Rolling bearing health index construction method based on stacked convolution denoising auto-encoder
The construction of rolling bearing health indicators through stacked convolutional denoising autoencoder solves the problem that traditional methods cannot accurately evaluate rolling bearing health status, realizes accurate assessment of rolling bearing health status and residual life prediction, and reduces operating costs.
Patent Information
- Application Number
- CN202510373977.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-27
- Publication Date
- 2025-07-25
AI Technical Summary
It is difficult for the prior art to accurately evaluate the health status of rolling bearings, especially in the nonlinear degradation process. Traditional health indicator construction methods cannot fully characterize their degradation trend, resulting in poor predictive maintenance results.
Using a stacked convolutional denoising autoencoder method, through signal acquisition, feature extraction, feature selection and model construction, health indicators that can fully characterize the health status of rolling bearings are constructed, including signal acquisition, time-frequency domain feature extraction, monotonicity, robustness and trend indicator weighted fusion, stacked convolutional denoising autoencoder model training and Bagging integrated learning network.
Accurate assessment of the health status of rolling bearings is achieved, repair and maintenance plans can be arranged reasonably, operating costs are reduced, and good data support is provided for the remaining service life forecast.
Smart Images
Figure CN120372242A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of health assessment of rolling bearings, and particularly to a method for constructing a health index of rolling bearings based on a stacked convolutional denoising autoencoder. Background Art
[0002] With the rapid development of computer and sensor technologies, various monitoring data of mechanical equipment have gradually provided extensive applications for research. As an indispensable important part of mechanical equipment in industries, military, aerospace and other fields, accurately evaluating the health status of rolling bearings is very necessary for subsequent predictive maintenance and replacement to reduce operating costs and accidents.
[0003] Currently, mechanical health prediction mainly includes model-based and data-driven methods. Model-based health prediction mainly predicts the degradation process by constructing a mechanism model of system behavior. Although this method has high accuracy in prediction, it requires a large amount of expert knowledge and human experience, and due to the complexity of industrial scenarios, its generality is low; data-driven health prediction uses machine learning and artificial intelligence technologies to mine the internal laws and patterns from a large amount of data, and constructs virtual indicators through neural networks to represent the health status of key components and perform regression prediction. Compared with the mechanism model method, this method has higher generality. For the nonlinear degradation process of bearings, the data-driven method has strong adaptive feature extraction ability and nonlinear fitting ability.
[0004] The data-driven health prediction method usually consists of three steps: data processing, health index construction, and remaining life prediction. Constructing a good health index is crucial for subsequent prediction of the remaining service life of machinery. Currently, in order to represent its health status, it is mainly constructed from two aspects: physical indicators and virtual indicators. Physical health indicators are related to the physical characteristics of mechanical faults. For example, for rolling bearings, parameters such as the root mean square, peak value in the time domain or the frequency mean value in the frequency domain are usually extracted from vibration signals as curves representing the health trend of the bearings. However, using only one feature often fails to accurately determine its degradation state due to the complexity of mechanical faults and the nonlinearity of trends. Virtual health indicators, by fusing multiple physical health indicators, contain multiple characteristics of the bearings compared with single features and can more accurately reflect the degradation trend of the bearings. Dimensionality reduction methods such as principal component analysis and linear discriminant analysis are often used to construct virtual health indicators, but for the nonlinear degradation process of bearings, the above methods are difficult to be convincing and cannot fully represent the health status of rolling bearings. Summary of the Invention
[0005] In view of this, the present invention provides a method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder. Through this method, a health index that can fully characterize the health state of a rolling bearing can be constructed. Based on this index, it is convenient to reasonably arrange equipment maintenance and repair plans, reduce operating costs, and is conducive to providing good data for subsequent prediction of the remaining service life.
[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:
[0007] In the first aspect, the present invention provides a method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder. The method includes the following steps:
[0008] Step 1, signal acquisition: Collect vibration signals, temperature signals, acoustic signals, etc. of the rolling bearing under different working conditions.
[0009] Step 2, feature extraction: Process the signals collected in Step 1, select the horizontal vibration signal for subsequent research, and extract time-domain, frequency-domain, and time-frequency domain features of the rolling bearing full-life cycle data as initial features.
[0010] Step 3, feature selection: Based on the initial features extracted in Step 2, construct a comprehensive index by weighted fusion of monotonicity, robustness, and trendiness indicators for feature selection. Select features with a threshold greater than 0.5 as the feature subset, and normalize the features.
