherwise, update the support set length L. SupS =L SupS +1, return to step ②;

[0123] ⑨ Based on the index number SupS of the non-zero terms in the sparse representation coefficients θ and the corresponding value θ SupS This yields the sparse representation coefficients θ. Except for the position numbered SupS, all other positions in coefficient θ have a value of 0.

[0124] ⑩ Represent the sparse coefficients θ∈RL Substitute x re In φ·θ, the repair signal x is obtained. re ∈R L .

[0125] Finally, substitute the obtained sparse representation coefficients θ into the following equation: x re =φ·θ; This yields the repair signal x. re ∈R L .

[0126] If a signal possesses compressibility and sparsity, or exhibits sparsity after transformation by a sparse representation matrix, it can be projected onto a low-dimensional space using an observation matrix independent of the sparse representation matrix, thus achieving signal reconstruction. This allows for the rapid acquisition of an effective repair signal x. re .

[0127] S2. Construct a closed-open self-complementary top-hat morphological filter FCO-NSTH to enhance the characteristic frequency amplitude of the vibration signal to be enhanced; the definition of the closed-open self-complementary top-hat morphological filter FCO-NSTH is as follows:

[0128]

[0129] In the formula, x enhance (l) represents the enhanced vibration signal; FCO-NSTH represents the closed-open self-complementary top cap morphological filter. This indicates that an opening operation is performed after a closing operation is performed on the repair signal; This indicates that the opening operation is performed on the repair signal; (x re ·g)(l) represents the closing operation on the repair signal; L represents the sampling length of the poor vibration data.

[0130] While basic morphological filters (composed of four basic morphological filters and their combinations, including erosion, dilation, opening, and closing operations) offer some noise reduction and feature enhancement, they still cannot fully meet the processing requirements of non-stationary signals. Therefore, this method constructs a closed-open self-complementary top-hat morphological filter, FCO-NSTH. Compared to conventional morphological filters, it not only increases the amplitude of the output signal but also simultaneously extracts positive and negative pulses and enhances detailed features, exhibiting a significant advantage in feature amplitude enhancement. This effectively enhances the amplitude of the characteristic frequencies of vibration signals from hot forging press components.

[0131] The main idea of ​​morphological filters is to analyze the relationships between different parts of the signal by continuously moving structural elements of different shapes within the signal to be processed. This allows for local matching or correction of the signal's geometric features, effectively extracting useful information, suppressing noise, and preserving the signal's main morphological characteristics. Morphological filters are mainly divided into noise reduction morphological filters and feature extraction morphological filters. Different morphological filters exhibit different feature extraction characteristics, and the performance of structural elements with different morphological features is decisively affected. Commonly used structural elements include flat, triangular, semi-circular, and (sine)cosine shapes. Currently, there is no clear standard for selecting the shape of structural elements; most methods select structural elements with shapes similar to the analyzed signal. Since the shape of cosine structural elements is similar to that of vibration signals, this method uses cosine structural elements for computation.

[0132] The expression for a cosine-type structuring element is:

[0133]

[0134] In the formula, l se L is the length of the cosine-type structuring element, used to adjust the period of the cosine function. se Less than the length L of the repair signal se ang se For the angle of the cosine-type structural element, ang se ∈0,...,π, used to adjust the height of the cosine function.

[0135] To better understand the computational process of cosine-type structural elements, a structure is constructed as follows: Figure 5 The diagram shows four different shapes of cosine-type structuring elements. As can be seen from the figure, the shapes of the cosine-type structuring elements constructed based on different parameter combinations vary. The longer the structuring element, the more complex its shape and the stronger its computational processing power, but the greater the computational load. Furthermore, different angles of the cosine-type structuring element affect the final value of the structuring element. (Comparison) Figure 5 (a) and Figure 5 (b) It can be seen that, when the length of the structuring element is the same, different angles of the cosine structuring element will cause different values. The smaller the angle, the denser the values, which in turn affects the different processing effects of the signal.

[0136] S3. Using a particle swarm optimization algorithm, the optimal structural element parameters of the closed-open self-complementary top-hat morphological filter are adaptively selected; the structural element parameters include length and angle.

