A Bearing Degradation Initiation Point Detection Method Based on Autoencoder and AOMM

Abstract

The present invention relates to the technical field of bearing fault diagnosis, and discloses a bearing degradation starting point detection method based on an autoencoder and an AOMM. The method includes collecting a real-time vibration signal of a bearing; obtaining a feature sequence of the real-time vibration signal using an autoencoder; obtaining a symbolic feature sequence of the real-time vibration signal based on an optimal number of symbols obtained through pre-training; obtaining a plurality of sliding windows based on an optimal window size obtained through pre-training; applying an AOMM method, adaptively adjusting the Markov order based on symbol distribution entropy and a polynomial regression model, obtaining a state stationary probability matrix; inputting the state stationary probability matrix and label data into a pre-trained adaptive enhancement classifier for prediction, and outputting a predicted degradation starting point. The present invention more accurately identifies the degradation starting point, solving the problems of incomplete feature extraction of the degradation starting point, being greatly affected by noise, and the problem that the degradation starting point detection relies heavily on the signal amplitude and ignores the dynamic behavior of the signal.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing fault diagnosis and relates to a bearing degradation starting point detection method based on an autoencoder and AOMM. Background Art

[0002] Bearings are critical components in mechanical systems, and their operating status directly impacts the reliability and safety of the entire system. In large-scale machinery such as wind turbines, aircraft engines, and high-end CNC machine tools, the service life of bearings critically affects the proper functioning of the equipment. Accurately predicting the remaining useful life (RUL) of bearings is crucial for preventing equipment failures, reducing maintenance costs, and improving production efficiency.

[0003] Identifying the degradation starting point is a crucial and fundamental step in bearing life prediction. Accurately identifying this point further confirms the bearing's degradation state and provides a key basis for RUL prediction. Identifying this point not only serves as a starting point for predicting bearing performance changes but also forms the foundation for developing subsequent life prediction models. Only by accurately identifying the degradation starting point can prediction errors caused by misjudgment be effectively avoided, providing a reliable time window for equipment maintenance and fault prevention.

[0004] However, existing methods for detecting the onset of degradation often rely on manually extracted features, making it difficult to fully and accurately capture the key features that reflect the onset of degradation. Furthermore, bearing vibration signals often contain a significant amount of noise, which can mask the true degradation characteristics and reduce detection accuracy. Another problem with existing methods for detecting the onset of degradation is that they rely heavily on signal amplitude to detect anomalies, while ignoring the dynamic behavior of the signal. This method, which relies on a single signal amplitude, cannot effectively detect the onset of degradation in some cases, especially when the signal changes are complex or there is significant noise interference. Summary of the Invention

[0005] The purpose of the present invention is to provide a bearing degradation starting point detection method and system based on an autoencoder and an Adaptive Order Markov Model (AOMM), which can more accurately identify the degradation starting point and solve the problems of incomplete feature extraction of the degradation starting point, which is greatly affected by noise, and the degradation starting point detection relying more on the signal amplitude and ignoring the dynamic behavior of the signal.

[0006] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions.

[0007] In a first aspect, the present invention proposes a bearing degradation starting point detection method based on an autoencoder and an AOMM, comprising:

[0008] Collect real-time vibration signals of bearings;

[0009] Preprocessing the collected real-time vibration signal to obtain a preprocessed real-time vibration signal;

[0010] The autoencoder is used to extract features from the preprocessed real-time vibration signal to obtain the feature sequence of the real-time vibration signal, remove invalid features, and extract multi-dimensional feature vectors;

[0011] Globally symbolize the feature sequence of the real-time vibration signal according to the optimal number of symbols obtained by pre-training to obtain a symbolic feature sequence of the real-time vibration signal;

[0012] The symbolic feature sequence of the real-time vibration signal is segmented according to the optimal window size obtained by pre-training to obtain several sliding windows;

[0013] Applying the AOMM method, the Markov order is adaptively adjusted based on the symbolic distribution entropy and the polynomial regression model to calculate the state stationary probability vector of each sliding window, and all the state stationary probability vectors are spliced into a state stationary probability matrix;

[0014] The state stationary probability matrix and label data are input into the pre-trained adaptive boosting classifier for prediction, and the predicted degradation starting point is output.

[0015] In combination with the first aspect, further, the AOMM method is applied to adaptively adjust the Markov order based on the symbolic distribution entropy and the polynomial regression model to calculate the state stationary probability vector of each sliding window, and all the state stationary probability vectors are spliced into a state stationary probability matrix, including:

[0016] Calculate the symbol distribution entropy of each window and compare it with the symbol distribution entropy of the previous window to obtain the change of symbol distribution entropy;

[0017] Based on the change of symbol distribution entropy and the polynomial regression model, the change value of symbol distribution entropy is predicted and the Markov order is dynamically adjusted;

[0018] Based on the dynamically adjusted Markov order, the state stationary probability vector of each sliding window is calculated, and all the state stationary probability vectors are spliced into a state stationary probability matrix.

