Rotating machine fault diagnosis method and system based on axis track confidence analysis

Abstract

The invention provides a rotating machine fault diagnosis method and system based on axis track confidence analysis, and relates to the technical field of rotating machine fault diagnosis. Two acceleration sensors which are orthogonal to each other are adopted to collect transverse vibration signals of a rotor, after preprocessing is carried out, an axis track is reconstructed, feature extraction is carried out to obtain a four-dimensional feature vector, a trained RBF neural network model is adopted to classify fault types, and a dual confidence evaluation method based on feature separation degree and track quality is established. The accuracy of axis track identification of the rotating machine rotor is improved, credibility quantitative identification of fault types is realized through confidence threshold judgment, a fault diagnosis result is obtained, multidimensional feature fusion of a two-dimensional track reconstruction set is realized, the accuracy of fault type identification is improved, double confidence quantitative evaluation is realized, the feasibility of the diagnosis result is improved, and the reliability of the system is improved. Therefore, the accuracy of fault diagnosis of the rotating machine rotor is improved.

Description

Technical Field

[0001] This invention relates to the field of rotating machinery fault diagnosis technology, and in particular to a rotating machinery fault diagnosis method and system based on shaft center trajectory confidence analysis. Background Technology

[0002] Rotating machinery (such as electric motors and turbines) is currently a core piece of equipment in industrial production, and its operating status directly affects production efficiency and safety. Fault diagnosis technology is crucial to ensuring stable equipment operation. Traditional diagnostic techniques and existing shaft trajectory correlation methods have significant limitations.

[0003] The core shortcomings of traditional rotating machinery fault diagnosis technology include: 1. Traditional methods rely on one-dimensional vibration signal analysis from a single sensor, which cannot meet the requirements for high-precision diagnosis. The main problems include: unidirectional vibration signals can only reflect the rotor's motion in a single direction, failing to fully describe the rotor's complex two-dimensional motion trajectory, leading to the loss of key motion information and difficulty in capturing the comprehensive characteristics of the fault; different fault types exhibit highly overlapping characteristics in one-dimensional vibration signals. For example, both imbalance and misalignment faults may manifest as an enhancement of the fundamental frequency component, which traditional methods struggle to accurately distinguish, easily leading to misdiagnosis; and diagnostic reliability is lacking: only a qualitative conclusion of "fault type" can be output, lacking a quantitative evaluation mechanism for the reliability of the diagnostic results.

[0004] To address the above technical issues, existing technologies have made some contributions, but current shaft center trajectory-based diagnostic methods still have the following shortcomings: First, while shaft center trajectories can intuitively reflect the two-dimensional motion state of the rotor and are an important carrier for fault diagnosis, existing methods based on shaft center trajectories still have two major bottlenecks: Second, insufficient trajectory reconstruction accuracy: most methods use a single sensor or non-orthogonal sensor configuration, making it difficult to accurately collect complete information on the rotor's lateral vibration; simultaneously, the signal preprocessing process does not effectively filter noise or extract key envelopes, leading to distortion of the reconstructed shaft center trajectory and an inability to accurately reflect the rotor's true motion state. Third, lack of fault confidence assessment: qualitative fault judgment is based solely on the trajectory's geometric shape (e.g., circular, elliptical), without establishing a quantitative mapping relationship between trajectory features and fault types, and lacking a comprehensive assessment of "trajectory quality" and "feature distinguishability," making it impossible to quantify the reliability of the diagnostic results, and still posing a risk of misdiagnosis and missed diagnosis. Summary of the Invention

[0005] The main objective of this invention is to provide a method and system for diagnosing rotating machinery faults based on shaft trajectory confidence analysis, thereby solving the technical problems of insufficient information dimensions, low diagnostic reliability, insufficient trajectory reconstruction accuracy, and low fault diagnosis precision in the prior art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for diagnosing rotating machinery faults based on shaft center trajectory confidence analysis, comprising the following steps: S1: Two accelerometers are used to collect vibration time-series signals in the horizontal and vertical directions at the same bearing position and within the same time period during the operation of the rotating machinery rotor. The two accelerometers are installed orthogonally to each other at the rotor bearing position. S2: Perform data preprocessing on the acquired vibration time-series signal, including: filtering with a Butterworth bandpass filter, extracting the signal envelope through Hilbert transform, and then reconstructing the shaft center trajectory; S3: Based on the reconstructed axis trajectory, feature extraction is performed, including four features: trajectory area, eccentricity, periodicity, and stability. S4: Construct a four-dimensional feature vector of four features, and use the dataset of four-dimensional feature vectors to train the RBF neural network model, and then use the trained RBF neural network model to classify the fault type. S5: Establish a dual-reset reliability assessment method based on feature separation degree and trajectory quality. A reliable quantitative identification of fault types is achieved through confidence threshold decision. The fault evolution trend is analyzed based on the fault severity index to obtain fault diagnosis results.

