Vacuum dry pump multi-fault prediction method and system based on improved C4.5

Abstract

The invention discloses a vacuum dry pump multi-fault prediction method and system based on improved C4.5, and the method comprises the steps: constructing a decision tree model through an improved C4.5 algorithm, and carrying out the training through the full-life vibration signal data of a vacuum dry pump, and obtaining a trained multi-fault prediction model; the method comprises the following steps: acquiring vibration signal data of real-time operation monitoring of a vacuum dry pump, and preprocessing the data; and inputting the preprocessed vibration signal data into the trained multi-fault prediction model to predict whether a fault occurs in the vibration signal data, and outputting a fault prediction result which comprises a fault type and fault occurrence time. Through the transparent decision process, real-time updating, fine feature extraction and strong multi-fault classification capability, the defects in the prior art are overcome, various overlapping faults of the vacuum dry pump can be effectively predicted, and timely decision support is provided for equipment maintenance.

Description

Technical Field

[0001] The present invention relates to equipment health monitoring, and in particular to a vacuum dry pump multi-fault prediction method and system based on improved C4.5. Background Art

[0002] During the operation of the vacuum dry pump, dust will accumulate in multiple locations such as the air inlet, upper and lower pumps. As the operating time increases, the amount of dust residue gradually increases. When it reaches a certain threshold, the squeezing effect of the dust may cause scratches between the mechanical parts inside the pump. This scratch will not only accelerate the wear of the equipment, but may also cause serious failures of the equipment, causing the vacuum dry pump to shut down. In addition, many external or internal interferences, including dust jamming, and faults caused at different locations of the pump will show multiple fault characteristics in the vibration signal data, making fault diagnosis extremely complicated. The superposition effect of multiple faults increases the difficulty of fault identification and prediction, and brings challenges to the normal operation and maintenance of the equipment.

[0003] At present, the fault prediction methods of vacuum dry pumps are mainly divided into physical models and data-driven methods. Physical models predict the status of equipment by establishing complex mathematical relationships. Although they have high accuracy under specific conditions, they usually rely on a lot of expert knowledge and are difficult to adapt to complex operating environments. Data-driven methods, such as deep learning, clustering, Transformer, XGBoost, etc., although the existing prediction methods have improved the fault detection capabilities to some extent, they still face problems such as real-time performance, lack of accuracy, many false positives and missed negatives, weak fault classification capabilities, and excessive reliance on historical data. In particular, when dealing with specific fault modes, the recognition capabilities of existing methods are still insufficient. Summary of the invention

[0004] The purpose of the present invention is to solve the problems in the prior art.

[0005] The technical solution adopted by the present invention to solve the technical problem is: to provide a vacuum dry pump multi-fault prediction method based on improved C4.5, comprising:

[0006] The decision tree model was constructed using the improved C4.5 algorithm and trained using the full life vibration signal data of the vacuum dry pump to obtain a trained multi-fault prediction model.

[0007] Acquiring vibration signal data of real-time operation monitoring of the vacuum dry pump, and preprocessing the data;

[0008] The preprocessed vibration signal data is input into the trained multi-fault prediction model to predict whether a fault occurs in the vibration signal data, and a fault prediction result is output, wherein the prediction result includes the fault type and the fault occurrence time.

[0009] Preferably, the method of constructing a decision tree model using the improved C4.5 algorithm and training the model using the full life vibration signal data of the vacuum dry pump comprises the following steps:

[0010] Use the improved C4.5 algorithm to build a decision tree model;

[0011] Obtain the full-life vibration signal data of the vacuum dry pump and perform preprocessing, label the preprocessed data, and obtain a data set including sample data and corresponding labels;

[0012] Extracting features from the data set to obtain a feature set; the features of the feature set include time domain features and frequency domain features;

[0013] The decision tree model is trained using the feature set.

[0014] Preferably, the use of the improved C4.5 algorithm to construct a decision tree model comprises the following steps:

[0015] The attribute weight mechanism is introduced into the original C4.5 algorithm to obtain the improved C4.5 algorithm;

[0016] The improved C4.5 algorithm selects the split point by the information gain rate to construct a decision tree model, which is expressed as:

[0017] M=C4.5(F train , Y train );

[0018] Among them, F train represents the training features, Y trai Denotes the training feature F train The information gain rate of the improved C4.5 algorithm is expressed as:

[0019] G ′ (A) = FII(A,t)·G(A);

[0020] FII(A,t)=Var(A,t)×Exp(A)×λ(t);

[0021]

[0022] Among them, FII(A,t) is the weight of attribute A, G(A) is the information gain rate of attribute A calculated by the original C4.5 algorithm, and G ′(A) is the information gain rate of attribute A calculated by the improved C4.5 algorithm; Var(A,t) represents the variance of attribute A in the time window t, reflecting its degree of change in the time period; Exp(A) represents the empirical importance index of attribute A evaluated by experts, which is set based on expert knowledge; λ(t) represents the time adjustment coefficient, which is used to dynamically adjust the feature weight according to time changes; tk is the starting point of the time window, and Var(A,t-1) is the variance of the previous time window. The adjustment coefficient reflects the change in the importance of the attribute in different time windows.

[0023] Preferably, the vibration signal data for real-time operation monitoring of the vacuum dry pump is obtained by installing multiple sensors at different positions on each vacuum dry pump, and each sensor independently collects the vibration signal of the position; each sensor represents a channel, thereby obtaining vibration data from multiple channels to form a multi-dimensional time series data set.

[0024] Preferably, the preprocessing includes: data dimension reduction, processing of missing values ​​and outliers, and normalizing the data to a uniform range.