[0019] In combination with the first aspect, further, the method for obtaining the optimal number of symbols and the optimal window size is:

[0020] Step S1, collecting the original vibration signal of the bearing throughout its life cycle;

[0021] Step S2, preprocessing the original vibration signal to obtain a preprocessed original vibration signal;

[0022] Step S3, using an autoencoder to extract features from the preprocessed original vibration signal to obtain a feature sequence of the original vibration signal;

[0023] Step S4, globally symbolizing the feature sequence of the original vibration signal to obtain a symbolized feature sequence of the original vibration signal;

[0024] Step S5, segmenting the symbolic feature sequence of the original vibration signal to obtain a plurality of sliding windows;

[0025] Step S6, applying the AOMM method, adaptively adjusting the Markov order based on the symbolic distribution entropy and the polynomial regression model, calculating the state stationary probability vector of each sliding window, and splicing all the state stationary probability vectors into a state stationary probability matrix;

[0026] Step S7: Input the state stationary probability matrix and the label data as training samples, train the adaptive boosting classifier, and obtain the probability that each training sample belongs to the abnormal category predicted by the adaptive boosting classifier;

[0027] In step S8, based on the probability that each training sample predicted by the adaptive boosting classifier belongs to the abnormal category, the regularized cross entropy loss function is used to optimize the sliding window size and the number of symbols, and the optimized number of symbols is fed back to the step S4, and the optimized sliding window size is fed back to the step S5, and finally the optimal number of symbols and the optimal window size are obtained.

[0028] In combination with the first aspect, further, the state stationary probability matrix and the label data are input as training samples to train the adaptive boosting classifier, and obtain the probability that each training sample predicted by the adaptive boosting classifier belongs to the abnormal category, including:

[0029] Initialize the weights of the training samples;

[0030] Based on the weights of the initialized training samples, the weak classifier is trained and the weighted classification error rate under the distribution of all training samples is calculated;

[0031] Calculating the weight of the weak classifier according to the classification weighted error rate to obtain the classification result of each training sample;

[0032] Update the weight of each training sample according to the classification result of the training sample;

[0033] Normalize all updated training sample weights to obtain the normalized weight of each training sample;

[0034] Determine whether the preset number of iterations has been reached. If not, return to the step of training the weak classifiers. If so, perform a linear combination of all weak classifiers according to the normalized weights to obtain the final strong classifier.

[0035] In combination with the first aspect, further, optimizing the sliding window size and the number of symbols by using a regularized cross entropy loss function includes:

[0036] Design loss function :

[0037] ;

[0038] in, represents the cross entropy loss function; represents the regularization function;

[0039] Cross Entropy Loss Function The mathematical expression is:

[0040] ;

[0041] in, is the number of training samples, It is The true label of the training sample is 0 or 1. is the first prediction of the adaptive boosting classifier The probability that a training sample belongs to the abnormal category;

[0042] Regularization function The mathematical expression is:

[0043] ;

[0044] in, is the number of adaptive boosting classifier weights, It is weights, is the regularization coefficient, which controls the strength of the regularization term;

[0045] According to the loss function Calculate the loss and output the loss value. The loss value is used to measure the degree of match between the prediction result of the adaptive boosting classifier and the true label. The smaller the loss value, the better the prediction performance of the adaptive boosting classifier.

[0046] Keeping the current number of symbols unchanged, adjust the window size in turn in each loop and use the adaptive boosting classifier for training. By comparing the loss values under different window sizes, determine whether the change in loss value is less than the set loss value tolerance. If the change in loss value is less than the set loss value tolerance, determine the optimal window size and corresponding optimal loss value under the current number of symbols, stop further adjusting the window size, and update the number of globally symbolized symbols.

[0047] Under the updated number of globally symbolized symbols, repeat the previous step and continue to optimize the window size until the change in the optimal loss value is less than the set optimal loss value tolerance. At this time, the entire optimization process is stopped and the optimal sliding window size is obtained. The corresponding updated number of globally symbolized symbols is the optimal number of symbols.

[0048] In combination with the first aspect, further, the extracting features of the preprocessed original vibration signal using the autoencoder to obtain a feature sequence of the original vibration signal includes:

[0049] make For the input data set, that is, the set of preprocessed raw vibration signals, define , Represents the input dataset A feature vector in is the input dataset The index of the eigenvector in , , Represents the input dataset This collection contains feature vectors, that is, the size of the input dataset is , express dimensional real space, Represents the dimension of the feature vector; encoder The high-dimensional input dataset Pass from the input layer to the hidden layer, input the data set in the hidden layer Compressed into low-dimensional space In the definition , Representing low-dimensional space A low-dimensional feature vector in is the compressed feature representation. is the index of the special vector, ; Representing low-dimensional space This collection contains feature vectors, express dimensional real space, Represents the dimension of the compressed feature vector; then the decoder The output layer remaps the compressed feature vector to the input data , Output result: encoder and decoder The basic mathematical expression is:

[0050] ;

[0051] ;

[0052] encoder Enter Convert to low-dimensional representation , then the decoder The low-dimensional representation Convert to output , to reconstruct the input .

[0053] In combination with the first aspect, further, globally symbolizing the feature sequence of the original vibration signal to obtain the symbolized feature sequence of the original vibration signal includes: globally symbolizing the feature sequence of the original vibration signal using a uniform partitioning method;

[0054] For feature sequences , is the sequence length, is the eigenvalue; initialization parameter , Indicates that the feature sequence How many symbols it is divided into, that is, the number of symbols;

[0055] Calculate feature sequence The minimum value in and maximum value , as the boundary of the partition interval;

[0056] The feature sequence The value range of intervals; the width of each interval is ;

[0057] The feature sequence Each eigenvalue in is mapped to the corresponding symbol to obtain a symbolic feature sequence.

[0058] In conjunction with the first aspect, further, real-time vibration signals of the bearing during operation are collected. The vibration signals are in both horizontal and vertical directions. The full life cycle refers to the entire process from the beginning of the use of the bearing to its scrapping and loss of operational capability.

[0059] In combination with the first aspect, the collected vibration signal is further preprocessed to obtain a preprocessed vibration signal, including: normalizing the collected vibration signal to obtain a normalized vibration signal, which can eliminate the dimensional influence between different signals and unify the data scale.