[0007] In the preferred embodiment, the two mutually orthogonal accelerometers in S1 collect the lateral vibration signal of the rotor, expressed as: ; In the formula, These are vibration displacement signals in the horizontal and vertical directions, respectively. These are the instantaneous amplitudes; The angular frequency of the shaft. ; For shaft rotation frequency; It is the instantaneous phase; For high-frequency measurement of noise components.

[0008] In the preferred embodiment, the data preprocessing in S2 specifically includes: A Butterworth bandpass filter with a passband range of [5, 200] Hz is used for filtering to eliminate the DC component and linear trend in the signal, and the signal-to-noise ratio (SNR) is calculated using the following formula: ; In the formula, The signal amplitude, This represents the noise amplitude. The signal envelope is extracted using the Hilbert transform, as shown in the formula: ; In the formula, H Here, j is the Hilbert transform operator, and j is the imaginary unit. For modular arithmetic of complex numbers; Then, the axis center trajectory is reconstructed, and the parametric equation of the axis center trajectory is: .

[0009] In the preferred embodiment, S3 performs feature extraction based on the reconstructed axis trajectory, using the following formula: Trajectory area: ; In the formula: Let be the enclosed area bounded by the trajectory; This represents the total number of discrete sampling points on the trajectory. For the first Two-dimensional coordinates of each trajectory sampling point; This is a condition for trajectory closure; Trajectory eccentricity: ; In the formula: Eccentricity, a dimensionless parameter, with a range of values. ; The length of the major semi-axis of the ellipse fitted to the trajectory; The length of the minor semi-axis of the ellipse fitted to the trajectory; Periodic indicators: ; In the formula: It is a periodic indicator; The length of the observation time window; The rotation period of the shaft, ; This represents the magnitude of the axis offset. ; Trajectory stability index: ; In the formula: As a stability index, the value range ; This represents the Euclidean distance between adjacent sampling points.

[0010] In the preferred embodiment, the expression for the S4 four-dimensional feature vector is: ; In the formula: It is a four-dimensional feature vector; Trajectory area features; Eccentricity characteristics; It is characterized by periodicity, and is used to detect intermittent faults. This is a characteristic of stability.

[0011] In the preferred scheme, the classifier expression based on the RBF neural network is: ; In the formula: For the first Network output values ​​for this type of fault; This represents the total number of RBF neurons. For the first The RBF neuron to the 1st The connection weights of each output node; For the first The center vector of each RBF neuron; For the first The radial width parameter of each RBF neuron; The Euclidean distance between the input feature vector and the center of the RBF; The probability of various faults is calculated using the following formula: ; In the formula: For the first Output probability of the type of fault; Total number of fault types; This is the probability normalization condition.

[0012] In the preferred embodiment, the double-reset reliability assessment in S5 is expressed as follows: Feature separation confidence score, based on the degree of separation of current sample features across different fault types, is calculated using the following formula: ; The formula for information entropy is: ; In the formula: The confidence level of feature separation; The classification uncertainty of the current sample; This is the theoretical maximum entropy value; The trajectory quality confidence score, based on the axis-centered trajectory reconstruction quality, is calculated using the following formula: ; In the formula: For trajectory quality confidence; The signal-to-noise ratio of the current signal; For dual-channel signal synchronization, ; This serves as a reference signal-to-noise ratio and synchronization degree under standard operating conditions. is the Pearson correlation coefficient for the two-channel signal; The formula for overall confidence level is: ; In the formula: For the final overall confidence level; These are the weighting coefficients; Weights for trajectory quality confidence.

[0013] In the preferred embodiment, step S5 involves a graded decision based on the confidence level, expressed as follows: .