[0011] Step 4, construct the training set and the test set: Based on the feature subset screened in Step 3, use the first two bearing data of each working condition as the training set, and the remaining bearings as the test set.
[0012] Step 5, construct the model: Construct an autoencoder model based on a convolutional neural network, construct an ensemble learning network through the Bagging method, add noise to the input data, input the training set into the stacked convolutional denoising autoencoder model for training, input the test set into the trained model, and perform feature dimensionality reduction through the encoder to obtain a one-dimensional latent variable Z as the health index characterizing the degradation trend of the rolling bearing.
[0013] Step 6, evaluate the health index constructed in Step 5: Evaluate the health index based on the trendiness and robustness indicators in Step 3.
[0014] Further, the specific content of Step 1 includes:
[0015] Use an acceleration sensor to collect the vibration signal of the rolling bearing, use an acoustic sensor to collect the acoustic signal, and use a temperature sensor to collect the temperature signal. The vibration signal includes the vibration signal in the horizontal direction and the vibration signal in the vertical direction. Select the horizontal vibration signal as the original data.
[0016] Further, step 2 specifically includes:
[0017] Extract multi-domain features of the horizontal vibration signal, including time-domain features, frequency-domain features, and time-frequency domain features as initial features. Among them, the time-domain features include 8 dimensional features: root mean square, peak value, peak-to-peak value, standard deviation, variance, absolute average amplitude, kurtosis, and skewness; 4 dimensionless features: kurtosis factor, waveform factor, pulse factor, and peak factor. The frequency-domain features include average frequency, center frequency, mean square frequency, and frequency variance. The time-frequency domain features include the energy ratios of 8 sub-signals obtained by 3-layer wavelet packet decomposition and the wavelet packet energy entropy obtained according to the energy proportion of each sub-signal.
[0018] Further, step 3 specifically includes:
[0019] Screen the initial features extracted in step 2 based on the comprehensive evaluation index obtained by the weighted fusion of monotonicity, robustness, and trendiness indicators, select the features with a threshold greater than 0.5 as the feature subset for subsequent model training, and normalize the data. The relevant index formulas are as follows:
[0020] Monotonicity:
[0021]
[0022] Among them, F represents the extracted feature, N represents the number of samples in the feature sequence; d / dF represents the difference in the feature sequence; NO.(d / dF>0) and NO.(d / dF<0) respectively represent the number of values greater than 0 and less than 0; the range of Mon(F) is [0, 1], and the closer it is to 1, the better the monotonicity of the feature.
[0023] Robustness:
[0024]
[0025] Among them, Fn represents the feature value corresponding to the nth sample; represents the average trend value obtained by smoothing the nth sample.
[0026] Trendiness:
[0027]
[0028] Among them, t n represents the time corresponding to the nth sample, F n represents the value of the feature at time tn, and the range of Tre(F, t) is between [-1, 1]. When the trend is closer to 1 or -1, it proves that the feature has a strong positive trend or negative trend over time.
[0029] In order to select the optimal feature subset, the present invention performs weighted fusion based on the above indexes, obtains a comprehensive index to evaluate the initial features, and the formula is as follows:
[0030] Cri = v1Mon(F) + v2Rob(F) + v3Tre(F, t)
[0031]
[0032] According to the influence of the health index on the prediction result of the remaining service life of the rolling bearing, the present invention assigns weights of 0.5, 0.2, and 0.3 to monotonicity, robustness, and trend respectively. Through screening, features with a comprehensive index greater than 0.5 are selected as the feature subset.
[0033] Use the [max, min] normalization method to make the data range of the feature subset between [0, 1]. The formula is as follows:
[0034]
[0035] Further, the specific steps of step 4 include:
[0036] Construct a training set. Select the feature subsets of the first two rolling bearings in each working condition as the training set {(X i , Y i )}, where X = [x1, x2... x9] N represents the feature subset obtained by screening through the weighted index in step 3, N is the number of samples, i is the bearing number, and X i = [1, 6]. y i is the life label of the i-th bearing.
[0037] Construct a test set. Select the feature subsets of the remaining rolling bearings as the test set {(X j )}, X j = [7, 17].
[0038] Further, the specific steps of step 5 include:
[0039] Construct a stacked convolutional denoising autoencoder: The model structure extracted by the present invention includes 14 layers: including 3 convolutional layers, 3 pooling layers, 1 flattening layer, 1 fully connected layer, 3 deconvolutional layers, and 3 de-pooling layers.