[0060] In combination with the first aspect, further, normalizing the collected vibration signal to obtain the normalized vibration signal includes: using maximum-minimum normalization processing:

[0061] ;

[0062] in, is the minimum value in the vibration signal, is the maximum value in the vibration signal, is each value in the vibration signal, is the normalized value so that all data points fall within the interval [0,1].

[0063] It should be noted that: the collected vibration signal is preprocessed above, and the vibration signal obtained in the preprocessed vibration signal includes a real-time vibration signal and an original vibration signal, and the method of preprocessing the real-time vibration signal and the original vibration signal is the same.

[0064] In a second aspect, the present invention proposes a bearing degradation starting point detection system based on an autoencoder and an AOMM, comprising:

[0065] A data acquisition module configured to acquire real-time vibration signals of the bearing;

[0066] A data preprocessing module is configured to preprocess the collected real-time vibration signal to obtain a preprocessed real-time vibration signal;

[0067] The autoencoder module is configured to extract features from the preprocessed real-time vibration signal using the autoencoder to obtain a feature sequence of the real-time vibration signal;

[0068] A symbolized feature sequence module is configured to globally symbolize the feature sequence of the real-time vibration signal according to the optimal number of symbols obtained through pre-training to obtain a symbolized feature sequence of the real-time vibration signal;

[0069] A sliding window segmentation module is configured to segment the symbolic feature sequence of the real-time vibration signal according to an optimal window size obtained through pre-training to obtain a plurality of sliding windows;

[0070] An AOMM method module is configured to apply the AOMM method, adaptively adjust the Markov order based on the symbolic distribution entropy and the polynomial regression model, calculate the state stationary probability vector for each sliding window, and concatenate all the state stationary probability vectors into a state stationary probability matrix;

[0071] The adaptive boosting classifier module is configured to input the state stationary probability matrix and label data into a pre-trained adaptive boosting classifier for prediction, and output a predicted degradation starting point.

[0072] In a third aspect, the present invention proposes a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for detecting the starting point of bearing degradation based on the autoencoder and AOMM are implemented.

[0073] In a fourth aspect, the present invention provides a computer device, comprising:

[0074] memory for storing computer programs;

[0075] A processor is used to execute the computer program to implement the steps of the above-mentioned bearing degradation starting point detection method based on autoencoder and AOMM.

[0076] In a fifth aspect, the present invention proposes a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for detecting the starting point of bearing degradation based on an autoencoder and an AOMM.

[0077] Compared with the prior art, the present invention has the following beneficial effects:

[0078] (1) The present invention can more accurately identify the degradation starting point, solving the problems of incomplete feature extraction of the degradation starting point, which is greatly affected by noise, and the degradation starting point detection relying more on the signal amplitude and ignoring the dynamic behavior of the signal.

[0079] (2) The present invention uses an autoencoder to automatically learn features in the data, without the need for manually designing feature extraction rules. It has a certain degree of robustness to noise and can effectively remove noise interference, thereby improving the efficiency and accuracy of feature extraction.

[0080] (3) The present invention models the symbol sequence by combining the symbolic feature sequence with the AOMM method, and adaptively adjusts the Markov order to accurately capture the dynamic change law of the signal, rather than relying solely on the amplitude information of the signal.

[0081] (4) The present invention dynamically adjusts the window size and the number of symbols to find the optimal window and the optimal number of symbols, ensuring that each window contains enough information to capture dynamic behavior while balancing the model complexity and detection performance.

[0082] (5) The present invention combines an adaptive enhanced classifier, which can quickly adapt to signal changes and accurately identify the starting point of bearing degradation. At the same time, it effectively reduces detection delay and provides a reliable time window for equipment maintenance and fault prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 Schematic diagram of the flow of the bearing degradation starting point detection method of the present invention;

[0084] Figure 2 Schematic diagram of the process of symbol number optimization, window size optimization and adaptive boosting classifier training in the present invention;

[0085] Figure 3 Schematic diagram of the structure of the autoencoder (AE) in the present invention;

[0086] Figure 4 Schematic diagram of the process of the adaptive enhanced classifier in the present invention;

[0087] Figure 5 Schematic diagram of the degradation starting point of the bearing 1 identified by the detection method of the present invention;

[0088] Figure 6 Schematic diagram of the degradation starting point of the bearing 2 identified by the detection method of the present invention. DETAILED DESCRIPTION

[0089] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0090] The term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. Additionally, the character " / " generally indicates an "or" relationship between the related objects.

[0091] Example 1

[0092] like Figure 1 As shown, the steps of the bearing degradation starting point detection method based on the autoencoder and AOMM in this embodiment are as follows:

[0093] Collect real-time vibration signals of bearings;

[0094] Preprocessing the collected real-time vibration signal to obtain a preprocessed real-time vibration signal;

[0095] The autoencoder is used to extract features from the preprocessed real-time vibration signal to obtain the feature sequence of the real-time vibration signal, remove invalid features, and extract multi-dimensional feature vectors;

[0096] Globally symbolize the feature sequence of the real-time vibration signal according to the optimal number of symbols obtained by pre-training to obtain a symbolic feature sequence of the real-time vibration signal;

[0097] The symbolic feature sequence of the real-time vibration signal is segmented according to the optimal window size obtained by pre-training to obtain several sliding windows;

[0098] Applying the AOMM method, the Markov order is adaptively adjusted based on the symbolic distribution entropy and the polynomial regression model to calculate the state stationary probability vector of each sliding window, and all the state stationary probability vectors are spliced into a state stationary probability matrix;

[0099] The state stationary probability matrix and label data are input into the pre-trained adaptive boosting classifier for prediction, and the predicted degradation starting point is output.

[0100] In this embodiment, a specific implementation method is as follows: Figure 2 As shown, the method for training the optimal number of symbols, the optimal window size, and the adaptive boosting classifier includes the following steps:

[0101] Step S1: collecting the original vibration signal of the bearing throughout its life cycle.