[0014] In the preferred embodiment, in step S5, the fault evolution trend is analyzed based on the fault severity index to obtain the fault diagnosis result, specifically as follows: First, define a fault severity index based on trajectory features: ; In the formula: This is an index representing the severity of the fault. The weight coefficients for each feature, and the constraints. ; The area of ​​the reference trajectory under normal operating conditions; The degree of lack of periodicity and stability; Further analysis of the fault evolution trend reveals that the fault development stages are divided based on the FSI value: ; Finally, trend analysis is performed, and the FSI rate of change is calculated using a sliding window: ; In the formula: The rate of failure development; For time intervals; These are the FSI values ​​for the current time and the previous time.

[0015] Secondly, a rotating machinery fault diagnosis system based on shaft center trajectory confidence analysis is provided, applicable to the aforementioned diagnosis method, including: The data acquisition module is used to collect vibration time-series signals in the horizontal and vertical directions at the same bearing position and within the same time period using two accelerometers during the operation of the rotating machinery rotor. The two accelerometers are orthogonally installed at the rotor bearing position. The preprocessing module is used to preprocess the acquired vibration time-series signals, including: filtering with a Butterworth bandpass filter, extracting the signal envelope through Hilbert transform, and then reconstructing the shaft center trajectory. The feature extraction module is used to extract features based on the reconstructed axis trajectory, including four features: trajectory area, eccentricity, periodicity, and stability. The model classification module is used to construct a four-dimensional feature vector of four features, and to train the RBF neural network model using the dataset of four-dimensional feature vectors. Then, the trained RBF neural network model is used to classify the fault types. The fault diagnosis module is used to establish a dual-reset reliability assessment method based on feature separation degree and trajectory quality. It achieves reliable quantitative identification of fault types through confidence threshold decision, analyzes fault evolution trends based on fault severity index, and obtains fault diagnosis results. This invention provides a method and system for diagnosing rotating machinery faults based on shaft center trajectory confidence analysis. It employs two orthogonal acceleration sensors to collect lateral vibration signals of the rotor, performs preprocessing, reconstructs the shaft center trajectory, extracts four-dimensional feature vectors, and uses a trained RBF neural network model to classify fault types. A dual-reset confidence assessment method based on feature separation and trajectory quality is established, improving the accuracy of shaft center trajectory identification for rotating machinery rotors. Quantitative identification of fault types is achieved through confidence threshold judgment, obtaining fault diagnosis results. The two-dimensional trajectory reconstruction combined with multi-dimensional feature fusion further improves the accuracy of fault type identification. The dual-reset confidence quantitative assessment enhances the feasibility of the diagnostic results, thereby improving the accuracy of rotating machinery rotor fault diagnosis. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a schematic diagram of the diagnostic method of the present invention; Figure 2 This is an overall schematic diagram of the rotary machine fault diagnosis method based on shaft trajectory confidence analysis of the present invention; Figure 3 This is a schematic diagram illustrating the overall implementation process of the diagnostic method of the present invention. Detailed Implementation

[0017] Example 1 like Figure 1-3 As shown, a method for diagnosing rotating machinery faults based on shaft center trajectory confidence analysis includes the following steps: S1: Two accelerometers are used to collect vibration timing signals in the horizontal and vertical directions at the same bearing position and within the same time period during the operation of the rotating machinery rotor. The two accelerometers are installed orthogonally to each other at the rotor bearing position.

[0018] S2: The acquired vibration time-series signal is preprocessed, including filtering with a Butterworth bandpass filter, extracting the signal envelope through Hilbert transform, and then reconstructing the shaft center trajectory.

[0019] S3: Based on the reconstructed axis trajectory, feature extraction is performed, including four features: trajectory area, eccentricity, periodicity, and stability.

[0020] S4: Construct a four-dimensional feature vector of four features, and use the dataset of four-dimensional feature vectors to train an RBF neural network model. Then, use the trained RBF (Radial Basis Function) neural network model to classify fault types. S5: Establish a dual-reset reliability assessment method based on feature separation degree and trajectory quality. A reliable quantitative identification of fault types is achieved through confidence threshold decision. The fault evolution trend is analyzed based on the fault severity index to obtain fault diagnosis results.