[0040] Construct a Bagging ensemble learning network. Process the input training set and test set to reduce the generalization error of the training model.
[0041] For the training set data X iAdd Gaussian noise and input it into the model for training. The encoding layer of the model mainly includes alternating connections of convolutional layers and pooling layers, deeply extracting the non-linear features in the input data to reduce noise interference.
[0042] The definition of convolution is as follows:
[0043]
[0044] Where represents the output at position i of the j-th channel in the l-th layer, represents the data participating in the convolution calculation in the l-th layer, represents the convolution kernel of size K in the l-th layer, b is the bias of the l-th layer, and * represents the convolution operation.
[0045] After the input data undergoes the convolution operation, an activation function is added to perform a non-linear transformation on the output value to better reflect the monotonicity of the data. In the present invention, tanh (hyperbolic tangent activation function) is selected to process the output value.
[0046] In order to reduce the subsequent computational complexity, a pooling layer is added to reduce the dimension of the high-dimensional features obtained from the previous convolutional layer to reduce the feature complexity. Currently, the main types of pooling layers are the max pooling layer, average pooling layer, global max pooling layer, and global average pooling layer. In the present invention, the max pooling layer is used. The formula is as follows:
[0047]
[0048] Where max represents taking the maximum value of the pooling layer window, represents the elements participating in the pooling layer calculation in the l-th layer, m is the size of the pooling layer window, and s is the moving step of the pooling layer window.
[0049] After sufficient training of the data through the alternating of multiple convolutional layers and pooling layers, it is then passed through a flattening layer and a fully connected layer, and the latent variable Z is obtained through the non-linear transformation of the fully connected layer.
[0050] Z l+1 = f(w l FC l + b)
[0051] Where f(·) is the activation function, w l represents the weight connecting the l-th layer and the l+1-th layer, FC l is the input of the fully connected layer in the l-th layer, FC = [c1, c2, c3…c n T , and FC is a column of n data.
[0052] Similarly, the process of the decoding layer is the opposite of the encoding process. Taking the latent variable Z as the input, it is reconstructed into a new data X' similar to the input data through the decoding layer composed of alternating transposed convolution layers and max unpooling layers.
[0053] Among them, based on the similarity between the input data and the reconstructed data, the formula of the model loss function is defined as follows:
[0054]
[0055] X and X' are the original data and the reconstructed data respectively.
[0056] The extracted health indicator Z is smoothed using the weighted exponential moving average method to improve the comprehensive performance of the health indicator. The formula is as follows:
[0057] E(t) = λZ(t) + (1 - λ)E(t - 1)
[0058] E(t) is the smoothed value at time t; Z(t) is the actual value at time t (t = 1, 2, 3,... n); λ is the smoothing factor (0 < λ < 1).
[0059] Further, the specific steps of step 6 include:
[0060] Based on the trend and robustness indicators in step 3, evaluate whether the health indicators extracted from the test set meet the requirements.
[0061] Perform an ablation experiment. Input the training set and test set data into a convolutional autoencoder, a convolutional neural network, and an autoencoder to extract health indicators, calculate the trend and robustness, and verify the effectiveness of the method of the present invention.
[0062] In a second aspect, the present invention also provides a storage medium, on which one or more programs readable by a computing device are stored. The one or more programs include instructions that, when executed by the computing device, cause the computing device to execute a method for constructing a rolling bearing health indicator based on a stacked convolutional denoising autoencoder as described above.
[0063] In a third aspect, an embodiment of the present invention also provides an electronic device, including a processor and a memory. The memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement a method for constructing a rolling bearing health indicator based on a stacked convolutional denoising autoencoder as described above.
[0064] Through the above technical solutions, compared with the prior art, the present invention discloses a method for constructing a rolling bearing health indicator based on a stacked convolutional denoising autoencoder, which has at least the following beneficial effects:
[0065] 1. The comprehensive index obtained by weighted fusion based on monotonicity, robustness, and trend can accurately screen out redundant features and low-correlation features that are irrelevant to the original vibration signal compared to traditional dimensionality reduction methods.