[0025] Preferably, the preprocessed data is labeled by performing a sliding window on the preprocessed full-life vibration signal data, first performing real-time normalization with progressive growth on the window and previous data, then extracting the window data, and constructing a multi-classification data set containing multiple window sample data and corresponding labels.

[0026] Preferably, the features of the feature set also include:

[0027] The zero-crossing rate is expressed as:

[0028]

[0029] Among them, N ZCR is the number of zero crossings between adjacent points, N win is the total number of data points in the window;

[0030] Stability is expressed as:

[0031]

[0032] Among them, N win is the total number of data points in the window, x i+1 is the i+1th data point in the signal, x i is the i-th data point in the signal;

[0033] The proportion of unstable data time is expressed as:

[0034]

[0035] Among them, N unstable is the number of adjacent points whose difference is greater than the set threshold, N win is the total number of data points in the window;

[0036] The proportion of high-frequency components is expressed as:

[0037]

[0038] Where, f represents the frequency, f c represents the high-frequency threshold, |X(f)| 2 represents the energy of high-frequency components, |X(D)| 2 Represents the total energy of the signal;

[0039] Hausdorff distance, expressed as:

[0040]

[0041] Among them, X a represents a sample of a certain type of fault, X b Represents a sample selected from other types of fault or normal data.

[0042] Preferably, the step of inputting the preprocessed vibration signal data into a trained multi-fault prediction model comprises the following steps:

[0043] For the preprocessed vibration signal data, by setting the window size W and the sliding step size S, the window and the previous data are first normalized in a progressive manner, and then the sliding window is extracted to form window data;

[0044] Perform feature extraction on each window data to obtain a feature matrix;

[0045] The feature matrix F real Input into the trained multi-fault prediction model for classification, and the model outputs the category label C pred It is expressed as:

[0046] C pred =C4.5(F real );

[0047] If the model identifies a fault, the output reports the fault, expressed as:

[0048] Fault Report = {C pred , T pred};

[0049] Among them, C pred Indicates the fault type, T pred Indicates the time of occurrence.

[0050] Preferably, the features included in the feature matrix are important features selected from the feature set, and the selection process includes the following steps:

[0051] Standardize the data of the feature set;

[0052] Calculate the covariance matrix C based on the standardized data:

[0053] Perform eigenvalue decomposition on the covariance matrix C and obtain the eigenvalue λ i and the corresponding eigenvector v i , expressed as:

[0054] Cv i =λ i v i ;

[0055] Among them, λ i v i The corresponding eigenvalue, v i is the i-th feature vector of the feature set;

[0056] The eigenvalue λ i Sort from largest to smallest and select the features corresponding to the first several eigenvalues ​​as important features.

[0057] The present invention also provides a vacuum dry pump multi-fault prediction system based on improved C4.5, characterized in that it includes:

[0058] The training module uses the improved C4.5 algorithm to build a decision tree model and uses the full-life vibration signal data of the vacuum dry pump for training to obtain a trained multi-fault prediction model;

[0059] An acquisition module is used to obtain vibration signal data of the real-time operation monitoring of the vacuum dry pump and perform data preprocessing on the data;

[0060] The recognition module inputs the preprocessed vibration signal data into the trained multi-fault prediction model to predict whether a fault occurs in the vibration signal data, and outputs a fault prediction result, which includes the fault type and the fault occurrence time.

[0061] The present invention has the following beneficial effects: by obtaining real-time vibration signal data during the operation of the vacuum dry pump and performing data preprocessing, the present invention can accurately predict possible faults, output specific fault types and occurrence times, and improve the accuracy of fault detection. In addition, the method adopts sliding window technology and real-time normalization with progressive growth, so that the model can quickly adapt to new operating conditions and fault modes, and realize real-time updates, thereby significantly improving the real-time and accuracy of predictions. In the feature engineering stage, not only the timing domain and frequency domain features, but also the zero-crossing rate, stability, the proportion of unsteady data time, the proportion of high-frequency components, Hausdorff distance, etc. are calculated, so as to effectively distinguish different types of faults and enhance the model's recognition ability for multiple fault modes. The improved C4.5 algorithm is suitable for multi-classification tasks and has good interpretability. By combining the information gain rate of the attribute weight to select the best partition attribute, multiple fault types can be accurately classified. The method can effectively identify multiple fault types caused by dust jamming and other reasons, provide accurate fault classification, and help maintenance personnel take timely measures to reduce equipment downtime and maintenance costs.

[0062] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments, but the present invention is not limited to the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A method step diagram of an embodiment of the present invention;

[0064] Figure 2 A diagram showing the steps of model training according to an embodiment of the present invention;

[0065] Figure 3 This is a flow chart of model training in an embodiment of the present invention;

[0066] Figure 4 A flow chart of a model test according to an embodiment of the present invention;

[0067] Figure 5 A system structure diagram of an embodiment of the present invention DETAILED DESCRIPTION

[0068] See also Figure 1 FIG. 1 is a diagram showing steps of a method according to an embodiment of the present invention, comprising the following steps:

[0069] S101, using the improved C4.5 algorithm to build a decision tree model and using the full life vibration signal data of the vacuum dry pump for training, to obtain a trained multi-fault prediction model;

[0070] S102, acquiring vibration signal data of real-time operation monitoring of the vacuum dry pump, and preprocessing the data;

[0071] S103, inputting the preprocessed vibration signal data into a trained multi-fault prediction model to predict whether a fault occurs in the vibration signal data and outputting a fault prediction result, wherein the prediction result includes a fault type and a fault occurrence time.

[0072] For details, see Figure 2 and Figure 3 As shown, it is a model training step diagram and a flow chart of an embodiment of the present invention, and the S101 includes the following steps:

[0073] A1, obtaining the full life vibration signal data of the vacuum dry pump, and performing data preprocessing, including data dimension reduction, processing missing values ​​and outliers, and normalizing the data to a uniform range. Specifically, steps A11 to A14 are included.