[0021] This embodiment uses two orthogonal accelerometers (S1-S5) to collect the lateral vibration signal of the rotor. After preprocessing, the shaft center trajectory is reconstructed, and a four-dimensional feature vector is extracted. A trained RBF neural network model is used to classify the fault type. A dual-reset reliability evaluation method based on feature separation degree and trajectory quality is established, which improves the accuracy of shaft center trajectory identification of rotating machinery rotors. The reliability of fault type identification is realized through confidence threshold judgment, and fault diagnosis results are obtained. The two-dimensional trajectory reconstruction is combined with multi-dimensional feature fusion, which improves the accuracy of fault type identification. The dual-reset reliability evaluation improves the feasibility of the diagnosis results, thereby improving the accuracy of rotating machinery rotor fault diagnosis.

[0022] like Figure 1 The flowchart shown in this embodiment is a method for identifying the axis trajectory of a rotating mechanical rotor. The following steps will be described in detail.

[0023] Before the accelerometer acquires the vibration time-series signal sequence, it first generates simulated shaft center trajectory signals for rotating machinery rotor faults of different fault types and densities. The shaft center trajectory patterns of the rotating machinery rotor faults are first classified, including circles and ellipses, double-elliptical figure-eight patterns (including inner and outer ring ellipses), divergence, etc. In this embodiment, the ellipse center trajectory is obtained by fitting the reconstructed shaft center trajectory point set with the semi-major axis length 'a' and the semi-minor axis length 'b' of the trajectory fitting ellipse.

[0024] This embodiment uses the diagnosis of motor rotor imbalance faults as an example. The equipment and sensor parameters used are shown in the table below.

[0025] Table 1 Equipment and Sensor Parameter Table

[0026] I. Data Acquisition and Preprocessing In the preferred embodiment, the two mutually orthogonal accelerometers in step S1 collect the lateral vibration signal of the rotor, expressed as: ; In the formula, These are vibration displacement signals in the horizontal and vertical directions, respectively. These represent the instantaneous amplitude, indicating the degree of axis offset; The angular frequency of the shaft. ; For shaft rotation frequency; It is an instantaneous phase that contains fault characteristic information; For high-frequency measurement of noise components; Substituting the parameters, we get: .

[0027] I. Reconstruction of the Axis Trajectory In the preferred scheme, the data preprocessing in step S2 specifically includes: A Butterworth bandpass filter with a passband range of [5, 200] Hz is used for filtering to eliminate the DC component and linear trend in the signal, and the signal-to-noise ratio (SNR) is calculated using the following formula: ; In the formula, The signal amplitude, This represents the noise amplitude.

[0028] The signal envelope is extracted using the Hilbert transform, as shown in the formula: ; In the formula, H Here, j is the Hilbert transform operator, and j is the imaginary unit. This is for modular arithmetic operations on complex numbers.

[0029] II. Detailed calculation process: 1. To Perform FFT transform to obtain the frequency domain signal .

[0030] 2. Constructing an analytical signal: (Positive frequency portion) (Negative frequency part).

[0031] 3. IFFT yields the analytic signal: .

[0032] 4. Extract the envelope: .

[0033] Numerical results (taking typical time points) ): ; ; .

[0034] Calculate similarly Then reconstruct the axis trajectory: .

[0035] III. Axial Trajectory Feature Extraction The axis trajectory can be represented by parametric equations: .

[0036] Then, the axis center trajectory is reconstructed, and the parametric equation of the axis center trajectory is: .

[0037] In this embodiment, the data preprocessing process effectively filters out high-frequency noise and DC interference, significantly reducing the impact of noise on trajectory reconstruction; it accurately separates the amplitude and phase information in the signal, and the reconstructed shaft center trajectory truly reflects the rotor motion state, reducing the trajectory distortion rate and improving the trajectory reconstruction accuracy.

[0038] IV. Dynamic Feature Calculation In the preferred embodiment, step S3 extracts key features reflecting fault characteristics based on the reconstructed shaft center trajectory, using the following formula: (a) Geometric features include trajectory area and trajectory eccentricity.

[0039] 1) Trajectory area: Discretize the trajectory of a complete cycle into... For each sampling point, apply the trapezoidal integral formula: ; In the formula: Let be the closed area enclosed by the trajectory, with dimensions . ; This represents the total number of discrete sampling points on the trajectory. For the first Two-dimensional coordinates of each trajectory sampling point; This is the condition for trajectory closure.