[0137] In specific implementation, when using the particle swarm optimization algorithm, the number of iterations and the population size are both set to 100, the acceleration factors c1 and c2 are set to 2, the weight coefficient is set to 1, the length optimization range is [2, 50], and the angle optimization range is [0, 180°].

[0138] The envelope entropy of the vibration signal enhanced by FCO-NSTH is used as the fitness function to select the optimal combination of structural element parameters. Envelope entropy reflects the periodic transient pulse sequence in the signal; the smaller the envelope entropy, the more pronounced the signal's impact characteristics after morphological filtering. This setting ensures the effectiveness of the obtained structural element parameters.

[0139] The formula for calculating the envelope entropy is as follows:

[0140]

[0141] In the formula, E evelope Represents the envelope entropy; a evelope (l) represents the characteristic amplitude enhancement signal x enhance (l) (l=1,2,...,L) is the envelope signal sequence obtained after demodulation by Hilbert; H[·] represents the Hilbert transform; κ l It is a evelope The normalized form of (l).

[0142] The process of adaptive selection of structure parameters based on particle swarm optimization algorithm is as follows: Figure 6 As shown.

[0143] A set of simulation signals was filtered using the FCO-NSTH morphological filtering method based on particle swarm optimization. The optimal structural element parameters obtained were a length of 14.3663 and an angle of 5.3373°. The length of the structural element must be an integer. If the optimization result of the structural element length is not an integer during the particle swarm optimization process, the length must be rounded to the nearest integer. Therefore, for the simulation signal, the final optimal structural element length is determined to be 14.

[0144] To verify the characteristic amplitude enhancement performance of the particle swarm optimization morphological filter, nine combinations of structural element parameters (length and angle) were manually selected: (2, 0°), (2, 90°), (2, 180°), (25, 0°), (25, 90°), (25, 180°), (50, 0°), (50, 90°), and (50, 180°). Morphological filters were constructed based on these parameter combinations to process the simulated signals, and the resulting spectra are shown below. Figure 7 As shown. Overall, Figure 7 (a) Figure 7 (b) and Figure 7The feature enhancement effect (c) is not good because the structuring element length is set too small, and the morphological filter does not extract useful information well. Figure 7 As can be seen in (d)-(j), the morphological filters constructed using these parameters all enhance the amplitude of the features to varying degrees. This demonstrates that the FCO-NSTH morphological filtering method proposed in this invention can effectively improve the amplitude of the feature frequencies when the structuring element length is appropriate. Figure 7 As can be clearly seen in (j), the signal amplitude is significantly enhanced after morphological filtering of the simulation signal using FCO-NSTH based on PSO optimization, especially in the frequency range of 0 to 1500 Hz. This indicates that the method proposed in this invention has a good adaptive characteristic amplitude enhancement effect.

[0145] S4. Use the FCO-NSTH closed-open self-complementary top-hat morphological filter with adaptive selection of optimal structuring element parameters to repair the signal x. re The characteristic frequency amplitude is enhanced to obtain an enhanced vibration signal.

[0146] S5. Input the enhanced vibration signal obtained in S4 into the intelligent model to identify and predict the wear state of hot forging press components.

[0147] This method utilizes a particle swarm optimization-optimized morphological filter to enhance the amplitude of characteristic frequencies of equipment, highlighting the amplitude of characteristic frequencies related to the equipment's operating state. Morphological filtering is a signal analysis method belonging to nonlinear filtering. This method uses structuring elements to perform translation matching and local correction on the signal from front to back, which can enhance the morphological features of the signal. Furthermore, morphological filtering only involves simple operations such as addition, subtraction, and finding extrema, making it computationally simple and efficient. The closed-open-self-complementary top-cap morphological filter FCO-NSTH constructed in this method, compared with conventional morphological filters, not only increases the amplitude of the output signal but also simultaneously extracts positive and negative pulses and enhances detailed features, exhibiting a significant advantage in characteristic amplitude enhancement. It can effectively enhance the amplitude of characteristic frequencies of vibration signals from hot forging press components. The repaired signal x after enhancement processing using the closed-open-self-complementary top-cap morphological filter FCO-NSTH is shown below. re This method can enhance the vibration characteristics of hot forging press components, eliminate / weaken the impact of vibrations from parts unrelated to the hot forging press components, thereby improving the performance and quality of equipment operation status recognition based on deep learning. Using this method, the wear condition of hot forging press components can be accurately identified by combining intelligent models.