[0066] 2. It is difficult for single features based on the time domain and frequency domain to fully characterize the health state of rolling bearings. The health index constructed by a stacked convolutional denoising autoencoder proposed in the present invention can fully characterize the degradation state of rolling bearings by deeply extracting the non-linear trends from multiple features. Based on the index, reasonable maintenance and repair plans can be arranged to reduce costs and provide good data for accurately predicting the remaining service life of rolling bearings subsequently.
[0067] 3. Aiming at the non-linear degradation process of rolling bearings, compared with traditional linear dimensionality reduction methods such as PCA and other non-linear dimensionality reduction methods, the method provided by the present invention can effectively learn non-linear features in the data.
[0068] Other features and advantages of the present invention will be described in the following specification, and part of them will be obvious from the specification or understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written specification and the drawings.
[0069] The technical solutions of the present invention will be further described in detail below through the drawings and embodiments. Description of the Drawings
[0070] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0071] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention.
[0072] Figure 1 It is a schematic flow chart of the method for constructing the health index of rolling bearings based on a stacked convolutional denoising autoencoder provided by the embodiment of the present invention.
[0073] Figure 2 It is a schematic diagram of the working principle of the method for constructing the health index of rolling bearings based on a stacked convolutional denoising autoencoder provided by the embodiment of the present invention.
[0074] Figure 3 It is a schematic diagram of the vibration signals of rolling bearings 1-3 in the time domain provided by the embodiment of the present invention.
[0075] Figure 4 Schematic diagram of the comprehensive evaluation index curve of the training set bearing provided by the embodiment of the present invention.
[0076] Figure 5 Schematic diagram of the health index curve constructed for bearings 1-3 in the test set provided by the embodiment of the present invention.
[0077] Figure 6 Schematic diagram of the health index curve of a part of the test set provided by the embodiment of the present invention.
[0078] Figure 7 Schematic diagram of the structure of the electronic device provided by the embodiment of the present invention. Detailed implementation manners
[0079] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention.
[0080] In the description of the present invention, it should be noted that in some processes described in the specification and drawings of this application, there are multiple operations that appear in a specific order, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. In addition, various serial numbers, etc. are only for descriptive purposes and cannot be understood as indicating or implying relative importance.
[0081] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.
[0082] See Figure 1 and Figure 2 As shown, the embodiment of the present invention provides a method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder. Through this method, a health index that can fully characterize the health state of the rolling bearing can be constructed, and based on the index, the equipment maintenance and repair plan can be reasonably arranged, the operation cost can be reduced, and good data can be provided for predicting the remaining service life in the future. Next, each step will be described in detail.
[0083] Step 1, signal acquisition: Collect vibration signals, temperature signals, sound signals, etc. of the rolling bearing under different working conditions: The signal acquisition method is a prior art, and in the present invention, the horizontal vibration signal of the rolling bearing is mainly collected as the initial data.
[0084] In the present invention, the rolling bearing accelerated life test data provided by the FEMTO-ST Research Institute of a certain country on the PRONOSTIA experimental platform is used to verify this method. This experiment is specifically designed for research on bearing fault diagnosis and prediction. By applying fixed loads and rotational speeds to the bearings, experiments are conducted on 17 rolling bearings under three working conditions. In this experiment, the bearing life is estimated through an accelerometer fixed on the outer ring of the rolling bearing. When the acceleration of the vibration signal exceeds 20g, the bearing is considered to have reached its final life. The specific parameters of the bearings are shown in Table 1:
[0085] Table 1 Basic parameters of the experimental bearings
[0086]
[0087] Among them, the data of the first two rolling bearings in working condition 1, working condition 2, and working condition 3 are used as the training set, and the remaining bearings are used as the test set.
[0088] Furthermore, horizontal vibration signals and vertical vibration signals of the rolling bearing are collected through vibration sensors, and temperature signals are collected through temperature sensors. Among them, the acceleration is sampled at a sampling frequency of 25.6KHz, and the temperature measurement is sampled at a sampling frequency of 0.1Hz. Each sampling interval is 10s, and each sampling time is 0.1s. Figure 3 It is the time-domain waveform diagram of the horizontal vibration signals of rolling bearings 1 - 3 for the feature set.
[0089] Step 2: Process the vibration signals collected in Step 1. In the embodiment of the present invention, the vibration signals of the rolling bearing collected by the horizontal sensor are used to extract the initial features of the full life cycle data. Time-domain, frequency-domain, and time-frequency domain features are extracted from the rolling bearing vibration signals as the initial features.