[0040] Detailed calculation steps: Data from some sampling points are shown in Table 3 (unit: mm): Table 3 Sampling Point Data Table

[0041] Add up all items: .

[0042] therefore: .

[0043] 2) Trajectory eccentricity: Fit the ellipse equation using the least squares method: .

[0044] Solving for elliptic parameters using singular value decomposition.

[0045] Numerical results: Center of the ellipse: Long half-axis: mm; Short half-shaft: mm.

[0046] Eccentricity calculation: ; In the formula: Eccentricity, a dimensionless parameter, with a range of values. ; The length of the major semi-axis of the ellipse fitted to the trajectory; The length of the minor semi-axis of the ellipse fitted to the trajectory; Represents a perfectly circular trajectory. This represents a highly elliptic trajectory.

[0047] (ii) Dynamic motion characteristics 1) Trajectory periodicity indicators: ; In the formula: As a periodic indicator, the value range is... A value close to 1 indicates high periodicity; The length of the observation time window; The rotation period of the shaft, ; This represents the magnitude of the axis offset. .

[0048] In this embodiment, wherein s. Among them s.

[0049] Intermediate calculation results: Average value: The mean of the cosine terms: .

[0050] therefore: .

[0051] 2) Trajectory stability index: ; In the formula: As a stability index, the value range ; The Euclidean distance between adjacent sampling points; This indicates that the trajectory is highly stable. This indicates that the trajectory fluctuates drastically.

[0052] Step-by-step calculation: Calculate the distance between adjacent points: .

[0053] Example calculation (first few points): .

[0054] Calculate the mean variance: .

[0055] Calculate the stability index: .

[0056] This embodiment constructs a multi-dimensional feature system, quantifies the severity of faults by trajectory area, distinguishes between imbalance and misalignment faults by eccentricity (e.g., imbalance fault E≈0.7, misalignment fault E<0.5), periodically detects intermittent faults, and identifies high-frequency bearing faults by stability (St), forming a comprehensive feature dimension that combines geometry and dynamics. This solves the problem that traditional single features cannot distinguish fault types. Furthermore, by defining the feature calculation logic in a formulaic way, the consistency and comparability of feature extraction under different operating conditions are ensured.

[0057] Fifth, assess the confidence level of the fault type and then establish a confidence mapping relationship between the axis trajectory characteristics and the fault type.

[0058] (a) The four features extracted in step S3 are combined into a four-dimensional feature vector, expressed as: ; In the formula: It is a four-dimensional feature vector; The trajectory area characteristic reflects the severity of the fault; Based on the eccentricity characteristic, it distinguishes between imbalance and misalignment faults; It is characterized by periodicity, and is used to detect intermittent faults. Based on stability characteristics, it identifies high-frequency faults such as those in bearings.

[0059] (II) RBF Neural Network Classification 1) Network parameter settings: Number of RBF neurons: .

[0060] Number of fault types: (Normal, unbalanced, misaligned, bearing failure).

[0061] 2) Pre-trained RBF centers and parameters (partial): (Normal state center).

[0062] (Imbalance fault center).

[0063] , .

[0064] Classification calculation process: Step S4 uses a classifier based on an RBF neural network, expressed as follows: ; In the formula: For the first Network output values ​​for this type of fault; This represents the total number of RBF neurons. For the first The RBF neuron to the 1st The connection weights of each output node; For the first The center vector of each RBF neuron; For the first The radial width parameter of each RBF neuron; The Euclidean distance between the input feature vector and the center of the RBF is given.

[0065] The probability of various faults is calculated using the following formula: ; In the formula: For the first Output probability of the type of fault; Total number of fault types; This is the probability normalization condition.

[0066] Calculate the distance between the current feature and each RBF center: .

[0067] Calculate RBF output: .

[0068] Calculate the network output (taking an imbalance fault as an example):

[0069] Calculate the probability of various types of failures: (normal); (unbalanced); (Not in the middle); (Bearing failure).

[0070] .

[0071] This embodiment utilizes a vector form to adapt the RBF neural network, constructing a nonlinear classification model through the radial basis function (Gaussian function) of the RBF neurons. Compared with traditional linear classification methods, it has a stronger ability to fit the complex mapping relationship of fault features, and can accurately distinguish multiple types of faults such as "normal, unbalanced, misaligned, and bearing fault". It accurately establishes the mapping relationship between features and fault types, thereby improving the classification accuracy.