[0148] Example 2

[0149] To help those skilled in the art better understand the effectiveness of this method, the following example is provided. In this embodiment, vibration signals from hot forging press components throughout their entire lifespan are used to experimentally verify a method for detecting and repairing substandard vibration data based on local outlier factors and compression sensing, as well as a method for identifying the wear state of hot forging press components based on a closed-open self-complementary top-cap morphological filter (FCO-NSTH) and a residual neural network. The proposed method is compared and its effectiveness analyzed from both qualitative and quantitative perspectives. It should be noted that the hot forging press component in this embodiment uses a brake as an example. If a clutch from a hot forging press is used as the component example, the experimental procedures are the same and will not be repeated here.

[0150] During data acquisition, the vibration signal of the hot forging press component collected by the piezoelectric accelerometer was first input into the data acquisition card via the sensor cable. After data conversion, it was stored in the portable hard drive. The parameters of the data acquisition equipment involved in the experiment are shown in Table 1.

[0151] Table 1. Parameters of the data acquisition equipment involved in the experiment.

[0152]

[0153] The time-domain waveforms and spectra of some of the collected vibration signals are as follows: Figure 8 As shown, there is no significant difference in the signal waveform at different time stages, but there are obvious differences in the spectrum.

[0154] Detection and repair of poor vibration data

[0155] To verify the performance of the proposed method for detecting and repairing substandard vibration signals during actual processing, a set of measured vibration data from a hot forging press component was used. One of the normal signals was artificially modified to contain data pulses, resulting in a substandard signal. The proposed method was then used to detect and repair the substandard signal, and the results were compared with those of the normal signal. The influence of parameter selection on the repair results was also discussed based on a set of simulated signals.

[0156] Poor data detection

[0157] A set of vibration signals x during normal processing normal Some data points in the middle section were replaced with low-quality data points, resulting in a low-quality signal x. poor like Figure 9 As shown, the local outlier factor algorithm is used to detect substandard data points. The parameter settings are the same as those used in Chapter 2 when repairing substandard simulation signals. The detection results are as follows. Figure 9 As shown in (a), the detected substandard data points were then set to zero, and the results are as follows. Figure 9As shown in (b). From Figure 9 As can be seen, the Local Outlier Factor algorithm can effectively detect inferior vibration data points, but some normal data points are also identified as inferior data points. This is because the relevant detection parameters in the Local Outlier Factor algorithm are not set appropriately. Experiments have shown that as long as the detection parameters are set within a reasonable range, they will not have a significant impact on the repair results. The question of how the setting of detection parameters affects the repair results will be discussed in Section 0 of this chapter.

[0158] Based on the index number of the zero-setting position, the corresponding row in the identity matrix is ​​removed to obtain the observation matrix. Then, the observation signal y is obtained using the observation matrix. 观测 ∈R 1509 .

[0159] Inferior data repair

[0160] Based on the obtained sparsity estimate SPAE = 73, the sparse representation coefficients θ are obtained using the MSAMP algorithm, and then the repair signal x is calculated using this method (MSAMP method). 修复 The result is as follows Figure 10 As shown.

[0161] The envelope spectra of the signal restored based on the MSAMP method, the normal signal, and the inferior signal were compared, and the results are as follows: Figure 11 As shown in the figure. (a) represents the normal signal; (b) represents the poor-quality signal; and (c) represents the MSAMP repair signal. It can be seen from the figure that within the 0-300Hz range ( Figure 11 As shown in the red dashed box, the vibration characteristic frequencies in low-quality signals are clearly submerged, especially in areas such as... Figure 11 (b) At the 40Hz position, the frequency amplitude of the inferior signal is significantly higher than that of the normal signal and the repaired signal. This makes it easy to regard 40Hz as the main characteristic frequency, affecting the judgment of the equipment's operating status. However, the envelope spectrum of the repaired signal using the MSAMP algorithm is almost identical to that of the normal signal, and the characteristic frequency distribution is the same.