[0102] The full life cycle refers to the entire process from the beginning of the use of the bearing to its scrapping and loss of operational capability. The direction of the vibration signal includes both horizontal and vertical directions.

[0103] It should be noted that the bearing in this embodiment is specifically a rolling bearing.

[0104] Step S2: normalize and preprocess the original vibration signals of the entire life cycle to eliminate the dimensional influence between different signals, unify the data scale, obtain the preprocessed original vibration signals, and further divide the preprocessed original vibration signals to obtain training sets and test sets.

[0105] The normalization preprocessing adopts the maximum-minimum normalization process:

[0106] ;

[0107] in, is the minimum value of the original vibration signal in the entire life cycle, is the maximum value of the original vibration signal over the entire life cycle. is each value in the original vibration signal of the whole life cycle, is the normalized value so that all data points fall within the interval [0,1].

[0108] Step S3: Use an autoencoder (AE) to extract features from the preprocessed original vibration signal, remove zero features and invalid features, extract multi-dimensional features, and obtain a feature sequence of the original vibration signal.

[0109] Among them, such as Figure 3 As shown, let For the input data set, that is, the set of preprocessed raw vibration signals, define , Represents the input dataset A feature vector in is the input dataset The index of the eigenvector in , , Represents the input dataset This collection contains feature vectors, that is, the size of the input dataset is , express dimensional real space, Represents the dimension of the feature vector; encoder The high-dimensional input dataset Pass from the input layer to the hidden layer, input the data set in the hidden layer Compressed into low-dimensional space In the definition , Representing low-dimensional space A low-dimensional feature vector in is the compressed feature representation. is the index of the special vector, ; Representing low-dimensional space This collection contains feature vectors, express dimensional real space, Represents the dimension of the compressed feature vector. The output layer remaps the compressed feature vector to the input data , is the output result. and decoder The basic mathematical expression is:

[0110] ;

[0111] ;

[0112] encoder Enter Convert to low-dimensional representation , then the decoder The low-dimensional representation Convert to output , to reconstruct the input .

[0113] Initialize the encoders separately and decoder Weight and bias , and the Sigmod function is used as the activation function , the autoencoder is trained as follows:

[0114] ;

[0115] ;

[0116] in, Represents the input dataset Elements in Representing low-dimensional space middle element; Indicates the output result Elements in represents the activation function of the encoder; represents the activation function of the decoder; represents the weight of the encoder; represents the weight of the decoder; Indicates the encoder bias; Denotes the bias of the decoder.

[0117] Use the mean squared error loss function , whose expression is:

[0118] ;

[0119] When training the autoencoder, use the Adam optimizer and adaptively adjust the learning rate. After extracting features, remove invalid features. Calculate the variance of each dimension and set a threshold (such as 0.0001) to remove features with variances below the threshold.

[0120] Step S4: globally symbolize the feature sequence obtained in step S3 using a uniform partitioning method to obtain a symbolized feature sequence.

[0121] For feature sequences , is the sequence length, is the eigenvalue. Initialization parameter , Indicates that the feature sequence The number of symbols divided into, that is, the number of symbols. Calculate the characteristic sequence The minimum value in and maximum value , as the boundary of the partition interval. The value range is evenly divided into intervals. The width of each interval is . The feature sequence Each eigenvalue in is mapped to a corresponding symbol. The symbol is an integer sequence starting from the simple numeric character "0". Finally, the feature sequence Convert to symbolic feature sequence ,in , For the symbols, ,and It is a symbol table consisting of a finite number of symbols.

[0122] Step S5: Use the sliding window method to perform the symbolic feature sequence obtained in step S4. Segment and generate multiple windows.

[0123] In step S51, the window size (window_size) and step size (step_size) are initialized. These are two key parameters of the sliding window method. The window size determines the number of symbols contained in each window, while the step size determines the step size of the window sliding across the symbolic feature sequence.

[0124] Step S52 , cyclically sliding the window. In the cycle, the window starts from the starting position of the symbolic feature sequence and slides one step each time until the end position of the window exceeds the sequence length of the symbolic feature sequence.

[0125] Step S6: Apply the AOMM method to each window, adaptively adjust the Markov order based on the symbolic distribution entropy and the polynomial regression model, calculate the state stationary probability vector of each sliding window, and splice all the state stationary probability vectors into a state stationary probability matrix.

[0126] The present invention proposes an AOMM method, namely an adaptive order Markov model, which can dynamically adjust the order of the Markov model according to the distribution of symbols. , which can more finely characterize the long-range dependencies and complex dynamic behaviors of sequences, reduce dependence on computing resources, and at the same time achieve better symbolic sequence feature extraction effects.

[0127] Step S61: Generate the state sequence of the Markov model for the symbolic feature sequence of each window. The state is transferred to The probability of a state is:

[0128] ;

[0129] in, is the state transition probability, is the state set, The state set states, Yes, the state is concentrated states, It is a probability function that is used to calculate the probability of an event occurring.

[0130] For a Markov model, the state transition probability matrix is the transition probability between all states and is a non-degenerate matrix. It describes the transition probability between all possible states.

[0131] Step S62: Calculate the symbol distribution entropy of the current window. Symbol distribution entropy is calculated by calculating the number of times each symbol appears in the sequence in the window and converting it into a probability distribution. The Shannon entropy formula is then used to measure the degree of dispersion of these probabilities. The mathematical formula is:

[0132] ;

[0133] in, represents the symbol distribution entropy, is the total number of symbols, It is The probability of a symbol appearing in the sequence.