[0072] This embodiment utilizes a vector form to adapt the RBF neural network, constructing a nonlinear classification model through the radial basis function (Gaussian function) of the RBF neurons. Compared with traditional linear classification methods, it has a stronger ability to fit the complex mapping relationship of fault features, and can accurately distinguish multiple types of faults such as "normal, unbalanced, misaligned, and bearing fault". It accurately establishes the mapping relationship between features and fault types, thereby improving the classification accuracy.

[0073] VI. Confidence Calculation In the preferred scheme, the double-reset reliability assessment in step S5 has the following expressions: 1) Feature separation confidence score, based on the degree of separation of current sample features across different fault types, is calculated using the following formula: ; The formula for information entropy is: ; In the formula: The feature separation confidence score, value range ; The classification uncertainty (information entropy) of the current sample; This is the theoretical maximum entropy value; The fault types are clearly distinguished.

[0074] Among them, the probability values ​​are: , , , .

[0075] .

[0076] Maximum entropy: .

[0077] Feature separation confidence level: .

[0078] 2) Trajectory quality confidence, based on the quality of axis-centered trajectory reconstruction, is calculated using the following formula: ; In the formula: For trajectory quality confidence; The signal-to-noise ratio of the current signal; For dual-channel signal synchronization, ; This serves as a reference signal-to-noise ratio and synchronization degree under standard operating conditions. is the Pearson correlation coefficient for the two-channel signal.

[0079] Signal quality parameters used: Current signal-to-noise ratio: dB; Reference signal-to-noise ratio: dB; dual-channel correlation coefficient: (synchronization) ); Reference synchronization: .

[0080] get: .

[0081] 3) The formula for the overall confidence level is: ; In the formula: For the final overall confidence level, the value range is... ; These are the weighting coefficients; Weights for trajectory quality confidence.

[0082] Take weight : .

[0083] In this embodiment, the feature separation confidence quantifies the clarity of distinguishing fault features between different types, and the trajectory quality confidence quantifies the reliability of trajectory reconstruction, thereby further improving the reliability of diagnosis.

[0084] In the preferred scheme, step S5 involves a graded decision based on the confidence level, expressed as follows: .

[0085] Then, a confidence level determination is made: .

[0086] Verdict: Diagnosis with moderate confidence.

[0087] (vii) Fault Evolution Analysis In the preferred scheme, step S5 involves analyzing the fault evolution trend based on the fault severity index to obtain the fault diagnosis result, specifically as follows: First, define the Fault Severity Index (FSI) based on trajectory features: ; In the formula: This is a fault severity index with a value range. ; The weight coefficients for each feature, and the constraints. ; The area of ​​the reference trajectory under normal operating conditions; This refers to the degree of lack of periodicity and stability.

[0088] Further analysis of the fault evolution trend reveals that the fault development stages are divided based on the FSI value: ; Finally, trend analysis is performed, and the FSI rate of change is calculated using a sliding window: ; In the formula: The fault development rate, with dimensions of ; The recommended time interval is 1-7 days. The FSI values ​​are the current time and the previous time; when At that time, it indicates the rapid development trend of the fault.

[0089] In this embodiment, weights are used: , , , .

[0090] Reference area: mm² (normal operating condition), then: .

[0091] Fault stage determination: .

[0092] Judgment: Early stage of failure.

[0093] The trend analysis is as follows: Assuming the FSI value was 1.56 7 days ago, the fault development rate is: .

[0094] because This indicates that the fault development is relatively slow.

[0095] Diagnostic conclusions and recommendations: The diagnostic results are summarized as follows: Fault type: Rotor imbalance fault (probability 82.1%); Confidence level: Medium confidence (0.645); Fault severity: Early stage of fault (FSI = 1.84); Development trend: Slow development (0.04 / day).

[0096] In this embodiment, the abstract confidence value is transformed into an intuitive diagnostic level. The FSI index achieves a quantitative assessment of the severity of the fault by weighted integration of four features. It is based on the FSI value to divide the fault into four stages: "normal, early, developing, and severe," accurately locating the fault status. The FSI change rate is used to predict the speed of fault development, reducing risk lag and minimizing equipment downtime losses. Example 2 Further illustrating with reference to Embodiment 1, a rotating machinery fault diagnosis system based on shaft center trajectory confidence analysis, applicable to the diagnosis method of Embodiment 1, includes: The data acquisition module is used to collect vibration time-series signals in the horizontal and vertical directions at the same bearing position and within the same time period using two accelerometers during the operation of the rotating machinery rotor. The two accelerometers are orthogonally installed at the rotor bearing position.