[0162] The effectiveness of this method is verified from the frequency domain perspective. Figure 12 The envelope spectra of the restored signal and the normal signal are shown using different reconstruction methods. Among them, (a) MSAMP; (b) OMP; (c) ROMP; (d) StOMP; (e) SP; (f) gOMP; (g) CoSaMP; (h) SAMP.

[0163] To clearly compare the differences in frequency domain information, the envelope spectrum in the 2000-3000Hz range was locally magnified. Figure 12As shown in the black circular box, at the characteristic frequency of 2240Hz, the signal restored using MSAMP and SAMP has the closest amplitude to the normal signal, while the restoration effects of other algorithms show significant amplitude differences at this position. Although the restoration effects based on SAMP and MSAMP are relatively similar, MSAMP still exhibits the best overall restoration performance.

[0164] The results of calculating the repair performance evaluation indicators are as follows: Figure 13 As shown, from the perspective of waveform similarity, the Pearson correlation coefficient and cosine similarity between the signal restored using MSAMP and the normal signal are significantly better than other reconstruction algorithms, both reaching 0.9813. The restoration effect based on the SAMP algorithm is the worst, with a value of 0.9615. Furthermore, in terms of error level, the signal restored using MSAMP also shows excellent results, with the normalized root mean square error and JS divergence being the best, at 0.0371 and 0.0069 respectively. In contrast, the restoration effect of SAMP is the worst, with values ​​of 0.0787 and 0.0121 respectively. This reflects that the restoration performance of MSAMP is significantly improved compared to SAMP.

[0165] The Manhattan distances between signals repaired using different methods and the normal signal, as well as between different repaired signals, were calculated, and the results are as follows: Figure 14 As shown. From Figure 14 As can be seen from the figure, compared with other repaired signals, the signal repaired using MSAMP has the smallest Manhattan distance with the normal signal, which is 76.81. At the same time, it can also be seen from the figure that although the MSAMP reconstruction algorithm is an improvement on the SAMP method, the Manhattan distance between the repaired signals of MSAMP and SAMP reaches 105.59, which is greater than the Manhattan distance between the repaired signals of MSAMP reconstruction algorithm and other reconstruction algorithms. This reflects that MSAMP has made a great improvement on SAMP, and the repair performance has been greatly improved.

[0166] By comparing the waveform similarity, error level, and Manhattan distance between the repaired signals, the signal detection and repair method based on local outlier factors and compressed sensing demonstrates, from a quantitative analysis perspective, that has better detection and repair performance.

[0167] The impact of parameter selection on repair results

[0168] When detecting and repairing poor vibration data, it is necessary to manually set the nearest neighbor number k and the number of poor points n. poor , set the threshold to zero Th zeroUsing a set of simulated signals and based on the normalized root mean square error index, the effects of selecting these three parameters on the restoration results of poor-quality vibration data are discussed. Furthermore, the restoration effects of different restoration methods on poor-quality vibration data under different signal-to-noise ratios are also discussed.

[0169] (1) The impact of the selection of the nearest neighbor number on the repair results

[0170] To discuss the impact of the nearest neighbor number k on the repair results, the number of defective points n is fixed. poor The value of k is 30, starting from 10, and the experiment is conducted every 10 numbers until 200, for a total of 20 sets of data. The results are as follows: Figure 15 As shown in the figure, within the range of k∈(0:20), the NMSE is greatly affected by the value of k. The effect of the parameter k on the NMSE exhibits a periodic pattern, and the NMSE is not significantly affected by the value of k, with small fluctuations. Setting the nearest neighbor number k=100 yields an NMSE of 0.0623, which, while not optimal, is generally within an acceptable range.