[0074] A11, the data obtained is the time series data of the vibration signal of the vacuum dry pump throughout its life. Part of the full-life data is used to extract window sample data to build a multi-classification data set for model training and testing, and the other part of the data is used to simulate real-time fault prediction, with a division ratio of 7:3. Multiple sensors are installed on each pump, and these sensors are distributed in different locations to ensure comprehensive monitoring of the operating status of the equipment. Each sensor represents a channel and can independently collect vibration signals at that location. This enables each pump to obtain vibration data from multiple channels to form a multi-dimensional time series data set. The full-life data obtained covers the entire process of the vacuum dry pump from the start of operation to the occurrence of downtime. This data records the operation of the equipment at different stages, including startup, normal operation, fault occurrence and final downtime. Different types of fault data may appear as specific vibration patterns and may appear in different forms in the same channel. Some faults may occur alone, while others may be superimposed on each other, resulting in complex signal characteristics.

[0075] A12, in the data dimension reduction stage, the sliding window method is used to process the continuous vibration data points. The window size is k and the step size is also k. The input vibration signal data is D = [d0, d1, ..., d n-1 ], where n represents the total number of data points in the original data set, and d i represents the i-th data point.

[0076] First, calculate the number of data points after dimensionality reduction, which can be determined by the following formula:

[0077]

[0078] Among them, m represents the total number of points of the data after dimensionality reduction, Indicates rounding up.

[0079] By taking the average of every k data points, the reduced - dimensional data R is calculated as follows:

[0080]

[0081] where R[j] represents the j - th data point after dimensionality reduction.

[0082] If the number of data points in the last window is less than k, assuming the starting index of this window is p, then calculate the average of all data points in the range from p to n - 1:

[0083] where p=(m - 1)k and p < n

[0084] A13. When dealing with outliers and missing values, values in the dataset that deviate significantly from other data points are regarded as outliers, usually manifested as particularly large or particularly small numerical values. A data point R[i] is defined as an outlier if it satisfies the following condition:

[0085] R[i]>μ + 3σ or R[i]<μ - 3σ

[0086] where μ represents the mean of the dataset and σ represents the standard deviation of the dataset.

[0087] A common method for dealing with outliers and missing values is mean imputation. The formula for mean imputation is as follows:

[0088] (where R j is a non - missing value)

[0089] where R i is the result after substituting the missing value, and N is the number of non - missing values.

[0090] A14. Finally, in the data normalization stage, the min - max normalization method is adopted, and its calculation formula is as follows:

[0091]

[0092] where r represents the normalized value, R is the original data, R min and R max represent the minimum and maximum values in the dataset respectively.

[0093] After dimensionality reduction processing, the specification size of the output data R is m, that is, the total number of data points after dimensionality reduction. Usually, m < n. The reduced - dimensional data can be represented as a vector:

[0094] R = [r0, r1, …, r m-1

[0095] ​A2, by setting the sliding window size and step size, sliding the window on the full-life vibration signal data, first normalizing the window and the previous data with progressive growth, and then extracting the data of the window, thereby extracting multiple window sample data. Label the samples using expert annotation, divide the faults for each sample, and construct a multi-classification data set containing multiple window sample data and corresponding labels. Finally, randomly divide the data set into a training set and a test set. Specifically including steps A21 to A25.

[0096] A21, after processing, the vibration signal data R = [r0, r1, ..., r m-1 ], set the window parameters to extract samples. The window size W is the number of consecutive data points contained in each sample, and the sliding step S is the number of points that the window slides forward each time a new sample is extracted. The number of windows M that can be extracted from a vibration signal data is calculated by the formula:

[0097]

[0098] A22 uses a normalization method with progressive growth to better meet the real-time requirements. Before extracting the window data, for the current window position j, the window data and all previous data, i.e., R′=[r0,r1,…,r j·s+w-1 ] is normalized, and the normalization formula is:

[0099]

[0100] A23, for the data R', extract samples using the set window parameters W and S. The specific operation is to extract a window X at each position j of the data R. j As a new sample:

[0101] X j =[R ′ [j.S],R ′ [j·S+1],…,R ′ [j·S+W-1], (j=0, 1,..., M-1)

[0102] A24, each window sample X j Manual labeling will be performed based on expert experience to generate the corresponding label Y j Experts evaluate each window sample based on historical fault records and the operating status of the equipment, determine its corresponding fault type or normal state, and record it as label Y j .

[0103] For the vibration signal data R of different vacuum dry pumps, each window sample X j and its corresponding label Y jStored in a data structure to form sample pairs (X j , Y j ). By dividing the window sample X j and the corresponding label Y j Combine and construct a dataset V containing sample data and labels:

[0104] V={(X j , Y j )∣j=0,1,2,…,L-1}

[0105] Among them, L is the number of samples in the data set.

[0106] A25, randomly divide the data set V into training set V train and the test set V test The division ratio is set to p = 0.7, and the specific division method is as follows:

[0107] Randomly select p·L samples from the following data set V as the training set V train , the remaining samples are used as the test set V test .

[0108] V train =(X train , Y train )

[0109] V test =(X test , Y test )

[0110] A3, in feature engineering, multiple features are extracted and a feature set of each sample data is constructed. In addition to calculating some commonly used time domain features and frequency domain features, some new features are also calculated, including zero crossing rate, stability, proportion of unstable data time, proportion of high-frequency components, etc. At the same time, the Hausdorff distance is calculated to evaluate the difference between a certain type of fault and other types of fault data and normal data. Specifically, it includes steps A31 to A34.