[0097] The preprocessing module is used to preprocess the acquired vibration time-series signals, including filtering with a Butterworth bandpass filter, extracting the signal envelope through Hilbert transform, and then reconstructing the shaft center trajectory.

[0098] The feature extraction module is used to extract features based on the reconstructed axis trajectory, including four features: trajectory area, eccentricity, periodicity, and stability.

[0099] The model classification module is used to construct a four-dimensional feature vector of four features, and to train the RBF neural network model using the dataset of four-dimensional feature vectors. Then, the trained RBF neural network model is used to classify the fault types. The fault diagnosis module is used to establish a dual-reset reliability assessment method based on feature separation degree and trajectory quality. It realizes the reliable quantitative identification of fault types through confidence threshold decision, analyzes the fault evolution trend based on the fault severity index, and obtains fault diagnosis results.

[0100] This embodiment provides a method for diagnosing rotating machinery faults based on shaft trajectory confidence analysis. The working process, working details and technical effects can be found in Embodiment 1, and will not be repeated here.

[0101] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for diagnosing rotating machinery faults based on shaft center trajectory confidence analysis, characterized in that, Includes the following steps: S1: Two accelerometers are used to collect vibration time-series signals in the horizontal and vertical directions at the same bearing position and within the same time period during the operation of the rotating machinery rotor. The two accelerometers are installed orthogonally to each other at the rotor bearing position. S2: Perform data preprocessing on the acquired vibration time-series signal, including: filtering with a Butterworth bandpass filter, extracting the signal envelope through Hilbert transform, and then reconstructing the shaft center trajectory; S3: Based on the reconstructed axis trajectory, feature extraction is performed, including four features: trajectory area, eccentricity, periodicity, and stability. S4: Construct a four-dimensional feature vector of four features, train an RBF neural network model using the dataset of four-dimensional feature vectors, and then use the trained RBF neural network model to classify fault types. S5: Establish a dual-reset reliability assessment method based on feature separation degree and trajectory quality. A reliable quantitative identification of fault types is achieved through confidence threshold decision. The fault evolution trend is analyzed based on the fault severity index to obtain fault diagnosis results.

2. The rotating machinery fault diagnosis method based on shaft center trajectory confidence analysis according to claim 1, characterized in that, The two orthogonal accelerometers in S1 collect the lateral vibration signal of the rotor, expressed as: ； In the formula, These are vibration displacement signals in the horizontal and vertical directions, respectively. These are the instantaneous amplitudes; The angular frequency of the shaft. ; For shaft rotation frequency; It is the instantaneous phase; For high-frequency measurement of noise components.

3. The rotating machinery fault diagnosis method based on shaft center trajectory confidence analysis according to claim 1, characterized in that, The data preprocessing in S2 specifically includes: A Butterworth bandpass filter with a passband range of [5, 200] Hz is used for filtering to eliminate the DC component and linear trend in the signal, and the signal-to-noise ratio (SNR) is calculated using the following formula: ； In the formula, The signal amplitude, This represents the noise amplitude. The signal envelope is extracted using the Hilbert transform, as shown in the formula: ； In the formula, H Here, j is the Hilbert transform operator, and j is the imaginary unit. For modular arithmetic of complex numbers; Then, the axis center trajectory is reconstructed, and the parametric equation of the axis center trajectory is: 。 4. The rotating machinery fault diagnosis method based on shaft center trajectory confidence analysis according to claim 3, characterized in that, S3 performs feature extraction based on the reconstructed axis trajectory, using the following formula: Trajectory area: ； In the formula: Let be the enclosed area bounded by the trajectory; This represents the total number of discrete sampling points on the trajectory. For the first Two-dimensional coordinates of each trajectory sampling point; This is a condition for trajectory closure; Trajectory eccentricity: ； In the formula: Eccentricity, a dimensionless parameter, with a range of values. ; The length of the major semi-axis of the ellipse fitted to the trajectory; The length of the minor semi-axis of the ellipse fitted to the trajectory; Periodic indicators: ； In the formula: It is a periodic indicator; The length of the observation time window; The rotation period of the shaft, ; This represents the magnitude of the axis offset. ; Trajectory stability index: ； In the formula: As a stability index, the value range ; This represents the Euclidean distance between adjacent sampling points.