[0171] (2) The impact of setting the number of defective points on the repair results

[0172] To discuss the number of defective points n poor The impact on the repair results, with the nearest neighbor number k fixed at 100, n poor The value of is taken starting from 10, and the experiment is conducted every 10 numbers until 200, resulting in a total of 20 sets of data. The results are as follows: Figure 16 As shown in the figure. It can be seen from the figure that the n set in this invention... poor =30 is the optimal position, and the obtained error result is 0.0623. As the number of defective points k increases, the error NMSE of the repair signal increases, reflecting that the setting of the number of defective points will affect the repair result.

[0173] This invention primarily focuses on data restoration. Defect detection is only necessary for constructing the observation matrix and observing signals. As long as the defect detection algorithm can detect all defective data points without omission, it is sufficient. If normal data points are incorrectly detected in addition to defective ones, multiple experiments have shown that as long as the parameters are set within acceptable ranges, this will not significantly affect data restoration; the signal can still be restored using compressed sensing technology. It should be noted that the number of defective data points, n... poor The value should not be set too high, as an excessive number of low-quality data points can easily result in null values ​​for the observed signal. Therefore, to avoid extreme cases, n is chosen. poor The suggested range is 0-100. If quality defects still exist after processing, the method proposed in this invention can be repeated multiple times to process the signal. Further research is needed to determine how to calculate the optimal parameter values.

[0174] (3) The impact of setting the zero threshold on the repair results

[0175] Choose a nearest neighbor number k = 100 and a bad point number n. poor Set the threshold to 30 and set Th to zero. zero The value of is taken starting from 10, and the experiment is conducted every 10 numbers until 200, resulting in a total of 20 sets of data. The results are as follows: Figure 17 As shown in the figure. It can be seen from the figure that as the threshold Th is set to zero... zero As Th increases, the repair error shows an upward trend. zero When the value is greater than 60, the normalized mean square error changes significantly, and the set Th value... zero =20 is within a reasonable range.

[0176] (4) The impact of different signal-to-noise ratios on the restoration results

[0177] The impact of noise on the restoration results was investigated using simulated signals with signal-to-noise ratios of 5dB and 10dB. In addition, the restoration effects of different restoration methods under different signal-to-noise ratio conditions were studied.

[0178] ①Signal-to-noise ratio 5dB

[0179] The repair performance of the proposed method was verified using a simulated signal with a signal-to-noise ratio of 5 dB. The time and frequency domains of the signal are as follows: Figure 18 and Figure 19 As shown. Figure 18 In the diagram, (a) is a normal signal; (b) is a poor signal; (c) is a poor data zeroing signal; and (d) is a repair signal. Figure 19 In the figure, (a) is the normal signal; (b) is the poor-quality signal; and (c) is the repaired signal. It can be seen from the figure that the presence of poor-quality data points severely interferes with the identification of characteristic frequencies, while the repaired signal has a high degree of similarity to the normal signal. The characteristic frequencies of the repaired signal are clearly visible, especially in the frequency range of 0-600Hz, where the characteristic frequencies in the envelope spectrum of the poor-quality signal are severely interfered with.

[0180] ②Signal-to-noise ratio 10dB

[0181] The repair performance of the proposed method was verified using a simulated signal with a signal-to-noise ratio of 10 dB. The time and frequency domains of the signal were analyzed as follows: Figure 20 and Figure 21 As shown. Figure 20 In the diagram, (a) is a normal signal; (b) is a poor signal; (c) poor data is set to zero; and (d) is a repair signal. Figure 21 In the image, (a) is the normal signal; (b) is the poor-quality signal; and (c) is the repaired signal. Similar to the 5dB simulation signal, the repaired signal shows good similarity to the normal signal, and the characteristic frequencies of the repaired signal are clearly defined.

[0182] (3) Model comparison and quantitative analysis

[0183] Different repair methods were used to repair poor-quality simulation signals with different signal-to-noise ratios, and the normalized mean square error was obtained as follows: Figure 22 As shown in the figure, under different signal-to-noise ratio conditions, the NMSE between the signal restored by the proposed method and the normal signal is minimized, which demonstrates the effectiveness of the proposed restoration method.