[0111] A31, in the time domain feature extraction stage, the input data is X, X i is the i-th data point of the signal, and N is the total number of data points of signal X.

[0112] The maximum value is the highest level of the signal, reflecting the strength of the signal. The formula is: X max =max(X)

[0113] The minimum value is the lowest level of the signal, indicating the bottom limit of the signal. The formula is: X min =min(X)

[0114] The peak-to-peak value represents the distance from the maximum value to the minimum value of the signal. The formula is: Xpp =X max -X min

[0115] The mean value represents the average level of the signal, and the formula is:

[0116] The variance quantifies the degree of volatility of the signal, and the formula is:

[0117] The square root of the standard deviation is:

[0118] The mean square value reflects the energy of the signal, and the formula is:

[0119] The root mean square value is used to evaluate the signal strength, and the formula is:

[0120] Kurtosis reflects the influence of outliers, and the formula is:

[0121]

[0122] Skewness indicates the symmetry of the signal distribution and is given by:

[0123] The waveform factor reflects the sharpness of the signal, and the formula is:

[0124] The crest factor is the ratio of the maximum value to the mean value of the signal. The formula is:

[0125] The ratio of the maximum value of the pulse factor signal to the mean square value is as follows:

[0126] The margin factor is the ratio of the maximum value to the minimum value of the signal. The formula is:

[0127] A32, in the frequency domain feature extraction stage, the frequency component of the signal is analyzed by Fourier transform, and the signal X(t) is Fourier transformed to obtain the frequency domain representation X(f), where f represents frequency and f high and f low Indicates the upper and lower limits of the frequency range.

[0128] The power spectrum density analyzes the energy distribution of the signal, expressed as: P(f) = |X(f)| 2

[0129] The center of gravity frequency describes the frequencies at which the signal energy is concentrated. The formula is:

[0130] The mean square frequency represents the frequency distribution characteristics of the signal, and the formula is:

[0131] The root mean square frequency is a measure of the strength of a frequency and is given by:

[0132] The frequency variance quantifies the degree of discreteness of the frequency distribution, and the formula is:

[0133] The frequency standard deviation indicates the degree of dispersion of the frequency distribution, and the formula is:

[0134] Bandwidth is the frequency range occupied by the signal in the frequency domain, the formula is: BW = f high -f low

[0135] The dominant frequency is usually the frequency with the largest energy in the spectrum, and the formula is:

[0136] Spectral kurtosis is used to describe the sharpness of the spectrum, and the formula is:

[0137] Spectral skewness indicates the symmetry of the spectrum distribution, and the formula is:

[0138] Total energy represents the total energy of the signal, and the formula is: E total =∑ t |X(t)| 2

[0139] Spectral entropy is used to measure the uncertainty of the spectrum, and the formula is:

[0140] Where E = ∑ f P(f).

[0141] Relative energy indicates the ratio of the energy of a certain frequency band to the total energy. The formula is:

[0142] A33, in the extraction of other features, some uncommon indicators are used. These indicators are designed to analyze the characteristics of vibration signals more deeply and better distinguish different types of faults in the case of overlapping occurrences of different faults. Specifically, the zero-crossing rate distinguishes faults with fast-changing characteristics from relatively stable normal data by measuring the number of times the signal crosses the zero point. A high zero-crossing rate usually indicates that the signal has fluctuated violently. Stability reflects the stability of the signal and helps identify violently changing faults. It focuses on measuring the overall change trend of the signal rather than instantaneous fluctuations. The proportion of unsteady data time calculates the duration of the fault, distinguishes between trend changes and mutations caused by the fault, and helps identify progressive faults from sudden faults. The proportion of high-frequency components distinguishes faults caused by high-frequency vibration from faults dominated by low-frequency vibration. The Hausdorff distance calculates the similarity of the curve trends between samples. A large distance indicates that the samples are very different and the fault types are different; a small distance indicates that the samples are similar, indicating that they are close in feature change trends and have the same fault type.

[0143] The zero-crossing rate calculates the ratio of the zero-crossing point in the window data X. When the product of the values ​​of adjacent data points is less than zero, it means that the signal crosses the zero point at this point. The zero-crossing rate can capture the features of frequent changes in the signal, and is particularly suitable for detecting high-frequency noise or sudden changes in the signal. Its calculation formula is:

[0144]

[0145] Among them, N ZCR is the number of zero crossings between adjacent points, N win is the total number of data points in the window. The adjacent point condition crossing zero is:

[0146] x i ·x i+1 <0

[0147] Stationarity evaluates the stationarity of a signal by calculating the average of the differences of all adjacent points in the window data X. Stationarity reflects the temporal consistency of a signal and can identify unstable components in the signal. The calculation formula is:

[0148]

[0149] The proportion of unstable data time measures the proportion of the duration of the unstable segment in the window data X relative to the total time. It is achieved by calculating the number of adjacent data point values ​​whose difference is greater than a certain threshold. The calculation formula is:

[0150]

[0151] Among them, N unstable It is the number of adjacent points whose difference is greater than the set threshold.

[0152] The proportion of high-frequency components refers to the frequency greater than a certain threshold frequency f c The ratio of signal energy to total frequency energy is calculated as:

[0153]

[0154] Where, f represents the frequency, f c represents the high-frequency threshold, |X(f)| 2 represents the energy of high-frequency components, |X(D)| 2 Represents the total energy of the signal.

[0155] The Hausdorff distance is used to calculate the difference between a certain type of fault and other types of fault data and normal data. The formula is:

[0156]

[0157] Among them, X a represents a sample of a certain type of fault, X b Indicates selecting a sample from other types of fault or normal data respectively.