[0090] In this embodiment, the initial features of the vibration signals include 12 time-domain features, 4 frequency-domain features, and 9 time-frequency domain features. Among them, the time-domain features include 8 dimensional features: root mean square, peak value, peak-to-peak value, standard deviation, variance, absolute average amplitude, kurtosis, and skewness; 4 dimensionless features: kurtosis factor, waveform factor, impulse factor, and peak factor. The frequency-domain features include average frequency, centroid frequency, mean square frequency, and frequency variance. The time-frequency domain features include the energy ratios of 8 sub-signals obtained through 3-layer wavelet packet decomposition and the wavelet packet energy entropy obtained according to the energy proportion of each sub-signal.
[0091] Step 3: Perform feature selection on the initial features extracted in Step 2 based on the comprehensive index obtained by weighted fusion of monotonicity, robustness, and trendiness indicators, and screen out the feature subset highly correlated with the original data; the specific process is as follows:
[0092] Step 31. Feature selection, as one of the important steps in data processing, mainly aims to eliminate variables with relatively low relevance and redundant variables, and reduce the impact on the subsequent construction of health indicators. Therefore, good feature parameters need to change with the change of the performance degradation process of the rolling bearing. Monotonicity, robustness, and trendiness can well reflect the relevant characteristics of the features. The main formulas are as follows:
[0093] Monotonicity:
[0094]
[0095] Among them, F represents the extracted feature, N represents the number of samples in the feature sequence; d / dF represents the difference of the feature sequence; N O .(d / dF>0) and N O .(d / dF<0) respectively represent the number of values greater than 0 and less than 0; the range of Mon(F) is [0, 1], and the closer it is to 1, the better the monotonicity of the feature.
[0096] Robustness:
[0097]
[0098] Among them, F n represents the feature value corresponding to the nth sample; represents the average trend value obtained by smoothing the nth sample.
[0099] Trendiness:
[0100]
[0101] Among them, t n represents the time corresponding to the nth sample, F n represents the value of the feature at t n . The range of Tre(F, t) is between [-1, 1]. When the trend is closer to 1 or -1, it proves that the feature has a strong positive or negative trend over time.
[0102] Step 32. In order to select the best feature subset, the present invention performs weighted fusion based on the above three indicators to obtain a comprehensive indicator to evaluate the initial features. The formula is as follows:
[0103] Cri = v1Mon(F) + v2Rob(F) + v3Tre(F, t)
[0104]
[0105] The present invention assigns weights of 0.5, 0.2, and 0.3 to monotonicity, robustness, and trend respectively according to the influence of the health indicator on the prediction result of the remaining service life of the rolling bearing. The results are asFigure 4 As shown. Sort the features according to the size of the comprehensive index, and select the features with a threshold greater than 0.5 after comprehensive consideration, that is, these nine features, namely F21, F16, F15, F10, F6, F4, F1, F14, and F13, constitute the feature subset.
[0106] Step 33, use the [max, min] normalization method for the filtered feature subset to make its data range between [0, 1]. The formula is as follows:
[0107]
[0108] Step 4, construct the training set and the test set. The specific process is as follows:
[0109] Step 41, construct the training set: Select the features obtained from the comprehensive index evaluation of the first two bearing data of each working condition as the training set data. That is, T Train ={(X1, Y1), (X2, Y2)...(X6, Y6)}, where X = [x1, x2…x9] N represents the 9 features obtained by screening in Step 3, and N represents the number of samples in the feature subset X.
[0110] Step 42, set the life label. For example: The total life time length T of a bearing, that is:
[0111] Y = y i / T, y i represents the current life length.
[0112] Step 43, use the remaining bearings as the test set. T Test ={(X7), (X8)...(X 17 )}.
[0113] Step 5, construct a stacked convolutional denoising autoencoder model, train and test the model, and construct a health index. The specific process is as follows:
[0114] Step 51, construct the model.
[0115] The autoencoder is an unsupervised neural network model, which consists of an encoder and a decoder. By inputting the data set into the encoder for encoding, it learns the non-linear features of the data to achieve the purpose of data compression.
[0116] As an improved autoencoder model, the stacked convolutional denoising autoencoder is characterized in that both the encoder and decoder parts use convolutional layers to replace the fully connected layers therein. Compared with the ordinary autoencoder, the stacked convolutional denoising autoencoder can greatly improve the extraction of the non-linear feature trend by the model, so that the health index obtained after dimensionality reduction can fully characterize the degradation trend of the rolling bearing.