[0134] Step S63: Dynamically adjust the order of the Markov model in each window First, calculate the symbol distribution entropy change between the current window and the previous window. Add the current symbol distribution entropy change value to the historical symbol distribution entropy list. If the symbol distribution entropy change exceeds the set threshold, it means that the data complexity has increased and the order of the Markov model needs to be increased. At this time, if the historical symbol distribution entropy data is sufficient (at least 3 data points), the historical symbol distribution entropy change value is used as the training data, and the polynomial regression model is used to fit the historical symbol distribution entropy change trend and predict the symbol distribution entropy change of the next window. This application selects a 4th-order polynomial regression model, which can capture more complex curve relationships and is suitable for situations where the data has a certain degree of smoothness and trend. The mathematical expression of the 4th-order polynomial regression model is as follows:

[0135] ;

[0136] in, Indicates the change in symbol distribution entropy, represents the window index, is the intercept, indicating that the polynomial regression model The predicted value at are the coefficients of the polynomial regression model, corresponding to the weights of the linear, quadratic, cubic, and quartic terms, respectively. is the error term.

[0137] When predicting the change in the symbol distribution entropy of the next window, this application uses a fitted polynomial regression model, substitutes the index of the next window into the polynomial regression model, obtains the predicted symbol distribution entropy change value, and adjusts the order of the Markov model according to the prediction result If the predicted symbol distribution entropy change still exceeds the preset symbol distribution entropy threshold, the order of the Markov model is greatly increased. , that is, increase by 2; otherwise, slightly increase the order of the Markov model , which is incremented by 1. If the historical symbol distribution entropy data is insufficient, the order of the Markov model is increased slightly. , which is incremented by 1. If the symbol distribution entropy change is less than or equal to the preset symbol distribution entropy threshold, the order of the current Markov model is maintained. Unchanged. Order The adjustment strategy is as follows:

[0138] ;

[0139] in, Indicates the The order of the Markov model corresponding to the window; Indicates the The entropy change between a window and the previous window; Indicates the preset symbol distribution entropy change threshold; Represents the number of data points of historical symbol distribution entropy; Indicates the The predicted value of the change in the entropy of the symbol distribution in each window.

[0140] Step S64: for the symbolic feature sequence , the window slides to the right by The length of the size, the statistical length is and The number of symbol strings is counted as and , Indicates count, Indicates a length of The symbol string, Indicates a length of A string of symbols, where It is before A sequence of symbols, It is a symbol that follows immediately. Indicates the first symbol in the sequence Symbols , subscripts Used to identify the index in the symbol sequence. , then the state The probability of occurrence is 0. , then the transition probability of a certain state is calculated as:

[0141] ;

[0142] The corresponding state is represented as and .

[0143] State transition probability matrix The left eigenvector corresponding to the unit eigenvalue of is the state stationary probability vector, representing the state distribution of the bearing operating state represented by the symbolic feature sequence under stable conditions. The state stationary probability vector, serving as the feature vector for each window, changes significantly when the bearing operating state transitions from normal to abnormal. Finally, the state stationary probability vectors for each window are concatenated into a matrix, the state stationary probability matrix. The state stationary probability matrix is used as the input feature of the adaptive boosting classifier. Label data is also generated for the training set, marking the time ranges for normal and abnormal states. Feature sequence windows before bearing degradation occur are labeled as normal (labeled 0), and feature sequence windows after bearing degradation occur are labeled as abnormal (labeled 1). This labeled feature data is used as training samples to train the adaptive boosting classifier, thereby enabling anomaly detection in real-time vibration signals during bearing operation.

[0144] Step S7: Using the state stationary probability matrix and label data as input for model training of the adaptive boosting classifier.

[0145] Among them, such as Figure 4 As shown in Figure 1, the adaptive boosting classifier is an adaptive ensemble learning algorithm. Its core concept is to construct a strong classifier by combining multiple weak classifiers, thereby leveraging the strengths of each base classifier and overcoming the shortcomings of a single base classifier. The adaptive boosting classifier uses a forward distribution algorithm to rapidly reduce the loss function and gradually approach the optimal solution, thus simplifying the complexity of ensemble learning.

[0146] First, initialize the training sample weights, assuming that training samples, and initially each training sample weight is assigned the same value:

[0147] ;

[0148] in, Indicates the training samples; Indicates the initial The weight of the training samples.

[0149] In each round of iteration, a weak classifier is trained based on the current training sample distribution (the weight of the current training sample) , i.e., the decision stump, Indicates the Weak classifiers trained in rounds of iterations.

[0150] Compute weak classifiers Weighted classification error rate under the current training sample distribution (weight of the current training sample):

[0151] ;

[0152] in, Indicates the The weighted error rate of the weak classifier, Indicates the Wheel The weight of the training samples, It is an indicator function that takes the value 1 when the condition is met and 0 otherwise.

[0153] Calculate weak classifiers based on weighted error rates of classification Weight , Indicates the The weight of the weak classifier in round t. The lower the weighted error rate of the classification, the higher the weight of the weak classifier:

[0154] ;

[0155] For each training sample, its weight is updated according to the classification result.

[0156] ;

[0157] in, Indicates the next round In the round of iteration The weight of the training samples, is the true label of the training sample. The weight of the training sample that is misclassified will increase, and the weight of the training sample that is correctly classified will decrease.

[0158] In order to keep the sum of the weights of the training samples to 1, the weights of all training samples are normalized:

[0159] ;

[0160] in, Represents the temporary index variable when summing the weights of all training samples in the denominator.

[0161] The obtained weak classifiers are linearly combined according to the weights to obtain the final strong classifier:

[0162] ;

[0163] in, The output of the final strong classifier is is the total number of weak classifiers, Indicates the The weight of the round weak classifier, No. The output of the weak classifier is Represents the training sample data input to the weak classifier. =1 means that the strong classifier predicts that the input feature sequence window is abnormal. =0 means that the strong classifier predicts that the input feature sequence window is normal.