[0158] A34, finally, the features extracted from the jth sample will be combined into a feature vector F j , the specific form of the vector can be expressed as:

[0159] F j ={X max , X min , X pp , X mean , X var , X std , X ms , X rms ,K,S,WF,PF,IF,MF,f c , f rms ,Var(f),f std , ZCR, K, P unstable , HF Ratio, d H}

[0160] Through the above comprehensive feature extraction method, the extracted feature vector F j With the corresponding label Y j Form a complete data set V s , which can be expressed as:

[0161] V s ={(F j , Y j )∣j=0,1,2,…,L-1}

[0162] The training set and test set are:

[0163]

[0164] A4, using the improved C4.5 algorithm to build a decision tree model, input the feature set of the training set samples and their corresponding labels into the model for training, and build a decision tree on the training set. Then, input the test set samples into the model for classification prediction, and use confusion matrix, accuracy, recall rate, F1 score and other indicators to evaluate the model performance. Figure 4 As shown, it specifically includes steps A41 to A44.

[0165] A41, C4.5 algorithm mainly selects split points and constructs decision trees by information gain rate. As a classic decision tree construction method, C4.5 algorithm selects the features that can best divide the data set for splitting, thereby generating a decision tree. Based on the original C4.5 algorithm, in the process of calculating information gain and information gain ratio for each attribute, a dynamically adjusted feature importance index (FII) mechanism is introduced to obtain the improved C4.5 algorithm; in machine learning and decision tree algorithms, attributes are used to describe and distinguish the characteristics of data samples, usually representing a specific physical quantity or category, and attributes can refer to equipment status indicators after feature extraction. The FII method dynamically updates the importance weight of the attribute according to its performance under different fault modes; in multidimensional feature time series data, for each attribute A, its dynamic feature importance index (FII) is calculated in different time periods. The index comprehensively considers three factors: variance, expert experience, and time adjustment coefficient. Variance reflects the fluctuation of features in different time windows, and features with larger fluctuations are usually more likely to reveal faults. Expert experience gives higher weights to certain features based on the expert's understanding of equipment failure modes. The time adjustment factor dynamically adjusts the importance of the feature as the equipment runs for a certain period of time to account for the effects of equipment aging or wear. The formula is as follows:

[0166] FII(A,t)=Var(A,t)×Exp(A)×λ(t)

[0167] Among them, FII(A,t) represents the feature importance index of attribute A in the time window of t; Var(A,t) represents the variance of attribute A in the time window of t, reflecting its degree of change in this time period; Exp(A) represents the empirical importance index of attribute A evaluated by experts, which is set based on expert knowledge; λ(t) represents the time adjustment coefficient, which is used to dynamically adjust the feature weight according to time changes. The weight of the attribute is dynamically adjusted by calculating the ratio of the variance of the current time window to the variance of the previous time window. The formula is as follows:

[0168]

[0169] Where tk is the starting point of the time window, and Var(A,t-1) is the variance of the previous time window. The adjustment coefficient reflects the change in the importance of the attribute in different time windows.

[0170] Taking attributes A and B as an example, assume that the variance of attribute A in the current time window is 0.064, while the variance of the previous time window is 0.05; the variance of attribute B in the current time window is 0.02, while the variance of the previous time window is 0.018. In addition, based on the experience of experts, we set experience weights for attributes A and B, which are 0.7 and 0.5 respectively.

[0171] In order to dynamically adjust the weight of the attribute, we first calculate the variance ratio of the current time window to the previous time window. For attribute A, the variance ratio is 1.28, that is, For attribute B, the variance ratio is 1.11, that is, This ratio reflects how the attribute changes over different time periods and is used to adjust the weights.

[0172] By introducing the variance ratio, the calculation formula of the attribute weight becomes:

[0173] W(A,t)=Var(A,t)×Exp(A)×λ(t)

[0174] W(B,t)=Var(B,t)×Exp(B)×λ(t)

[0175] After substituting the known data, the final weights of attributes A and B are obtained:

[0176] FII(A,t)=0.064×0.7×1.28=0.057344

[0177] FII(B,t)=0.02×0.5×1.11=0.0111

[0178] The improved C4.5 algorithm selects the split point by the information gain rate to construct a decision tree model, which is expressed as:

[0179] M=C4.5(F train , Y train );

[0180] F train represents the training features, Y train Denotes the training feature F train The information gain rate of the improved C4.5 algorithm is expressed as:

[0181] G ′ (A) = FII(A,t)·G(A);

[0182] Among them, FII(A,t) is the weight of attribute A, G(A) is the information gain rate of attribute A calculated by the original C4.5 algorithm, and G ′ (A) is the information gain rate of attribute A calculated by the improved C4.5 algorithm.

[0183] This method is significantly different from traditional decision tree methods such as the C4.5 algorithm and random forest. The core innovation lies in the introduction of a dynamic feature importance index (FII), which greatly enhances the adaptability and real-time performance of the model. Traditional C4.5 and random forest algorithms rely on static information gain ratios in the feature selection process, which makes it difficult to maintain efficiency and accuracy when dealing with dynamic changes in equipment operating status and failure modes. In contrast, the present invention introduces the FII mechanism, combines the variance of each feature in different time windows, expert experience and time adjustment coefficient, and dynamically adjusts the weight of the feature, so that the model can flexibly adjust the importance of each feature according to the long-term change trend and instantaneous fluctuations of the equipment.

[0184] FII not only considers the performance of features in different time periods, but also integrates the changes in the actual working environment and operating status of the equipment, so that the model can respond to fluctuations under various working conditions in real time. This dynamic adjustment mechanism, especially by introducing the variance ratio to optimize the feature selection process, significantly improves the accuracy and flexibility of fault prediction compared to traditional static methods. Unlike traditional methods such as the C4.5 algorithm and random forest, the present invention can dynamically adjust the importance and selection process of features according to the real-time status of the equipment, thereby more accurately capturing early signs of equipment failure and updating fault prediction results in real time. This method is particularly suitable for complex industrial equipment fault monitoring systems, which can respond to equipment operation changes in real time and improve fault diagnosis and prediction capabilities.