[0184] Vibration characteristic frequency amplitude enhancement

[0185] A set of vibration signals from hot forging press components collected during actual processing was selected. This vibration data includes vibration data from the bearings supporting the rotation of the hot forging press spindle, as well as environmental noise. The time-domain waveforms and spectra of the original vibration signals and the vibration signals enhanced by morphological filters are shown below. Figure 23 As shown in the figure. (a) Original signal; (b) Enhanced signal; (c) Original signal spectrum; (d) Enhanced signal spectrum. From... Figure 23 (c) and Figure 23 As can be seen in (d), the amplitude of the enhanced signal spectrum is increased, indicating that the improved morphological filter has excellent characteristic amplitude enhancement performance.

[0186] This experiment verifies the performance of the proposed method for detecting and repairing substandard data from the perspective of practical engineering applications. First, vibration signals from components of a hot forging press were collected. Second, a method based on local outlier factors and compressed sensing was used to detect and repair substandard vibration data present in the vibration signals. The repair performance of the proposed method was discussed from both qualitative and quantitative evaluation perspectives. Experimental results show that the method has higher waveform similarity and lower error levels. Specifically, the Pearson correlation coefficient between the signal repaired by the proposed method and the normal signal is 0.9813, and the normalized root mean square error is only 0.0371.

[0187] This method uses vibration signals as the data basis. After completing preprocessing work such as improving the quality of poor vibration data, it further proposes a vibration characteristic frequency amplitude enhancement method based on an improved morphological filter.

[0188] To address the problem that substandard vibration signals, caused by random interference from the processing environment and performance degradation of data acquisition equipment, can affect the accurate assessment of equipment operating status, a method for detecting and repairing substandard vibration data based on local outlier factor and compressed sensing is proposed. This method uses a local outlier factor algorithm to locate and zero-reset substandard data points. Based on this, compressed sensing technology is used to repair the missing data. The effectiveness of the proposed method is verified by comparing simulated and measured signals. The normalized root mean square error between the repaired vibration signal of the hot forging press component and the normal signal is only 0.0371, and the Pearson correlation coefficient reaches 0.9813. This indicates that the proposed method performs well in detecting and repairing substandard vibration data in terms of both error level and waveform similarity, and can be used for substandard vibration data repair tasks.

[0189] To address the issue that the collected vibration signals contain vibration signal components unrelated to the equipment under study, such as bearings, a novel morphological filter based on particle swarm optimization is proposed to enhance the vibration characteristic frequency amplitude, thereby reducing the impact of vibrations from other components and environmental noise on the equipment's vibration characteristics.

[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit the technical solutions. Those skilled in the art should understand that any modifications or equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for enhancing feature signals for identifying wear conditions of components in a hot forging press, characterized in that, Includes the following steps: S1. Inferior vibration signal x for hot forging press components poor Sampling, and the sampled poor-quality vibration signal x poor Perform defective data detection and repair processing to obtain the repair signal x. re ; S2. Construct a closed-open self-complementary top-hat morphological filter FCO-NSTH to enhance the characteristic frequency amplitude of the vibration signal to be enhanced; the definition of the closed-open self-complementary top-hat morphological filter FCO-NSTH is as follows: In the formula, x enhance (l) represents the enhanced vibration signal; FCO-NSTH represents the closed-open self-complementary top cap morphological filter. (l) indicates that an opening operation is performed on the repair signal after a closing operation; (l) indicates that the repair signal is opened; (x) re ·g)(l) represents performing a closing operation on the repair signal; L represents the sampling length of the poor-quality vibration data; S3. The optimal structural element parameters of the closed-open self-complementary top-hat morphological filter are adaptively selected through the particle swarm optimization algorithm. S4. Use the FCO-NSTH closed-open self-complementary top-hat morphological filter with adaptive selection of optimal structuring element parameters to repair the signal x. re The characteristic frequency amplitude is enhanced to obtain an enhanced vibration signal; S5. Input the enhanced vibration signal obtained in S4 into the intelligent model to identify and predict the wear state of hot forging press components.

2. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 1, characterized in that: In S2, within the constructed FCO-NSTH, cosine-type structuring elements are used for corresponding processing; the expression for the cosine-type structuring element is: In the formula, l se L is the length of the cosine-type structuring element, used to adjust the period of the cosine function. se Less than the length L of the repair signal se ang se For the angle of the cosine-type structural element, ang se ∈0,...,π, used to adjust the height of the cosine function.

3. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 2, characterized in that: In S3, the structural element parameters include length and angle.

4. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 3, characterized in that: In S3, when using the particle swarm optimization algorithm, the number of iterations and the population size are both set to 100, the acceleration factors c1 and c2 are set to 2, the weight coefficient is set to 1, the length optimization range is [2, 50], and the angle optimization range is [0, 180°].

5. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 1, characterized in that: When using the particle swarm optimization algorithm, the envelope entropy of the vibration signal enhanced by FCO-NSTH is used as the fitness function to select the optimal combination of structural element parameters.

6. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 5, characterized in that: The formula for calculating the envelope entropy is as follows: In the formula, E evelope Represents the envelope entropy; a evelope (l) represents the characteristic amplitude enhancement signal x enhance (l) (l=1,2,...,L) is the envelope signal sequence obtained by Hilbert demodulation; H[·] denotes the Hilbert transform; κ l It is a evelope The normalized form of (l).

7. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 1, characterized in that: In S1, for the inferior vibration signal x poor The process of detecting substandard data includes: ① Input poor-quality vibration signal x poor ∈R L Where L represents the sampling length of the poor-quality vibration data; ②Let i = 1; ③ Calculate the i-th vibration data point p i The k-th distance neighbor NBR k (p i ); ④ Calculate vibration data points o∈NBR k (p i ) to data point p i The k-th reachable distance RD k (o,p i ); ⑤ Calculate the i-th vibration data point p i Locally achievable density (LRD) k (p i ); ⑥ Calculate the i-th vibration data point p i The k-th local outlier LOF k (p i ); ⑦ If i ≤ L, then let i = i + 1 and repeat steps ③-⑥; otherwise, proceed to step ⑧. ⑧ Find the largest n in step ⑦ poor The position index number corresponding to the vibration data point of each LOF value is put into the set PosIndex = {pos_index} i |i=1,2,…,n poor }middle; 9. Based on pos_index i ∈PosIndex, the data range [pos_index] i -Th zero pos_index i +Th zero The data within ] is set to zero, where Th zero This indicates the threshold value set to zero. ⑩ The signal obtained after the defective data detection is denoted as the defective data detection signal x. LOF .

8. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 7, characterized in that: In S1, the repair process includes: First, using a sparse transformation matrix φ∈R L×L The inferior detection signal x LOF ∈R L Sparse representation is θ∈R L As shown in the following formula: x LOF =φ·θ; where θ is the sparse representation coefficient; Then, using the observation matrix The inferior detection signal x LOF Compressed to a low-dimensional space, as shown in the following equation: y observe =ψ·x LOF In the formula, For the observed signal, L observe Indicates the data length of the observed signal; Then, x LOF =φ·θ Substitute into y observe =ψ·x LOF , get y observe =ψ·φ·θ=SM·θ, and solve for the sparse representation coefficient θ; where, For the perception matrix; Finally, substitute the obtained sparse representation coefficients θ into the following equation: x re =φ·θ; This yields the repair signal x. re ∈R L . If a signal possesses compressibility and sparsity, or exhibits sparsity after transformation by a sparse representation matrix, it can be projected onto a low-dimensional space using an observation matrix independent of the sparse representation matrix, thus achieving signal reconstruction. This allows for the rapid acquisition of an effective repair signal x. re .

9. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 8, characterized in that: For inferior detection signals x LOF When performing sparse representation, the discrete cosine transform matrix is ​​used as the sparse representation matrix φ. The element in the i-th row and j-th column of this matrix is ​​calculated by the following formula:

10. The feature signal enhancement method for identifying wear conditions of components in a hot forging press as described in claim 8, characterized in that: The process of constructing the observation matrix includes: after detecting poor-quality vibration data, the location numbers of the vibration data to be set to zero are placed into the index set PosIndex = {pos_index}. i |i=1,2,…,n poor In the set, the data length is LL. observe Based on the position index numbers in the index set PosIndex, the identity matrix I∈R is... L×L Delete the corresponding row in the matrix to obtain the observation matrix.

Images