[0164] Step S8: Optimize the sliding window size and the number of symbols using the regularized cross entropy loss function to obtain the optimal number of symbols and the optimal window size.

[0165] Step S81, design loss function Cross-Entropy Loss is an indicator that measures the difference between the model's predicted probability distribution and the true label distribution. For binary classification problems, the cross-entropy loss function is The mathematical expression is:

[0166]

[0167] in, is the number of training samples, It is The true labels (0 or 1) of the training samples, is the first prediction of the adaptive boosting classifier The probability that a training sample belongs to the abnormal category.

[0168] Regularization is a technique to prevent overfitting by adding a regularization term to the cross entropy loss function to constrain the complexity of the adaptive enhancement classifier weights. The mathematical expression is:

[0169]

[0170] in, is the number of adaptive boosting classifier weights, It is weights, is the regularization coefficient, which controls the strength of the regularization term.

[0171] The regularized cross entropy loss function combines the cross entropy loss and the regularization term to form the final loss function :

[0172]

[0173] Step S82: Calculate the loss based on the final loss function. The regularized cross-entropy loss function takes the predicted probability distribution and the true label distribution as input, calculates the difference between the two, and outputs a specific loss value. This loss value measures the degree of match between the adaptive boosting classifier's predictions and the true labels. A smaller loss value indicates better prediction performance of the adaptive boosting classifier.

[0174] In step S83, the current number of symbols remains unchanged. In each loop, the sliding window size is adjusted sequentially, and training is performed using the adaptive boosting classifier. By comparing the loss values for different sliding window sizes, it is determined whether the change in loss value is less than a set loss tolerance (e.g., 0.001). If the change in loss value is less than the set loss tolerance, the optimal sliding window size and corresponding optimal loss value for the current number of symbols are determined. Further adjustment of the sliding window size is stopped, and the number of globally symbolized symbols is updated.

[0175] In step S84, step S83 is repeated using the global symbolized number updated in step S83 to continue optimizing the sliding window size. This process continues until the change in the optimal loss value is less than a set optimal loss value tolerance (e.g., 0.0001). At this point, the entire optimization process is terminated, and the optimal sliding window size is obtained. The global symbolized number updated in step S83 is the optimal symbol number.

[0176] Through steps S81 to S84, the optimal number of symbols and the optimal sliding window size can be determined, thereby providing the most ideal parameter configuration for subsequent bearing degradation starting point detection.

[0177] Step S9: Use the trained adaptive boosting classifier to predict the test set data to identify the starting point of the degradation process.

[0178] Combining the optimal number of symbols and the optimal window size, the trained adaptive boosting classifier is applied to predict the test set data to obtain the probability that each window belongs to the anomaly category.

[0179] Degradation typically occurs gradually. The degradation starting point is often the first significant change in the bearing's operating state from normal. As mentioned above, when the bearing's operating state transitions from normal to abnormal, the state stationary probability vector changes significantly, significantly increasing the probability that the corresponding window will be identified as abnormal by the adaptive boosting classifier. Therefore, when a window has the highest probability of belonging to the abnormal category, it effectively captures the degradation starting point.

[0180] The present invention selects the window with the highest probability and takes its starting position as the degradation starting point. The experimental results of bearing 1 and bearing 2 are as follows: Figure 5 and Figure 6 As shown, the feature sequence extracted by the autoencoder is represented by a gray solid line. Figure 5 and Figure 6 Statistical methods based on normal distribution are also introduced Threshold value is used as a comparison method. The threshold is marked with a gray dotted line, and the area outside this range is considered abnormal. The light gray area marks the time period when the data enters the degradation stage, and the degradation starting points predicted by the method proposed in this invention are respectively Figure 5 1358 in Figure 6 The value in the figure is 3663, indicating the degradation starting point predicted by the adaptive boosting classifier in the present invention. The results show that the proposed method can accurately identify the starting point of bearing degradation.

[0181] Example 2

[0182] Based on the same inventive concept as Example 1, this example introduces a bearing degradation starting point detection system based on an autoencoder and an AOMM, including:

[0183] A data acquisition module configured to acquire real-time vibration signals of the bearing;

[0184] A data preprocessing module is configured to preprocess the collected real-time vibration signal to obtain a preprocessed real-time vibration signal;

[0185] The autoencoder module is configured to extract features from the preprocessed real-time vibration signal using the autoencoder to obtain a feature sequence of the real-time vibration signal;

[0186] A symbolized feature sequence module is configured to globally symbolize the feature sequence of the real-time vibration signal according to the optimal number of symbols obtained through pre-training to obtain a symbolized feature sequence of the real-time vibration signal;

[0187] A sliding window segmentation module is configured to segment the symbolic feature sequence of the real-time vibration signal according to an optimal window size obtained through pre-training to obtain a plurality of sliding windows;

[0188] An AOMM method module is configured to apply the AOMM method, adaptively adjust the Markov order based on the symbolic distribution entropy and the polynomial regression model, calculate the state stationary probability vector for each sliding window, and concatenate all the state stationary probability vectors into a state stationary probability matrix;

[0189] The adaptive boosting classifier module is configured to input the state stationary probability matrix and label data into a pre-trained adaptive boosting classifier for prediction, and output a predicted degradation starting point.

[0190] Example 3

[0191] Based on the same inventive concept as other embodiments, this embodiment introduces a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned steps of detecting the starting point of bearing degradation based on the autoencoder and AOMM are implemented.

[0192] Example 4

[0193] Based on the same inventive concept as other embodiments, this embodiment introduces a computer device, including: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the above-mentioned bearing degradation starting point detection method based on autoencoder and AOMM.