[0185] A43, when building a decision tree, the improved C4.5 algorithm selects the optimal features to divide the nodes, thereby forming a decision tree, divides the training data according to the selected features to form child nodes, and repeats the above process for each child node until the stop condition is met. At this time, the model M will be expressed by the following formula:

[0186] M=C4.5(F train , Y train )

[0187] After the model training is completed, the test set feature F test Input into the trained decision tree model M for classification prediction and generate the prediction result Y pred :

[0188] Y pred =M(F test )

[0189] A44,To ensure the effectiveness and reliability of the model, the performance of the model is evaluated using indicators such as confusion matrix, precision, recall, and F1 score. The confusion matrix is ​​a summary of the model's prediction results, including true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN).

[0190] Among them, the true positive (TP) is the number of samples correctly predicted as positive, the false positive (FP) is the number of negative samples incorrectly predicted as positive, the false negative (FN) is the number of positive samples incorrectly predicted as negative, and the true negative (TN) is the number of samples correctly predicted as negative.

[0191] The calculation formula of relevant indicators is as follows, and the accuracy can be expressed as:

[0192]

[0193] The recall rate is:

[0194]

[0195] The F1 score is the harmonic mean of precision and recall, and the formula is:

[0196]

[0197] A5, constructing a feature subset through feature dimensionality reduction analysis, first calculates the covariance matrix between the features in the data set, and extracts the eigenvectors and eigenvalues ​​to determine the main components. Select the first k features according to the size of the eigenvalue to form a new feature subset. Specifically includes steps A51 to A55.

[0198] A51, the entire data set can be represented as a matrix Where L is the number of samples and K is the number of features.

[0199] In order to eliminate the dimensional effects of different features, the data is first standardized. The calculation formula of the standardized data matrix Z is:

[0200]

[0201] Among them, F ij represents the value of the i-th sample on the j-th feature, μ j and σ j are the mean and standard deviation of the jth feature respectively. A52, based on the standardized data Z, calculate the covariance matrix C:

[0202]

[0203] The element C of the covariance matrix C pq Defined as the covariance between the p-th feature and the q-th feature.

[0204] A53, perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalue λ i and the corresponding eigenvector v i , satisfying the following relationship:

[0205] Cv i =λ i v i

[0206] Among them, λ i is the i-th eigenvalue, v i is the corresponding feature vector.

[0207] A54, sort the eigenvalues ​​from large to small to get the sorted eigenvalue array λ (1) ,λ (2) , …, λ (K) , and rank the eigenvectors accordingly. Select the first k eigenvalues ​​(λ (1) ,λ (2) , …, λ (k) ) and its corresponding eigenvector (v (1) , v (2) , …, v (k) ) to form a new feature subset.

[0208] The selected eigenvectors form a matrix The new feature matrix F s It can be calculated by the following formula:

[0209]

[0210] Among them, F s It is the feature matrix after dimensionality reduction, which contains the values ​​of L samples in the new k feature dimensions.

[0211] A55, the model will be trained and tested using a subset of features and the model predictions will be evaluated to determine whether these principal components are effective.

[0212] A6, perform sliding window and real-time fault prediction on the vibration signal data of the entire life, set the window size and step size consistent with A2 on the vibration signal data, and perform progressive normalization and sliding window extraction. Calculate the features of the data in each window, input the C4.5 model for classification, and determine which type of fault data or normal data the window data belongs to. If the prediction is a fault, report the fault in time and mark the corresponding fault type and occurrence time, so as to achieve effective prediction of multiple fault types of vacuum dry pumps. Specifically including steps A61 to A65.

[0213] A61, input full life vibration signal data D real After preprocessing, the data preprocessing process is the same as that described in claim 3. Subsequently, a window size W and a sliding step size S consistent with claim 4 are set on the vibration signal data, and the window and the previous data are first normalized with progressive growth, and then a sliding window extraction is performed to form the window data X real , X real It is expressed as:

[0214] X real ={x i , x i+1 , …, x i+w-1}

[0215] Among them, i is the window starting index.

[0216] A62, for each window data X real , calculate the feature matrix F real , the dimension of the feature matrix is ​​1×k, where k is the number of selected features. The feature matrix can be represented as follows:

[0217] F real =[f1, f2, …, f k ]

[0218] A63, the feature matrix F real Input into the improved C4.5 model for classification to determine whether the window belongs to a certain type of fault feature. The category label C output by the model pred It can be expressed as:

[0219] C pred =C4.5(F real )

[0220] A64, if the model identifies a fault, the system will report the fault in a timely manner and record the corresponding fault type C pred and occurrence time T pred , which can be expressed as:

[0221] Fault Report = {C pred , T pred}

[0222] A65, finally evaluate the accuracy of fault type prediction. If the fault type C output by the model is pred With the real fault type C true If the prediction is correct, then it is wrong. The evaluation formula is:

[0223]

[0224] The accuracy of time interval prediction is calculated by calculating the fault occurrence time T predicted by the model. pred The actual time T true The evaluation formula is:

[0225]

[0226] Among them, ΔT is the allowable time error range.

[0227] When both the fault type and time predictions are correct, it is recorded as a successful prediction. The total number of correct predictions is N. correct The calculation formula is:

[0228]

[0229] Where N is the actual number of faults.