[0117] The stacked convolutional denoising autoencoder strengthens the feature extraction ability of the model by inputting data and adding noise to it. The output of the previous convolutional layer after passing through the pooling layer serves as the input for the next convolutional layer. The encoding layer consists of 3 convolutional layers and 3 max pooling layers. Through a flattening layer and a fully connected layer, the latent variable Z is output. The decoding layer consists of a reconstruction layer, 3 transposed convolutional layers, and 3 max unpooling layers, finally obtaining reconstructed data similar to the original data.
[0118] Step 52: Construct a Bagging ensemble learning network. By introducing randomness, the diversity of model learning is enhanced, improving the accuracy and generalization ability of the model prediction results.
[0119] Step 53: Add Gaussian noise to the input data.
[0120] Step 54: Input the noisy data into the constructed stacked convolutional denoising autoencoder model for training.
[0121] Step 55: Input the test set data into the trained model and output the required health indicators. Figure 5 Health indicators constructed for bearings 1 - 3 in the test set.
[0122] Step 56: Smooth the health indicators. In the present invention, the exponential weighted moving average method is used to smooth the health indicators. The formula is as follows:
[0123] E(t) = λX(t) + (1 - λ)E(t - 1)
[0124] E(t) is the smoothed value at time t; X(t) is the actual value at time t (t = 1, 2, 3,... n); λ is the smoothing factor (0 < λ < 1)
[0125] Figure 6 Shows some of the health indicators of the test set obtained after the above smoothing method.
[0126] Step 6: Evaluate the health indicators. Table 2 shows the trends and robustness of the health indicators constructed for bearings 1 - 3 in the test set under four methods: stacked convolutional denoising autoencoder (SCDAE), convolutional autoencoder (CAE), convolutional neural network (CNN), and autoencoder (AE).
[0127] Table 2 Comparison results of the four methods
[0128]
[0129] The method for constructing health indicators proposed in the present invention has larger indicator values than those obtained by other methods, proving that this method can well construct health indicators that can characterize the operating conditions of rolling bearings.
[0130] In the present invention, the original features are selected by using a comprehensive index obtained by weighted fusion of monotonicity, trendiness, and robustness indicators, and the features highly correlated with the remaining data are selected as the feature subset according to the feature trend. A stacked convolutional denoising autoencoder based on the Bagging method is constructed to further reduce the dimension of the feature subset to construct a health index that can fully characterize the health state of the bearing. The present invention can accurately evaluate the health state of the bearing, and this method can be widely applied to the health assessment of key components of bearings in mechanical equipment in fields such as chemical engineering, electric power, and aerospace.
[0131] From the description of the above embodiments, those skilled in the art can know that the present invention provides a method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder by combining a convolutional neural network with an autoencoder model. This model can effectively learn non-linear features and construct a trend curve that can effectively characterize the health state of the bearing, which plays an important role in accurately predicting the remaining service life of the rolling bearing subsequently.
[0132] Furthermore, an embodiment of the present invention also provides a storage medium on which one or more programs readable by a computing device are stored. The one or more programs include instructions that, when executed by the computing device, cause the computing device to execute the above-mentioned method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder.
[0133] Examples of the computer-readable storage medium herein include: read-only memory (ROM), programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, DVD-ROM, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), cartridge memory (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk, and any other device configured to store a computer program and any associated data, data files, and data structures in a non-transitory manner and to provide the computer program and any associated data, data files, and data structures to a processor or computer such that the processor or computer can execute the computer program. The computer program in the computer-readable storage medium can run in an environment deployed in computer devices such as clients, hosts, proxy devices, servers, etc. In addition, in one example, the computer program and any associated data, data files, and data structures are distributed on a networked computer system such that the computer program and any associated data, data files, and data structures are stored, accessed, and executed in a distributed manner by one or more processors or computers.
[0134] Further, with reference to Figure 7 shown, an embodiment of the present invention further provides an electronic device that can execute the above method. The electronic device may include a processor 10, a memory 11, a communication bus 12, and a communication interface 13, and may further include a computer program stored in the memory 11 and executable on the processor 10.
[0135] Among them, the processor 10 may be composed of integrated circuits in some embodiments. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same or different functions packaged together, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting various components of the entire electronic device through various interfaces and lines, and executing various functions of the electronic device and processing data by running or executing programs or modules stored in the memory 11 and calling data stored in the memory 11.