[0194] Example 5

[0195] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned bearing degradation starting point detection method based on autoencoder and AOMM.

[0196] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0197] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0198] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0199] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0200] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms under the guidance of the present invention, which all fall within the protection of the present invention.

Claims

1. A bearing degradation starting point detection method based on autoencoder and AOMM, characterized in that: include: Collect real-time vibration signals of bearings; Preprocessing the collected real-time vibration signal to obtain a preprocessed real-time vibration signal; The autoencoder is used to extract features from the preprocessed real-time vibration signal to obtain a feature sequence of the real-time vibration signal; Globally symbolize the feature sequence of the real-time vibration signal according to the optimal number of symbols obtained by pre-training to obtain a symbolic feature sequence of the real-time vibration signal; The symbolic feature sequence of the real-time vibration signal is segmented according to the optimal window size obtained by pre-training to obtain several sliding windows; Applying the AOMM method, the Markov order is adaptively adjusted based on the symbolic distribution entropy and the polynomial regression model to calculate the state stationary probability vector of each sliding window, and all the state stationary probability vectors are spliced into a state stationary probability matrix; The state stability probability matrix and label data are input into the pre-trained adaptive boosting classifier for prediction, and the predicted degradation starting point is output; The method of adaptively adjusting the Markov order based on the symbolic distribution entropy and the polynomial regression model includes: Dynamically adjust the order of the Markov model in each window First, calculate the symbol distribution entropy change between the current window and the previous window, and add the current symbol distribution entropy change value to the historical symbol distribution entropy list. If the symbol distribution entropy change exceeds the set threshold, it means that the data complexity increases and the order of the Markov model needs to be increased. At this time, if the historical symbol distribution entropy data contains at least 3 data points, the historical symbol distribution entropy change value is used as the training data, and the polynomial regression model is used to fit the historical symbol distribution entropy change trend and predict the symbol distribution entropy change of the next window; When predicting the change in the symbol distribution entropy of the next window, use the fitted polynomial regression model, substitute the index of the next window into the polynomial regression model, obtain the predicted symbol distribution entropy change value, and adjust the order of the Markov model according to the prediction result. If the predicted symbol distribution entropy change still exceeds the preset symbol distribution entropy threshold, the order of the Markov model is greatly increased. , that is, increase by 2; otherwise, slightly increase the order of the Markov model , that is, it increases by 1; if the historical symbol distribution entropy data is insufficient, the order of the Markov model is directly increased slightly , that is, it increases by 1; if the symbol distribution entropy change is less than or equal to the preset symbol distribution entropy threshold, the order of the current Markov model is maintained constant.

2. The bearing degradation starting point detection method based on autoencoder and AOMM according to claim 1 is characterized in that: The method for obtaining the optimal number of symbols and the optimal window size is: Step S1, collecting the original vibration signal of the bearing throughout its life cycle; Step S2, preprocessing the original vibration signal to obtain a preprocessed original vibration signal; Step S3, using an autoencoder to extract features from the preprocessed original vibration signal to obtain a feature sequence of the original vibration signal; Step S4, globally symbolizing the feature sequence of the original vibration signal to obtain a symbolized feature sequence of the original vibration signal; Step S5, segmenting the symbolic feature sequence of the original vibration signal to obtain a plurality of sliding windows; Step S6, applying the AOMM method, adaptively adjusting the Markov order based on the symbolic distribution entropy and the polynomial regression model, calculating the state stationary probability vector of each sliding window, and splicing all the state stationary probability vectors into a state stationary probability matrix; Step S7: Input the state stationary probability matrix and the label data as training samples, train the adaptive boosting classifier, and obtain the probability that each training sample belongs to the abnormal category predicted by the adaptive boosting classifier; In step S8, based on the probability that each training sample predicted by the adaptive boosting classifier belongs to the abnormal category, the regularized cross entropy loss function is used to optimize the sliding window size and the number of symbols, and the optimized number of symbols is fed back to the step S4, and the optimized sliding window size is fed back to the step S5, and finally the optimal number of symbols and the optimal window size are obtained.

3. The bearing degradation starting point detection method based on autoencoder and AOMM according to claim 2 is characterized in that: The state stationary probability matrix and label data are input as training samples to train the adaptive boosting classifier, and the probability of each training sample belonging to the abnormal category predicted by the adaptive boosting classifier is obtained, including: Initialize the weights of the training samples; Based on the weights of the initialized training samples, the weak classifier is trained and the weighted classification error rate under the distribution of all training samples is calculated; Calculating the weight of the weak classifier according to the classification weighted error rate to obtain the classification result of each training sample; Update the weight of each training sample according to the classification result of the training sample; Normalize all updated training sample weights to obtain the normalized weight of each training sample; Determine whether the preset number of iterations has been reached. If not, return to the step of training the weak classifiers. If so, perform a linear combination of all weak classifiers according to the normalized weights to obtain the final strong classifier.