[0230] In order to verify the effectiveness of the model and model solution proposed in this invention, the vacuum dry pump data with full-life vibration signal data of a certain process condition of a certain company was selected as the research object. The data information contains the full-life vibration signal data of 49 vacuum dry pumps. The data of 34 vacuum dry pumps were selected to construct a multi-classification data set, and the data of 15 vacuum dry pumps were selected to simulate online real-time fault prediction.

[0231] For the vibration data of 34 vacuum dry pumps, by setting the sliding window size and step size, a sliding window is performed on the full-life vibration signal data. The window and the previous data are first normalized with progressive growth, and then the data of the window is extracted, thereby extracting multiple window sample data. The samples are labeled by expert annotation, and each sample is divided into faults to construct a multi-classification data set containing multiple window sample data and corresponding labels. Finally, the data set is randomly divided into a training set and a test set. The experimental parameter settings of the present invention are shown in Table 1 below:

[0232] Table 1. Experimental parameter settings

[0233] Parameter name Setting Value Sliding window size 128 data points Step Length 16 data points C4.5 Maximum Depth 15 C4.5 minimum number of sample partitions 5 C4.5 gain threshold 0.01 Model training rounds 100 Predicted number of fault types 4 categories (3 categories of failure + 1 category of normal) Real-time forecast update frequency Every 5 seconds

[0234] The vacuum dry pump multi-fault prediction training and testing model based on the improved C4.5 of the present invention is adopted to observe its performance effect, and the confusion matrix of the test set results in the multi-classification data set is obtained as shown in Table 2.

[0235] Table 2. Confusion matrix

[0236] Prediction Category 1 Prediction Category 2 Prediction Category 3 Prediction Normal Actual Category 1 140 31 15 14 Actual Category 2 19 138 22 21 Actual Category 3 5 12 79 5 Actual normal 27 15 8 350

[0237] According to the confusion matrix, the accuracy of multi-classification prediction is calculated to be 72.66%.

[0238] The recall and F1 scores of multi-classification predictions are calculated as shown in Table 3.

[0239] Table 3. Recall and F1 score results

[0240] category Recall F1 score Category 1 68.48％ 70.17％ Category 2 70.85％ 69.58％ Category 3 79.45％ 74.75％ normal 87.50％ 87.50％

[0241] The real-time fault prediction is performed on the vibration data of 15 vacuum dry pumps, and its accuracy is shown in Table 4.

[0242] Table 4. Real-time fault prediction accuracy

[0243] Fault type The actual number of occurrences The number of predicted failures Accuracy Fault Type 1 147 104 70.73％ Fault Type 2 95 62 65.26％ Fault Type 3 49 37 75.51％ Overall accuracy 291 203 69.74％

[0244] For details, see Figure 5 FIG. 1 is a system structure diagram of an embodiment of the present invention, including:

[0245] A training module 501 uses an improved C4.5 algorithm to construct a decision tree model and uses the full life vibration signal data of the vacuum dry pump for training to obtain a trained multi-fault prediction model;

[0246] The acquisition module 502 acquires vibration signal data of the real-time operation monitoring of the vacuum dry pump and performs data preprocessing on the data;

[0247] The identification module 503 inputs the preprocessed vibration signal data into the trained multi-fault prediction model to predict whether a fault occurs in the vibration signal data, and outputs a fault prediction result, which includes the fault type and the fault occurrence time.

[0248] In summary, compared with the prior art, the decision tree structure generated by the C4.5 algorithm in the present invention is clear, easy to understand and explain, so that engineers can intuitively analyze the prediction results and have high operability, while other black box models such as deep learning usually lack transparency and it is difficult to trace the basis for decision making. Secondly, the present invention combines sliding window technology and progressive normalization to effectively capture the dynamic changes in the vibration signal, ensuring that the model can be updated in real time and adapt to new operating conditions. Compared with clustering and Transformer methods that may ignore local details, it has stronger adaptability. In terms of feature engineering, this method significantly improves the accuracy of fault classification by introducing time-frequency domain and other features such as stationarity, high-frequency component proportion and Hausdorff distance, while algorithms such as Transformer, XGBoost and Catboost usually rely on automatic feature selection and may not be able to fully process specific local features. The improved C4.5 algorithm can also update the prediction model in real time, adapt to new data, and maintain high accuracy, while methods such as Transformer usually require retraining the entire model, which increases computational costs and time overhead. In addition, this method is particularly targeted at complex faults such as dust jamming and other faults caused by multiple factors. It can effectively distinguish different types of faults and reduce false alarm and missed alarm rates. Compared with other methods such as fault prediction methods based on Transformer, XGBoost, and Catboost, it has stronger recognition capabilities when dealing with complex faults. Finally, by combining the information gain rate based on attribute weights, the contribution of each feature in the prediction is ensured, thereby improving the prediction accuracy. Methods such as Transformer and XGBoost may rely on automatic feature selection strategies and cannot fully consider the importance of each feature under different fault modes. In general, the present invention solves the shortcomings of the existing technology through a transparent decision-making process, real-time updates, fine feature extraction, and powerful multi-fault classification capabilities, and has higher practical value.

[0249] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A vacuum dry pump multi-fault prediction method based on improved C4.5, characterized in that: include: The decision tree model was constructed using the improved C4.5 algorithm and trained using the full life vibration signal data of the vacuum dry pump to obtain a trained multi-fault prediction model. Acquiring vibration signal data of real-time operation monitoring of the vacuum dry pump, and preprocessing the data; The preprocessed vibration signal data is input into the trained multi-fault prediction model to predict whether a fault occurs in the vibration signal data, and a fault prediction result is output, wherein the prediction result includes the fault type and the fault occurrence time.