5. The rotating machinery fault diagnosis method based on shaft center trajectory confidence analysis according to claim 1, characterized in that, The expression for the S4 four-dimensional feature vector is: ； In the formula: It is a four-dimensional feature vector; Trajectory area features; Eccentricity characteristics; It is characterized by periodicity, and is used to detect intermittent faults. This is a characteristic of stability.

6. The rotating machinery fault diagnosis method based on shaft center trajectory confidence analysis according to claim 5, characterized in that, The classifier expression based on the RBF neural network is as follows: ； In the formula: For the first Network output values ​​for this type of fault; This represents the total number of RBF neurons. For the first The RBF neuron to the 1st The connection weights of each output node; For the first The center vector of each RBF neuron; For the first The radial width parameter of each RBF neuron; The Euclidean distance between the input feature vector and the center of the RBF; The probability of various faults is calculated using the following formula: ； In the formula: For the first Output probability of the type of fault; Total number of fault types; This is the probability normalization condition.

7. The rotating machinery fault diagnosis method based on shaft center trajectory confidence analysis according to claim 1, characterized in that, The double-reset reliability assessment in S5 is expressed as follows: Feature separation confidence score, based on the degree of separation of current sample features across different fault types, is calculated using the following formula: ； The formula for information entropy is: ； In the formula: The confidence level of feature separation; The classification uncertainty of the current sample; This is the theoretical maximum entropy value; The trajectory quality confidence score, based on the axis-centered trajectory reconstruction quality, is calculated using the following formula: ； In the formula: For trajectory quality confidence; The signal-to-noise ratio of the current signal; For dual-channel signal synchronization, ; This serves as a reference signal-to-noise ratio and synchronization degree under standard operating conditions. is the Pearson correlation coefficient for the two-channel signal; The formula for overall confidence level is: ； In the formula: This is the final overall confidence level; These are the weighting coefficients; Weights for trajectory quality confidence.

8. The rotating machinery fault diagnosis method based on shaft center trajectory confidence analysis according to claim 7, characterized in that, The graded decision based on the confidence level in S5 is expressed as follows: 。 9. The rotating machinery fault diagnosis method based on shaft center trajectory confidence analysis according to claim 1, characterized in that, In step S5, the fault evolution trend is analyzed based on the fault severity index to obtain the fault diagnosis result, specifically as follows: First, define a fault severity index based on trajectory features: ； In the formula: This is an index representing the severity of the fault. The weight coefficients for each feature, and the constraints. ; The area of ​​the reference trajectory under normal operating conditions; The degree of lack of periodicity and stability; Further analysis of the fault evolution trend reveals that the fault development stages are divided based on the FSI value: ； Finally, trend analysis is performed, and the FSI rate of change is calculated using a sliding window: ； In the formula: The rate of failure development; For time intervals; These are the FSI values ​​for the current time and the previous time.

10. A rotating machinery fault diagnosis system based on shaft center trajectory confidence analysis, characterized in that, The diagnostic method applicable to any one of claims 1-9 includes: The data acquisition module is used to collect vibration time-series signals in the horizontal and vertical directions at the same bearing position and within the same time period using two accelerometers during the operation of the rotating machinery rotor. The two accelerometers are orthogonally installed at the rotor bearing position. The preprocessing module is used to preprocess the acquired vibration time-series signals, including: filtering with a Butterworth bandpass filter, extracting the signal envelope through Hilbert transform, and then reconstructing the shaft center trajectory. The feature extraction module is used to extract features based on the reconstructed axis trajectory, including four features: trajectory area, eccentricity, periodicity, and stability. The model classification module is used to construct a four-dimensional feature vector of four features, and to train the RBF neural network model using the dataset of four-dimensional feature vectors. Then, the trained RBF neural network model is used to classify the fault types. The fault diagnosis module is used to establish a dual-reset reliability assessment method based on feature separation degree and trajectory quality. It realizes the reliable quantitative identification of fault types through confidence threshold decision, analyzes the fault evolution trend based on the fault severity index, and obtains fault diagnosis results.

Images