[0136] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, electronic devices, computer program products, etc. Therefore, the present invention can take the form of a complete software embodiment, a complete hardware embodiment, or an embodiment combining software and hardware aspects.
[0137] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The present invention can be implemented by means of hardware including several different components and by means of a suitably programmed computer.
[0138] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other.
[0139] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for constructing a health index of a rolling bearing based on a stacked convolutional denoising autoencoder, characterized in that, The method includes the following steps: Step 1: Collect signals of rolling bearings under different working conditions; Step 2: Process the collected signals, and extract multi-domain features of the horizontal vibration signals of the rolling bearings as initial features; Step 3: Construct a comprehensive evaluation index based on monotonicity, robustness, and trend indexes, screen the initial features through the comprehensive evaluation index, and select the features that meet the preset conditions as the feature subset; Step 4: Based on the feature subset obtained by screening, construct a training set and a test set; Step 5: Construct a stacked convolutional denoising autoencoder model, input the training set into the stacked convolutional denoising autoencoder model for training, input the test set into the trained model, and perform feature dimensionality reduction through the encoder to obtain potential one-dimensional variables, which are used as health indexes representing the degradation trend of the rolling bearings; Step 6: Evaluate the health index based on robustness and trend indexes.
2. The method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder according to claim 1, wherein In Step 1, sensors are used to collect vibration signals, acoustic signals, and temperature signals of the rolling bearings. Among them, the vibration signals include horizontal vibration signals and vertical vibration signals, and the horizontal vibration signals are selected as the original data.
3. A method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder according to claim 1, characterized in that In Step 2, the multi-domain features include time-domain features, frequency-domain features, and time-frequency domain features, where: The time-domain features include dimensional features and dimensionless features; the dimensional features include: root mean square, peak value, peak-to-peak value, standard deviation, variance, absolute average amplitude, kurtosis, and skewness; the dimensionless features include: kurtosis factor, waveform factor, pulse factor, and peak factor; The frequency-domain features include average frequency, center frequency, mean square frequency, and frequency variance; The time-frequency domain features include the energy ratio of sub-signals obtained by wavelet packet decomposition and the wavelet packet energy entropy obtained according to the energy proportion of each sub-signal.
4. A method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder according to claim 1, characterized in that In Step 3, a comprehensive evaluation index is constructed by weighted fusion of monotonicity, robustness, and trend indexes. The expression of the comprehensive evaluation index is: Cri = v1Mon(F) + v2Rob(F) + v3Tre(F,t) where, Cri represents the value of the comprehensive evaluation index; Mon(F) represents the value of the monotonicity index; Rob(F) represents the value of the robustness index; Tre(F,t) represents the value of the trend index; v1, v2, and v3 are the weights of the monotonicity, robustness, and trend index values respectively; Select the features with a threshold greater than 0.5 as the feature subset for subsequent model training, and normalize the features.
5. A method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder according to claim 1, characterized in that In Step 4, based on the feature subset obtained by screening in Step 3, the first two bearing data of each working condition are used as the training set, and the remaining bearings are used as the test set, and life labels are added to the training set data.
6. The method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder according to claim 1, characterized in that Step 5 includes the following steps: Construct a stacked convolutional denoising autoencoder model, and the model includes: convolutional layer, pooling layer, flattening layer, fully connected layer, deconvolutional layer, and de-pooling layer; Construct an ensemble learning network based on the Bagging method and add noise to the input data; Input the training set data with added noise into the stacked convolutional denoising autoencoder model for training; Input the test set data into the trained model, and extract the potential one-dimensional variables, which are the health indexes; Use the weighted exponential moving average method to smooth the health indexes.
7. A method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder according to claim 6, characterized in that The smoothing of the health index using the weighted exponential moving average method has the following calculation formula: E(t) = λZ(t) + (1 - λ)E(t - 1) Where, E(t) is the smoothed value at time t; Z(t) is the actual value at time t, and λ is the smoothing factor.
8. A storage medium having stored thereon one or more programs readable by a computing device, characterized in that: The one or more programs include instructions that, when executed by a computing device, cause the computing device to execute a method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder as described in any one of claims 1 to 7.
9. An electronic device, characterized in that, It includes a processor and a memory, the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement a method for constructing a rolling bearing health index based on a stacked convolutional denoising autoencoder as described in any one of claims 1 - 7.