2. The vacuum dry pump multi-fault prediction method based on improved C4.5 according to claim 1 is characterized in that: The method of using the improved C4.5 algorithm to construct a decision tree model and using the full life vibration signal data of the vacuum dry pump for training includes the following steps: Use the improved C4.5 algorithm to build a decision tree model; Obtain the full-life vibration signal data of the vacuum dry pump and perform preprocessing, label the preprocessed data, and obtain a data set including sample data and corresponding labels; Extracting features from the data set to obtain a feature set; the features of the feature set include time domain features and frequency domain features; The decision tree model is trained using the feature set.

3. The vacuum dry pump multi-fault prediction method based on improved C4.5 according to claim 2 is characterized in that: The improved C4.5 algorithm is used to construct a decision tree model, comprising the following steps: The attribute weight mechanism is introduced into the original C4.5 algorithm to obtain the improved C4.5 algorithm; The improved C4.5 algorithm selects the split point by the information gain rate to construct a decision tree model, which is expressed as: M=C4.5(F train ,Y train ); Among them, F train represents the training features, Y train Denotes the training feature F train The information gain rate of the improved C4.5 algorithm is expressed as: G ′ (A)=FII(A,t)·G(A); FII(A,t)=Var(A,t)×Exp(A)×λ(t); Among them, FII(A,t) is the weight of attribute A, G(A) is the information gain rate of attribute A calculated by the original C4.5 algorithm, and G ′ (A) is the information gain rate of attribute A calculated by the improved C4.5 algorithm; Var(A,t) represents the variance of attribute A in the time window of t, reflecting its degree of change in this time period; Exp(A) represents the empirical importance index of attribute A evaluated by experts, which is set based on expert knowledge; λ(t) represents the time adjustment coefficient, which is used to dynamically adjust the feature weight according to time changes; tk is the starting point of the time window, and Var(A,t-1) is the variance of the previous time window, reflecting the change of the importance of the attribute in different time windows.

4. The vacuum dry pump multi-fault prediction method based on improved C4.5 according to claim 1 is characterized in that: The method of obtaining vibration signal data for real-time operation monitoring of the vacuum dry pump is specifically as follows: multiple sensors are installed at different positions on each vacuum dry pump, and each sensor independently collects the vibration signal of the position; each sensor represents a channel, thereby obtaining vibration data from multiple channels to form a multi-dimensional time series data set.

5. The vacuum dry pump multi-fault prediction method based on improved C4.5 according to claim 1 or 2, characterized in that: The preprocessing includes: reducing data dimension, processing missing values ​​and outliers, and normalizing data to a uniform range.

6. A vacuum dry pump multi-fault prediction method based on improved C4.5 according to claim 1, characterized in that: The labeling of the preprocessed data is specifically as follows: a sliding window is performed on the preprocessed full-life vibration signal data, the window and the previous data are firstly normalized in real time with progressive growth, and then the data of the window is extracted to construct a multi-classification data set containing multiple window sample data and corresponding labels.

7. A vacuum dry pump multi-fault prediction method based on improved C4.5 according to claim 2, characterized in that: The features of the feature set also include: The zero-crossing rate is expressed as: Among them, N ZCR is the number of zero crossings between adjacent points, N win is the total number of data points in the window; Stability is expressed as: Among them, N win is the total number of data points in the window, x i+1 is the i+1th data point in the signal, x i is the i-th data point in the signal; The proportion of unstable data time is expressed as: Among them, N unstable is the number of adjacent points whose difference is greater than the set threshold, N win is the total number of data points in the window; The proportion of high-frequency components is expressed as: Where, f represents the frequency, f c represents the high-frequency threshold, |X(f)| 2 represents the energy of high-frequency components, |X(D)| 2 Represents the total energy of the signal; Hausdorff distance, expressed as: Among them, X a represents a sample of a certain type of fault, X b Represents a sample selected from other types of fault or normal data.

8. A vacuum dry pump multi-fault prediction method based on improved C4.5 according to claim 7, characterized in that: The step of inputting the pre-processed vibration signal data into the trained multi-fault prediction model comprises the following steps: For the preprocessed vibration signal data, by setting the window size W and the sliding step size S, the window and the previous data are first normalized in a progressive manner, and then the sliding window is extracted to form window data; Perform feature extraction on each window data to obtain a feature matrix; The feature matrix F real Input into the trained multi-fault prediction model for classification, and the model outputs the category label C pred It is expressed as: C pred =C4.5(F real ); If the model identifies a fault, the output reports the fault, expressed as: Fault Report = {C pred , T pred }; Among them, C pred Indicates the fault type, T pred Indicates the time of occurrence.

9. A vacuum dry pump multi-fault prediction method based on improved C4.5 according to claim 8, characterized in that: The features included in the feature matrix are important features selected from the feature set, and the selection process includes the following steps: Standardize the data of the feature set; Calculate the covariance matrix C based on the standardized data: Perform eigenvalue decomposition on the covariance matrix C and obtain the eigenvalue λ i and the corresponding eigenvector v i , expressed as: Cv i =λ i v i ; Among them, λ i v i The corresponding eigenvalue, v i is the i-th feature vector of the feature set; The eigenvalue λ i Sort from largest to smallest and select the features corresponding to the first several eigenvalues ​​as important features.

10. A vacuum dry pump multi-fault prediction system based on improved C4.5, characterized in that: include: The training module uses the improved C4.5 algorithm to build a decision tree model and uses the full-life vibration signal data of the vacuum dry pump for training to obtain a trained multi-fault prediction model; An acquisition module is used to obtain vibration signal data of the real-time operation monitoring of the vacuum dry pump and perform data preprocessing on the data; The recognition module inputs the preprocessed vibration signal data into the trained multi-fault prediction model to predict whether a fault occurs in the vibration signal data, and outputs a fault prediction result, which includes the fault type and the fault occurrence time.

Images