4. The method for detecting the starting point of bearing degradation based on autoencoder and AOMM according to claim 2, characterized in that: The regularized cross entropy loss function is used to optimize the sliding window size and the number of symbols, including: Design loss function : ； in, represents the cross entropy loss function; represents the regularization function; Cross Entropy Loss Function The mathematical expression is: ； in, is the number of training samples, It is The true label of the training sample is 0 or 1. is the first prediction of the adaptive boosting classifier The probability that a training sample belongs to the abnormal category; Regularization function The mathematical expression is: ； in, is the number of adaptive boosting classifier weights, It is weights, is the regularization coefficient, which controls the strength of the regularization term; According to the loss function Calculate the loss and output the loss value. The loss value is used to measure the degree of match between the prediction result of the adaptive boosting classifier and the true label. The smaller the loss value, the better the prediction performance of the adaptive boosting classifier. Keeping the current number of symbols unchanged, adjust the window size in turn in each loop and use the adaptive boosting classifier for training. By comparing the loss values under different window sizes, determine whether the change in loss value is less than the set loss value tolerance. If the change in loss value is less than the set loss value tolerance, determine the optimal window size and corresponding optimal loss value under the current number of symbols, stop further adjusting the window size, and update the number of globally symbolized symbols. Under the updated number of globally symbolized symbols, repeat the previous step and continue to optimize the window size until the change in the optimal loss value is less than the set optimal loss value tolerance. At this time, the entire optimization process is stopped and the optimal sliding window size is obtained. The corresponding updated number of globally symbolized symbols is the optimal number of symbols.

5. The method for detecting the starting point of bearing degradation based on autoencoder and AOMM according to claim 4, characterized in that: The method of extracting features from the pre-processed original vibration signal using an autoencoder to obtain a feature sequence of the original vibration signal includes: make For the input data set, that is, the set of preprocessed raw vibration signals, define , Represents the input dataset A feature vector in is the input dataset The index of the eigenvector in , , Represents the input dataset This collection contains feature vectors, that is, the size of the input dataset is , express dimensional real space, Represents the dimension of the feature vector; encoder The high-dimensional input dataset Pass from the input layer to the hidden layer, input the data set in the hidden layer Compressed into low-dimensional space In the definition , Representing low-dimensional space A low-dimensional feature vector in is the compressed feature representation. is the index of the eigenvector, ; Representing low-dimensional space This collection contains feature vectors, express dimensional real space, Represents the dimension of the compressed feature vector; then the decoder The output layer remaps the compressed feature vector to the input data , Output result: encoder and decoder The basic mathematical expression is: ； ； encoder Enter Convert to low-dimensional representation , then the decoder The low-dimensional representation Convert to output , to reconstruct the input .

6. The method for detecting the starting point of bearing degradation based on autoencoder and AOMM according to claim 2, characterized in that: The globally symbolizing the characteristic sequence of the original vibration signal to obtain the symbolized characteristic sequence of the original vibration signal includes: globally symbolizing the characteristic sequence of the original vibration signal using a uniform partitioning method; For feature sequences , is the sequence length, is the eigenvalue; initialization parameter , Indicates that the feature sequence How many symbols it is divided into, that is, the number of symbols; Calculate feature sequence The minimum value in and maximum value , as the boundary of the partition interval; The feature sequence The value range of intervals; the width of each interval is ; The feature sequence Each eigenvalue in is mapped to the corresponding symbol to obtain a symbolic feature sequence.

7. A bearing degradation starting point detection system based on autoencoder and AOMM, characterized in that: include: A data acquisition module configured to acquire real-time vibration signals of the bearing; A data preprocessing module is configured to preprocess the collected real-time vibration signal to obtain a preprocessed real-time vibration signal; The autoencoder module is configured to extract features from the preprocessed real-time vibration signal using the autoencoder to obtain a feature sequence of the real-time vibration signal; A symbolized feature sequence module is configured to globally symbolize the feature sequence of the real-time vibration signal according to the optimal number of symbols obtained through pre-training to obtain a symbolized feature sequence of the real-time vibration signal; A sliding window segmentation module is configured to segment the symbolic feature sequence of the real-time vibration signal according to an optimal window size obtained through pre-training to obtain a plurality of sliding windows; An AOMM method module is configured to apply the AOMM method, adaptively adjust the Markov order based on the symbolic distribution entropy and the polynomial regression model, calculate the state stationary probability vector for each sliding window, and concatenate all the state stationary probability vectors into a state stationary probability matrix; An adaptive boosting classifier module is configured to input the state stationary probability matrix and label data into a pre-trained adaptive boosting classifier for prediction, and output a predicted degradation starting point; The method of adaptively adjusting the Markov order based on the symbolic distribution entropy and the polynomial regression model includes: Dynamically adjust the order of the Markov model in each window First, calculate the symbol distribution entropy change between the current window and the previous window, and add the current symbol distribution entropy change value to the historical symbol distribution entropy list. If the symbol distribution entropy change exceeds the set threshold, it means that the data complexity increases and the order of the Markov model needs to be increased. At this time, if the historical symbol distribution entropy data contains at least 3 data points, the historical symbol distribution entropy change value is used as the training data, and the polynomial regression model is used to fit the historical symbol distribution entropy change trend and predict the symbol distribution entropy change of the next window; When predicting the change in the symbol distribution entropy of the next window, use the fitted polynomial regression model, substitute the index of the next window into the polynomial regression model, obtain the predicted symbol distribution entropy change value, and adjust the order of the Markov model according to the prediction result. If the predicted symbol distribution entropy change still exceeds the preset symbol distribution entropy threshold, the order of the Markov model is greatly increased. , that is, increase by 2; otherwise, slightly increase the order of the Markov model , that is, it increases by 1; if the historical symbol distribution entropy data is insufficient, the order of the Markov model is directly increased slightly , that is, it increases by 1; if the symbol distribution entropy change is less than or equal to the preset symbol distribution entropy threshold, the order of the current Markov model is maintained constant.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for detecting the starting point of bearing degradation based on an autoencoder and an AOMM as described in any one of claims 1 to 6 are implemented.

9. A computer device, characterized in that: include: Memory for storing computer programs; A processor is used to execute the computer program to implement the steps of the bearing degradation starting point detection method based on autoencoder and AOMM according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that: When the computer program is executed by a processor, the steps of the bearing degradation starting point detection method based on an autoencoder and an AOMM according to any one of claims 1 to 6 are implemented.

Images